THE ESTIMATION OF NEIGHBORHOOD DEPRIVATION AND PRETERM BIRTH USING LONGITUDINALLY LINKED NATALITY RECORDS By Cristin Elizabeth McArdle A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Epidemiology ŠDoctor of Philosophy 2019 ABSTRACT THE ESTIMATION OF NEIGHBORHOOD DEPRIVATION AND PRETERM BIRTH USING LONGITUDINALLY LINKED NATALITY RECORDS By Cristin Elizabeth McArdle This dissertation examined the association between neighborhood -level deprivation and perinatal outcomes. We studied the association between neighborhood poverty rate and pre-term birth (PTB; birth < 37 weeks) usi ng longitudinal maternally -linked natality files of women and their infants in Michigan during the period 1990 -2012. This study examine d the embodiment of place and role of maternal characteristics during pregnancy in an effort to understand how selection into neighborhoo d may bias our understanding of neighborhood level associations . We looked at pregnancy outcomes across multiple pregnancies for the same woman (the mother) as she change d neighborhoods , and levels of poverty between pregnancies. In the first study examining residential mobility between pregnancies, we report ed that approximately half of our sample changed residences between pregnancies. We further exploited our data structure to examine th e association with prior PTB on subsequent mobility in two sub -sample s restricted by parity : births 1 and 2, and births 2 and 3. We found the strongest risk factors for mobility were related to marital change (Divorce: births 1 to 2 OR: 2.5 95% CI: 2.4 -2.6, births 2 to 3 OR: 3.3, 95% CI: 3.1 -3.6); Married: births 1 to 2 OR: 2.8, 95% CI: 2.7 -2.8, births 2 to 3 OR: 1.9, 95% CI:1.9 -2.0) but not prior PTB ( prior PTB: births 1 to 2 OR: 1.0, 95% CI: 1.0-1.0, births 2 to 3 OR: 1.1 95%CI: 1.0 -1.1). In the second study, we report that most women did not experience a change in the level of neighborhood poverty, based on quartile of neighborhood poverty . Women who remained in the poorest neighborhoods experience d the high est percen tage of PTB across two births samples, Births 1 to 2 (11.4% PTB) and Births 2 to 3 (12.3% PTB) . We f ound increased odds of PTB for births 1 to 2 with strong downward neighborhood trajectory (OR 1.2, 95% CI 1.0 -1.3) but also increased odds of PTB among strong upward neighborhood poverty trajectory (OR 1.1, 95%CI: 1.1 - 1.2) compared to the static trajectory group of lowest neighborhood poverty quartile . In Study 3, we then employed a novel approach , maternal fixed effects, utilizing data linked over time to compare birth outcomes for the same mother under different exposures which allows the mother to act as her own control, analogous to a case -crossover design, while comparing the contextual effects of neighborhood deprivation on PTB. We conducted l ogistic regression, random effects and fixed effects analysis to evaluate n=2,191,063 eligible births during our study period. Because a fixed e ffects model relies on variation over time within a mother to identify the estimated association of nei ghborhood deprivation and PTB, the primary analytic sample was restricted (n=280,277 births to 103,328 women).We found a null association between neighborhood poverty and PTB when using a maternal fixed effects analysis (OR: 1.0, 95% CI: 1.0 -1.0). This was one of the first studies to profile the maternal neighborhood mobility patterns over a long period of time, between successive pregnancies and evaluated by neighborhood poverty rate . iv This dissertation is dedicated to my mother, you are simply the most intelligent and loving person and it is the great fortune of my life to call you Mom. I love you .v ACKNOWLEDGEMENTS I would like to thank my advisor and dissertation committee chair, Dr. Claire Margerison, for her invaluable suggestions and guidance from the beginning to end of this process. I am also very grateful to my dissertation committee members, Dr. Nigel Paneth, Dr. Zhehui Luo, and Dr. Sue Grady, for their encouragement and guidance throughout this process. Thank you to Dr. David Barondess and Nancy Beiber for your help navigating this process . I would like to thank Dr. Ellen Velie for starting me on this journey and the strong foundation she provided. I would like to express my appreciation to Glenn Copeland for his work in developing the maternally -linked dataset and Dr. Yu ‚Seashore™ Li , for her work in cleaning much of the initial datasets . I want to thank my many champions, none greater than my friends and family. Thank you for the strength and support when needed and for many hours of laughter and love. In particular, I want to thank my Dad, who spent his lifetime filling my head with sage advice and gentle encouragement Œ I will always hear your voice even if you cannot speak . LYC. Mom, you are the best. I love you. Mick and Tim, I love you too. For Joel, thank you for your love and time and many hours of discussion that made these papers and my life so much better. I love you. For Sammy and Jackie, you are my world and I hope to spend my life filling yours with all the love, sage advice, and gentle encouragement so that you too find yourselves in the delightful position of setting out and achiev ing a dre am. I am the luckiest. I love you. Thank you. vi TABLE OF CONTENTS LIST OF TABLES ....................................................................................................................... viii LIST OF FIGURES ....................................................................................................................... xi KEY TO ABBREVIATIONS ....................................................................................................... xii INTRODUCTION .......................................................................................................................... 1 Background ................................................................................................................................. 1 Preterm Birth and Disparities .................................................................................................. 1 Neighborhood as a Determinant of Adverse Birth Outcomes ................................................. 2 Neighborhood Context and Pathway ....................................................................................... 4 Selection Bias in Neighborhood Effects .................................................................................. 5 Study Objectives ......................................................................................................................... 6 REFERENCES ............................................................................................................................. 10 STUDY 1 ...................................................................................................................................... 17 Introduction ............................................................................................................................... 17 Methods ..................................................................................................................................... 19 Data and study population ..................................................................................................... 19 Measures ................................................................................................................................ 21 Statistical Analyses ................................................................................................................ 23 Results ....................................................................................................................................... 23 Discussion ................................................................................................................................. 25 APPENDIX ................................................................................................................................... 38 REFERENCES ............................................................................................................................. 48 STUDY 2 ...................................................................................................................................... 52 Introduction ............................................................................................................................... 52 Methods ..................................................................................................................................... 54 Study Data and Population .................................................................................................... 54 Measures ................................................................................................................................ 55 Statistical Analysis ................................................................................................................ 58 Results ....................................................................................................................................... 59 Discussion ................................................................................................................................. 61 APPENDIX ................................................................................................................................... 68 REFERENCES ............................................................................................................................. 75 STUDY 3 ...................................................................................................................................... 79 Introduction ............................................................................................................................... 79 Methods ..................................................................................................................................... 81 Study Population and Data .................................................................................................... 81 Measures ................................................................................................................................ 82 Statistical Analyses ................................................................................................................ 83 vii Results ....................................................................................................................................... 86 Primary Analyses ................................................................................................................... 86 Secondary Analyses ............................................................................................................... 87 Discussion ................................................................................................................................. 89 APPENDIX ................................................................................................................................... 98 REFERENCES ........................................................................................................................... 102 CONCLUSION ........................................................................................................................... 105 REFERENCES ........................................................................................................................... 108 viii LIST OF TABLES Table 1.1 Maternal characteristics at index birth stratified by residential status (mover vs. non -mover) at successive births restricted by parity, Births 1 to 2 and Births 2 to 3, singleton births recorded in Michigan 1990 -2010 ....................................................................................... 30 Table 1.2a Associations between maternal characteristics at index birth and residential mobility at succes sive births restricted to first and second singleton births recorded in Michigan 1990 -2010 .................................................................................................................................... 33 Table 1.2b Associations between maternal characteristics at index birth and residential mobility at successive births restricted to second and third singleton bi rths recorded in Michigan 1990 -2010..................................................................................................................................... 35 Table 1.3 Estimates from adjusted logis tic model of association between prior PTB and residential change for mothers at successive birth .............................................................. 37 Table A.1 Maternal characteristics at index birth stratified by geocoded status among singleton births in Michigan Birth File 1990 -2012 ............................................................................ 39 Table A.2 Geocoded missing status by infant year of birth in singleton Michigan births 1989 -2012..................................................................................................................................... 41 Table A.3 Paternal charact eristics overall and stratified by maternal PTB outcomes among singleton Michigan Births 1990 -2012 ................................................................................ 42 Table A.4 Maternal characteristics and poverty exposure by ordering of preterm birth outcomes across two pregnancies, births 1 to 2, and births 2 to 3 amo ng singleton births in Michigan 1990 -2010..................................................................................................................................... 43 Table A.5 Residential mobility status (mover, non -mover) and percent neighborhood poverty by ordering of preterm birth outcomes across two pregnancies, births 1 to 2, and births 2 to 3 among singleton births in Michigan 1990 -2010 ............................................................................. 45 Table A.6 Table poverty variable derivation and missing poverty values ......................... 46 Table 2.1 Neighborhood poverty trajectories at successive birth (upward, downward, static) stratified by residential status (mover, non -mover) in singleton maternally -linked Michigan births1990 -2012 .................................................................................................................. 65 Table 2.2 Percentage of PTB births at successive birth among singleton Michigan births by neighborhood poverty quartile at index and successive births ........................................... 66 ix Table 2.3 Risk per 100 live births of preterm birth (<37 weeks completed gestation) at birth 2 and birth 3 by m other™s neighborhood poverty trajectory from previous index birth, singleton Michigan Births 1990 -2012 ................................................................................................ 66 Table 2.4 Associations of Neighborhood Trajectory Trajectories and Successive Birth Outcomes, Bivariate and Multivariate Models in Singleton Michigan Births 1990 -2012 (n=895,214) ......................................................................................................................... 67 Table B.1 Datasets Mean, SD and Quartile Cut -points of Neighborhood Poverty ............ 69 Table B.2 Bivariate Associations of Quartiles of Neighborhood Poverty and PTB across all births, Singleton Michigan Births 1990 -2012 ..................................................................... 69 Table B.3 Poverty quartile of static neighborhood poverty trajectory (no change) and PTB, and residential mobility at birth 2 (n= 285,662) ........................................................................ 70 Table B.4 Associations of Improved Neighborhood Trajectory Trajectories and Successive Birth Outcomes, Bivariate and Multiv ariate Models in Singleton Michigan Births 1990 -2012 (n=895,214) Q1 Static Reference Group ............................................................................ 71 Table B. 5 Risk per 100 live births by calendar year of preterm birth (< 37 weeks completed gestation) at birth 2 by level of mother™s trajectory from birth 1, Singleton Michigan Births 1990 -2012..................................................................................................................................... 72 Table B. 6 Risk per 100 live births by calendar year of preterm birth (<37 weeks completed gestation) at birth 3 by level of mother ™s trajectory from birth 2, Singleton Michigan Births 1990 -2012..................................................................................................................................... 73 Table B.7 Gestational age of preterm births among singleton Michigan births 1990 -2012 ............................................................................................................................................. 74 Table B. 8 Mother™s (N=447,607) Neighborhood Poverty Trajectories by MSA (Metropolitan, Micropolitan , Rural) Change, Singleton Michigan Births 1 to 2, 1990 -2012 .................... 74 Table 3.1 Comparison of Samples for All Singleton Michigan Births 1990 -2012 ............ 93 Table 3.2 Comparison of multivariable -adjusted associations between neighborhood poverty rate and preterm birth ( PTB) using logistic regression and maternal fixed -effects analyses among singleton births in Michigan, 1990 -2012 ............................................................................ 95 Table 3.3 Maternal Fixed Effects Models of PTB with Interaction between Maternal Race/Ethnicity and Poverty ................................................................................................. 95 x Table 3.4 Maternal Fixed Effe cts Odds Ratios (OR) and 95% Confidence Intervals (CI) predicting the probability of preterm birth (PTB), maternal poverty quartile mobility levels (vs. no change) and covariates ................................................................................................... 96 Table 3.5 Comparison of Multivariable -Adjusted Odds Ratios for Lagged Preterm Birth Covariate in Models (Logistic and Fixed Effects) .............................................................. 97 Table 3.6 Table Robustness Checks for positive monotonic covariates and study time period ............................................................................................................................................. 97 Table 3.7 Comparison of multivariabl e-adjusted associations between neighborhood poverty rate by birth and preterm birth (PTB) using logistic regression and maternal fixed -effects analyses among singleton births in Michigan, 1990 -2012 ................................................................ 97 Table C.1 Crude Logistic Association of Neighborhood P overty Mobility and PTB ....... 99 Table C.2 Associations with Poverty Mobility Trajectories and PTB using a Prior PTB (lagged) dependent variable in model, Michigan Births 1990 -2012 ................................................. 99 Table C.3 Comparison of multivariable -adjusted associations between quartiles of neighborhood poverty rate by birth and preterm birth (PTB) using logistic regression, random effects and maternal fixed -effects analyses among singleton births in Michigan, 1990 -2012 ............. 99 Table C.4 Crude FE Model Response Pattern s .................................................................. 100 xi LIST OF FIGURES Figure 1 Analytic Sample Flow Chart for Inclusion Michigan Births ................................ 29 xii KEY TO ABBREVIATIONS ACS American Community Survey AGA Appropriate for gestational age CI Confidence Interval FE Fixed effect GA Gestational age HS High school LGA Large for gestational age MAUP Modifiable aerial unit problem MI Michigan MDHHS Michigan Department of Health and H uman Services NCDB Neighborhood Change Database NH Non -Hispanic AI/AN Non -Hispanic American Indian/ Alaska Native A/PI Non -Hispanic Asian/Pacific Islander NHB Non -Hispanic Black NHW Non -Hispanic White OR Odds Ratio PTB Preterm Birth Q1 Quartile 1 Q2 Quartile 2 Q3 Quartile 3 Q4 Quartile 4 RE Random effect xiii RR Relative Risk SGA Small for gestational age SES Socioeconomic status SD Standard deviation US Unites States 1 INTRODUCTION Background Preterm Birth and Disparities Adverse perinatal outcomes have potential negative lifelong consequences (Barker et al. 1993) and thus provide a salient opportunity for intervention with lasting impact. Preterm birth, or birth before 37 weeks of completed gestation, is the most significant cause of neonatal mortality and morbidity worldwide (Harrison and Goldenberg 2016) . The global incidence of PTB varies by location but is estimated at between 5 -15% representing 15 million births per year (Blencowe et al. 2013) . In the U.S. the incidence of PTB is approximately 11 -12% of live births. Compared with other developed countries the U.S. has the highest rates of preterm birth PTB account ing for 42% of the 1.2 million preterm births occurring in developed countries (Koullali et al. 2016) . This extreme discrepancy suggests a greater need for prevention (MacDorman and Mathews 2011) . Prevention can be a challenge given that in almost half of PTB cases there is no risk factor identified (Blencowe et al. 2013; Menon 2008) . The most significant indicator of risk for preterm birth is a maternal history of prior preterm birth, although the exact mechanism for this risk is not fully understood it may be related to genetic, epigenetic or environmental factors (Plunkett and Mugli a 2008; Yang et al. 2016) . Additional maternal risk factors for PTB include young or advanced maternal age (Muglia and Katz 2010 ), increased parity, shorter inter -pregnancy interval (DeFranco et al. 2007; Thiel de Bocanegra et al. 2014). Risk factors specific to pregnancy include smoking during pregnancy, adequacy of prenatal care, male infant sex, and multiple pregnancy (Committee on Practice Bulletins ŠObstetrics, The American College of Obstetricians and Gynecologists 2012) . Additional social risks include low socioeconomic status (Bertin et al. 2015; Smith et al. 2007) and Non -Hispanic Black race/ethnicity. The rates of PTB 2 vary by race and socioeconomic status, even after controlling for individual level medical and behavioral f actors, hinting at an underlying social process (Culhane and Goldenberg 2011; Lhila and Long 2012; Thoma et al. 2019) . There are enormous societal and medical costs related to prematurity, which affects survival and quality of life of the infant that extends beyond infancy. Recent estimates report that the cost of care in a single year for the 1 in 10 infants born prematurely was $6 billion, as the most conservative estimate based on employer -sponsored plans (Grosse et al. 2017) . Racial and ethnic disparities in preterm birth have long been the subject of study. The black -white disparity in risk f or preterm birth is well established but remains particularly stubborn (Kramer et al. 2010; McKinnon et al. 2016) . Possible explanations have focused on fetal programming potential reproductive outcom es(Burris and Collins 2010) and cumulative weathering as a physical manifestation of lifelong accumulative of deleterious social exposures (Geronimus et al. 2006) . The complex and multifactorial pathways are incompletely understood, and may not be the same for individual birth outcomes and health disparities, but the consequences of PTB remain severe and well documented including being the foremost cause for infant mortality and development of chronic comorbidities (MacDorman and Mathews 2011) . Neighborhood as a Determinant of Adverse Birth