THESOS \ ‘ s . I cyl- ‘o I" II Ir 7 .- t y” This is to certify that the thesis entitled Lead Poisoning in Young Children: Determining Risk Factors and Exposure Sources - An Environmental Justice Approach presented by Sean Frost has been accepted towards fulfillment of the requirements for the M.A. degree in Sociologx Am: tail—JV ' Major rrgt'esior’s Signature Li a! (.x LL l I I Date MSU is an Affinnative ActiorVEquaI Opportunity Institution LIBRARY Michigan State University PLACE IN RETURN Box to remove this checkout from your record. TO AVOID FINES return on or before date due. MAY BE RECALLED with earlier due date if requested. DATE DUE DATE DUE DATE DUE 6/01 c:/CIRC/DateDue.p65-p.15 LEAD POISONING IN YOUNG CHILDREN: DETERMINING RISK FACTORS AND EXPOSURE SOURCES — AN ENVIRONMENTAL JUSTICE APPROACH By Sean William Frost A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of MASTER OF ARTS Department of Sociology 2004 ABSTRACT LEAD POISONING IN YOUNG CHILDREN: DETERMINING RISK FACTORS AND EXPOSURE SOURCES — AN ENVIRONMENTAL JUSTICE APPROACH By Sean William Frost In 2001 , data from the Michigan Department of Community Health, Childhood Lead Poisoning Prevention Project Showed that nearly 4,800 or 5.5% of those children tested, exhibited elevated blood lead levels (EBLL). EBLL is defined by the medical community as blood lead level (BLL) being at or above 10 micrograms per deciliter (> 10 ug/dL). The purpose of this study was to evaluate the predictiveness of the elements Of the existing statewide screening program, as well as to identify other non-invasive measures which can best predict who is at greatest risk for elevated blood lead levels (EBLL) as well as define which geographic/environmental, socioeconomic, and behavioral characteristics put some children at more risk than others. An eighteen item survey was administered to the caregivers of 4,1 77 children younger than age 6 that were being tested for lead poisoning at pediatric offices and WIC departments across the state of Michigan. A number of statistical techniques were applied to the data, including sensitivity and specificity tests, OLS regression, and logistic regression. The research presented here found that the best predictors for EBLL included the presence of lead pipes in the home, low household income, proximity to others with EBLL, the presence of peeling paint, child’s race, and location of residence. The inclusion of Census Block Group data in future research will greatly increase the ability to accurately predict those children more at risk for EBLL. This thesis is dedicated to my wife, Becca, and my children, for their continuing support and never-ending encouragement. The inspiration I receive from them is immeasurable and irreplaceable. iii TABLE OF CONTENTS LIST OF TABLES v INTRODUCTION Current State of Lead Poisoning In Michigan 1 Health Concerns Related To Lead Poisoning THEORY An Environmental Justice Approach to the Problem 4 Environmental Justice, Race, and Lead Poisoning 4 ENVIRONMENTAL SOURCES AND SOCIODEMOGRAPHIC CORRELATES Lead Exposure Sources 6 Sociodemographic Correlates Of B11 7 NEEDS FOR RESEARCH ON LEAD POISONING Current Statewide Screening Measures 10 Reasons for Continuing Research on Lead Poisoning 12 Project Goals 13 RESEARCH METHODS 15 RESEARCH HYPOTHESES 17 SAMPLING, DATA COLLECTION, AND DESCRIPTIVES Sample Selection and Data Collection Process 19 Sample Population Descriptive Characteristics 21 DATA ANALYSIS Data Screening Techniques 26 Data Analysis I 28 Lead Poisoning: Predictive Model Building 35 Logistic Regression Analysis of Lead Poisoning 40 CONCLUSION Discussion 45 Future Research Needs 47 APPENDICES 51 REFERENCES 58 iv LIST OF TABLES Number of children tested, Rates of EBLL, number of children with EBLL Blood Lead Level Results — Survey Respondents Race/Ethnicity of Respondents Age of Respondents Skewness and Kurtosis of non-transformed and transformed BLL Sensitivity and Specificity of predictor variables for BLL >= 10 ug/dL OLS Regression — Independent variables in Research Hypotheses Comparison of regression models Regression Model 2: Coefficients, Significance, and Unique variance explained Logistic regression coefficients, Wald, and Odds-ratio with 95% CI 2 21 23 24 28 32 37 43 Introduction The Current State of Lead Poisoning in Michigan In 2001, data from the Michigan Department of Community Health, Childhood Lead Poisoning Prevention Project showed that nearly 4,800 children younger than age six had elevated blood lead levels (EBLL). The Centers for Disease Control and