Outcomes Maternal characteristics, measured at the individual -level, fail to explain persistent disparities in preterm b irth outcomes which may be driven by structural conditions (Lhila and Long 2012) . One example of a grow ing body of literature that examines structural conditions is that of the neighborhood effects on birth outcomes. Research focusing on neighborhood as a determinant of adverse birth often defines the neighborhood context in relation to deprivation, constra ined socioeconomic position measured at the area -level, such as high poverty, or crime, or 3 as a composite score of neighborhood deprivation income and other neighborhood contextual factors (e.g. % college education and % housing owner -occupied). Neighborho ods are often defined using administrative boundaries such as census tracts. In the U.S. census tracts are relatively homogenous and permanent geographical groupings encompassing between 2,500 -8,000 people (Bureau n.d.). Using census tract neighborhood definitions evidence consistently shows a modest association between neighborhood deprivation and PTB (Holzman et al. 2009; Messer et al. 2008; O™Campo et al. 2008; Yang et al. 2016; Vos et al. 2014) . The association between neighborhood deprivation and increased risk for adverse perinatal outcomes persists even after adjustment for maternal covariates (Luo et al. 2006; Schempf, Strobino, and O™Campo 2009; Janevic et al. 2010) . A recent meta -analyses of neighborhood deprivation and PTB showed a 27% higher risk for PTB [OR 1.27, 95%CI:1.16,1.39] for mothers living in the most (compared to the least) deprived neighborhoods (Ncube et al. 2016) . These studies of neighborhood deprivation have primarily measured exposure to neighborhood deprivation at a cross section single point in time. More recent sch olarship builds on the study of neighborhood deprivation by evaluating trajectory patterns in and out of deprived areas over at least two time points (Collins, Rankin, and David 2015; Bruckner, Kane, and Gailey 2019) using longitudinal data linked to census geographic data. Generally these trajectory patterns show the direction of neighborhood poverty exposure where up ward trajectory reflects decreased neighborhood poverty, downward trajectory reflects increased neighborhood poverty, and static trajectory represents no change in neighborhood poverty level. These trajectory patterns explore exposures in a single individu al and across families. Deprivation trajectories are thus measured inter -generationally (Collins, Mariani, and Rankin 2018; Collins, Rankin, and David 2015; Pearl et al. 2018) and within mothers (Bruckner, Kane, and Gailey 2019) . Similar to 4 cro ss sectional findings, studies of neighborhood trajectories report a modest association between neighborhood deprivation and adverse birth outcomes. The magnitudes of the reported effects are related to the magnitude of the trajectory change, with greater trajectory changes representing stronger associations between neighborhood deprivation and PTB. Upward trajectory reduced preterm birth among upper born white women while downward mobility was more deleterious for birth outcomes (Collins, Rankin, and David 2015) . The authors also report a stronger effect of downward mobility among women who were born into more impoverished areas. Neighborhood Context and Pathway While the neighborhood effects literature consistently shows modest association with birth outcomes there is less consistency in identifying or directly addressing the pathway b y which neighborhood context affects health (Kane et al. 2017) . The neighborhood cont ext itself can represent several different pathways by which birth outcome risk operates with a variety of measures represented including crime (Messer, Vinikoor -Imler, and Laraia 2012) , income (Farley et al. 2006; Metcalfe et a l. 2011) , housing tenure (Morris, Manley, and Sabel 2018) , built environment (Miranda, Messer, and Kroeger 2012) , greenspace (Cusack et al. 2018) and walkability (Messer, Vinikoor -Imler, and Laraia 2012) , residential segregation (A. H. Schempf et al. 2011) and racial isolation (Anthopolos et al. 2011) , and social environment (Messer, Vinikoor -Imler, and Laraia 2012) . While these area -level measures may each contribute to a full contextual understanding of the associations with neighborhoods and birth outcomes we chose to focus on neighborhood deprivation because we believe structural conditions may be addressed through policy action and because it is one of the most studied, making methodological comparisons more informative. 5 For birth outcomes, there may be a two -fold mechanism by which neigh borhood deprivation is thought to affect birth outcomes either through psychosocial factors or through access to material resources or both simultaneously. Psychosocial factors related to neighborhood deprivation include stress (allostatic load, susceptibi lity to infection), as well as poor coping behaviors leading to worse birth outcomes (smoking, alcohol use). Material resources include housing, proximity to health care, food security and access, as well as the physical environment. A mother™s neighborhoo d environment informs her health through these factors; however, the pathway may not be the same for each birth outcome. Here we focus on the material resources pathway by examining poverty although there may be unmeasured psychosocial factors related to m oving and poverty. In fact, we structure our analysis in response to the absence of a complete understanding of the mechanism of these neighborhood effects. Selection Bias in Neighborhood Effects A major problem in the identification of causal ity in studies reporting neighborhood effects is related to selection bia s (Jencks and Mayer 1990; Verheij et al. 1998; Bergström and Van Ham 2010; Hedman and van Ham 2012; Ha et al. 2016; Chetty, Hendren, and Katz 2016; van Ham, Boschman, and Vogel 201 8). However, few epidemiological studies include controls for selection bias in the estimation of neighborhood effects despite a firm consensus on the potential interference with causality. In our studies, selection bias occurs when the mechanism of liv ing in a neighborhood (selection) is not independent from preterm birth. As an example, if Non - Hispanic Black women are more likely to move into a neighborhood with high poverty compared to Non -Hispanic White women and this selection mechanism is not contr olled for adequately we may incorrectly observe a correlation between maternal race/ethnicity and PTB as a neighborhood effect. Therefore, the validity of findings from studies examining associations 6 between neighborhood deprivation and PTB depends upon th e extent to which the neighborhood measure represents the exposures of the population measured. Unobserved maternal characteristics are particularly problematic when examining selection bias as they are not explicitly measured and therefore we cannot contr ol for them the way we can for race/ethnicity. If there is a large amount of selection bias then the measurement reflects the factors that cause a woman to live in a neighborhood. If these are also associated with her risk of PTB we may have confounding in our estimates. In order to better understand selection w e examine 3 mechanisms by which there may be selection bias in the study of the association between neighborhood poverty and preterm birth (1) residential mobility, selective sorting by changing res idential locations (2) poverty trajectory, selective sorting into specific levels of poverty and (3) unobserved maternal characteristics, selective sorting into neighborhoods by maternal characteristics which are not measured as covariates in our data. In order to better control for selection mechanisms we need to better understand the selective sorting patterns before we can distinguish between the causal effects of neighborhood and the results of neighborhood selection. Study Objectives Main Research Question : The overall goal of this dissertation is to characterize maternal neighborhood mobility between pregnancies, to determine the association between neighborhoo d poverty mobility and PTB, and to assess whether previously observed associations between neighborhood poverty and PTB are due to selection of mothers into neighborhoods . Our studies aim to answer the following question s related to the association between neighborhood deprivation and PTB. 7 Study Aim 1a: To characterize and descriptivel y summarize the extent to which mothers move neighborhoods between pregnancies within the state of Michigan. Hypothesis 1a: Based on a previous stud y in a similar population , we hypothesize that approximately 25% of our sample will move neighborhoods betw een pregnancies. This analysis will create and characterize a sample to use in a subsequent analysis of neighborhood selection bias using matern al fixed effects study design. Study Aim 1b: To describe the differences in sociodemographic and health characteristics between mothers who move between pregnancies and those that do not move. Hypothesis 1b: We hypothesize movers will be more likely to be younger, have more children, experience lo wer poverty, and more likely to smoke . This descriptive analysis is important because it will help us understand whether differences in measured maternal characteristics between mothers that move neighborhoods between pregnancies may lead to bias in neigh borhood effects study designs that do not account for these factors, which may also be related to PTB. Study Aim 1c: Are prior adverse birth outcomes associated with residential mobility changes neighborhoods between pregnancies? Hypothesis 1c: We hypoth esize that a mother is less likely to move if she has an adverse birth outcome , consistent with the healthy migrant theory (Collins, 2011) which postulates that migrants are more likely to be healthy . Study Aim 2a: Do mothers experience neighborhood pover ty trajectory across their successive pregnancies? 8 Hypothesis 2a: Based on previous studies of neighborhood poverty changes, we expect the mothers in our sample will experience neighborhood poverty trajectory changes across their successive pregnancies, b ut this change will be limited to a 1 -quartile change in poverty level. Study Aim2b: Do movers experience greater neighborhood poverty trajectory changes than non -movers? Hypothesis 2b: There has not been enough previous scholarship on this topic to draw a firm hypothesis. We hypothesize that there may be systematic differences between movers and non -movers that warrant this investigation. Study Aim 2c: Is neighborhood poverty mobility associated with PTB? Hypothesis2c: Consistent with previous studies of maternal mobility ( Lupo et al. 2010; Sundquist et al. 2011; Bell and Belanger 2012; Miller, Siffel, and Correa 2010) , we hypothesize that will find a modestly protective association between upward neighborhood poverty trajectory and PTB . We further hyp othesize the strongest association will occur with the greatest change in neighborhood poverty trajectory, measured by at least a 3 quartile change in neighborhood poverty , in either direction (downward or upward) . Study Aim 3: To determine the associat ion between neighborhood poverty and PTB using a maternal fixed effects method to control for maternal characteristics, both measured and unmeasured, that may be associated with selection into the neighborhood and PTB. Hypothesis 3: We hypothesize using a maternal fixed effects approach will yield attenuated association between neighborhood poverty and PTB compared to traditional logistic and random effects approaches. 9 In summary, in this dissertation we hypothesize that there are characteristics of mother s that determine what neighborhood they live in. Some of those characteristics are measured and can be controlled for Œ in Study 1, we are characterizing what measured characteristics are associated with moving as a way to begin to address this issue. In S tudy 2, we are looking at whether the type/direction of move is associated with PTB. In Study 3, we are trying to account for the unmeasured characteristics that might also determine what neighborhood women live in and thus account for some of the relation ship between neighborhood and PTB. 10 REFERENCE S 11 REFERENCES Anthopolos, Rebecca, Sherman A. James, Alan E. Gelfand, and Marie Lynn Miranda. 2011. fiA Spatial Measure of Neighborhood Level Racial Isolation Applied to Low Birthweight, Preterm Birth, and Birthweight in North Carolina.fl Spatial and Spatio -Temporal Epidemiology 2 (4): 235Œ46. https://doi.org/10.1016/j.sste.2011.06.002. Barker, D.J.P, K.M Godfrey, P.D Gluckman, J.E Harding, J.A Owens, and J.S Robinson. 1993. fiFetal Nutrition and Cardiovascular Disease in Adult Life.fl The Lancet 341 (8850): 938 Œ41. htt ps://doi.org/10.1016/0140 -6736(93)91224 -A. Bell, Michelle L., and Kathleen Belanger. 2012. fiReview of Research on Residential Mobility during Pregnancy: Consequences for Assessment of Prenatal Environmental Exposures.fl Journal of Exposure Science and Envir onmental Epidemiology 22 (5): 429 Œ38. https://doi.org/10.1038/jes.2012.42. Bergström, Lina, and Maarten Van Ham. 2010. fiUnderstanding Neighbourhood Effects: Selection Bias and Residential Mobility.fl Bertin, Mélanie, Jean -François Viel, Christine Monfort, S ylvaine Cordier, and Cécile Chevrier. 2015. fiSocioeconomic Disparities in Adverse Birth Outcomes in Urban and Rural Contexts: A French Mother ŒChild Cohort.fl Paediatric and Perinatal Epidemiology 29 (5): 426 Œ35. https://doi.org/10.1111/ppe.12208. Blencowe, Hannah, Simon Cousens, Doris Chou, Mikkel Oestergaard, Lale Say, Ann -Beth Moller, Mary Kinney, Joy Lawn, and the Born Too Soon Preterm Birth Action Group (see acknowledgement for full list). 2013. fiBorn Too Soon: The Global Epidemiology of 15 Million Prete rm Births.fl Reproductive Health 10 (1): S2. https://doi.org/10.1186/1742 -4755-10-S1-S2. Bruckner, Tim A., Jennifer B. Kane, and Samantha Gailey. 2019. fiStrong Upward Neighborhood Mobility and Preterm Birth: A Matched -Sibling Design Approach.fl Annals of Epi demiology 36 (August): 48 -54.e1. https://doi.org/10.1016/j.annepidem.2019.05.005. Bureau, US Census. n.d. fiGeography Program.fl Accessed November 9, 2019. https://www.census.gov/programs -surveys/geography.html. Burris, Heather H., and James W. Collins. 2010 . fiRace and Preterm Birth --the Case for Epigenetic Inquiry.fl Ethnicity & Disease 20 (3): 296 Œ99. Chetty, Raj, Nathaniel Hendren, and Lawrence F. Katz. 2016. fiThe Effects of Exposure to Better Neighborhoods on Children: New Evidence from the Moving to Oppor tunity Experiment.fl The American Economic Review 106 (4): 855 Œ902. 12 Collins, James W., Allison Mariani, and Kristin Rankin. 2018. fiAfrican -American Women™s Upward Economic Mobility and Small for Gestational Age Births: A Population -Based Study.fl Maternal an d Child Health Journal 22 (8): 1183 Œ89. https://doi.org/10.1007/s10995 -018-2503-4. Collins, James W., Kristin M. Rankin, and Richard J. David. 2011. fiAfrican American Women™s Lifetime Upward Economic Mobility and Preterm Birth: The Effect of Fetal Programm ing.fl American Journal of Public Health 101 (4): 714 Œ19. https://doi.org/10.2105/AJPH.2010.195024. ŠŠŠ. 2015. fiDownward Economic Mobility and Preterm Birth: An Exploratory Study of Chicago -Born Upper Class White Mothers.fl Maternal and Child Health Journal 19 (7): 1601 Œ7. https://doi.org/10.1007/s10995 -015-1670-9. Committee on Practice Bulletins ŠObstetrics, The American College of Obstetricians and Gynecologists. 2012. fiPractice Bulletin No. 130: Prediction and Prevention of Preterm Birth.fl Obstetrics and G ynecology 120 (4): 964 Œ73. https://doi.org/10.1097/AOG.0b013e3182723b1b. Culhane, Jennifer F., and Robert L. Goldenberg. 2011. fiRacial Disparities in Preterm Birth.fl Seminars in Perinatology 35 (4): 234 Œ39. https://doi.org/10.1053/j.semperi.2011.02.020. Cusack, Leanne, Hind Sbihi, Andrew Larkin, Angela Chow, Jeffrey R. Brook, Theo Moraes, Piush J. Mandhane, et al. 2018. fiResidential Green Space and Pathways to Term Birth Weight in the Canadian Healthy Infant Longitudinal Development (CHILD) Study.fl Internat ional Journal of Health Geographics 17 (1): 43. https://doi.org/10.1186/s12942 -018-0160-x. DeFranco, Emily A., David M. Stamilio, Sarah E. Boslaugh, Gilad A. Gross, and Louis J. Muglia. 2007. fiA Short Interpregnancy Interval Is a Risk Factor for Preterm Birth and Its Recurrence.fl American Journal of Obstetrics and Gynecology 197 (3): 264.e1 -264.e6. https://doi.org/10.1016/j.ajog.2007.06.042. Farley, Thomas A., Karen Mason, Janet Rice, Joanna D. Habel, Richard Scribner, and Deborah A. Cohen. 2006. fiThe Relationship between the Neighbourhood Environment and Adverse Birth Outcomes.fl Paediatric and P erinatal Epidemiology 20 (3): 188 Œ200. https://doi.org/10.1111/j.1365 -3016.2006.00719.x. Geronimus, Arline T., Margaret Hicken, Danya Keene, and John Bound. 2006. fi‚Weathering™ and Age Patterns of Allostatic Load Scores Among Blacks and Whites in the Unite d States.fl American Journal of Public Health 96 (5): 826 Œ33. https://doi.org/10.2105/AJPH.2004.060749. Grosse, Scott D., Norman J. Waitzman, Ninee Yang, Karon Abe, and Wanda D. Barfield. 2017. fiEmployer -Sponsored Plan Expenditures for Infants Born Preterm. fl Pediatrics 140 (4). https://doi.org/10.1542/peds.2017 -1078. 13 Ha, Sandie, Hui Hu, Liang Mao, Dikea Roussos -Ross, Jeffrey Roth, and Xiaohui Xu. 2016. fiPotential Selection Bias Associated with Using Geocoded Birth Records for Epidemiologic Research.fl Annals of Epidemiology 26 (3): 204 Œ11. https://doi.org/10.1016/j.annepidem.2016.01.002. Ham, Maarten van, Sanne Boschman, and Matt Vogel. 2018. fiIncorporating Neighborhood Choice in a Model of Neighborhood Effects on Income.fl Demography 55 (3): 1069 Œ90. https://d oi.org/10.1007/s13524 -018-0672-9. Harrison, Margo S., and Robert L. Goldenberg. 2016. fiGlobal Burden of Prematurity.fl Seminars in Fetal & Neonatal Medicine 21 (2): 74 Œ79. https://doi.org/10.1016/j.siny.2015.12.007. Hedman, Lina, and Maarten van Ham. 2012. fiUnderstanding Neighbourhood Effects: Selection Bias and Residential Mobility.fl In Neighbourhood Effects Research: New Perspectives , edited by Maarten van Ham, David Manley, Nick Bailey, Ludi Simpson, and Duncan Maclennan, 79 Œ99. Dordrecht: Springer Nether lands. https://doi.org/10.1007/978 -94-007-2309-2_4. Holzman, Claudia, Janet Eyster, Mary Kleyn, Lynne C. Messer, Jay S. Kaufman, Barbara A. Laraia, Patricia O™Campo, Jessica G. Burke, Jennifer Culhane, and Irma T. Elo. 2009. fiMaternal Weathering and Risk o f Preterm Delivery.fl American Journal of Public Health 99 (10): 1864 Œ1871. Janevic, T, CR Stein, DA Savitz, JS Kaufman, SM Mason, and AH Herring. 2010. fiNeighborhood Deprivation and Adverse Birth Outcomes among Diverse Ethnic Groups.fl Annals of Epidemiolog y 20 (6): 445 Œ51. https://doi.org/10.1016/j.annepidem.2010.02.010. Jencks, Christopher, and Susan E. Mayer. 1990. fiThe Social Consequences of Growing up in a Poor Neighborhood.fl Inner -City Poverty in the United States 111: 186. Kane, Jennifer B., Gandarvak a Miles, Jennifer Yourkavitch, and Katherine King. 2017. fiNeighborhood Context and Birth Outcomes: Going beyond Neighborhood Disadvantage, Incorporating Affluence.fl SSM - Population Health 3 (August): 699 Œ712. https://doi.org/10.1016/j.ssmph.2017.08.003. Koullali, B., M. A. Oudijk, T. A. J. Nijman, B. W. J. Mol, and E. Pajkrt. 2016. fiRisk Assessment and Management to Prevent Preterm Birth.fl Seminars in Fetal and Neonatal Medicine , Prediction and prevention of preterm birth and its sequelae, 21 (2): 80 Œ88. https://doi.org/10.1016/j.siny.2016.01.005. Kramer, Michael R., Hannah L. Cooper, Carolyn D. Drews -Botsch, Lance A. Waller, and Carol R. Hogue. 2010. fiMetropolitan Isolation Segregation and Black -White Disparities in Very Preterm Birth: A Test of Mediating Pathways and Variance Explained.fl Social Science & Medicine (1982) 71 (12): 2108 Œ16. https://doi.org/10.1016/j.socscimed.2010.09.011. 14 Lhila, Aparna, and Sharon Long. 2012. fiWhat Is Driving the Black ŒWhite Difference in Low Birthweight in the US?fl Health Ec onomics 21 (3): 301 Œ15. https://doi.org/10.1002/hec.1715. Luo, Zhong -Cheng, Russell Wilkins, Michael S. Kramer, and Fetal and Infant Health Study Group of the Canadian Perinatal Surveillance System. 2006. fiEffect of Neighbourhood Income and Maternal Educat ion on Birth Outcomes: A Population -Based Study.fl CMAJ: Canadian Medical Association Journal = Journal de l™Association Medicale Canadienne 174 (10): 1415 Œ20. https://doi.org/10.1503/cmaj.051096. Lupo, Philip J., Elaine Symanski, Wenyaw Chan, Laura E. Mitc hell, D. Kim Waller, Mark A. Canfield, and Peter H. Langlois. 2010. fiDifferences in Exposure Assignment between Conception and Delivery: The Impact of Maternal Mobility.fl Paediatric and Perinatal Epidemiology 24 (2): 200 Œ208. https://doi.org/10.1111/j.1365 -3016.2010.01096.x. MacDorman, Marian F., and T. J. Mathews. 2011. fiUnderstanding Racial and Ethnic Disparities in U.S. Infant Mortality Rates.fl NCHS Data Brief , no. 74 (September): 1 Œ8. McKinnon, Britt, Seungmi Yang, Michael S. Kramer, Tracey Bushnik, Ama nda J. Sheppard, and Jay S. Kaufman. 2016. fiComparison of Black -White Disparities in Preterm Birth between Canada and the United States.fl CMAJ: Canadian Medical Association Journal = Journal de l™Association Medicale Canadienne 188 (1): E19 -26. https://doi .org/10.1503/cmaj.150464. Menon, Ramkumar. 2008. fiSpontaneous Preterm Birth, a Clinical Dilemma: Etiologic, Pathophysiologic and Genetic Heterogeneities and Racial Disparity.fl Acta Obstetricia Et Gynecologica Scandinavica 87 (6): 590 Œ600. https://doi.org/1 0.1080/00016340802005126. Messer, Lynne C., Lisa C. Vinikoor, Barbara A. Laraia, Jay S. Kaufman, Janet Eyster, Claudia Holzman, Jennifer Culhane, Irma Elo, Jessica G. Burke, and Patricia O™Campo. 2008. fiSocioeconomic Domains and Associations with Preterm B irth.fl Social Science & Medicine 67 (8): 1247 Œ1257. Messer, Lynne C., Lisa C. Vinikoor -Imler, and Barbara A. Laraia. 2012. fiConceptualizing Neighborhood Space: Consistency and Variation of Associations for Neighborhood Factors and Pregnancy Health across Multiple Neighborhood Units.fl Health & Place 18 (4): 805 Œ13. https://doi.org/10.1016/j.healthplace.2012.03.012. Metcalfe, Amy, Parabhdeep Lail, William A. Ghali, and Reg S. Sauve. 2011. fiThe Association between Neighbourhoods and Adverse Birth Outcomes: A Systematic Review and Meta -Analysis of Multi -Level St udies.fl Paediatric and Perinatal Epidemiology 25 (3): 236 Œ45. https://doi.org/10.1111/j.1365 -3016.2011.01192.x. Miller, Assia, Csaba Siffel, and Adolfo Correa. 2010. fiResidential Mobility during Pregnancy: Patterns and Correlates.fl Maternal and Child Healt h Journal 14 (4): 625 Œ34. https://doi.org/10.1007/s10995 -009-0492-z. 15 Miranda, Marie Lynn, Lynne C. Messer, and Gretchen L. Kroeger. 2012. fiAssociations between the Quality of the Residential Built Environment and Pregnancy Outcomes among Women in North Car olina.fl Environmental Health Perspectives 120 (3): 471 Œ77. https://doi.org/10.1289/ehp.1103578. Morris, Tim, David Manley, and Clive E. Sabel. 2018. fiResidential Mobility.fl Progress in Human Geography 42 (1): 112 Œ33. https://doi.org/10.1177/030913251664945 4. Muglia, Louis J., and Michael Katz. 2010. fiThe Enigma of Spontaneous Preterm Birth.fl The New England Journal of Medicine 362 (6): 529 Œ35. https://doi.org/10.1056/NEJMra0904308. Ncube, Collette N., Daniel A. Enquobahrie, Steven M. Albert, Amy L. Herrick, and Jessica G. Burke. 2016. fiAssociation of Neighborhood Context with Offspring Risk of Preterm Birth and Low Birthweight: A Systematic Review and Meta -Analysis of Population -Based Studies.fl Social Science & Medicine (1982) 153 (March): 156 Œ64. https://do i.org/10.1016/j.socscimed.2016.02.014. O™Campo, Patricia, Jessica G. Burke, Jennifer Culhane, Irma T. Elo, Janet Eyster, Claudia Holzman, Lynne C. Messer, Jay S. Kaufman, and Barbara A. Laraia. 2008. fiNeighborhood Deprivation and Preterm Birth among Non -Hispanic Black and White Women in Eight Geographic Areas in the United States.fl American Journal of Epidemiology 167 (2): 155 Œ163. Pearl, Michelle, Jennifer Ahern, Alan Hubbard, Barbara Laraia, Bina Patel Shrimali, Victor Poon, and Martin Kharrazi. 2018. fiLife -Course Neighbourhood Opportunity and Racial -Ethnic Disparities in Risk of Preterm Birth.fl Paediatric and Perinatal Epidemiology 32 (5): 412 Œ19. https://doi.org/10.1111/ppe.12482. Plunkett, Jevon, and Louis J. Muglia. 2008. fiGenetic Contributions to Preterm Birth: Implications from Epidemiolo gical and Genetic Association Studies.fl Annals of Medicine 40 (3): 167Œ79. https://doi.org/10.1080/07853890701806181. Schempf, Ashley H., Jay S. Kaufman, Lynne C. Messer, and Pauline Mendola. 2011. fiThe Neighborhood Contribution to Black -White Perinatal Di sparities: An Example from Two North Carolina Counties, 1999 -2001.fl American Journal of Epidemiology 174 (6): 744 Œ52. https://doi.org/10.1093/aje/kwr128. Schempf, Ashley, Donna Strobino, and Patricia O™Campo. 2009. fiNeighborhood Effects on Birthweight: An Exploration of Psychosocial and Behavioral Pathways in Baltimore, 1995 Œ1996.fl Social Science & Medicine 68 (1): 100 Œ110. https://doi.org/10.1016/j.socscimed.2008.10.006. Smith, L. K., E. S. Draper, B. N. Manktelow, J. S. Dorling, and D. J. Field. 2007. fiSo cioeconomic Inequalities in Very Preterm Birth Rates.fl Archives of Disease in Childhood -Fetal and Neonatal Edition 92 (1): F11 Œ14. https://doi.org/10.1136/adc.2005.090308. 16 Sundquist, Jan, Kristina Sundquist, Sven -Erik Johansson, Xinjun Li, and Marilyn Wink leby. 2011. fiMothers, Places and Small for Gestational Age Births: A Cohort Study.fl Archives of Disease in Childhood 96 (4): 380 Œ85. https://doi.org/10.1136/adc.2009.180042. Thiel de Bocanegra, Heike, Richard Chang, Mike Howell, and Philip Darney. 2014. fiInterpregnancy Intervals: Impact of Postpartum Contraceptive Effectiveness and Coverage.fl American Journal of Obstetrics and Gynecology 210 (4): 311.e1 -311.e8. https://doi.org/10.1016/j.ajog.2013.12.020. Thoma, Marie E., Laura B. Drew, Ashley H. Hirai, Ther esa Y. Kim, Andrew Fenelon, and Edmond D. Shenassa. 2019. fiBlack ŒWhite Disparities in Preterm Birth: Geographic, Social, and Health Determinants.fl American Journal of Preventive Medicine 57 (5): 675 Œ86. https://doi.org/10.1016/j.amepre.2019.07.007. Verheij , R. A., H. D. van de Mheen, D. H. de Bakker, P. P. Groenewegen, and J. P. Mackenbach. 1998. fiUrban -Rural Variations in Health in The Netherlands: Does Selective Migration Play a Part?fl Journal of Epidemiology & Community Health 52 (8): 487 Œ93. https://doi .org/10.1136/jech.52.8.487. 2014. fiDeprived Neighborhoods and Adverse Perinatal Outcome: A Systematic Review and Meta -Analysis.fl Acta Obstetricia Et Gynecologica Sc andinavica 93 (8): 727 Œ40. https://doi.org/10.1111/aogs.12430. Wallace, Maeve, Emily Harville, Katherine Theall, Larry Webber, Wei Chen, and Gerald Berenson. 2013. fiNeighborhood Poverty, Allostatic Load, and Birth Outcomes in African American and White Wom en: Findings from the Bogalusa Heart Study.fl Health & Place 24 (November): 260 Œ66. https://doi.org/10.1016/j.healthplace.2013.10.002. Yang, Juan, Rebecca J. Baer, Vincenzo Berghella, Christina Chambers, Paul Chung, Tumaini Coker, Robert J. Currier, et al. 2016. fiRecurrence of Preterm Birth and Early Term Birth.fl Obstetrics and Gynecology 128 (2): 364 Œ72. https://doi.org/10.1097/AOG.0000000000001506. 17 STUDY 1 Introduction There is substantial evidence that neighborhood deprivation is associated with adverse birth outcomes, specifically preterm birth (PTB; <37 weeks gestation) and small -for-gestational age (SGA; <10 percentile birth weight) (Sundquist et al. 2011; Janevic et al. 2010; Arcaya et al. 2012; Schempf and Kaufman 2012; O™Campo et al. 2008; Messer et al. 2008) . However, the role of residential mobility, a change in residential address, in this association of neighborhood effects and birth outcomes has received limit ed consideration. In order to elucidate the role of maternal selection into neighborhoods we started with a foremost concept of entry and exit into neighborhoods: residential mobility. Mobility represents a change in place of residence based on residentia l address reported on the birth certificate. Research on mobility and poverty has found that mobility to lower -poverty areas resulted in improved safety, physical health, and mental health (Chetty, Hendren, and Katz 2016) . Chetty e t al. found residential mobility to better neighborhoods had a causal effect on children™s outcomes but that the effect of mobility decl ines with a child™s age at move. Research on residential mobility and pregnancy suggests there are several demographic features associated with movement. Movers tend to be younger (Canfield et al. 2006; Khoury et al. 1988) with some estimates showing 3.39 OR (95% CI: 2.12 -5.4, 20-24 v >30) (Miller, Siffel, and Correa 2010) ; more likely to be smokers (1.46 OR 95% CI 1.01 -2.12)(Miller, Siffel, and Correa 2010) while non -movers tend to have greater household income (Fell, Dodds, and King 2004) and lower parity (Canfield et al. 2006) . Characteristics of mobility, such as distance moved, may be an important aspect in understanding the relationship betw een residential mobility and health (Clark and Huang 2003) . Most literature of residential mobility and 18 pregnancy found pregnant women move shorter distances of less than 3 miles (Lupo et al. 2010; Tang et al. 2018) . Shorter distance moves represent uninte nded movement while long distance moves show an upward pattern of mobility and planning (Morris, Manley, and Sabel 2018; Clark and Huang 2003). Having children and employment may actually be constraints to mobility (Lee and Waddell 2010; Kim, Pagliara, and Preston 2003) . Maternal residential mobility is not uncommon but also not well characterized over the life course . For pregnant women there are multiple time periods to consider when evaluating mobility (1) across a single pregnancy, usually measured at trimester time points; or (2) across multiple pregnancies. Presently, evidence estimates between 12 -30% of women move during pregnancy (Lupo et al. 2010; Sundquist et al. 2011; Bell and Belanger 2012; Miller, Siffel, and Correa 2010) . Studies of residentia l mobility and pregnancy overwhelmingly focus on the effect of environmental exposure misclassification during a single pregnancy (Bell and Belanger 2012; Chen et al. 2010; Canfield et al. 2006) . Two studies which followed women up to the first year of bir th reported even higher mobility with up to 42% of women mobility at least once (Saadeh et al. 2013; Urayama et al. 2009) . Even less is known about the socioeconomic status (SES) mobility patterns of women postpartum with one study reporting 7% of women mo ve to areas of different SES (Margerison -Zilko et al. 2016) and another reported 24% moved to different SES (Saadeh et al. 2013), similarly split in direction of mobility. Moreover, no studies to date have specifically looked at the effect of birth outcomes on subsequent mobility patterns. Although, other studies have examined motivations for mobili ty related to birth outcomes. Residential mobility research links childbirth as a fitrigger eventfl for mobility and suggest increased parity is associated with increased likelihood of residential mobility (Kulu and Washbrook 2014), likely due to space needs of a growing family. There is 19 mixed evidence on the effect of children on mo bility with some researchers arguing parents are less likely to move in an effort to maintain established neighborhood ties to benefit their children (Morris, Manley, and Sabel 2018) . We are unaware of any studies that have investigated the association of adverse prior birth outcome on either (1) mobility or (2) neighborhood poverty mobility. The objective of this study is to describe residential mobility patterns and birth outcomes among Michigan -resident women between their successive births over multipl e years in order to better understand inform our empirical design for testing residual confounding due to selection bias in our subsequent fixed effects analysis. We described study demographics and differences in residential mobility status, while future papers will examine the neighborhood poverty mobility (downward, static, and upward ) and PTB. We used maternally -linked data to follow the same woman across multiple pregnancies and evaluated her residential mobility biography, maternal characteristics, mo bility correlates, and birth outcomes. We hypothesized that a large proportion of women move between births and that they are demographically distinct compared to non -movers (more likely to smoke, be younger, have lower educational attainment, and have a shorter time in census tract). We hypothesized non-movers would have poorer birth outcomes across all levels of parity with increased occurrence of PTB. Finally, we looked at how prior birth outcomes influence residential mobility changes. Methods Data and study population We used birth certificate data from the Michigan Department of Health and Human Services (MDHHS) Department of Vital Statistics for all births to Michigan -resident women during the period 1989 -2012 linked by mother. Birth certificate data was geocoded by MDHHS 20 based on self -reported maternal residence at birth for all births between 1994 to 2012. Births that occurred prior to 1994 were geocoded by the Children™s Environmental Health Initiative also using self -reported residence at birth. A s shown in Figure 1 , birth records were included in the study if the women were between 15 and 44 years of age, had a singleton birth, and had geocoded information. Records were excluded if they did not meet the inclusion criteria and if they had any missing covariate inform ation. Our analysis is restricted to the reproductive age of the mother and as such do es not capture her movement from childhood. Mothers will be left truncated for births before 1990 and right truncated for births after 2012, although we do evaluate parit y and birth year to account f or this aspect of study design . In the overall sample, we found almost a quarter of the first recorded births were not nulliparous. As prior PTB has been associated with subsequent PTB (Blencowe et al. 2013) and inter -state bir th records were not available, we further restricted our sample to first, second, and third births recorded in Michigan. This study was approved by the MDHHS Institutional Review Board. Geolytics™ Neighborhood Change Database (NCDB) Information on income level was measured from the census tract level poverty data collected from the Geolytics™ Neighborhood Change Database for the period 1990 -2010. This database collected information from 3 decennial census years 1990, 2000, and 2010 an d adjusts data to the 2010 census tract boundaries to allow comparison across multiple census years. US Census American Community Survey (ACS) Census tract level poverty rates for 2012 were collected from the US Census American Community Survey. They are adjusted to the 2010 Census Tract boundaries and therefore compatible with the boundary definitions from the Geolytics™ census tracts. 21 US Census Tiger/Line Shapefiles The U.S. Census Tiger/Line Shapefiles contain geographic entity codes which can be lin ked to census tract in the birth certificate data by census year. Longitude and latitude coordinates were extracted from this file and merged with the birth certificate data by census tract id and census year. Measures Poverty Neighborhood poverty rates were based on census tract poverty rates collected every 10 years. As we were interested in examining maternally -linked data over a long study time period we used normalized census data, Geolytics™ Neighborhood Change Database (NCDB), to acc ount for geographic boundary changes between decennial census years. Intercensal census tract poverty rates were then linearly interpolated across calendar years for the period 1990 -2010 using a join method. For the period 2011 -2012, poverty rates were ext racted from the American Community Survey, Michigan 2012 for each census tract. We then performed a linear interpolation between 2010 and 2012 data for the 2011 census tract poverty rates. Neighborhood poverty rates were merged with data from the birth cer tificate based on infant™s birth year and mother™s census tract at the time of delivery using the longitudinal crosswalk reference tool (Logan, Xu, and Stults 2014) . Neighborhood poverty rate is repor ted categorically as quartile distributions, based quartile cut -points from the linearly interpolated NCDB data, and as a continuous mean. Residential Mobility Status Residential mobility status is measured across two successive births by a change in census tract geocoded t o residential address provided on the birth certificate. Distance moved Information on distance between census tract centroid was computed using latitude and longitude coordinates from the U.S. Census Tiger/Line Shapefiles. In order to 22 compute census tract distance, each census tract was assigned a longitude and latitude for each birth record from the Tiger/Line Shapefiles by census tract and census year resulting in 2 sets of longitude and latitude per women for each analytic sample res tricted by parity (Births 1 to 2, and Births 2 to 3). For non -movers there was no difference in distance, but for movers we then used the GEODIST function in SAS to compute the geodetic distance in miles between a mother™s two latitude and longitude coordi nates using input values in degrees. Time in census tract The time period of residence in census tract was calculated based on time between births rec ord in the birth certificate files in days. Adverse birth outcomes Our main outcome was PTB (< 37 weeks completed gestation) following recent evidence linking suggesting a protective effect of strong upward neighborhood mobility on PTB outcomes (Bruckner, Kane, and Gailey 2019) . Maternal Characteristics Maternal demographic information from birth certificate data included continuous age; age categorized (<20, 20 -24, 25-29, 30-24, 35-40, >40); marital status (married or unmarried); nativity by country (United State or Foreign) and state (Michigan or not Michigan). A previous analysis evaluated mater nal education and race/ethnicity for consistency across maternally -linked dataset and excluded improbable observations (for example, negative educational attainment). Maternal race included race/ethnicity using birth certificate date (Non -Hispanic White, N on-Hispanic Black, Hispanic, Non -Hispanic American Indian, Non -Hispanic Asian/Pacific Islander, and Unknown). Maternal educational attainment, also using birth certificate data, was collapsed into 6 categories (less than high school; 9 th-12 grade/no diplom a; high school Graduate/GED; Some College/Associate; Bachelor™s Degree; Professional Degree). Paternal demographic fa ctors are limited to age, race/ ethnicity and not reported in this analysis 23 but presented in supplemental tables (Appendix A.3). Maternal risk factor information from birth certificate data included tobacco use during pregnancy recorded as yes/no or unknown. Delivery and birth information from the birth certificate included source of payment (private insurance, Medicaid, self -pay, other, and unknown) that may act as a proxy for individual socioeconomic status. Statistical Analyses Women who moved between births (movers) were compared with women who did not move between their birt hs (non -movers). Univariate analysis compared the demographic differences and mean poverty rate between movers and non -movers and tested for statistical significance using X 2 test for general association. Modified Poisson regression with standard robust er ror variances were used to compute the relative risk of residential mobility by maternal characteristics (Zhao, n.d.) . We used Poisson regression to model this risk as outcome was a binary count variable. As Poisson is better suited to count data and does not have as many convergence challenges as a binomial -log model we used this regression strategy. We used a logistic regression to examine the association between prior PTB and odds of mobility adjusted for confounders related to PTB and mobility (maternal age, parity, marital changes, smoking, and maternal race /ethnicity). There were 38,957 PTB outcomes at birth 1 and 14,181 at birth 2. All index birth s marked the onset of risk for subsequent mobility. Results Table 1.1 shows the analytic subsample of births 1 to 2 (n= 895,214 births to 447,607 women), and births 2 to 3 (n=355,912 births to 117,956) where both births occurred in Michigan stratified by residential mobility status (mover and non -mover). Notably, a large pe rcentage of women were classified as movers for both birth samples (50.2% and 51.1%, for births 1 to 2 and births 2 to 3, respectively). Movers were a greater percentage white, younger, and U.S. Born 24 compared to non -movers. Movers had a slightly higher me an poverty rate compared to non -movers (15% compared to 11%; 16% compared to 12%) and the poverty rates increased as parity increased. Among movers , the majority of moves were within the same county and the mean distance moved was 15.9 and 14.5 miles in ea ch birth sample. Table 1.2a shows the risk ratio of mobility by maternal characteristics for births 1 and 2. Table 1.2b sho ws the risk ratio of mobility by maternal characteristics for births 2 and 3. Movers and non -movers were significantly different for all characteristics except infant sex and educational attainment. The unadjusted risk of mobility among Non -Hispanic Black women was 1.4 times the risk (95% CI: 1.4 -1.4) of mobility among Non -Hispanic white women, and after adjusted the risk was 1.2 (95% C I 1.2 -1.2). Unmarried women had 1.6 times the risk (95% CI: 1.6-16) of mobility compared to married women and when adjusted for all other covariates in the models was 1.2 times the risk (95%CI 1.2 -1.2). Women who smoked also had a high risk of mobility co mpared to women who did not report smoking during pregnancy (unadjusted RR: 1.3, 95% CI: 1.3 -1.3, adjusted RR: 1.1, 95% CI: 1.1 -1.2). Finally, Medicaid as the insurance pay source at the time of birth had a higher risk for mobility (RR: 1.5, 95% CI: 1.5 -1.5) compared with private payer. The risk of residential mobility by maternal characteristics remained fairly consistent across two levels of parity from births 1 to 2 and from births 2 to 3. In Table 1.3 we exploit our data structure to examine the associa tion between prior PTB and subsequent residential mobility using logistic regression. We report on odds ratio for mobility by the time of your successive birth following a prior preterm birth. The odds for mobility after a PTB compared to a term birth at b irth 1, after adjustment for race, marital change and age are null (OR: 1.0, 95% CI: 1.0-1.0) at birth 2 and modestly increased at birth 3 (OR: 1.06, 95% CI:1.03 to 1.10). This corresponds to a 6% increase in the odds of mobility between 25 births 2 and 3 given you have a preterm birth at birth 2 compared to a term second birth. Our model of mobility showed a greater magnitude of odds for marital change and race/ethnicity. This suggests that a change in marital status between births may more strongly associated with mobility for divorce (Births 1 to 2 OR: 2.49 , 95% CI:2.37 -2.62 ; Births 2 to 3 OR: 3.34 , 95% CI: 3.13 -3.55) and for married (Births 1 to 2 OR: 2.79 , 95% CI: 2.71 -2.83 ; Births 2 to 3 OR:1.92 , 95% CI: 1.85 -2.00). Maternal race/e thnicity also had significant odds, with significant ORs for NHB 1.99 and 2.19 compared to NHW, at births 2 and 3 respectively. All other maternal race ethnicities, except NH Native American had increased odds of mobility compared to NHW, although they wer e higher at birth 2 compared to birth 3. Discussion Our major finding is that a larger percentage of women moved (50%) than we hypothesized (25%). The distribution of maternal characteristics is systemati cally different between movers and non -movers acros s demographic, risk factor, birth outcomes, and neighborhood characteristics. The risk of mobility is higher among women who are NHB, smokers, unmarried, and had Medicaid as a source of insurance. The diff erences between movers and non -movers remain consis tent across multiple levels of recorded births. We also looked at across two pairs of births and observe increasing levels of poverty and PTB as parity increased. We found that prior preterm birth had a nu ll association with mobility. Compared to previous reports which ranged from 12 -45% and measured primarily mobility during pregnancy we report almost half of our sample moved between successive births. We found little evidence that prior PTB was associated with mobility by birth 2, con trary to our hypothesis that it would increase the risk of mobility. We found that changes related to marital 26 status, getting divorced or married, were the strong est correlates of mobility and may be important predictors in future studies. We did not find a significant difference between movers and non -movers for educati onal attainment, Table 1.2. We would have expected movers to be more likely to have higher educational attainment as that might be related to job opportunity and mobility. Our sample only captures women during the childbearing years. This may indicate that this is not a very dynamic time for changes in educational attainment. Further, this may be due to limitation of a single state capture whereby movers with different educational attainment are more likely to leave the state and not be collected in subsequ ent Michigan birth records. The mean poverty status for non -movers was higher than expected and increased with increased parity in our analytic subsamples. Non -movers may therefore exper ience more poverty change than we anticipated and may also be a viable population of interest in examining residual confounding of neighborhood effects using a continuous poverty measure. Finally we report that non -movers experience a statistically signifi cant (X 2 test for association p -value <.0001) higher proportion of large -for-gestational age (>90% percentile birthweight) compared to movers (Table 1.1). This was an unexpected finding and we could find no previous reports of this association. This may b e due a more proximal mechanism (obesity promoting neighborhood) on the causal pathway. Future investigations may be useful in examining this outcome stratified by mobility status. Limitations Our study is one of the first to report on the mobility patterns of women between pregnancies over a partial life -course. While our study provides new information it is important 27 to highlight some of the study limitations. The Michigan singleton birth file is missing 18% of the geocoded census tract data not at random which could bias our results. The data are more missing for earlier years of the study period when geocoding was not as systematically performed at the time of birth. Future studies using this data should conduct a sensitivity analysis of the data using more reliable later years of data (2000 -2012) to examine meaningful change in results. However, birth outcomes were not systematically different among the missing and not -missing geocoded data. The measurement of the poverty variable at the time of birth makes us unable to measure the timing of the mobility between births or to have uniform duration of exposure. We therefore must interpret our results as the effect of index birth outcome on mobil ity patterns at the time of the successive birth. There may be shorter inter -pregnancy mobility patterns that we are not able to capture with our study design. However, we did evaluate the duration of each interval (measured as the time between the index b irth and the successive birth in days). The mean time spent in census tract between movers (1,409 days) and non -movers (992 days) showed movers had longer duration of time in census tract. However, this is measurement strictly captures time between births and does not fully detail the duration of time in the census tract irrespective of birth. Our study captures a relatively short period of time for each woman. There may be some errors in the linkage criterion performed by Michigan Department of Health and Human Services although we did find evidence of successful linkage u sing the tabulation proposed by Adams and Kirby (Adams and Kirby 2007) (Appendix A). We were unable to include any information on housing tenure, renting or home ownership status, as it was not included in our data. Previous studies on the effects of population 28 mobility and health control for the home ownership or renting status of individ uals as they are likely to influence mobility. This may be particularly useful information as it related to the role of public housing and in relation to one particular risk factor, smoking. In our data we do observe a difference in smoking by residential mobility status with movers having higher risk. In summary, our study found that a large percentage of women move between pregnancies and they are demographically distinct compared to non -movers. Therefore, studies of neighborhood effects that rely on a si ngle measurement may be capturing a weighted average of non-movers and movers. Future studies, should incorporate multiple time points and residential mobility status to remove the potential for misclassification of neighborhood exposure. 