Prevention (CDC) has had an active role in determining the toxicity levels of lead. “Until the late 1960’s acceptable upper limits in blood lead levels were 80 ug/dL (micrograms per deciliter) for adults and 60 ug/dL for children” (Phelps 1999, p 478). The US. Surgeon General, in 1971, reduced the level considered toxic to 40 ug/dL, as well as encouraged “the identification and remediation of exposure sources” (Ibid, p. 478). Finally, in 1991 the CDC reduced the level of lead toxicity, measured in micrograms of lead per deciliter of blood (ug/dL), from 25 ug/dL to 10 ug/dL. Furthermore, the same guidelines from the CDC “recommended that all children be tested for blood lead level at 12 months of age and again, if possible, at 24 mon ” (Kazal 1997, p 515). Since that time thresholds as low as 2.5 ug/dL have been considered, by some, to be toxic levels, though this has been a hotly debated issue. Implementation of a universal screening measure was recommended at regular intervals between the ages of 6 and 72 months. A five-item questionnaire was used to gauge a child’s risk for lead poisoning and a single "yes” response would signal that child as ‘high risk’ and warrant testing that child’s blood for lead (CDC 1991). In spite of these recommendations from the CDC, of the nearly 815,000 children under age six residing in Michigan, only 87,875, or 11%, were tested for lead poisoning. SO, while only 5.5% of children tested in Michigan had elevated blood lead levels (BLL), the current level of lead poisoning among children in Michigan is troubling. The Third National Health and Nutrition Examination Survey (NHANES 111), Phase 11 identified a national average of 4.4% testing positive for lead poisoning. This puts Michigan above the national average. Since 1997, the state health department has been collecting test results for lead on every child in Michigan. In many cases a capillary, or fingerstick, blood sample is drawn, however a more reliable method is a venipuncture blood draw. The drawback of using capillary testing is that they often result in false positive findings, and therefore any elevated capillary test should be confirmed by a venous test. (Farmer 2001; Kaplowitz, et al 2000) However, it is highly likely that an elevated capillary blood lead level represents a child affected by lead poisoning, even if the micrograms per deciliter is slightly lower than recorded. Therefore, the analysis and findings of this research will include the rates of EBLL for all children tested, regardless of the type of blood draw administered. With this compilation of statewide data, an examination of the recent trends in rates of lead poisoning and testing is possible. Table 1: Number of children tested, Rates of EBLL, number of children with EBLL“ 1998 1999 2000 2001 Number of children < 6 tested 73862 78876 78043 87920 % children < 6 with EBLL (BLL > 10 ug/dL) 13.6% 9.96% 7.88% 7.36% Number of children < 6 with EBLL 10045 7856 6149 6470 "‘ Source: Michigan Department of Community Health, Childhood Lead Poisoning Prevention Project From a cursory examination of the proportion of children in Michigan under age 6 it appears as if the proportion of children with EBLL is a diminishing problem. However, an increase of over 12% in the number of children tested explains the decreased proportion with EBLL. Upon closer examination the data shows that the actual number of children with EBLL increased slightly from 2000 to 2001. While this is not cut and dried evidence that there is an epidemic in Michigan of lead poisoning, at the very least it provides an impetus for finding more reliable identifiers to predict those at risk for elevated levels of blood lead. Health Concerns Related to Lead Poisoning Why are we concerned about young children having lead poisoning anyway? Age is a primary concern because young children’s bodies are more apt to absorb greater levels of lead than are adults or even older children “due to a higher prevalence of iron deficiency” (Farmer 2001, p. 2). Furthermore, in addition to a greater absorption rate children are likely to have more contact with lead contaminants and have greater organ sensitivity (Jackson, Cummins, Tips, and Rosenblum 1998). With this increased absorption rate comes a host of deleterious health consequences. Low-level lead exposure (< 10 ug/dL) can create a danger to the developing nervous systems in young children, as well as impair hearing and reduce stature (Mielke 1999; Jacobs et a1 2002). Lead poisoned children (those over 10 > ug/dL) can incur learning disabilities