29 Figure 1 Analyt ic Sample Flow Chart for Inclusion Michigan Births Excluded Total Michigan Births 1989 -2012 N=3,137,483 Singleton births to women 15 -44 years of age n= 3,027,162 Analytic sample for RE analysis n=2,275,735 Available, plausible data on birth weight and gestational age n =2,826,100 -non-singleton births (n=101,256) -births to women <15, >44 years of age (n=9,065) - misclassified singletons, same or birthdate <120 days apart (n= 170) -missing geocoded census tract data (n=505,625; 18.17 % ) -missing c ensus tract poverty information (n=1,810; 0.08%) -missing gestational age (n=6,804) -missing birth weight (n=2,409) -gestational age <22 or >44 weeks or implausible combination of GA and BW (n=191,849) -missing data on maternal age, parity, marital sta tus, education, infant sex, diabetes, or nativity (n=42,760) All maternal covariates complete for all births n=2,783,340 Analytic sample >1 birth and meets all inclusion criteria n=1,555,075 - <2 births exclusions (n=720,660) Parity Restricted Birth samples (all births recorded in Michigan) Births 1 to 2 Births 2 to 3 n=895,214 n=355,912 Analytic sample for FE analysis n=292,152 30 Table 1.1 Maternal characteristics at index birth stratified by residential status (mover vs. non -mover) at successive births restricted by parity, Births 1 to 2 and Births 2 to 3, singleton births recorded in Michigan 1990 -2010a Births 1 to 2 Births 2 to 3 Overall Movers Non -Movers Overall Movers Non -Movers n=447,607 n=224,602 n= 223,005 n=177,956 n =90,909 n= 87,047 Maternal Characteristics at first birth n n % n % n n % n % Race NHW 333,946 154,388 80.5 179,558 68.7 125,983 58,182 64.0 67,801 77.9 NHB 66,762 44,055 10.2 22,707 19.6 32,358 21,921 24.1 10,437 12.0 Hispanic 22,310 12,845 4.2 9,465 5.7 10,341 5,750 6.3 4,591 5.3 NH AI/AN 673 359 0.1 314 0.2 281 147 0.2 134 0.2 NH A/PI 16,913 8,802 3.6 8,111 3.9 6,389 3,420 3.8 2,969 3.4 Non -Hispanic Mixed race / other/ missing 7,003 4,153 1.3 2,850 1.9 2,604 1,489 1.6 1,115 1.3 Age <20 100,578 69,382 30.9 31,196 14.0 18,511 13,148 14.5 5,363 6.2 20-24 129,016 76,995 34.3 52,021 23.3 62,501 39,025 42.9 23,476 27.0 25-29 134,828 52,826 23.5 82,002 36.8 57,284 25,636 28.2 31,648 36.4 30-34 68,264 21,504 9.6 46,760 21.0 33,630 11,481 12.6 22,149 25.4 35-40 14,226 3,747 1.7 10,479 4.7 5,782 1,565 1.7 4,217 4.8 >40 695 148 0.1 547 0.3 248 54 0.1 194 0.2 Nativity US born 414,358 209,031 93.1 205,327 92.1 165,513 85,493 94.0 80,020 91.9 Foreign born 33,249 15,571 6.9 17,678 7.9 12,443 5,416 6.0 7,027 8.1 Insurance Payer Private 297,402 126,588 56.4 170,814 76.6 109,119 47,102 51.8 62,017 71.3 Medicaid 142,457 93,998 41.9 48,459 21.7 65,428 42,122 46.3 23,306 26.8 Self -pay 2,977 1,483 0.7 1,494 0.7 1,545 748 0.8 797 0.9 Other 430 247 0.1 183 0.1 195 112 0.1 83 0.1 31 Table 1.1 (cont™d) Unknown 4,341 2,286 1.0 2,055 0.9 1,669 825 0.9 844 1.0 Infant sex -male 230,113 115,303 51.3* 114,810 51.5* 91,176 46,559 51.2* 44,617 51.3* Marital Status Married 273,956 105,036 46.8 168,920 75.8 115,261 48454 53.3 66807 76.7 Unmarried 173,651 119,566 53.2 54,085 24.3 62,695 42455 46.7 20240 23.3 Education Less than HS 11,952 6,027 2.7 5,925 2.7 5,118 2,609 2.9* 2,509 2.9* Some HS 66,423 33,654 15.0 32,769 14.7 29,008 14,949 16.4* 14,059 16.2* High School (HS) 137,181 69,881 31.1 67,300 30.2 59,422 30,397 33.4* 29,025 33.3* Some college 105,353 53,547 23.8 51,806 23.2 42,870 21,958 24.2* 20,912 24.0* College graduate 68,374 33,966 15.1 34,408 15.4 26,787 13,573 14.9* 13,214 15.2* Greater than college 36,285 18,260 8.1 18,025 8.1 13,439 6,803 7.5* 6,636 7.6* Smoke No 386,651 187,271 83.4 199,380 89.4 148,943 72,721 80.0 76,222 87.6 Yes 56,497 35,147 15.7 21,350 9.6 27,671 17,513 19.3 10,158 11.7 Successive Birth outcomes PTB No 408,650 204,051 90.9 206,873 92.8 163,775 82,838 91.1 80,937 93.0 Yes 38,957 20,551 9.2 16,132 7.2 14,181 8,071 8.9 6,110 7.0 SGA No 394,039 195,250 86.9 198,789 89.1 162,009 81,497 89.7 80,512 92.5 Yes 53,568 29,352 13.1 24,216 10.9 15,947 9,412 10.4 6,535 7.5 LGA No 415,971 209,728 93.4 206,243 92.5 160,773 83,009 91.3 77,764 89.3 Yes 31,636 14,874 6.6 16,762 7.5 17,183 7,900 8.7 9,283 10.7 Mean (sd) Mean (sd) Mean (sd) Mean (sd) Mean (sd) Mean (sd) 32 Table 1.1 (cont™d) Mean neighborhood poverty rate 0.13 (0.12) 0.15 (0.13) 0.11 (0.11) 0.15 (0.13) 0.16 (0.14) 0.12 (0.11) Mean census tract time in days 1201.8 (803.6) 1410.0 (935.1) 992.1 (572.5) 1240.6 (819.8) 1431.1 935.7 1041.7 618.1 Mean Distance moved in miles -- 15.9 (36.4) -- -- -- 14.5 35.1 -- -- Inter -county Movers -- 56,801 26.0 -- -- -- 21,797 24.0 aThe analytic sample was restricted to births recorded in Michigan and presented here grouped into first and second birth, and second and third birth. *Not statistically significant All other variables significant at p<0.001 using Chi -Square tests 33 Table 1.2a Associations between maternal characteristics at index birth and residential mobility at successive births restricted to first and second singleton births recorded in Michigan 1990 -2010 Births 1 to 2 Non -Movers n=223,00 5 Movers n=224,602 Unadjusted RR a (95% CI) Adjusted RR a,b (95% CI) Maternal Characteristics at first birth % % RR 95% CI RR 95% CI Race Non -Hispanic White 80.5 68.7 1.0 1.0 Non -Hispanic Black 10.2 19.6 1.4 1.4 1.4 1.2 1.2 1.2 Hispanic 4.2 5.7 1.3 1.2 1.3 1.1 1.1 1.1 Non -Hispanic American Indian/ Alaska Native 0.1 0.2 1.2 1.1 1.2 1.0 0.9 1.0 Non -Hispanic Asian/ Pacific Islander 3.6 3.9 1.1 1.1 1.1 1.2 1.1 1.2 Non -Hispanic Mixed race / other/ missing 1.3 1.9 1.3 1.3 1.3 1.2 1.1 1.2 Age <20 30.9 14.0 0.9 0.9 1.0 0.9 0.8 0.9 20-24 34.3 23.3 1.0 1.0 25-29 23.5 36.8 0.8 0.8 0.8 0.9 0.9 0.9 30-34 9.6 21.0 0.6 0.6 0.6 0.8 0.8 0.8 35-40 1.7 4.7 0.6 0.6 0.6 0.8 0.8 0.8 >40 0.1 0.3 0.6 0.6 0.6 0.8 0.7 0.8 Nativity US born 93.1 92.1 1.0 1.0 Foreign born 6.9 7.9 0.9 0.9 0.9 1.0 0.9 1.0 Insurance Payer Private 56.4 76.6 1.0 1.0 Medicaid 41.9 21.7 1.5 1.5 1.5 1.2 1.2 1.2 Self -pay 0.7 0.7 1.2 1.1 1.2 1.1 1.0 1.1 Other 0.1 0.1 1.4 1.4 1.5 1.2 1.1 1.3 Unknown 1.0 0.9 1.1 1.0 1.1 0.9 0.9 1.0 Infant sex -male 51.3 51.5 1.0 1.0 1.0 1.0 1.0 1.0 Marital Status Married 46.8 75.8 1.0 1.0 Unmarried 53.2 24.3 1.6 1.6 1.6 1.2 1.2 1.2 Education Less than HS 2.7 2.7 1.0 1.0 1.0 1.0 1.0 1.0 Some HS 15.0 14.7 1.0 1.0 1.0 1.0 1.0 1.0 High School (HS) 31.1 30.2 1.0 1.0 1.0 1.0 1.0 1.0 Some college 23.8 23.2 1.0 1.0 1.0 1.0 1.0 1.0 34 Table 1.2a (cont™d) College graduate 15.1 15.4 1.0 1.0 1.0 1.0 1.0 1.0 Greater than college 8.1 8.1 1.0 1.0 Smoke No 83.4 89.4 1.0 1.0 Yes 15.7 9.6 1.3 1.3 1.3 1.1 1.1 1.2 Hypertension No 90.9 1.0 1.0 Yes 9.2 1.0 1.0 1.0 1.1 1.0 1.1 a RR calculated using modified Poisson regression with robust error variances (Zhao, K.) b Adjusted for all other covariates listed. 35 Table 1.2b Associations between maternal characteristics at index birth and residential mobility at successive births restricted to second and third singleton births recorded in Michigan 1990 -2010 Births 2 to 3 Non -Movers n= 87,047 Movers n= 90,909 Unadjusted RRa (95% CI) Adjusted RR a,b (95% CI) Maternal Characteristics at first birth % % RR 95% CI RR 95% CI Race Non -Hispanic White 77.9 64 1.0 1.0 Non -Hispanic Black 12.0 24.11 1.5 1.5 1.5 1.2 1.2 1.2 Hispanic 5.3 6.33 1.2 1.2 1.2 1.1 1.1 1.1 Non -Hispanic American Indian/ Alaska Native 0.2 0.16 1.1 1.0 1.1 1.0 0.9 1.1 Non -Hispanic Asian/ Pacific Islander 3.4 3.76 1.2 1.1 1.2 1.1 1.1 1.2 Non -Hispanic Mixed race / other/ missing 1.3 1.64 1.2 1.2 1.2 1.1 1.1 1.2 Age <20 6.2 14.5 1.0 1.0 1.1 0.9 0.9 1.0 20-24 27.0 42.9 1.0 25-29 36.4 28.2 0.9 0.9 0.9 1.0 1.0 1.0 30-34 25.4 12.6 0.7 0.7 0.7 0.9 0.9 0.9 35-40 4.8 1.7 0.7 0.7 0.7 0.9 0.9 0.9 >40 0.2 0.1 0.7 0.7 0.7 1.0 0.9 1.0 Nativity US born 91.9 94.0 1.0 1.0 Foreign born 8.1 6.0 0.8 0.8 0.9 0.9 0.9 0.9 Insurance Payer Private 71.3 51.8 1.0 1.0 Medicaid 26.8 46.3 1.4 1.4 1.5 1.2 1.2 1.2 Self -pay 0.9 0.8 1.1 1.0 1.1 1.0 1.0 1.1 Other 0.1 0.1 1.4 1.3 1.5 1.2 1.1 1.3 Unknown 1.0 0.9 1.0 1.0 1.1 1.0 0.9 1.0 Infant sex -male 51.3 51.2 1.0 1.0 1.0 1.0 1.0 1.0 Marital Status Married 76.75 53.3 1.0 1.0 Unmarried 23.25 46.7 1.6 1.6 1.6 1.3 1.3 1.3 Education Less than HS 2.9 2.9 1.0 1.0 1.0 1.0 1.0 1.0 Some HS 16.2 16.4 1.0 1.0 1.0 1.0 1.0 1.0 High School (HS) 33.3 33.4 1.0 1.0 1.0 1.0 1.0 1.0 Some college 24.0 24.2 1.0 1.0 1.0 1.0 1.0 1.0 College graduate 15.2 14.9 1.0 1.0 1.0 1.0 1.0 1.0 Greater than college 7.6 7.5 1.0 1.0 36 Table 1.2b (cont™d) Smoke No 87.6 80.0 1.0 1.0 Yes 11.7 19.3 1.3 1.3 1.4 1.2 1.1 1.2 Hypertension No 1.0 Yes 1.0 1.0 1.1 1.0 1.0 1.1 aRR calculated using modified Poisson regression with robust error variances (Zhao, K.) bAdjusted for all other covariates listed. 37 a ORs derived using logistic regression models adjusted for all other covariates in the model Table 1.3 Estimates from adjusted logistic model of association between prior PTB and residential change for mothers at successive birth a Birth 2 Birth 3 OR 95% CI OR 95% CI PTB At Index Birth 1.0 1.0 1.0 1.1 1.0 1.1 Age <20 0.8 0.8 0.8 1.0 0.9 1.0 20-24 (ref) 1.0 1.0 25-29 0.7 0.7 0.7 0.8 0.8 0.9 30-34 0.5 0.5 0.6 0.7 0.6 0.7 35-40 0.5 0.5 0.5 0.6 0.6 0.7 >40 0.5 0.5 0.5 0.7 0.6 0.7 Race/Ethnicity NHW 1.0 1.0 NHB 2.0 2.0 2.0 2.2 2.1 2.3 Hispanic 1.4 1.4 1.5 1.4 1.3 1.4 NH Native American 1.0 0.9 1.2 0.9 0.7 1.2 NH Asian/Pacific Is. 1.4 1.4 1.4 1.3 1.3 1.4 Mixed/Other/Missing 1.5 1.5 1.6 1.4 1.3 1.5 Marital Change from Index Birth No change (ref) 1.0 1.0 Divorce 2.5 2.4 2.6 3.3 3.1 3.6 Married 2.8 2.7 2.8 1.9 1.9 2.0 Smoke No 1.0 1.0 Yes 1.6 1.6 1.7 1.7 1.7 1.7 Unknown 1.0 0.9 1.1 1.1 1.0 1.3 38 APPENDIX 39 Table A.1 Maternal characteristics at index birth stratified by geocoded status among singleton births in Michigan Birth File 1990 -2012 Geocode Complete Geocode not complete a n= 2,277,545 n= 505,625 Maternal Characteristics at first birth n % n % Race Non -Hispanic White 1,643,120 72.1 408,827 80.9 Non -Hispanic black 432,728 19.0 63,462 12.6 Hispanic 103,006 4.5 17,323 3.4 Non -Hispanic American Indian/ Alaska Native 10,343 0.5 4,373 0.9 Non -Hispanic Asian/Pacific Islander 60,698 2.7 10,572 2.1 Non -Hispanic Mixed race / other/ missing 27,650 1.2 1,068 0.2 Age <20 239,525 10.6 59,988 11.9 20-24 547,415 24.2 139,347 27.7 25-29 670,769 29.7 155,810 31.0 30-34 548,042 24.3 106,687 21.2 35-40 229,105 10.1 37,927 7.5 >40 24,966 1.1 3,269 0.7 Nativity US born 2,073,456 91.0 485,620 96.0 Foreign born 204,089 9.0 20,005 4.0 Insurance Payer Private 1,427,744 62.7 321,967 63.7 Medicaid 802,473 35.2 165,960 32.8 Self -pay 22,589 1.0 9,313 1.8 Other 3,837 0.2 567 0.1 Missing 20,902 0.9 7,818 1.6 Infant sex -male 1,166,988 51.2 259,474 51.3 Marital Status Married 1469135 64.5 369,931 73.2 Unmarried 808410 35.5 135,694 26.8 Education Less than HS 59,854 2.6 11,350 2.2 Some HS 329,160 14.5 76,480 15.1 High School (HS) 733,860 32.2 197,159 39.0 Some college 593,514 26.1 124,462 24.6 College graduate 354,419 15.6 62,515 12.4 Greater than college 206,738 9.1 33,659 6.7 Parity Nulliparous (0 live births) 908,095 39.9 210,833 42 Primiparous (1 previous live birth) 733,341 32.2 165,697 33 40 Table A.1 (cont™d) Multiparous ( >1 previous live births) 636,109 27.9 129,095 26 Birth outcomes PTB No 2,065,969 90.7 460,844 91.1 Yes 211,576 9.3 44,781 8.9 SGA No 2,029,387 89 455,150 90.0 Yes 248,158 11 50,475 10.0 LGA No 2,071,814 91.0 454,698 89.9 Yes 205,731 9.0 50,927 10.1 a census tract missing, not Michigan, or incomplete 41 Table A.2 Geocoded missing status by infant year of birth in singleton Michigan births 1989-2012 Geocoded Missing Geocode Infant Birth Year n % n % 1989 - 0 131,357 100 1990 96,169 70.2 40,782 29.8 1991 93,784 70.4 39,481 29.6 1992 90,833 70.6 37,768 29.4 1993 88,145 70.5 36,901 29.5 1994 83,220 68.8 37,819 31.3 1995 102,232 87.0 15,293 13.0 1996 78,613 67.8 37,331 32.2 1997 77,323 67.1 37,995 33.0 1998 81,727 70.8 33,666 29.2 1999 80,437 69.8 34,740 30.2 2000 114,292 97.8 2,565 2.2 2001 113,455 97.9 2,467 2.1 2002 111,098 98.1 2,153 1.9 2003 111,276 98.0 2,275 2.0 2004 109,054 98.0 2,175 2.0 2005 107,908 97.9 2,266 2.1 2006 109,461 98.3 1,941 1.7 2007 108,327 98.2 1,964 1.8 2008 109,860 99.0 1,085 1.0 2009 102,232 99.2 842 0.8 2010 100,068 99.3 745 0.7 2011 104,089 99.0 1,059 1.0 2012 103,942 99.1 955 0.9 Total 2,277,545 81.8 505,625 18.2 42 Table A.3 Paternal characteristics overall and stratified by maternal PTB outcomes among singleton Michigan Births 1990 -2012 Overall PTB Paternal Characteristics No Yes n % n % Race Non -Hispanic White 1,452,507 1,345,472 92.6 107,035 7.4 Non -Hispanic black 224,592 195,235 86.9 29,357 13.1 Hispanic 90,021 82,350 91.5 7,671 8.5 Non -Hispanic American Indian/ Alaska Native 9,196 8,413 91.5 783 8.5 Non -Hispanic Asian/Pacific Islander 53,407 49,476 92.6 3,931 7.4 Non -Hispanic Mixed race / other/ missing 446,012 383,356 86.0 62,656 14.1 Age Missing 34,776 31,784 91.4 2,992 8.6 <20 57,065 50,506 88.5 6,559 11.5 20-24 276,403 250,963 90.8 25,440 9.2 25-29 513,948 473,898 92.2 40,050 7.8 30-34 561,571 519,529 92.5 42,042 7.5 35-40 314,991 289,401 91.9 25,590 8.1 >40 516,981 448,221 86.7 68,760 13.3 43 Table A.4 Maternal characteristics and poverty exposure by ordering of preterm birth outcomes across two pregnancies, births 1 to 2, and births 2 to 3 among singleton births in Michigan 1990 -2010 PTB Outcomes Births 1 to 2 Births 2 to 3 Maternal Characteristics at first birth Term - Term Preterm - Term Term - Preterm Preterm - Preterm Term - Term Preterm - Term Term - Preterm Preterm - Preterm n n n n n n n n Total 762,954 62,494 54,346 15,420 151,747 11,102 12,028 3,079 Race NHW 290,891 21,012 17,383 4,660 110,856 6,443 7,177 1,507 NHB 50,920 6,842 6,739 2,261 24,388 3,357 3,368 1,245 Hispanic 18,870 1,646 1,427 367 8,702 667 809 163 NH AI/AN 576 47 44 6 237 25 18 1 NH A/PI 14,344 1,210 1,072 287 5,380 440 455 114 Non -Hispanic Mixed race / other/ missing 5,876 490 508 129 2,184 170 201 49 Age <20 80,733 9,135 8,344 2,366 14,163 2,033 1,708 607 20-24 110,209 8,621 8,085 2,101 52,081 4,404 4,751 1,265 25-29 118,716 7,935 6,366 1,811 50,389 2,754 3,421 720 30-34 59,264 4,396 3,467 1,137 29,837 1,578 1,809 406 35-40 11,992 1,099 856 279 5,062 319 321 80 >40 563 61 55 16 215 14 18 1 Nativity US born 352,561 29,198 25,312 7,287 140,893 10,386 11,286 2,948 Foreign born 28,916 2,049 1,861 423 10,854 716 742 131 Insurance Payer Private 257,753 19,204 15,868 4,577 95,531 5,853 6,254 1,481 Medicaid 117,168 11,443 10,861 2,985 53,294 5,029 5,577 1,528 Self -pay 2,499 252 159 67 1,327 106 86 26 Other 362 34 25 9 167 11 13 4 Unknown 3,695 314 260 72 1,428 103 98 40 Infant sex -male 194,878 16,940 14,095 4,200 77,436 5,997 6,107 1,636 44 Table A.4 (cont™d) Marital Status Married 240,252 16,492 13,395 3,817 102,062 5,646 6,230 1,323 Unmarried 141,225 14,755 13,778 3,893 49,685 5,456 5,798 1,756 Education Less than HS 10,119 887 730 216 4,361 323 340 94 Some HS 56,578 4,730 3,950 1,165 24,689 1,850 1,964 505 High School (HS) 116,695 9,623 8,491 2,372 50,627 3,771 4,008 1,016 Some college 89,970 7,370 6,181 1,832 36,522 2,656 2,951 741 College graduate 58,352 4,759 4,125 1,138 22,937 1,622 1,759 469 Greater than college 31,006 2,538 2,123 618 11,553 775 898 213 Smoke No 330,665 26,746 22,756 6,484 128,048 8,929 9,546 2,420 Yes 47,059 4,133 4,162 1,143 22,578 2,085 2,376 632 Birth outcomes SGA no 338,127 27,471 21,916 6,525 138,991 10,237 10,076 2,705 yes 43,350 3,776 5,257 1,185 12,756 865 1,952 374 LGA no 354,003 28,760 25,823 7,385 136,866 9,758 11,261 2,888 yes 27,474 2,487 1,350 325 14,881 1,344 767 191 Mean neighborhood poverty rate 0.13 (0.12) 0.15 (0.13) 0.16 (0.14) 0.16 (0.14) 0.14 (0.13) 0.18 (0.14) 0.18 (0.14) 0.20 (0.15) 45 Table A.5 Residential mobility status a (mover, non -mover) and percent neighborhood poverty by ordering of preterm birth outcomes across two pregnancies, births 1 to 2, and births 2 to 3 among singleton births in Michigan 1990 -2010 PTB Outcomes Births 1 to 2 Births 2 to 3 Term - Term Preterm - Term Term - Preterm Preterm - Preterm Term - Term Preterm - Term Term - Preterm Preterm - Preterm Movers 189,375 16,476 14,676 4,075 189,375 16,476 14,676 4,075 Mean poverty rate birth 1 0.143 (0.124) 0.166 (0.136) 0.174 (0.138) 0.182 (0.142) 0.143 (0.124) 0.166 (0.136) 0.174 (0.138) 0.182 (0.142) Mean poverty rate birth 2 0.143 (0.126) 0.167 (0.136) 0.175 (0.141) 0.180 (0.146) 0.143 (0.126) 0.167 (0.136) 0.175 (0.141) 0.180 (0.146) Mean poverty rate change 0.000 (0.123) 0.001 (0.135) 0.001 (0.137) -0.001 (0.139) 0.000 (0.123) 0.001 (0.135) 0.001 0.137) -0.001 (0.139) Non -Movers 192,102 14,771 12,497 3,635 192,102 189,375 14,771 3,635 Mean poverty rate Birth 1 0.110 (0.103) 0.127 (0.117) 0.140 (0.127) 0.139 (0.127) 0.110 (0.103) 0.127 (0.117) 0.140 (0.127) 0.139 (0.127) Mean poverty rate birth 2 0.117 (0.107) 0.133 (0.120) 0.147 (0.129) 0.145 (0.130) 0.117 (0.107) 0.133 (0.120) 0.147 (0.129) 0.145 (0.130) Mean poverty rate change 0.007 (0.030) 0.006 (0.033) 0.006 0.033 0.006 (0.032) 0.007 (0.030) 0.006 (0.033) 0.006 (0.033) 0.006 (0.032) aDetermined from geocoded residential address census tract change between births 46 Table A.6 Table poverty variable derivation and missing poverty values NCDB Ncdb.sas7bdat Variable Year Decisions Missing code N Missing Total Census Tracts Percent Missing POVRAT9 1990 -- 0 65 2813 2.0% POVRATA 2000 -- 0 62 2813 2.0% POVRAT1A 2010 -- -999 65 2813 2.0% Total -- 200 8439 2.0% NCDB recode ncdbrecode.sas7bdat POV90 1990 if povrat9= 0 and ninetiesD= 0 then pov90= .; else pov90=povrat9; if povrat9= 0 and povrata> 0 then pov90=povrata . 51 2813 1.8% POV00 2000 if povrata= 0 and aughtsD= 0 then pov00= .; else if povrata= 0 and povrat9> 0 then pov00=povrat9; else pov00=povrata; . 53 2813 1.8% POV10 2010 if povrat1a= -999 and povrata<= 0 then pov10= .; else if povrat1a= -999 and povrata> 0 then pov10=povrata; else pov10=povrat1a; . 52 2813 1.8% Total 8439 1.8% ACS 2012 acs2012.sas7bdat pov 2012 -- . 68 2813 2.4% 47 Table A. 6 (cont™d) Linear Interpolation interpolatedpov.sas7bdat pov 1990-2012 Proc expand data=work.addyear2 out=work.year from=year10 to=year method=join; by geo2010 notsorted; id yearD; run ; proc expand data=pov2 out=pova method=join; by geo2010; ID BXYEAR; run ; . 1,254 64,699 1.9% Full eligible dataset NPfinal.sas7dbat pov proc sql ; create table poverty as select L.*, R. pov from normalizedcensus L left join interpolatedpov R on l.nbxyear=r.bx and l.long10=r.geo; quit ; . 1,810 Birth records missing poverty 2,277,545 total Birth Records 0.08% Final Eligible Births NPmibirths.sas7bdat 2275735 pov 1990-2012 >1 birth after poverty missing and all other inclusion criteria 0 0% 48 REFERENCES 49 REFERENCES Adams, Melissa M., and Russell S. Kirby. 2007. fiMeasuring the Accuracy and Completeness of Linking Certificates for Deliveries to the Same Woman.fl Paediatric and Perinatal Epi demiology 21 Suppl 1 (July): 58 Œ62. https://doi.org/10.1111/j.1365 -3016.2007.00838.x. Arcaya, Mariana, Mark Brewster, Corwin M. Zigler, and S. V. Subramanian. 2012. fiArea Variations in Health: A Spatial Multilevel Modeling Approach.fl Health & Place 18 (4): 824 Œ831. Bell, Michelle L., and Kathleen Belanger. 2012. fiReview of Research o n Residential Mobility during Pregnancy: Consequences for Assessment of Prenatal Environmental Exposures.fl Journal of Exposure Science and Environmental Epidemiology 22 (5): 429 Œ38. https://doi.org/10.1038/jes.2012.42. Blencowe, Hannah, Simon Cousens, Dori s Chou, Mikkel Oestergaard, Lale Say, Ann -Beth Moller, Mary Kinney, Joy Lawn, and the Born Too Soon Preterm Birth Action Group (see acknowledgement for full list). 2013. fiBorn Too Soon: The Global Epidemiology of 15 Million Preterm Births.fl Reproductive He alth 10 (1): S2. https://doi.org/10.1186/1742 -4755-10-S1-S2. Bruckner, Tim A., Jennifer B. Kane, and Samantha Gailey. 2019. fiStrong Upward Neighborhood Mobility and Preterm Birth: A Matched -Sibling Design Approach.fl Annals of Epidemiology 36 (August): 48 -54.e1. https://doi.org/10.1016/j.annepidem.2019.05.005. Canfield, Mark A., Tunu A. Ramadhani, Peter H. Langlois, and D. Kim Waller. 2006. fiResidential Mobility Patterns and Exposure Misclassification in Epidemiologic Studies of Birth Defects.fl Journal of Ex posure Science & Environmental Epidemiology 16 (6): 538 Œ43. https://doi.org/10.1038/sj.jes.7500501. Chen, Lei, Erin M. Bell, Alissa R. Caton, Charlotte M. Druschel, and Shao Lin. 2010. fiResidential Mobility during Pregnancy and the Potential for Ambient Ai r Pollution Exposure Misclassification.fl Environmental Research 110 (2): 162 Œ68. https://doi.org/10.1016/j.envres.2009.11.001. Chetty, Raj, Nathaniel Hendren, and Lawrence F. Katz. 2016. fiThe Effects of Exposure to Better Neighborhoods on Children: New Evi dence from the Moving to Opportunity Experiment.fl The American Economic Review 106 (4): 855 Œ902. Clark, William A V, and Youqin Huang. 2003. fiThe Life Course and Residential Mobility in British Housing Markets.fl Environment and Planning A: Economy and Spac e 35 (2): 323 Œ39. https://doi.org/10.1068/a3542. 50 Fell, Deshayne B., Linda Dodds, and Will D. King. 2004. fiResidential Mobility during Pregnancy.fl Paediatric and Perinatal Epidemiology 18 (6): 408 Œ14. https://doi.org/10.1111/j.1365 -3016.2004.00580.x. Janevi c, T, CR Stein, DA Savitz, JS Kaufman, SM Mason, and AH Herring. 2010. fiNeighborhood Deprivation and Adverse Birth Outcomes among Diverse Ethnic Groups.fl Annals of Epidemiology 20 (6): 445 Œ51. https://doi.org/10.1016/j.annepidem.2010.02.010. Khoury, M. J., W. Stewart, A. Weinstein, S. Panny, P. Lindsay, and M. Eisenberg. 1988. fiResidential Mobility during Pregnancy: Implications for Environmental Teratogenesis.fl Journal of Clinical Epidemiology 41 (1): 15 Œ20. Kim, Jae -Hong, Francesca Pagliara, and John Pres ton. 2003. fiAn Analysis of Residential Location Choice Behaviour in Oxfordshire, Uk: A Combined Stated Preference Approach.fl International Review of Public Administration 8 (1): 103 Œ14. https://doi.org/10.1080/12294659.2003.10805021. Kulu, Hill, and Elizab eth Washbrook. 2014. fiResidential Context, Migration and Fertility in a Modern Urban Society.fl Advances in Life Course Research 21 (September): 168 Œ82. https://doi.org/10.1016/j.alcr.2014.01.001. Lee, Brian H. Y., and Paul Waddell. 2010. fiResidential Mobil ity and Location Choice: A Nested Logit Model with Sampling of Alternatives.fl Transportation 37 (4): 587 Œ601. https://doi.org/10.1007/s11116 -010-9270-4. Logan, John R., Zengwang Xu, and Brian J. Stults. 2014. fiInterpolating U.S. Decennial Census Tract Data from as Early as 1970 to 2010: A Longitudinal Tract Database.fl The Professional Geographer 66 (3): 412 Œ20. https://doi.org/10.1080/00330124.2014.905156. Lupo, Philip J., Elaine Symanski, Wenyaw Chan, Laura E. Mitchell, D. Kim Waller, Mark A. Canfield, and Peter H. Langlois. 2010. fiDifferences in Exposure Assignment between Conception and Delivery: The Impact of Maternal Mobility.fl Paediatric and Perinatal Epidemiology 24 (2): 200 Œ208. https://doi.org/10.1111/j.1365 -3016.2010.01096.x. Margerison -Zilko, Clai re, Catherine Cubbin, Jina Jun, Kristen Marchi, and Paula Braveman. 2016. fiPost -Partum Residential Mobility Among a Statewide Representative Sample of California Women, 2003 -2007.fl Maternal and Child Health Journal 20 (1): 139 Œ48. https://doi.org/10.1007/s 10995-015-1812-0. Messer, Lynne C., Lisa C. Vinikoor, Barbara A. Laraia, Jay S. Kaufman, Janet Eyster, Claudia Holzman, Jennifer Culhane, Irma Elo, Jessica G. Burke, and Patricia O™Campo. 2008. fiSocioeconomic Domains and Associations with Preterm Birth.fl Social Science & Medicine 67 (8): 1247 Œ1257. 51 Miller, Assia, Csaba Siffel, and Adolfo Correa. 2010. fiResidential Mobility during Pregnancy: Patterns and Correlates.fl Maternal and Child Health Journal 14 (4): 625 Œ34. https://doi.org/10.1007/s10995 -009-0492-z. Morris, Tim, David Manley, and Clive E. Sabel. 2018. fiResidential Mobility: Towards Progress in Mobility Health Research.fl Progress in Human Geography 42 (1): 112 Œ33. https://doi.org/10.1177/03091 32516649454. O™Campo, Patricia, Jessica G. Burke, Jennifer Culhane, Irma T. Elo, Janet Eyster, Claudia Holzman, Lynne C. Messer, Jay S. Kaufman, and Barbara A. Laraia. 2008. fiNeighborhood Deprivation and Preterm Birth among Non -Hispanic Black and White Wom en in Eight Geographic Areas in the United States.fl American Journal of Epidemiology 167 (2): 155 Œ163. Saadeh, Frances B., Melissa A. Clark, Michelle L. Rogers, Crystal D. Linkletter, Maureen G. Phipps, James F. Padbury, and Patrick M. Vivier. 2013. fiPregn ant and Moving: Understanding Residential Mobility during Pregnancy and in the First Year of Life Using a Prospective Birth Cohort.fl Maternal and Child Health Journal 17 (2): 330 Œ43. https://doi.org/10.1007/s10995 -012-0978-y. Schempf, Ashley H., and Jay S. Kaufman. 2012. fiAccounting for Context in Studies of Health Inequalities: A Review and Comparison of Analytic Approaches.fl Annals of Epidemiology 22 (10): 683 Œ90. https://doi.org/10.1016/j.annepidem.2012.06.105. Sundquist, Jan, Kristina Sundquist, Sven -Erik Johansson, Xinjun Li, and Marilyn Winkleby. 2011. fiMothers, Places and Small for Gestational Age Births: A Cohort Study.fl Archives of Disease in Childhood 96 (4): 380 Œ85. https://doi.org/10.1136/adc.2009.180042. Tang, Zhongfeng, Hanru Zhang, Haiya Bai, Ya Chen, Nan Zhao, Min Zhou, Hongmei Cui, et al. 2018. fiResidential Mobility during Pregnancy in Urban Gansu, China.fl Health & Place 53 (September): 258 Œ63. https://doi.org/10.1016/j.healthplace.2018.08.021. Urayama, Kevin Y., Julie Von Behren, Peggy Reyno lds, Andrew Hertz, Monique Does, and Patricia A. Buffler. 2009. fiFactors Associated with Residential Mobility in Children with Leukemia: Implications for Assigning Exposures.fl Annals of Epidemiology 19 (11): 834 Œ40. https://doi.org/10.1016/j.annepidem.2009 .03.001. Zhao, Kechen. n.d. fiProper Estimation of Relative Risk Using PROC GENMOD in Population Studies,fl 14. 52 STUDY 2 Introduction Neighborhood poverty at the time of delivery is considered a risk factor for adverse birth outcomes (David and Messer 2011; Croteau, Marcoux, and Brisson 2007; Braveman et al. 2010; Meyer, Warren, and Reisine 2007) . Neighborhood poverty may also be increasing over time. Using U.S. census data, one study reported that in the U.S. between 1980 and 2014, a combination of growing income segregation and muted economic performance have increased the risk for living in high -poverty neighborhoods (Iceland and Hernandez 2017) . Alarmingly, the extreme poverty has grown at a disproportionate rate meaning a greater percentage of people ar e experiencing the highest level s of neighborhood poverty (Jargowsky 1998) . The association between neighborhood poverty and birth outcomes have been extensively studied (Messer et al. 2006; O™Campo et al. 2008; Holzman et al. 2009a; Schempf et al. 2011; J. W. Collins, Rankin, and David 2015; Pearl et al. 2018; Bruckner, Kane, and Gailey 2019). The associations between neighborhood poverty trajectories, changes in level of poverty over time, have studied less . Primarily, this type of socioeconomic trajecto ry has been studied as an individual trajectory in relation to small for gestational age (Love et al. 2010; Osypuk et al. 2016; Slaughter -Acey et al. 2016; J. W. Collins, Mariani, and Rankin 2018) and low birth weight (Spencer 2004; Colen CG et al. 2006; Love et al. 2010; Osypuk et al. 2016) . Fewer studies measure neighborhood poverty trajectory and PTB (Love et al. 2010; J. J. Collins et al. 2007; Collins, Rankin, and David 2015; Kramer, Dunlop, and Hogue 2014) . One study (Collins, Rankin, and David 2011) used inter -generationally linked records of African American women comparing economic trajectory from childhood to adulthood in the association of PTB. They reported decreased risk for PTB among African American women who experienced neighborhood poverty in childhood with strong upward trajectory as adults (0.7 RR (95% CI:0.6, 53 0.8)) and modest upward trajectory (0.8 RR (95% CI: 0.7, 0.9) and weak upward trajectory (0.9 RR (95% CI: 0.8, 0.9)) compared to those who staye d in high neighborhood poverty . They report increased risk of PTB for white women who experience downward poverty trajectory , starting from low neighborhood poverty in childhood and experienced slight 1.2 (95% CI: 1.0, 4.0), moderate RR 1.6 (95% CI: 1.3, 1.9), and extreme RR 1.9 (95% CI 1.3, 2.6) downward poverty, respectively , compared to white women who continued in low neighborhood poverty (J. W. Collins, Rankin, and David 2015) . Another study that used longitudinally linked births in Georgia to evaluate cumulative neighborhood deprivation reported a modest association between neighborhood deprivation, using the Neighborhood Deprivation Index (NDI) and PTB (adjusted RD 0.95 (95% CI: 0.79 Š1.12) with strong effect modification by history of prior PT B and significant differences by race for black but not white women with increasing age (Kramer, Dunlop, and Hogue 2014) . Though limited, increasingly studies of neighborhood poverty based trajectory consistently show that large changes in neighborhood poverty, in either direction, are associated with PTB . Upward trajectory, moving away from poverty, has been associated with a modest reduction in preterm birth although this has not been consistent for all race/ethnicities and may be subject to effect mod ification by mother ™s own birth history (Collins, Rankin, and David 2011) . One potential explanation for the paucity of this type of investigation is the need for longitudinally da ta linked in order to evaluate changes measured over multiple time points . We previously reported on residential mobility in Michigan using a maternally -linked longitudinal dataset (Study 1) where we f ound that approximately half of women in the Michigan birth dat a move between pregnancies . As this type of investigation may be increasingly utilized as more longitudinal datasets become available with area -based geocoding, we were interested in investigating how residential mobility would modify the association between neighborhood 54 poverty traj ectory and PTB . In that sense, n eighborhood poverty trajectory occurred either as a result of moving to a neighborhood with a different level of poverty or as the neighborhood poverty conditions either deteriorate d or improve d. We aim ed to determine wheth er mother s experience d neighborhood poverty trajectory changes between successive pregnancies and if those changes are differen t for movers compared to non -movers . We aimed to address if poverty trajectory changes were driven more by the act of moving or b y the neighborhoods changing in order to evaluate how these changes related to preterm birth outcomes . We hypothesized that the poverty changes, measured by at least a quartile change in neighborhood poverty status across births, will be greater for movers than non-movers. We then aim ed to examine the association between neighborhood poverty trajectory and PTB. We hypothesize d consistent with previous finding there will be a modest association between upward trajectory and a reduction in PTB. We then further control led for individual -level characteristics associated with adverse birth outcomes and evaluate the role of poverty trajectory in these relationships. Methods Study Data and Population Birth data files were linked by the Michigan Department of Health and Human Services to create maternally -linked births files that include a mother and all of her suc cessive pregnancies in the state of Michigan during the study time period, 1990 -2012. Women who had a previous birth outside of Michigan were excluded from this analysis as we did not have access to inter -state birth records. We further restrict our analys is to at least 2 births to compare across birth time points. As a final restrict ion we limit birth records including up to the third birth to control for confounding of the association between neighborhood poverty and PTB by parity and as this captures the majority of our full eligible sample had three or less births. 55 We used a normalized dataset as the source for our poverty rate, the Neighborhood Change Database. This dataset allowed comparisons of census tract poverty rates across different census years by normalizing tract boundary changes to the 2010 decennial census tract boundaries. For our purposes, the NCDB allows comparison across 3 decennial census years in our study period: 1990, 2000, and 2010. Census tracts are designed to be stable statistic al groupings that reflect natural geographic boundaries and homogenous groups that encompass the fineighborhoodfl (Geography n.d.) However, they do evolve over time and may be split due to population growth or merged as a result of substantial population decline. In Michigan there were 2,552 census tracts recorded in the 1990 census, 2,717 census tracts in the 2000 census, and 2,813 census tracts recorded in the American Community Survey and 2010 census (Bureau n.d.; n.d.). We performed linear interpolation of census tract poverty rate for each infant birth year between decennial census periods and using American Community Survey data . We used a longitudinal database as a crosswalk reference between the 1990 to 2010 and 2000 to 2010 to allow linkage between the normalized poverty data and Michigan Birth files for census tracts that had been merged or split (Logan, Xu, and Stults 2014) . Measures Residential Mobility We define residential mobility as successive changes in census tract based on geocoding of maternal residence at live birth. Residential mobility is categorized as mover (changed census tract) or non -mover (did not change census tract) at the time of the successive live birth. Neighborhood Poverty Trajectory Previous studies have examined changes in neighborhood poverty rates across multiple time points as a determinant in birth outcomes. There is however no standard appro ach for how to measure these socioeconomic changes. The use of cut points to define socioeconomic changes vary from relative definitions such as quartiles 56 (Janevic et al. 2010; J. W. Collins, Rankin, and David 2015; Slaughter -Acey et al. 2016; Bruckner, Ka ne, and Gailey 2019) , absolute poverty levels (Margerison -Zilko et al. 2015) or dichotomous high/low (Pearl et al. 2018; Heinonen et al. 2013) . The measurement of socioeconomic change range from individual socioeconomic position (Osypuk et al. 2016; Slaugh ter -Acey et al. 2016) to occupational status (Heinonen et al. 2013) to neighborhood deprivation indices (O™Campo et al. 2008; Holzman et al. 2009b; Bruckner, Kane, and Gailey 2019), or neighborhood poverty rate (Margerison -Zilko et al. 2015) . The defined study periods for examining socioeconomic changes vary as well depending on the hypothesis being tested, but are conceptually within the individual (e.g., childhood to adulthood, birth 1 to birth 2) or intergenerational (e.g., grandmother to grandchild). F or this study, we seek to understand the relationship between socioeconomic change, measured as changes in neighborhood poverty rate that we term ‚neighborhood poverty trajectory™, and birth outcomes for an individual during a mother™s adulthood (years 15 -44) over successive births, or times. This type of change has been previously examined in relation to PTB using intergenerational births (Pearl et al. 2018; J. W. Collins, Mariani, and Rankin 2018; J. W. Collins, Rankin, and David 2015) and across an individual mother™s successive births (Bruckner, Kane, and Gailey 2019) . Based on previous studies, we calculated a neighborhood poverty trajectory score based a mother™s level of neighborhood poverty trajectory away from, or into poverty. W e first compared neighborhood poverty rates of maternal address at each birth (census tract) based on quartile cut points from the linearly interp olated neighborhood poverty (Q1 : <0.05, Q2: >=0.05 and <0.09 , Q3: >=0.09 and <0.017 , Q4: >0.17). These quartile categories represent neighborhoods with the least percentage of poverty (Q1) to those with the highest percentage of poverty (Q4). We then compared changes in neighborhood poverty rates across time points between two successive 57 pregnancies restricted to within state migration. We defined neighborhood poverty trajectory as any quartile change in census tract poverty between two successive pregnancies in Michigan based on quartile cut points from the linearly interpolated poverty data (Q1,: 0 .05, Q2: >=0.05 and <0.09 , Q3: >=0.09 and <0.017 , Q4: >0.17 ). We created neighborhood poverty trajectory categorical variables representing these changes: strong downward (Q1 to Q4), moderate downward (a 2 -quartile move into higher percentage poverty), a nd weak downward (a 1 -quartile move into higher poverty), strong upward (Q4 to Q1), moderate upward (a 2 -quartile move to away from poverty), weak upward (a 1 -quartile move away from poverty), and static mobility. Again consistent with previous studies (Br uckner, Kane, and Gailey 2019) , the static trajectory category serves as a reference category for neighborhood poverty trajectory but represents both women who did not move and did not experience a quartile change in neighborhood poverty rate and women who moved to neighborhood with similar (within the same poverty quartile) neighborhood poverty rate. Finally, women who did not move could be classified in an upward or downward trajectory if the neighborhood changed poverty levels between her successive live births. Maternal characteristics Maternal characteristics included age (<20, 20 -24, 25-29, 30 -34, 35-40, >40), race/ethnicity (non -Hispanic white [NHW], non -Hispanic black [NHB], Hispanic, non-Hispanic American Indian / Alaska Native [AI/AN], non-Hispanic Asian/Pacific Islander [A/PI], and non -Hispanic mixed race/other), nativity (US born, foreign born), parity (nulliparous, primiparous, multiparous), marital status (married, not married), educational attainment (some high school, high school, some college, college, greater than college), insurance pay source at birth (private, Medicaid, self -pay, other, unknown), smoking during pregnancy (yes, no, unknown) and hypertension during pregnancy (yes, no). 58 Preterm Birth PTB is our dependent variabl e, defined as less than 37 completed weeks at delivery. This is a binary (yes/no) variable and is reported at index, or initial, birth and successive, or subsequent, births for a mother . The gestational age is listed on the birth certificate based on clini cal estimate of gestation. We also examine gestational age of PTB (<32, 32 -36, 37) and fetal growth (SGA: infants with <10 th percentile birthweight, AGA: infants with between 10 -90th percentile birthweight, and LGA: infants with >90 th percentile birthweigh t) in a sensitivity analysis but these are not in our main analysis as there may be distinct etiologies related to mechanistic pathway. Statistical Analysis We first reported univariate and bivariate analysis for neighborhood poverty trajectory and residential mobility. Our strategy was to understand how much of neighborhood poverty trajectory was driven by residential mobility. We calculated unadjusted relative risk and 95% confidence intervals for the relationship between neighborhood trajectory an d residential mobility, using static trajectory as the reference group similar to previous work . Successive PTB rates were then calculated within each quartile of poverty. Crude PTB rate per 1000 live births were calculated by neighborhood poverty trajecto ry. We then performed modified Poisson regression with robust standard error multivariable adjusted models and neighborhood poverty trajectories and PTB. The models were adjusted for a priori covariates (maternal age, parity, educational attainment and sm oking during pregnancy) thought to confound the association between neighborhood mobility trajectory and PTB. We also adjusted for PTB at the index birth which may cause an issue with selection if it influences the women™s neighborhood poverty mobility. Fo r example, if a heathier mother who has a lower risk of PTB shows stronger upward mobility compared with a mother at higher risk for PTB then we may expect our model to be confounded by not including PTB at the index birth. This is also 59 consistent with the healthy migrant theory which postulates that healthier women are more likely to move, however in our study we focused on health relative to prior PTB. In sensitivity analyses we performed modified Poisson regression with robust standard error adjuste d models to assess the association between neighborhood poverty quartile and PTB. We then tested the static trajectory group for association o f poverty quartile category and PTB using chi -square test for general association and using Mantel -haenszal chi -square test for trend. Although this group represents no change in neighborhood mobility trajectory it also represents varying levels of neighborhood poverty by quartile. We sought to examine if there was increasing PTB with increasing neighborhood poverty a mong this group as it serves as the reference group for neighborhood trajectory. We then examined the associations between our neighborhood poverty trajectory categories using only static trajectory of the lowest poverty group (Q1) as the reference . Final ly, we also examined PTB by levels of gestational age (< 28, 28-<32, 32-<34, 34-<37). Results Table 2.1 shows the distribution of neighborhood poverty trajectories stratified by residential mobility. Most women in both birth samples (births 1 to 2, n=447,607; births 2 to 3, n=117,956) experienced static neighborhood poverty trajectory. Of the static traject ory category, 65.1% were non -movers compared to 34.9% movers. Movers experienced greater risk of both upward and downward poverty trajectory (compared to static trajectory) than non -movers. The relative risk of moving mirrored in either direction, where mo vers had 2.8 times the risk compared to non -movers for both strong upward and downward poverty trajectory. This mirror pattern remained consistent across all measured levels of parity, although the risk was slightly reduced by the 3 rd birth (RR : 2.6). 60 Table 2.2 arrays the percent PTB births at the successive birth by quartile of neighborhood poverty at index birth (horizontal) and successive birth (vertical) across births 1 to 2 and births 2 to 3. The bold diagonal on these arrays shows women whose neighbo rhood poverty stayed in the same quartile of poverty across both births and includes both lateral movers, in that the residence changed but the poverty quartile did not, and non -movers who did not experience a quartile change in poverty. The remaining valu es show neighborhood poverty trajectory, which accounts for approximately 36% of women at birth 2 and 35% of women at birth 3. The quartile representing the static trajectory (Q4 and Q4) with the highest poverty rate also has the highest rate of PTB in bot h birth samples (11.4% births 1 to 2; 12.3% births 2 to 3). Only 7.4% of women experienced PTB who exhibited strong upward trajectory (Q4 to Q1) from births 1 to 2, and 8.7% of women who had strong upward trajectory experience PTB from birth 2 to 3. In co mparison, women who had strong downward trajectory (Q1 to Q4) experience d 8.6% of PTB from births 1 to 2, and 10.6% for births 2 to 3. Table 2.3 shows the crude PTB rates per 100 live births by level of mother™s trajectory from previous live birth (Appendix tables B. 5 and B. 6 present the PTB rates per 100 live births per calendar year). Strong downward trajectory was associated with the highest r ate of PTB per live birth at both the second (8.65 cases/100 live births) and third birth (10.60 cases/live births). Strong upward trajectory did not show a reduction in the crude rate of PTB. The lowest crude rate of PTB at the second birth occurred in the weak downward trajectory category while at the third birth it occurred in the weak upward category. The majority of cases of PTB occurred in the static trajectory category, representing no change in neighborhood poverty. Table 2.4 shows the multivariable adjusted association of neighborhood poverty trajectories with PTB at the successive birth. In crude and adjusted models, strong upward 61 poverty trajectory was modestly associated with a slight reduction in odds of PTB but the association was not significa nt (Crude OR 0.93, 95% CI : 0.83 -1.03; Adjusted OR: 0.94, 95% CI: 0.84-1.06). Weak downward trajectory between births was significantly associated with a 10% decrease in the odds of PTB at the successive birth (OR: 0.90, 95% CI: 0.87 -0.93) in crude models and adjusted models (OR: 0.90, 95% CI: 0.87 -0.93). In the adjusted model, the direction of the association is inconsistent with the direction of trajectory. For example, weak upward trajectory showed slightly decreased odds of PTB at successive birth (OR: 0.96, 95% CI: 0.93 -0.99), while moderate upward trajectory had an increased odds of PTB (OR: 1.05, 95% CI: 1.00 -1.12), and strong upward trajectory was again protective. The strongest association was seen in the strong downward trajectory between births 2 and 3 with a 27% increase in the odds of PTB (OR : 1.27 , 95%CI 1.07 -1.50). Discussion We had prev iously reported a large percentage of women move between pregnancies, but we fin d here that there is limited trajectory across poverty levels (movers move to similar levels of poverty) and that there is only a modest association between neighborhood poverty trajectory and PTB during our study time period. The majority of PTB cases oc curred in the static trajectory category (table 2 .1) which is comprised of more non -movers than movers (63.0% versus 37.0%). This is consistent with our findings that the majority of women experience static trajectory and that neighborhood poverty trajecto ry was only modestly related to PTB. However, we also saw the highest rates of PTB among the poorest quartile of neighborhood poverty. Overall, women who started in low poverty had a lower percentage of PTB compared to women who started in high poverty. Our findings were similar across two levels of birth parity . We hypothesized that upward trajectory would be associated with a reduction in PTB consistent with other published results. We found evidence of a modest protective association 62 between strong upward neighborhood poverty trajectory and PTB between birth 1 and 2 . We also found a significant protective association with PTB among women who experience d weak downward neighborhood poverty trajectory between birth 1 and 2 . As this group is comprised of more non -movers than any other level of neighborhood poverty trajectory there may be some confounding by residential mobility status. We may also have measured two simultaneous associations between moving itself and a change in poverty level . It may be us eful to examine this further by stratifying by moving status in the association between poverty trajectory and PTB. We might also examine women who had the same change (e.g., Q1 to Q3) who moved and who did not move. In the association between neighborhood poverty trajectory and PTB we found the strongest association with strong downward trajectory between birth 2 and 3 with OR 1.2 (95%CI: 1.0 -1.3). As downward poverty trajectory represents an unfavorable poverty condi tion and this group represents an increase in the total number of children we suspect this may represent a more unexpected downtur n or stress pathway. We also used a static neighborhood poverty trajectory as our reference group, consistent with another study of neighborhood poverty trajectory (Bruckner, 2019). However, our choice in reference group may have introduced confounding into our model as it sta tic included all levels of neighborhood poverty that did not change. In sensitivity analysis we compare the association of neighborhood poverty trajectory using a restricted reference group. This group is comprised of the lowest risk group, the static Q1 l ow poverty group. We found increased odds of PTB for births 1 to 2 with strong downward neighborhood trajectory (OR 1. 2, 95% CI 1. 0-1.3) but also increased odds of PTB among strong upward neighborhood poverty trajectory (OR 1.1, 95%CI: 1.1- 1.2) compared to the static trajectory group of lowest neighborhood poverty quartile(Table 63 B.4) . This is more consistent with the estimation methods used by Collins et al. when they used reference groups of lowest risk. Limitations While this study offers important contributions to our understanding of neighborhood poverty trajectories and PTB , there are several important limitations to interpreting and generalizing our results. First, our sample excludes women with only one birth. This limits the generalizability of our study and may not truly define the PTB risk if mothers with a first PTB delayed or discontinued subsequent childbearing. We reported the proportion of PTB by quartile of neighborh ood poverty rate at the index and successive births. However, we did not investigate the distribution of maternal characteristics by quartile of neighborhood poverty rate. We therefore cannot conclusively determine the role of structural confounding in our model (Messer, Oakes and Mason, 2010) . If for example, our Q4 category, the highest poverty level, also disproportionately represents white women or women with high educational attainment, such that their exposure differs within the covariate strata then we may have structural confounding associated with social stratification . Future investigations using this poverty quartile structure should delineate the categorical distribution of maternal characteristics . This is necessary in order to allow for meaningful causal contrasts by level of poverty exposure, or exchangeability (Greenland and Robins , 2009 ; Messer, Oakes and Mason, 2010 ), Previous studies measured neighborhood poverty trajectory across a larger tim e period from childhood to adulthood for early -life exposure to poverty which may bias our results if, for example, women in our study were born into high poverty but now reside in low poverty. We do not have information in our study about childhood povert y exposure. Our study period may not be early enough to measure critical windows of exposure for subsequent 64 PTB outcomes related to poverty trajectory. While our study time frame is broad it may represent too acute a time period for change in trajectory t o be associated with PTB. We also do not have information on the mother™s own birth histories which may be relevant a s a previous report suggest women who had adverse birth events as infants may be more likely to have PTB with increased neighborhood povert y exposure (J. W. Collins, Rankin, and David 2011) . Further examination of this potential for residual confounding is warranted. We also do not have detailed socioeconomic data on the mother or her family unit, such as her employment and partner™s socioeco nomic status. We therefore cannot make inference about the motivations for trajectory and the relationship to birth outcomes. Finally, we used a relative measure of poverty trajectory, quartiles based on the population neighborhood poverty rate, but an alt ernate specification, such as absolute poverty levels, may be more informative as it relates to policy decisions (such as the U.S. Census guidelines for high poverty >20% (Bureau n.d.) ). In summary, this study extends the work on associations between neig hborhood poverty trajectories and adverse birth outcomes using a large sample of maternally -linked births in Michigan during the period 1990 -2012. We find that non -movers experience some poverty change through neighborhood changes but the majority of pover ty trajectory comes in the form of moving. Even among movers, there is limited trajectory across poverty groups with the majority of women remaining in the same quartile of poverty across successive births. PTB is associated with high levels of neighborhoo d poverty, but not with neighborhood poverty trajectory across poverty quartiles. 65 Table 2.1 Neighborhood poverty trajectories at successive birth (upward, downward, static) stratified by residential status (mover, non -mover) in singleton maternally -linked Michigan births1990 -2012 Births 1 to 2 (n=895,214) Neighborhood Trajectory a Residential Mobility Overall Non -Movers Movers n % n % n % RRb 95%CI Total Mothers 447,607 100 223,005 49.8 224,602 50.2 Strong Downward 3,955 0.9 144 3.6 3,811 96.4 2.8 2.7 2.8 Moderate Downward 17,133 3.8 950 5.5 16,183 94.5 2.7 2.7 2.7 Weak Downward 69,421 15.5 26,119 37.6 43,302 62.4 1.8 1.8 1.8 Static 285,662 63.8 186,094 65.1 99,568 34.9 1.0 Weak Upward 50,265 11.2 9,188 18.3 41,077 81.7 2.3 2.3 2.4 Moderate Upward 16,567 3.7 424 2.6 16,143 97.4 2.8 2.8 2.8 Strong Upward 4,604 1.0 86 1.9 4,518 98.1 2.8 2.8 2.8 Births 2 to 3 (n=355,912) Neighborhood Trajectory a Residential Mobility Overall Non -Movers Movers n % n % n % RRb 95% CI Total Mothers 177,956 100 87,047 48.9 90,909 51.1 Strong Downward 1,405 0.8 51 1.9 1,354 98.1 2.6 2.5 2.6 Moderate Downward 6,337 3.6 463 2.6 5,874 97.4 2.5 2.4 2.5 Weak Downward 27,556 15.5 10,927 17.8 16,629 82.3 1.6 1.6 1.6 Static 115,193 64.7 71,971 62.5 43,222 37.5 1.0 Weak Upward 19,375 10.9 3,440 39.7 15,935 60.4 2.2 2.2 2.2 Moderate Upward 6,265 3.5 161 7.3 6,104 92.7 2.6 2.6 2.6 Strong Upward 1,825 1.0 34 3.6 1,791 96.4 2.6 2.6 2.6 a Quartile cut points from linearly interpolated census data 1990 -2012: 0.05, 0.09, and 0.17 . Neighborhood poverty Trajectories are categories of change away, or into poverty based on a quartile change. The strongest change represents Strong Downward (Q1 to Q4) and Strong Upward (Q4 to Q1). b Modified Poisson regression of movers versus non -movers by neighborhood poverty trajectory with static mobility as reference group 66 Table 2.2 Percentage of PTB births at successive birth among singleton Michigan births by neighborhood poverty quartile a at index and successive births Births 1 to 2 (n=447,607) Neighborhood Poverty Birth 1 (Index Birth) Q1 Q2 Q3 Q4 Birth 2 (Successive Birth) Q1 6.1% (4,735/78,159) 6.3% (1,164/18,427) 7.2% (573/7,979) 7.4% (340/4,604) Q2 6.2% (1,438/23,089) 6.6% (4,443/67,439) 7.3% (1,164/15,963) 9.3% (795/8,588) Q3 6.6% (543/8,197) 6.9% (1,727/25,193) 6.2% (4,425/70,865) 9.4% (1,488/15,875) Q4 8.6% (342/3,955) 9.0% (802/8,936) 8.7% (1,836/21,139) 11.4% (9,068/79,199) Births 2 to 3 (n=177,956) Neighborhood Poverty Birth 2 (Index birth) Q1 Q2 Q3 Q4 Birth 3 (Successive Birth) Q1 5.6% (1,537/27,252) 6.5% (438/6,771) 7.8% (226/2,880) 8.7% (159/1,825) Q2 6.9% (570/8,228) 6.8% (1,660/24,337) 8.1% (478/5,906) 10.4% (352/3,385) Q3 7.9% (222/2,823) 7.8% (792/10,101) 7.6% (1,844/24,384) 9.7% (652/6,698) Q4 10.6% (149/1,405) 9.5% (333/3,514) 9.5% (875/ 9,227) 12.3% (4,820/39,220) aNeighborhood Poverty Quartiles: Q1 Very Low: <0.05 Q2 Low: >=0.05 and <0.09 Q3 High: >=0.09 and <0.017 Q4 Very High: >=0.17 Table 2.3 Risk per 100 live births of preterm birth (< 37 weeks completed gestation) at birth 2 and birth 3 by mother™s neighborhood poverty trajectory from previous index birth, singleton Michigan Births 1990 -2012 Preterm Births at Birth 2 (n=447,607) Preterm Births at Birth 3 (n=177,956) Neighborhood Poverty Trajectory Cases Cases per 100 live births Cases Cases per 100 live births Strong downward 342 8.7 149 10.6 Moderate Downward 1,345 7.9 555 8.8 Weak Downward 5,001 7.2 2,237 8.1 Static 22,671 7.9 9,861 8.6 Weak upward 3,816 7.6 1,568 8.1 Moderate upward 1,368 8.3 578 9.2 Strong upward 340 7.4 159 8.7 67 Table 2.4 Associations of Improved Neighborhood Trajectory Trajectories and Successive Birth Outcomes, Bivariate and Multivariate Models in Singleton Michigan Births 1990 -2012 (n=895,214) Neighborhood Poverty Trajectory PTB Crudea Adjusted b OR (95% CI) OR (95% CI) Poverty Trajectory Birth 1 to 2 Strong Downward 1.1 1.0 1.2 1.0 0.9 1.2 Moderate Downward 1.0 0.9 1.1 1.0 0.9 1.1 Weak Downward 0.9* 0.9 0.9 1.0 0.9 1.0 Static (ref) 1.0 1.0 Weak Upward 1.0 0.9 1.0 1.0 1.0 1.0 Moderate Upward 1.0 1.0 1.1 1.1 1.0 1.1 Strong Upward 0.9 0.8 1.0 0.9 0.8 1.0 Poverty Trajectory Birth 2 to 3 Strong Downward 1.3 1.1 1.5 1.2 1.0 1.5 Moderate Downward 1.0 0.9 1.1 1.1 1.0 1.2 Weak Downward 0.9 0.9 1.0 1.0 1.0 1.1 Static (ref) 1.0 1.0 Weak Upward 0.9 0.9 1.0 1.0 1.0 1.1 Moderate Upward 1.1 1.0 1.2 1.1 1.0 1.2 Strong Upward 1.0 0.9 1.2 1.1 0.9 1.2 aCrude is bivariate model. bAdjusted for a priori covariates age, parity, maternal education, race/ethnicity, smoking and prior PTB. PTB preterm birth <37 weeks *significant p<.0001 68 APPENDIX 69 Table B.2 Bivariate Associations of Quartiles of Neighborhood Poverty a and PTB across all births, Singleton Michigan Births 1990 -2012 Quartiles Births 1 to 2 Births 2 to 3 PTB Odds Ratio 95% CI PTB Odds Ratio 95% CI Q2 vs Q1 1.1 1.1 1.1 1.2 1.1 1.2 Q3 vs Q1 1.2 1.2 1.2 1.3 1.3 1.4 Q4 vs Q1 1.6 1.6 1.7 2.0 2.0 2.1 aNeighborhood Poverty Quartiles: Q1 Very Low: <0.05 Q2 Low: >=0.05 and <0.09 Q3 High: >=0.09 and <0.017 Q4 Very High: >=0.17 Table B.1 Datasets Mean, SD and Quartile Cut -points of Neighborhood Poverty Dataset Mean SD Quartile Cut -point 1 Quartile Cut -point 2 Quartile Cut -point 3 NCDB Linear Interpolated* 0.137120 0.13092 0.0492723 0.0919302 0.1736361 All births 0.145801 0.12861 0.0527120 0.0991436 0.1992184 All births >1 0.146822 0.12975 0.0525170 0.0990175 0.2020000 MI births 1 to 2 0.132156 0.11947 0.0494155 0.0893607 0.1724290 MI births 2 to 3 0.147351 0.12925 0.0537776 0.1005618 0.2020000 * Used as Quartile cut points for all analyses 70 Table B.3 Poverty quartile of static neighborhood poverty trajectory (no change) and PTB, and residential mobility at birth 2 (n= 285,662) PTB at Successive Birth Residential Mobility Quartile of Neighborhood Poverty a at index and successive births No YES X2 M-H X2 Non -Mover Mover X2 M-H X2 n % n % n % n % Q1-Q1 (n=78,159 ) 73,424 93.9 56,675 6.1 <.0001 <.0001 56,675 72.5 21,484 27.5 <.0001 <.0001 Q2-Q2 (n=67,439 ) 62,996 93.4 48,354 6.6 48,354 71.7 19,085 28.3 Q3-Q3 (n=60,865) 56,440 92.7 41,970 7.3 41,970 69.0 18,895 31.0 Q4-Q4 (n=79,199) 70,131 88.6 39,095 11.5 39,095 49.4 40,104 50.6 aNeighborhood Poverty Quartiles: Q1 Very Low: <0.05 Q2 Low: >=0.05 and <0.09 Q3 High: >=0.09 and <0.017 Q4 Very High: >=0.17 bX2 test for general association cMantel -haenszel test for trend 71 Table B.4 Associations of Improved Neighborhood Trajectory Trajectories and Successive Birth Outcomes, Bivariate and Multivariate Models in Singleton Michigan Births 1990 -2012 (n=146,518) Q1 Static Reference Group Neighborhood Poverty Trajectory PTB Crudea Adjusted b OR (95% CI) OR (95% CI) Poverty Trajectory Birth 1 to 2 Strong Downward 1.5 1.3 1.7 1.2 1.0 1.3 Moderate Downward 1.3 1.2 1.4 1.1 1.0 1.2 Weak Downward 1.3 1.2 1.4 1.1 1.0 1.1 Static (ref) Q1 only 1.0 1.0 Weak Upward 1.3 1.2 1.4 1.1 1.1 1.2 Moderate Upward 1.4 1.3 1.5 1.2 1.1 1.3 Strong Upward 1.2 1.1 1.4 1.0 0.9 1.1 aCrude is bivariate model. bAdjusted for a priori covariates age, parity, maternal education, race/ethnicity, smoking and prior PTB. PTB preterm birth <37 weeks *significant p<.0001 72 Table B.5 Risk per 100 live births by calendar year of preterm birth (< 37 weeks completed gestation) at birth 2 by level of mother™s trajectory from birth 1, Singleton Michigan Births 1990 -2012 Birth Year Neighborhood Poverty Trajectory by Quartile Change a No change Downward (inclusive) Strong Downward Moderate Downward Weak Downward Upward (inclusive) Weak Upward Moderate Upward Strong Upward 1990 47.83 66.67 . 66.67 . 0.00 0.00 . . 1991 12.56 10.66 6.67 13.04 10.71 16.14 17.49 9.26 23.53 1992 8.52 7.03 17.74 4.73 6.43 8.47 8.38 10.63 4.35 1993 8.72 10.05 14.04 13.78 8.74 7.38 6.98 8.15 9.43 1994 8.10 6.84 12.24 5.86 6.55 6.44 6.02 7.09 9.90 1995 8.06 6.64 5.75 6.08 6.83 7.53 6.94 8.49 10.99 1996 8.33 7.41 5.62 9.58 6.94 7.46 6.90 8.87 9.38 1997 8.15 7.39 6.67 8.57 7.16 8.34 8.60 8.23 5.92 1998 8.39 8.89 9.43 8.67 8.91 7.89 7.49 8.25 11.23 1999 8.14 9.90 7.69 7.43 10.76 8.87 8.43 11.44 5.58 2000 8.39 8.36 8.20 8.83 8.24 7.82 7.48 8.89 7.55 2001 9.15 7.79 13.14 8.88 7.27 7.37 7.00 8.82 5.81 2002 7.93 7.74 6.52 7.17 7.93 8.25 8.08 8.33 9.70 2003 8.58 8.50 8.76 8.22 8.55 9.24 9.47 9.33 6.07 2004 6.87 6.53 7.07 7.88 6.24 7.83 7.70 8.44 6.83 2005 7.47 6.77 8.75 7.11 6.60 7.62 7.40 8.67 5.65 2006 7.19 6.21 9.41 7.06 5.85 7.09 6.76 8.57 5.02 2007 6.66 6.81 7.54 6.71 6.80 7.36 7.13 7.82 8.00 2008 7.33 7.11 8.98 7.82 6.84 7.44 7.07 8.52 6.92 2009 6.92 6.69 8.14 7.25 6.47 7.80 8.29 6.94 6.29 2010 6.41 6.39 7.08 6.50 6.30 6.27 6.46 6.48 3.65 2011 8.92 8.58 9.09 9.43 8.31 7.84 7.84 7.07 10.76 2012 9.26 7.69 9.11 8.42 7.40 7.81 8.09 6.90 7.61 Total 7.94 7.39 8.65 7.85 7.20 7.73 7.59 8.26 7.38 aQuartile Q1 Very Low: <0.05 ; Q2 Low: >=0.05 and <0.09 ; Q3 High: >=0.09 and <0.017 ; Q4 Very High: >=0.17 73 Table B. 6 Risk per 100 live births by calendar year of preterm birth (< 37 weeks completed gestation) at birth 3 by level of mother™s trajectory from birth 2, Singleton Michigan Births 1990 -2012 Birth Year Neighborhood Poverty Trajectory by Quartile Change a No change Downward (inclusive) Strong Downward Moderate Downward Weak Downward Upward (inclusive) Weak Upward Moderate Upward Strong Upward 1990 0 . . . . . . . . 1991 22.17 6.67 0.00 0.00 10.00 0.00 0.00 0.00 0.00 1992 13.86 11.70 33.33 5.56 11.43 11.11 10.23 17.39 0.00 1993 10.26 11.41 7.69 14.29 11.19 9.72 11.43 5.45 5.88 1994 8.74 11.00 5.88 13.33 10.92 6.28 6.42 2.74 13.79 1995 7.30 8.37 8.33 8.70 8.29 6.99 6.16 9.15 10.00 1996 8.52 6.44 12.00 6.06 6.17 7.19 5.38 10.29 16.98 1997 8.87 10.29 16.67 12.22 9.64 8.16 7.26 10.81 9.80 1998 8.77 9.60 9.38 10.48 9.38 7.61 8.44 6.42 2.67 1999 9.46 8.54 6.25 8.89 8.51 8.51 7.75 12.30 4.94 2000 8.72 8.97 10.81 5.88 9.62 8.16 7.85 8.72 9.52 2001 8.62 8.46 11.29 7.47 8.57 8.70 8.14 9.71 10.00 2002 9.71 8.04 12.31 7.07 8.04 11.36 12.16 10.29 6.67 2003 9.11 9.27 12.66 11.42 8.68 7.39 7.75 7.08 3.70 2004 8.41 8.52 12.22 8.52 8.35 7.40 7.17 8.52 5.75 2005 8.10 8.09 6.82 5.67 8.65 9.53 8.97 10.44 12.77 2006 7.30 7.87 12.09 9.26 7.38 8.04 8.16 7.57 8.65 2007 7.80 6.87 4.21 8.44 6.66 9.16 8.12 12.17 8.57 2008 7.67 7.87 9.00 8.68 7.63 8.50 8.28 9.05 8.61 2009 7.58 7.63 12.75 7.87 7.30 7.02 6.59 7.43 9.63 2010 7.48 7.01 7.25 8.86 6.50 8.09 8.03 9.13 4.63 2011 9.48 9.65 13.64 9.76 9.34 8.71 8.25 9.19 12.17 2012 9.64 9.09 12.40 10.32 8.59 8.92 8.32 10.34 11.71 Total 8.56 8.33 10.60 8.76 8.12 8.39 8.09 9.23 8.71 aQuartile Q1 Very Low: <0.05 ; Q2 Low: >=0.05 and <0.09 ; Q3 High: >=0.09 and <0.017 ; Q4 Very High: >=0.17 74 Table B. 8 Mother™s (N=447,607) Neighborhood Poverty Trajectories by MSA (Metropolitan, Micropolitan, Rural) Change, Singleton Michigan Births 1 to 2, 1990-2012 Neighborhood Poverty Trajectories MSA Change from Birth 1 to Birth 2 Micro to Metro Metro to Rural No change Rural to Metro Metro to Micro Strong Downward 4 189 3,723 77 36 Moderate Downward 33 922 15,820 544 253 Weak Downward 142 1,978 66,469 1,508 1,171 Static 141 2,461 283,603 2,529 2,911 Weak Upward 59 1,227 47,733 1,683 804 Moderate Upward 6 406 15,725 687 125 Strong Upward 1 57 4,508 114 12 Table B.7 Gestational age of preterm births among singleton Michigan births 1990 -2012 PTB Gestational age (weeks) Births 1 to 2 (n= 73,840 ) Births 2 to 3 (n=29,288) n % n % Extreme Preterm <28 2,940 4.0 971 3.3 Very Preterm >28 to 32 5,572 7.6 2,028 6.9 Moderate Preterm >32 to 34 8,624 11.7 3,418 11.7 Late Preterm >34 to 37 56,704 76.8 22,871 78.1 75 REFERENCES 76 REFERENCES Braveman, Paula A., Catherine Cubbin, Susan Egerter, David R. Williams, and Elsie Pamuk. 2010. fiSocioeconomic Disparities in Health in the United States: What the Patterns Tell Us.fl American Journal of Public Health 100 Suppl 1 (April): S186 -196. https://doi.org/10.2105/AJPH.2009.166082 . Bruckner, Tim A., Jennifer B. Kane, and Samantha Gailey. 2019. fiStrong Upward Neighborhood Mobility and Preterm Birth: A Matched -Sibling Design Approach.fl Annals of Epidemiology 36 (August): 48 -54.e1. https://doi.org/10.1016/j.annepidem.2019.05.005 . Bureau, US Census. n.d. fiAbout.fl Accessed March 4, 2019a. https://www.census.gov/topics/income -poverty/poverty/about.html. ŠŠŠ. n.d. fiAmerican Community Survey (ACS).fl Accessed October 24, 2016b . https://www.census.gov/programs -surveys/acs. ŠŠŠ. n.d. fiGeography Program.fl Accessed November 9, 2019c. https://www.census.gov/programs -surveys/geography.html . Colen CG, Geronimus AT , Bound J, and James SA. 2006. fiResearch and Practice. Maternal Upward Socioeconomic Mobility and Black -White Disparities in Infant Birthweight.fl American Journal of Public Health 96 (11): 2032 Œ39. https://doi.org/10.2105/AJPH.2005.076547 . Collins, James W., Allison Mariani, and Kristin Rankin. 2018. fiAfrican -American Women™s Upward Economic Mobility and Small for Gestational Age Births: A Population -Based Study.fl Maternal and Child Health Journal 22 (8): 1183 Œ89. https://doi.org/10.1007/s10995 -018-2503-4. Collins, James W., Kristin M. Rankin, and Richard J. David. 2011. fiAfrican American Women™s Lifetime Upward Economic Mobility and Preterm Birth: The Effect of Fetal Programming.fl American Journal of Public Health 101 (4): 714 Œ19. https://doi.org/10.2105/AJPH.2010.195024 . ŠŠŠ. 2015. fiDownward Economic Mobility and Preterm Birth: An Exploratory Study of Chicago -Born Upper Class White Mothers.fl Maternal and Child Health Journal 19 (7): 1601 Œ7. https://doi.org/10.1007/s10995 -015-1670-9. Collins, Jr JW, R. J. David, D. M. Simon, and N. G. Prachand. 2007. fiPreterm Birth among African American and White Women with a Lifelong Residence in High -Income Chicago Neighborhoods: An Exploratory Study.fl Ethnicity & Disease 17 (1): 113 Œ17. Croteau, Agathe, Sylvie Marcoux, and Chantal Brisson. 2007. fiWork Activity in Pregnancy, Preventive Measures, and the Risk of Preterm Delivery.fl American Journal of Epidemio logy 166 (8): 951 Œ65. https://doi.org/10.1093/aje/kwm171. David, Richard, and Lynne Messer. 2011. fiReducing Disparities: Race, Class and the Social Determinants of Health.fl Maternal and Child Health Journal 15 Suppl 1 (September): S1 -3. https://doi.org/10.1007/s10995 -011-0878-6. 77 Geography, US Census Bureau. n.d. fiGuide to State and Local Geography - Michigan.fl Accessed January 11, 2017. https://www.census.gov/geo/reference/guidestloc/st26_mi.html . Greenland, Sander, and James M Robins. 2009. fiIdentifiability, Exchangeability and Confounding Revisited.fl 6 (September): 4. https://doi.org/10.1186/1742 -5573-6-4. Heinonen, Kati, Johan G. Eriksson, Eero Kajantie, Anu -Katriina Pesonen, David J. Barker, Clive Osmond, and Katri Raikkonen. 2013. fiLate -Preterm Birth and Lifetime Socioeconomic Attainments: The Helsinki Birth Cohort Study.fl Pediatrics 132 (4): 647 Œ55. https://doi.org/10.1542/peds.2013 -0951. Holzman, Claudia, Janet Eyster, Mary Kleyn, Lynne C. Messer, Jay S. Kaufman, Barbara A. Laraia, Patricia O™Campo, Jessica G. Burke, Jennifer Culhane, and Irma T. Elo. 2009a. fiMaternal Weathering and Risk of Preterm Delivery.fl American Journal of Public Health 99 (10): 1864 Œ1871. ŠŠŠ. 2009b. fiMaternal Weathering and Risk of Preterm Delivery.fl American Journal of Public Health 99 (10): 1864 Œ71. https://doi.org/10.2105/AJPH.2008.151589 . Iceland, John, and Erik Hernandez. 2017. fiUnderstanding Trends in Concentrat ed Poverty: 1980 to 2014.fl Social Science Research 62 (February): 75 Œ95. https://doi.org/10.1016/j.ssresearch.2016.09.001 . Janevic, T, CR Stein, DA Savitz, JS Kaufman, SM Mason, and AH Herri ng. 2010. fiNeighborhood Deprivation and Adverse Birth Outcomes among Diverse Ethnic Groups.fl Annals of Epidemiology 20 (6): 445 Œ51. https://doi.org/10.1016/j.annepidem.2010.02.010 . Jargowsky, Paul A. 1998. Poverty and Place: Ghettos, Barrios, and the American City . New York: Russell Sage Foundation. Kramer, Michael R., Anne L. Dunlop, and Carol J. R. Hogue. 2014. fiMeasuring Women™s Cumulative Neighborhood Deprivation Exposure Using Longitudin ally Linked Vital Records: A Method for Life Course MCH Research.fl Maternal and Child Health Journal 18 (2): 478 Œ87. https://doi.org/10.1007/s10995 -013-1244-7. Logan, John R., Zengwang Xu, and Brian J. Stults. 2014. fiInterpolating U.S. Decennial Census Tract Data from as Early as 1970 to 2010: A Longitudinal Tract Database.fl The Professional Geographer 66 (3): 412 Œ20. https://doi.org/10.1080/00330124.2014.905156. Love, Catherine, Richard J. David, Kristin M. Rankin, and James W. Collins. 2010. fiExploring Weathering: Effects of Lifelong Economic Environment and Maternal Age on Low Birth Weight, Small for Gestational Age, and Preterm Birth in African -American and Whi te Women.fl American Journal of Epidemiology 172 (2): 127 Œ34. https://doi.org/10.1093/aje/kwq109 . 78 Margerison -Zilko, Claire, Catherine Cubbin, Jina Jun, Kristen Marchi, Kathryn Fingar, and Paula Braveman. 2 015. fiBeyond the Cross -Sectional: Neighborhood Poverty Histories and Preterm Birth.fl American Journal of Public Health 105 (6): 1174 Œ80. https://doi.org/10.2105/AJPH.2014.302441 . Messer, Lynne C., J ay S. Kaufman, Nancy Dole, David A. Savitz, and Barbara A. Laraia. 2006. fiNeighborhood Crime, Deprivation, and Preterm Birth.fl Annals of Epidemiology 16 (6): 455 Œ62. https://doi.org/10.1016/j. annepidem.2005.08.006 . Messer, Lynne C., J. Michael Oakes, and Susan Mason. 2010. fiEffects of Socioeconomic and Racial Residential Segregation on Preterm Birth: A Cautionary Tale of Structural Confounding.fl American Journal of Epidemiology 171 (6): 664 Œ73. https://doi.org/10.1093/aje/kwp435 . Meyer, John D., Nicholas Warren, and Susan Reisine. 2007. fiJob Control, Substantive Complexity, and Risk for Low Birth Weight and Preterm Delivery: An Analysis from a State Birth Registry.fl American Journal of Industrial Medicine 50 (9): 664 Œ75. https://doi.org/10.1002/ajim.20496 . O™Campo, Patricia, Jessica G. Burke, Jennifer Culhane, Irma T. Elo, Janet Eyster, Claudia Holzman, Lynne C. Messer, Jay S. Kaufman, and Barbara A. Laraia. 2008. fiNeighborhood Dep rivation and Preterm Birth among Non -Hispanic Black and White Women in Eight Geographic Areas in the United States.fl American Journal of Epidemiology 167 (2): 155 Œ163. Osypuk, Theresa L., Jaime Slaughter -Acey, Rebecca Kehm, and Dawn P. Misra. 2016. fiLife Course Social Mobility and Reduced Risk of Adverse Birth Outcomes.fl American Journal of Preventive Medicine 51 (6): 975 Œ82. https://doi.org/10.1016/j.amepre.2016.09.008 . Pearl, Michelle, Jennife r Ahern, Alan Hubbard, Barbara Laraia, Bina Patel Shrimali, Victor Poon, and Martin Kharrazi. 2018. fiLife -Course Neighbourhood Opportunity and Racial -Ethnic Disparities in Risk of Preterm Birth.fl Paediatric and Perinatal Epidemiology 32 (5): 412 Œ19. https://doi.org/10.1111/ppe.12482 . Schempf, Ashley H., Jay S. Kaufman, Lynne C. Messer, and Pauline Mendola. 2011. fiThe Neighborhood Contribution to Black -White Perinatal Disparities: An Example from Two North Carolina Counties, 1999 -2001.fl American Journal of Epidemiology 174 (6): 744 Œ52. https://doi.org/10.1093/aje/kwr128 . Slaughter -Acey, Jaime C., Claudia Holzman, Danuelle Calloway, and Yan Tian. 2016. fiMovin ™ on Up: Socioeconomic Mobility and the Risk of Delivering a Small -for-Gestational Age Infant.fl Maternal and Child Health Journal 20 (3): 613 Œ22. https://doi.org/10.1007/s10995 -015-1860-5. Spencer, N. 2004. fiAccounting for the Social Disparity in Birth Weight: Results from an Intergenerational Cohort.fl Journal of Epidemiology & Community Health 58 (5): 418 Œ19. https://doi.org/10.1136/jech.2003.012757. 79 STUDY 3 Introduction Individuals who live in high -poverty areas fare worse compared to individuals who live in low -poverty neighborhoods on a broad range of outcomes not just limited to health but including economic and educational outcomes (Sampson 2013; Leventhal and Brooks -Gunn 2000). Evidence from neighborhood associations examining deprivation suggests a persistent association with adverse birth outcomes that remains even after adjustment for neighborhood -level and individual -level covariates. Increasingly, there is eviden ce of an association between upward mobility, moving away from higher poverty areas, and a modest protective effect on PTB (Kramer 2016; Collins, Mariani, and Rankin 2018; Pearl et al. 2018) . However, this protective effect is not consistent across racial groups and may be modified by early -life poverty status (Collins, Mariani, and Rankin 2018; Collins, Rankin, and David 2015) . Further, neighborhoods are not discrete entities but reflect larger societal patterns such that Non -Hispanic Black women are more likely than Non -Hispanic White women to reside in neighborhoods with high deprivation, increased crime, and diminished housing quality (Culhane and Goldenberg 2011). As we previously reported, there is a large amount of residential mobility among women between pregnancies with approximately 50% of women in our previous study changing residence between births. In the estimation of the association between neighborhood poverty and birth outcomes , it is therefore important to evaluate these residential mobility patterns as a potential source for bias, particularly if they are also related to the successive birth outcome. Even non -movers may be subject to selection bias if they have unobserved conf ounding that selects them into their original neighborhood and impacts PTB outcomes. These observed societal patterns may challenge the validity of causal inference in the association between neighborhood poverty and preterm birth if there are conditions that select a 80 woman into a neighborhood which also influence her pregnancy outcome. The resulting selection bias, when there are maternal characteristics which differential ly select a woman into a neighborhood , are associated with her pregnancy outcomes, is then incorrectly attributed as neighborhood exposure. Prior studies have recognized the potential for neighborhood compositional changes due to maternal selection and controlled for maternal covariates such as race/ethnicity, age and parity. However, the re may be unmeasured factors, such as stressors, health, prenatal practices, which also impact selection into neighborhoods. Maternal factors Œ both measured and unmeasured Œ can either be time -invariant, where they do not substantially vary over time suc h as gender, race/ethnicity, or time -varying, when there is a change in value over time such as parity or marital status. Prior studies of neighborhood associations with birth outcomes have adjusted for measured confounders by including measured maternal c ovariates. Prior studies of neighborhood associations with birth outcomes have also focused on adjustments for confounding by unmeasured neighborhood -level characteristics using multi -level modeling or time -varying area -level characteristics. However, to o ur knowledge there have been scant studies that have addressed confounding by unmeasured maternal characteristics (Margerison, Luo, and Li 2019; Bruckner, Kane, and Gailey 2019) . One such method that addresses unmeasured factors inherent to the mother tha t do not vary substantially over time is a maternal fixed effects analysis, analogous to a case crossover study where a mother is used as her own control to account for stable maternal characteristics, both observed and unobserved. This is also termed a w ithin -mother analysis or matched sibling design as it examines the variation within the mother while discarding the between person variation that is likely contaminated by the unmeasured maternal characteristics resulting in unbiased estimates (Allison 200 5). 81 We utilized a longitudinal dataset linked by a mother™s recorded births in Michigan during the study period 1990 to 2012 to examine the association between neighborhood poverty and preterm birth (PTB) using a maternal fixed effects analysis. We examin ed changes in neighborhood poverty between successive pregnancies measured at the time of infant birth and PTB. We then compared these findings to conventional applications of neighborhood associations of poverty and PTB, logistic regression and a random e ffects approach using data from the full maternally -linked Michigan births data 1990 -2012. Methods Study Population and Data We used birth certificate data from the State of Michigan linked by the mother during the study time period 1990 -2012. Figure 1.1 shows the study inclusion and exclusion criteria. The final study sample includes a total of 2,199,206 births to 1,181,640 unique mothers. We then linked yearly neighborhood poverty rates obtained from linearly interpolated Neighborhood Change Database (NCDB) poverty rates at the census tract level. NCDB information is taken from 3 decennial census years 1990, 2000, and 2010 all normalized to census tract boundaries of 2010 to allow for longitudinal comparisons. In order to use a maternal fixed effects (FE) analysis we further limited the sample to mothers who had a least two births and had discordant PTB outcomes. This left 279 ,150 births to 103,180 mothers in the FE analytic sample. Previous research using the Michigan Birth file reported recoding inconsistent maternal data, such as race/ethnicity differed across births, when available or excluding inconsistent data that did no t meet recoding criteria. We use the corrected covariates in our analysis using the methodology detailed in (Margerison, Luo, and Li 2019) . 82 Measures Neighborhood poverty rate We measured area -level poverty conditions using census tract poverty rates (i.e. the number of poor persons divided by the number of persons in a census tract) measured at the time of birth by residential address linked to geocoded census tract. Neighborhood poverty rates were merged with data from the birth certificate based on infant™s birth year and mother™s census tract at the time of delivery. Neighborhood poverty rates are a frequently used proxy for area -level socioeconomic conditions in research on bi rth outcomes, and in neighborhood effects in general. One of our primary objectives i n utilizing a fixed effects analysis is to compare with traditional logistic associations used in previous studies of neighborhood deprivation and PTB . Thus our primary an alyses uses neighborhood poverty rate. There are 2,813 census tracts and 83 counties in the state of Michigan. Our analyses focused on poverty conditions at the time of birth because 1) they have been previously associated with adverse birth outcomes and 2) they represent a critical window of exposure for the child, which may be considered for future analysis. Adverse birth outcomes Our primary outcome was PTB (<37 weeks completed gestation) following recent evidence (Bruckner, Kane, and Gailey 2019) that strong upward neighborhood poverty were associated with decreased PTB. We completed secondary analysis of small -for-gestational age ( SGA : <10 th percentile) and large -for-gestational -age ( LGA :> 10th percentile) outcomes but did not develop a priori hypotheses about the relationship with neighborhood poverty other than it would be inverse of the PTB association (downward mobility is associated with increased SGA). Maternal characteristi cs Maternal characteristics included age (<20, 20 -24, 25-29, 30 -34, 35-40, >40), race/ethnicity (non -Hispanic white [NHW], non -Hispanic black [NHB], Hispanic, non-Hispanic American Indian / Alaska Native [AI/AN], non -Hispanic Asian/Pacific Islander 83 [A/PI], and non -Hispanic mixed race/other), nativity (US born, foreign b orn), parity (nulliparous, primiparous, multiparous), marital status (married, not married), educational attainment (some high school, high school, some college, college, greater than college), insurance pay source at birth (private, Medicaid, self -pay, ot her, unknown), smoking during pregnancy (yes, no, unknown) and hypertension during pregnancy (yes, no). Inter -pregnancy interval period was calculated between th e conception date and the previous birth date of birth. Conception date was calculated based on gestational days estimation and date of birth. Statistical Analyses Primary Analyses First, we first conducted a traditional logistic regression for the association between the neighborhood poverty rate and PTB (logistic). We adjuste d these for race/ethnicity and nativity. Second, we conducted a random effects analysis with a mother specific random effect that accounts for differences in maternal outcomes with a random distribution. The RE model is as follows: |=++ Here is the outcome for mother i™s birth j. is the neighborhood poverty variable of interest for birth j. is a vector of control variables that include the month and year of birth, and maternal characteristics. is a mother -specific random intercept that is independent of , but could possible covary with . G is a logit link function for the outcome, PTB. The parameter of interest is , presented as a transformation for odds ratio (Allison 2005) . For the RE model this type of mixed effect works well with our longitudinal data for several reasons. First, time is treated as a continuous variable allowing for variations in time across subjects, useful when inter -birth intervals vary for each woman in the data set. Both time 84 and time -varying covariates can be included in the model. Therefore the outcome is modelled due both the mother™s stable characteristics (e.g., race or gender) and those that changes over time (e.g., number of children). We controlled for observed maternal characteristics hypothesized to influence and woman™s likelihood PTB Œ maternal age, race/ethnicity, nativity, marital status, marital status, parity, education, and infant sex. These variabl es can be included as birth -specific controls in the model. We included indicator variables seasonality (birth month) and secular trends (year of birth). Third, we conducted fixed analysis using a maternal fixed effect model (conditional logistic model) a djusted for the same maternal characteristics as the RE model. The FE model, in contrast to the RE model, controls for unobserved heterogeneity of time -invariant maternal covariates (measured and unmeasured). We added an indicator variable for each birth r ecord to control for increasing parity. The FE (with in mother) analysis controls for time -invariant characteristics shared by the mothers across births . The FE model is similar to the RE model except that the variable is not independent and does not f ollow a specific distribution. This parameter then represents the unobserved confounding factors that do not change over time for the mother i. It may be correlated with the exposure and . The fixed effects logit model uses the conditional ma ximum likelihood method to eliminate the nuisance parameter, , and estimate other coefficients ( ). We considered the use of a lagged preterm birth control in our fixed effects model as prior PTB is considered a risk factor for subsequent PTB The re are multiple ways to conceptualize the risk of prior PTB. The risk may be due to inherent characteristics of the mother, which our model would capture, or the risk may be due to the impact of a non -term 85 pregnancy on the gestational term of a subsequent pregnancy, which our model would not capture. We are currently not aware of a consensus on the mechanism of the risk of prior PTB on subsequent birth, but to the extent that is related to the mother it would be possible to include a covariate to control fo r prior PTB. However, this may introduce some bias related to our fixed effect model. In fixed effects analysis we chose not to include a lagged dependent PTB variable primarily because it introduced a form of bias, Nickell™s bias, into the models (Nickell 1981). This occurs particularly when there is a small time T, large N as in our case with a large sample size (N) but each individual woman contributes only a few births (T). The estimate of the coefficient of the lagged dependent variable, lagged prior PTB, is biased the first birth because observations are alway s 0 (T-1 observations are always 0 ) and which mean the mean error contains bias as it is subtracted from each . Increasing births do not mitigate this error and the result is termed Nickell™s bias. Including these poorly specified variable lagged var iables in the model comes at a risk of introducing this bias . In our primary analysis we excluded prior, or lagged, PTB but i n secondary analysis we included models with a la gged PTB variable to evaluate the covariate specification and the effect on our po verty variable in our models. We then performed a RE model using the FE sample to determine if differences in the measure of association are due to differences in our analytic sampling or modeling. All data management was conducted using SAS 9.4 (Cary, N C) and analyses were performed using Stata 15 (College Station, TX). Secondary Analyses In supplementary analyses, we replicated models 1 and 2 replacing continuous poverty with poverty mobility calculated based on three cut points from the interpolated poverty source data and differences in poverty quartiles across births. We did this in an effort to replicate recent findings that strong upward trajectory of quartile -based neighborhood 86 deprivation was protective for PTB using a sibling -based, or maternal effects, design (Bruckner, Kane, and Gailey 2019) . We also examine the differences in neighborhood poverty scale in an effort to in part examine the susceptibility of our data to Modifiable Areal Unit Problem (MAUP) (Openshaw and Taylor, 1979). This occurs when spatial measures are aggregated into spatial boundaries which group the population into observed associations that depend on the boundary location and scale of aggregation, introducing statistical bias. We tested for maternal race/ethnicity ef fect modification by including an interaction with race and neighborhood poverty in our FE model. While you cannot adjust for time invariant maternal characteristics in this model, such as gender or race, as they will be omitted but you can include the cov ariate as an interaction term with time or a time -varying variable . We also examine residential moving status in a stratified analysis. We perform robustness checks on our model in several ways. First, we compare our fully adjusted results to those that d o not include positively correlated monotonic covariates such as age and parity, which only increase with each birth. We then also compare our model to one that is restricted to 2000 -2012 to control for the period effects of our missing geographical data, which is more missing for the 1990 -1999 census years. Results Primary Analyses Table 3.1 describes analytic samples for the RE analysis (n= 2,199,206 births) and the FE analysis (n=279,150 births) of discordant PTB outcomes. 1,223,040 mothers (1,920,056 births) did not have discordant outcomes and were not included in the fixed effects analysis. In the RE sample, approximately 70.4% of births were to Non -Hispanic White mothers and 18.5% were to Non -Hispanic Black mothers. The maternal FE analytic sample consists of mothers with at least one PTB and thus differs from the RE sample. In the FE sampl e, approximately 56.1 % 87 of births were to NHW mothers and 31.7 were to NHB mothers. Overall, mothers in the FE sample were more likely to be NHB, and have Medicai d as an insurance pay source and more likely to have an inter -pregnancy interval of <18 month (38.7% versus 14.2%) on average. The mean neighborhood poverty rate was higher in the FE sample 18.4% (14.6 SD) compared with 14.6% (12.9 SD) in the RE sample of all births. PTB births in the FE sample were higher 41.3% compared to the RE sample 9.3%. Table 3.2 compares the odds ratios for the variable of interest, neighborhood poverty rate percentage using different model specifications (logistic, random effect , fixed effect, and random effect in the fixed effect sample) and adjustments (crude, adjusted for year and month of birth, maternal age, parity, marital status, education, infant sex, and additionally adjusted for a lagged term of prior PTB). The Hausman test results reject ed the null hypothesis that the Random Effects model is preferred in favor o f using the Fixed Effects model (X 2 P<.0001) . In all models there was a null or modest effect. In the random effects models there was a modest OR of 1.01 (95% CI: 1.07 -1.12) meaning that for a one percentage point increase in neighborhood poverty between b irths there is a 1% increase in the odds of preterm birth. The fixed effects model did not show a change in the odds for a one percenta ge point change in poverty (OR: 1.00, 95% CI: 1.00-1.00). Estimates from the RE model using the smaller FE sample were identical to those of the FE model. Secondary Analyses Table 3.3 shows the association of poverty and PTB with an interaction covariate f or poverty and maternal race/ethnicity. This adjustment allows us to model the effect of a stable maternal characteristic, i.e., race/ethnicity. Overall we did not find strong statistical interaction between neighborhood poverty rate and maternal race/ethn icity. Native American/Alaska Native maternal race/ethnicity has a slightly reduced odds of PTB for a 1 percentage point increase in 88 neighborhood poverty between births (OR: 0.99 95% CI: 0.96-1.02) but it is not significant. Asian/Pacific Islander maternal race/ethnicity had a 1% increase in the odds of PTB for a one percentage point increase in the neighborhood poverty rate between births (OR: 1.01, 95% CI: 1.00-1.01). Table 3.4 shows fixed effects analysis with a different exposure classification by using neighborhood poverty trajectory across quartiles of poverty, similar to recently published studies. Here the effect of poverty is captured over a larger change (quartile change versus a single percentage p oint change). We also exclude the birth year and birth month adjustments to allow for comparison to a recently published maternal fixed effects analysis. We see slightly stronger associations of neighborhood poverty trajectory level and PTB compared to the continuous poverty measure. There is a modest protective association with PTB of the most extreme decline in poverty quartile, strong upward mobility moving from >17% neighborhood poverty to <5% neighborhood poverty, which is associated with a decrease in odds of PTB of 7% (OR: 0.93 , 95%CI: 0.84 to 1.03). Similarly, an extreme incline in neighborhood poverty is associated with an increased in the odds for PTB (OR: 1.09, 95% CI : 0.98 to 1.20). That is, for a strong downward change in poverty quartile from t he lowest level of poverty to the highest level of neighborhood poverty the odds of PTB increased by 9% between births. Increasing maternal age lead to increased odds of PTB with women who were greater than 40 having the highest odds of PTB (OR 1.40, 95% C I: 1.28 to 1.51). Table 3.5 shows the covariates for poverty and lagged -PTB (PTB at previous birth, 0 for all first births). We demon strate the potential limitation of including the lagged -dependent covariate in a fixed effects model by comparing the esti mate to alternate models. In the logistic model, the poverty variable is unchanged but the lagged -PTB OR is significant and has a large 89 magnitude (OR: 2.88, 95% CI: 2.83 -2.93). We then control for the effect of the lagged variable equal to zero at the firs t birth by removing those births from the model and still find a strong, significant association (OR 3.03 , 95% CI:2.97 -3.09). This suggests in both these models the odds of PTB is much more related to prior PTB events than to neighborhood poverty. The sam e is true, although with a reduction in the magnitude of the odds of prior PTB for the Random Effects model. However, when we include d the lagged term as an explanatory variable in the Fixed Effects Models the Prior PTB variable had an OR: 0.4, 95% CI 0.4 - 0.5) which would suggest a strong protective effect of prior preterm birth in the association of poverty and PTB. This table shows the association between prior PTB and PTB diminishes across the model. T he problem we highlight ed he re including or not including a lagged PTB estimate in the model. The association between neighborhood poverty and PTB does not seem to be affected by controlling for prior PTB by including it in the model, but it may introduce bias due to the fixed effect s method and we chose not to include it. In our robustness checks (Table 3.6) we see no evidence that parity and age, which both increase as our number of births increase, influence the association of the model. The estimates usi ng a restricted data sample that includes only 2000 -2012 data, eliminating the 1990 census poverty variables also did not change the association. We see no significant difference in our associations when we stratify by movers and non -movers (Table 3.7). Discussion Our results show ed that using a fixed effects method shows little to modest association between neighborhood poverty and PTB. However, we also see this null or modest association in the crude logistic model and random effects models. Unlike our hypothesis, which stated that the results would be attenuated in fixed effects compared to logistic and random effects models, we 90 see a consistent null association. We modeled our exposure as a continuous 1 percentage point change in poverty. This approach used a maternal fixed effec ts analysis to examine the association between neighborhood poverty conditions and preterm birth using a large dataset that included all the births in the state of Michigan over the study period. We aimed to use this approach to reduce the problem of diffe rential selection bias into neighborhoods in the association of neighborhood poverty and birth outcomes. This approach allows control for time -invariant maternal characteristics (both measured and unmeasured) that are associated with adverse birth and neig hborhood poverty exposure. Previous studies have used this approach to examine macroeconomic conditions during preconception and adverse pregnancy outcomes in this study population (Margerison, Luo, and Li 2019) . We also report ed our results using a simila r parametrization of poverty mobility (quartile changes downward, static, and upward) for another study using California birth data using a maternal fixed effects (Bruckner, Kane, and Gailey 2019).However, our studies are not entirely comparable as they employ a lagged PTB variable while computing a fixed effects difference model , which we chose not to do given the introduction of bias . Interestingly, when we include a lagged PTB term in our models of poverty mobility ( Secondary analysis Appendix Table) we find increased odds of PTB but at all levels of mobility in both directions. This is not advisable as we hypothesize the lagged PTB covariate in the fixed effects analysis to be biased. Finally, we observed different sized odds of PTB given different scale of aggregation for the poverty variable (quartile versus continuous). This may indicate that our data is susceptible to the MAUP. Further analysis that incorporates re -zoning of the data by changing the boundary lines and adjusts for spatial autocorrelati on would be warranted in order to more definitively determine the effect of MAUP . 91 There are important limitations to the maternal fixed effects models to consider. Both the RE and FE models assume the coefficients of the same covariate remain equal across all numbers of birth. Another restriction is that the unexplained variance stays the same over time; even if an individual changes over time the error variance does not. The latent time -invariant variables must either correlated with all the covariates (F E) or be uncorrelated with all the covariates (RE). This means that we assume the latent time invariant variables are uncorrelated with the observed covariates (Bollen and Brand 2010) . Finally, the RE and FE models do not report any test statistics to allow identification of overall fit (the Hausman test just compares the two models). FE allows for latent time -invariant variables to correlate to time -varying variables but the analysis does not report any information on the magnitude of the correlations. One potential model to consider for future work would be a structural equation modeling framework using fixed effects. This is particularly salient given the prior birth history of prete rm birth. Previous models have either used a lagged -PTB in their model or shown a strong association between a prior PTB and subsequent PTB. A possible test for this association is in the RE model by seeing if RE model shows similarities to the logistic m odel estimates, after controlling for the prev ious PTB. This means that after controlling for the woman™s unobserved characteristics, as in the random effects model, the odds ratio for the previous PTB is strongly reduced from 2.83 to 1.4, with a significa nt p value (p<0.000). The magnitude of the logistic model may instead represent the omission of her time -constant characteristics the affect the probability of PTB at all of her pregnancies. Our primary analyses focused on neighborhood poverty at the tim e of birth. There is evidence that early pregnancy exposure may be the most strongly associated with birth outcomes (Margerison, Luo, and Li 2019) and that would only be captured in our analysis if the women 92 resided in the same census tract for the duratio n of her pregnancy and during the preconception period. It stands to reason that among non -movers between pregnancies we can estimate this exposure, but not for the remainder of our sample. In summary, our findings suggest that 1) increases in neighborhoo d poverty rate between pregnancies are not associated with increases in the odds of PTB, and 2) this association does not change when using a maternal fixed effect analysis to control for time -invariant and time -varying characteristics and 3 ) the use of la gged -dependent variables in one -way fixed analysis is inappropriate and introduces bias into models that are specifically designed to remove bias. 93 Table 3.1 Comparison of Samples for All Singleton Michigan Births 1990 -2012 Maternal Characteristics Random Effects Sample Fixed Effects Sample N % N % Number of Births 2,199,206 100 279,150 100 Race Non -Hispanic White 1,546,767 70.3 156,602 56.1 Non -Hispanic black 409,412 18.6 88,407 31.7 Hispanic 118,627 5.4 17,139 6.1 Non -Hispanic American Indian/ Alaska Native 5,434 0.3 410 0.2 Non -Hispanic Asian/Pacific Islander 87,988 4.0 12,533 4.5 Non -Hispanic Mixed race / other/ missing 30,978 1.4 4,059 1.5 Age <20 230,135 10.5 39,907 14.3 20-24 525,803 23.9 77,228 27.7 25-29 647,576 29.5 77,161 27.6 30-34 530,767 24.1 57,063 20.4 35-40 223,224 10.2 23,648 8.5 >40 41,701 1.9 4,143 1.5 Nativity US born 2,003,503 91.1 259,141 92.8 Foreign born 195,703 8.9 20,009 7.2 Insurance Payer Private 1,393,352 63.4 149,205 53.5 Medicaid 780,460 35.5 126,630 45.4 Self -pay 21,774 1.0 2,834 1.0 Other 3,620 0.2 481 0.2 Infant sex -male 1,126,803 51.2 144,950 51.9 Marital Status Married 1,417,529 64.5 145,067 52.0 Unmarried 781,677 35.5 134,083 48.0 Education Less than HS 59,801 2.7 8,180 2.9 Some HS 334,941 15.2 46,181 16.5 High School (HS) 732,080 33.3 94,260 33.8 Some college 555,673 25.3 66,950 24.0 College graduate 338,449 15.4 42,981 15.4 Greater than college 178,262 8.1 20,598 7.4 Parity Nulliparous (0 live births) 874,597 39.8 80,634 28.9 Primiparous (1 previous live birth) 701,943 31.9 85,595 30.7 Multiparous ( >1 previous live births) 622,666 28.3 112,921 40.5 94 Table 3.1 (cont™d) Smoke No 1,809,509 82.3 219,740 78.7 Yes 369,576 16.8 57,090 20.5 Unknown 20,121 0.9 2,320 0.8 Pregnancy Hypertension No 2,110,892 96.0 265,084 95.0 Yes 88,314 4.0 14,066 5.0 Delivery Vaginal 1,656,945 75.3 211,489 75.8 Caesarean 542,261 24.7 67,661 24.2 Birth outcomes PTB No 1,994,585 90.7 163,876.0 58.7 Yes 204,621 9.3 115,274.0 41.3 SGA No 1,959,489 89.1 239,885 85.9 Yes 239,717 10.9 39,265 14.1 LGA No 2,000,154 91.0 257,022 92.1 Yes 199,052 9.1 22,128 7.9 Inter -pregnancy Interval <18 Months No 1,886,752 85.8 107,933 61.3 Yes 312,454 14.2 68,037 38.7 Mean Gestational Age in Weeks 39.0 2.3 37.2 3.2 Mean neighborhood poverty rate 14.6 12.9 18.4 14.6 The FE sample was different from the RE sample with respect to having a higher percentage of NHB women, Medicaid payers, and greater number of women with an inter -pregnancy interval of <18 months and higher PTB and SGA outcomes. 95 Table 3.3 Maternal Fixed Effects Models of PTB with Interaction between Maternal Race/Ethnicity and Poverty FE Sample (n=279,150) OR 95% CI Maternal Race/Ethnicity*poverty Non -Hispanic White 1.00 1.00 1.00 Non -Hispanic black 1.00 1.00 1.00 Hispanic 1.00 1.00 1.01 Non -Hispanic American Indian/ Alaska Native 0.99 0.96 1.02 Non -Hispanic Asian/Pacific Islander 1.01 1.00 1.01 Non -Hispanic Mixed race / other/ missing 1.00 0.99 1.01 All analyses adjusted for year of birth, month of birth, maternal age, parity, marital status, education, and infant sex. Table 3.2 Comparison of multivariable -adjusted a associations between neighborhood poverty rate b and preterm birth (PTB) using logistic regression and maternal fixed -effects analyses among singleton births in Michigan, 1990 -2012 All Michigan Births Logistic Full Model RE FE RE using FE sample Neighbor -hood Poverty Percentage OR 95% CI OR 95% CI OR 95% CI OR 95% CI OR Model 1 Crude 1.00 1.02 1.02 1.02 1.02 1.02 1.00 1.00 1.00 1.00 1.00 1.00 Model 2 Adjusted 1 1.00 1.00 1.01 1.01 1.07 1.12 1.00 1.00 1.00 1.00 1.00 1.00 Hausman Test: Ho: Random Effects is better X2 Pr<.0001, Reject H o (Fixed Effects better model) X2 Pr<.0001, Reject H o (Fixed Effects better model) aAll analyses adjusted for year of birth, month of birth, maternal age, parity, marital status, education, and infant sex. Logistic regression models additionally adjusted for maternal race/ethnicity. bLinearly interpolated Neighborhood poverty percentage from Neighborhood Change Database. 96 Table 3.4 Maternal Fixed Effects Odds Ratios (OR) a and 95% Confidence Intervals (CI) predicting the probability of preterm birth (PTB), maternal poverty quartile mobility levels (vs. no change) and covariates OR 95% CI Mobility Strong Upward (Q4 to Q1) 0.93 0.84 1.03 Modest Upward 1.03 0.98 1.09 Weak Upward 0.99 0.96 1.02 Static (ref) 1.00 Weak Downward 0.97 0.94 1.00 Modest Downward 1.00 0.95 1.05 Strong Downward (Q1 to Q4) 1.09 0.98 1.20 Insurance Private (ref) 1.00 Medicaid 0.99 0.97 1.01 Self -pay 1.31 1.20 1.42 Other 1.16 0.95 1.40 Parity 1 birth 0.85 0.79 0.90 2 births 0.72 0.68 0.77 3 births 0.79 0.74 0.83 4 births 0.89 0.84 0.94 5 births 0.94 0.89 0.99 6+ births (ref) 1.00 Infant male 1.12 1.10 1.14 Not Married 1.05 1.02 1.09 Foreign Born 0.89 0.79 1.01 Maternal Age <20 1.07 1.04 1.10 20-24 (ref) 1.00 25-29 1.02 0.99 1.04 30-34 1.09 1.05 1.13 35-40 1.20 1.15 1.27 >40 1.39 1.28 1.51 Education Less than HS 1.03 0.96 1.11 Some HS 1.04 0.99 1.09 High School (HS) 1.00 0.96 1.05 Some college 1.01 0.97 1.06 College graduate 1.01 0.97 1.06 Greater than college 1.00 aAll odds ratios in the table are adjusted for every other variable in the model. Not adjusted as previous models for seasonal and period effects. 97 Table 3.7 Comparison of multivariable -adjusted a associations between neighborhood poverty rate b by birth and preterm birth (PTB) using logistic regression and maternal fixed -effects analyses among singleton births in Michigan, 1990 -2012 All Michigan Births Neighborhood Poverty Logistic Full Model RE FE OR 95% CI OR 95% CI OR 95% CI Movers 1.00 1.00 1.00 1.00 1.00 1.00 Non -Movers 1.00 1.00 1.00 0.99 0.99 1.00 aAll analyses adjusted for year of birth, month of birth, maternal age, parity, marital status, education, and infant sex. Logistic regression models additionally adjusted for maternal race/ethnicity. bNeighborhood poverty rates from Neighborhood Change Dat abase. Table 3.5 Comparison of Mutlivariable -Adjusted a Odds Ratios for Lagged Preterm Birth Covariate in Models (Logistic and Fixed Effects) All Births (N=279,150 births to 103,180 mothers) Logistic a RE Model a FE model b Covariates of interest OR 95% CI OR 95% CI OR 95% CI Poverty 1.00 1.00 1.00 1.01 1.00 1.01 1.00 1.00 1.00 Lagged PTB 2.88 2.83 2.93 1.41 1.36 1.46 0.04 0.04 0.05 Births >1 (N=88,325 births to 33,562 mothers) Logistic a RE Model a FE model b Covariates of interest OR 95% CI OR 95% CI OR 95% CI Poverty 1.00 1.00 1.01 1.00 1.00 1.01 1.00 1.00 1.00 Lagged PTB 3.03 2.97 3.09 2.93 2.87 3.00 0.08 0.08 0.08 a Adjusted for lagged PTB, race/ethnicity, year of birth, month of birth, maternal age, parity, marital status, education, and infant sex. b Adjusted for lagged PTB, year of birth, month of birth, maternal age, parity, marital status, education, and infant sex. Table 3.6 Table Robustness Checks for positive monotonic covariates and study time period Fixed Effects All Michigan Births No Parity or Age a N=279,150 births (103,180 mothers) Fully Adjusted b 2000-2012 N= 140,185 births (54,820 mothers) OR 95% CI OR 95% CI Neighborhood Poverty c 1.00 1.00 1.00 1.00 1.00 1.00 aAdjusted for year of birth, month of birth, marital status, education, and infant sex. bFully adjusted for year of birth, month of birth, maternal age, parity, marital status, education, and infant sex. Logistic regression models additionally adjusted for materna l race/ethnicity cNeighborhood poverty rates from Neighborhood Change Database. 98 APPENDI X 99 Table C .1 Crude Logistic Association of Neighborhood Poverty Mobility and PTB Mobility OR 95% CI Strong Upward (Q4 to Q1) 0.90 0.83 0.96 Modest Upward 0.98 0.94 1.02 Weak Upward 0.90 0.88 0.92 Static (ref) Weak Downward 0.87 0.85 0.89 Modest Downward 0.95 0.91 0.98 Strong Downward (Q1 to Q4) 1.10 1.02 1.19 Table C .2 Associations with Poverty Mobility Trajectories and PTB using a Prior PTB (lagged) dependent variable in model, Michigan Births 1990 -2012 OR 95% CI Mobility Strong Upward (Q4 to Q1) 1.2 1.1 1.4 Modest Upward 1.3 1.2 1.4 Weak Upward 1.3 1.2 1.3 Static (ref) Weak Downward 1.2 1.2 1.3 Modest Downward 1.3 1.2 1.3 Strong Downward (Q1 to Q4) 1.3 1.2 1.5 Prior PTB (lagged) Table C. 3 Comparison of multivariable -adjusted a associations between quartiles of neighborhood poverty rate b by birth and preterm birth (PTB) using logistic regression, random effects and maternal fixed -effects analyses among singleton births in Michigan, 1990-2012 All Michigan Births Quartiles of Neighborhood Poverty 1 Logistic Full Model RE FE RE using FE sample OR 95% CI OR 95% CI OR 95% CI OR 95% CI Q2 1.1 1.1 1.1 1.1 1.1 1.1 1.0 1.0 1.0 1.0 1.0 1.0 Q3 1.1 1.1 1.1 1.1 1.1 1.2 1.0 1.0 1.1 1.0 1.0 1.1 Q4 1.2 1.2 1.2 1.2 1.2 1.2 1.0 1.0 1.1 1.0 1.0 1.0 a Q1 referent. All analyses adjusted for year of birth, month of birth, maternal age, parity, marital status, education, and infant sex. Logistic and Random Effects regression models additionally adjusted for maternal race/ethnicity. bLinearly interpolated Neighborhood poverty percentage from Neighborhood Change Database. 100 Table C. 4 Crude FE Model Response Patterns Response Patterns PTB Strata Frequency 1 0 1 0 1 640,461 640,461 2 1 0 68,715 68,715 3 0 2 342,839 685,678 4 1 1 54,974 109,948 5 2 0 7,946 15,892 6 0 3 111,446 334,338 7 1 2 25,185 75,555 8 2 1 5,267 15,801 9 3 0 947 2,841 10 0 4 26,293 105,172 11 1 3 8,017 32,068 12 2 2 2,352 9,408 13 3 1 591 2,364 14 4 0 136 544 15 0 5 5,454 27,270 16 1 4 2,385 11,925 17 2 3 817 4,085 18 3 2 283 1,415 19 4 1 109 545 20 5 0 28 140 21 0 6 1,462 8,772 22 1 5 689 4,134 23 2 4 341 2,046 24 3 3 119 714 25 4 2 44 264 26 5 1 19 114 27 6 0 4 24 28 0 7 439 3,073 29 1 6 220 1,540 30 2 5 115 805 31 3 4 42 294 32 4 3 19 133 33 5 2 11 77 34 6 1 6 42 35 7 0 1 7 36 0 8 156 1,248 37 1 7 81 648 38 2 6 50 400 39 3 5 27 216 40 4 4 10 80 41 5 3 6 48 101 Table C.4 (cont™d) 42 6 2 3 24 43 7 1 1 8 44 0 9 55 495 45 1 8 23 207 46 2 7 15 135 47 3 6 6 54 48 4 5 4 36 49 5 4 5 45 50 6 3 2 18 51 7 2 1 9 52 8 1 1 9 53 0 10 18 180 54 1 9 9 90 55 2 8 5 50 56 3 7 3 30 57 4 6 2 20 58 5 5 2 20 59 6 4 1 10 60 0 11 10 110 61 1 10 5 55 62 2 9 1 11 63 3 8 3 33 64 10 1 1 11 65 0 12 3 36 66 1 11 1 12 67 0 14 1 14 68 1 15 1 16 102 REFERENCES 103 REFERENCES Allison, Paul D. 2005. Fixed Effects Regression Methods for Longitudinal Data Using SAS . Sas Institute. https://books.google.com/books?hl=en&lr=&id=OIPExEh -tcMC &oi=fnd&pg=PA1&dq=fixed+effects+regression+methods+for+longitudinal+data+using+ sas&ots=HpFkcTQAph&sig=eMEB4_irtoTbbRUw_Fd8MI_5_gE . Bollen, Kenneth A., and Jennie E. Brand. 2010. fiA GENERAL PANEL MODEL WITH RANDOM AND FIXED EFFECTS: A STRUCTURAL EQUATIONS APPROACH.fl Social Forces; a Scientific Medium of Social Study and Interpretation 89 (1): 1 Œ34. https://doi.org/10.1353/sof.2010.0072 . Bruckner, Tim A., Jennifer B. Kane, and Samantha Gailey. 2019. fiStrong Upward Neighborhood Mobility and Preterm Birth: A Matched -Sibling Design Approach.fl Annals of Epidemiology 36 (August): 48 -54.e1. https://doi.org/10.1016/j.annepidem.2019.05.005 . Collins, James W., Allison Mariani, and Kristin Rankin. 2018. fiAfrican -American Women™s Upward Economic Mobility and S mall for Gestational Age Births: A Population -Based Study.fl Maternal and Child Health Journal 22 (8): 1183 Œ89. https://doi.org/10.1007/s10995 -018-2503-4. Collins, James W., Kristin M. Rankin, and R ichard J. David. 2015. fiDownward Economic Mobility and Preterm Birth: An Exploratory Study of Chicago -Born Upper Class White Mothers.fl Maternal and Child Health Journal 19 (7): 1601 Œ7. https://doi.o rg/10.1007/s10995 -015-1670-9. Culhane, Jennifer F., and Robert L. Goldenberg. 2011. fiRacial Disparities in Preterm Birth.fl Seminars in Perinatology 35 (4): 234 Œ39. https://doi.org/10.1053/j.sem peri.2011.02.020 . Kramer, Michael R. 2016. fiRace, Place, and Space: Ecosocial Theory and Spatiotemporal Patterns of Pregnancy Outcomes.fl In Recapturing Space: New Middle -Range Theory in Spatial Demography , 275Œ299. Springer. http://link.springer.com/chapter/10.1007/978 -3-319 -22810-5_14. Leventhal, T., and J. Brooks -Gunn. 2000. fiThe Neighborhoods They Live in: The Effects of Neighborhood Residence on Child and Adolescent Outcomes .fl Psychological Bulletin 126 (2): 309Œ37. https://doi.org/10.1037/0033 -2909.126.2.309 . Margerison, Claire E., Zhehui Luo, and Yu Li. 2019. fiEconomic Conditions during Pregnancy and Preterm Birth : A Maternal Fixed -Effects Analysis.fl Paediatric and Perinatal Epidemiology 33 (2): 154 Œ61. https://doi.org/10.1111/ppe.12534 . Nickell, Stephen. 1981. fiBiases in Dynamic Models with Fixed Effects.fl Econome trica 49 (6): 1417Œ26. https://doi.org/10.2307/1911408 . 104 Openshaw, Stan, and Peter J. Taylor. 1979. fiA Million or so Correlation Coefficients: Three Experiments on the Modifiable Areal Unit Problem.fl Statistic al Applications in the Spatial Sciences21: 127 Œ144. Pearl, Michelle, Jennifer Ahern, Alan Hubbard, Barbara Laraia, Bina Patel Shrimali, Victor Poon, and Martin Kharrazi. 2018. fiLife -Course Neighbourhood Opportunity and Racial -Ethnic Disparities in Risk of Preterm Birth.fl Paediatric and Perinatal Epidemiology 32 (5): 412 Œ19. https://doi.org/10.1111/ppe.12482 . Sampson, Robe rt J. 2013. Great American City: Chicago and the Enduring Neighborhood Effect . Reprint edition. Chicago; London: University of Chicago Press. 105 CONCLUSION Our studies examine d the residential movement, poverty mobility, and maternal fixed effects in associations with PTB using a maternally -linked sample of singleton births in Michigan during the period 1990 -2012. We hypothesized that movement would change neighborhood compositi on and lead to misclassification bias in the association between poverty and PTB. However, we observed that most women do not change poverty levels, even those who change residence. There are extreme movers across levels of poverty, but they represent a se lect minority of our study population. The static trajectory group encompasses the most women, although at varying poverty levels from low to high poverty . We find there is a modest association between downward neighborhood poverty trajectory and increased PTB. However, there may be unobserved maternal characteristics that lead a woman to select her neighborhood that may also be associated with her risk of P TB. We therefore examine the association with neighborhood poverty and PTB using a maternal fixed effect approach that allows for control of both measure and unmeasured maternal characteristics. We find a null association. We highlight the need for careful selection of covariates in this model by showing how a prior PTB control may introduce bias . This does raise concerns for how much a prior PTB itself is a risk versus due to the mothers own characteristics. Future models to consider include dynamic panel models and structural equations modeling with fixed effects which can better account for both the maternal characteristics and the prior PTB events (as a time -varying covariate) (Gunasekara et al. 2014) . Our study used neighborhood poverty as an exposure v ariable because we hypothesized it would have a strong association with PTB in conventional models and because it has been used in previous research. However, we did not observe a strong association. This may be due to our time period of study, a mother™s adulthood during her period of fertility. Previous work that 106 observed strong associations did so over the life course from childhood to adulthood (Collins, Rankin, and David 2015; Pearl et al. 2018) . In our study we are not able to do that for the mothers. However, this work could be foundational to a study th at examines the birth outcomes of the infants in this study. Such inter -generational linkage has been previously done in select populations (Spencer 2004; Collins, Rankin, and David 2011) . This would afford an opportunity for a state wide study and Michigan is an ideal state to capture this information as there is limited out-migration. Previous researc h also found the strongest associations between neighborhood disadvantage and adverse birth events among mothers who had adverse births as infants (Collins, Rankin, and David 2011 ). This suggests a component of heritability in the association, and one of the reasons a fixed effects analysis may be most appropriate. Another way to test this would be to examine epigenetic effects among the future inter -generational sample. One challenge of modeling large data stems from the absence of changes in the significa nce level of covariates. These significance levels are typically used in model building and evaluation to inform the model design. However, when working with a large sample size there is ample power to make most covariates significant. We relied on covaria tes chosen based on our a priori hypothesis and reference groups but additional covariates or different reference groups may be warranted for future analysis. Finally, in our study population we report high levels of poverty that increased over the study period. In a study examining birth outcomes , we would be remiss to not remark that not only are these infants born into poverty but they may experience limited mobility out of poverty without intervention. Alarmingly , we also observe increasing rates of po verty over our s tudy period. We hypothesize that while the mothers may not experience a critical window during our 107 study period, the data captured in our studies reflects a critical window for the infants whose birth outcomes we include. We see our current work as foundational and f uture research efforts would be wise to focus on this group and make every effort to work towards appropriate interventions in the reduction of poverty conditions. 108 REFERENCES 109 REFERENCES Collins, James W., Kristin M. Rankin, and Richard J. David. 2015. fiDownward Economic Mobility and Preterm Birth: An Exploratory Study of Chicago -Born Upper Class White Mothers.fl Maternal and Child Health Journal 19 (7): 1601 Œ7. https://doi.org/10.1007/s10995 -015-1670-9. Gunasekara, Fiona Imlach, Ken Richardson, Kristie Carter, and Tony Blakely. 2014. fiFixed Effects Analysis of Repeated Measures Data.fl International Journal of Epidemiology43 (1): 264Œ69. https://doi.org/10.1093/ije/dyt221 . Pearl, Michelle, Jennifer Ahern, Alan Hubbard, Barbara Laraia, Bina Patel Shrimali, Victor Poon, and Ma rtin Kharrazi. 2018. fiLife -Course Neighbourhood Opportunity a nd Racial -Ethnic Disparities in Risk of Preterm Birth.fl Paediatric and Perinatal Epidemiology 32 (5): 412 Œ19. https://doi.org/10.1111/ppe.12482.