and behavioral problems, including Attention Deficit Disorder and juvenile delinquency (Farmer 2001). But perhaps the most serious health consequence of living with lead poisoning for children is a decrease in intelligence. Different studies have disagreed about the amount of decrease in IQ, but they have all agreed that there is an associated decrease. It is generally agreed upon that the decrease in IQ is estimated between 2- to 7- points for each 10 ug/dL (Lamphear et a1 1996; Farmer 2001; Phelps 1999). For these reasons it is clear that lead poisoning is still an important public health issue. Theory An Environmental Justice Approach to the Problem As will become clear from a discussion of the various risk factors and exposure sources of lead, an environmental justice framework is helpful when trying to understand the problem of lead poisoning in children. An environmental justice perspective on public health and/or social issues is not a new approach. It has been nearly twenty-five years since an African-American community in Houston filed suit against Southwestern Waste Management, to protest the selection of a site in their community for a solid waste landfill (Lee 2002). This case, with the national attention that it received, facilitated a growing “awareness about environmental conditions in low-income, people of color, and tribal communities” (Lee 2002, p142). It is these groups of people, those at greater disadvantage in society, that suffer more of the consequences of environmental degradation. The lack of access to adequate health care, safe housing, and other physical, social, cultural, and spiritual factors have detrimental affects on the health of members of a community, both individually and collectively (Lee 2002). Environmental Justice. Race, 4nd Lead Poisoning According to the Census Bureau, Black children are at least three times more likely than their White counterparts to live in poverty (Census Bureau 2000; Weintraub 1997). The poverty rates vary greatly, with Whites having a 7.6% poverty rate while Blacks have a poverty rate nearing 23%. Furthermore, the inadequate housing and poorer living conditions of Blacks and the poor increase their risk level. Since the connection between old housing, lead-based paint, and increased elevated blood lead levels have been empirically established, the poor and minorities are placed at the highest risk. Not only does this population lack the necessary financial resources to afford basic healthcare they are further disadvantaged by their exposure to environmental risk factors. These environmental risk factors include the age and location of housing, proximity to former industrial development (i.e. inner cities — former site of manufacturing before much of this industry moved to the suburbs and/or overseas and to Mexico), proximity to high- traffic roads and more. In short, this condition seems to be a much greater risk for the low-income population, especially those without access to good housing and medical insurance. Environmental Sources and Sociodemographic Correlates Lead Exposure Sources With such awareness of the health consequences from lead poisoning it is somewhat surprising that our society continues to have such a high rate of children with EBLL. In the 1990’s the CDC called lead poisoning “one of the most common pediatric health problems in the United States today... [and added that the problem was]...entirely preventable” (Mielke 1999, p. 62). We must then illuminate the potential sources of lead which children are exposed to. The classification of various risk factors (aka sources of lead) can be conflated to three categories: environmental, behavioral/cultural, and socioeconomic; though these three categories should not be considered mutually exclusive from one another. The primary source of contamination comes from exposure to various environmental sources of lead. The most frequently cited, and well documented, environmental source of lead that children are most likely to be exposed to is lead-based paint. The allowable amount of lead in paint was reduced in 1950 and completely banned in 1978. This has not, however, eliminated the continuing exposure to this type of paint. Given the long history of using lead-based paint there are still many people living in residences which have lead-based paint. Markowitz and Rosner, in their book Deceit and Denial (2002), estimated that as recently as the 1950’s a can of interior house paint was half lead, containing an estimated 12 — 15 pounds of lead. Obviously, many of the homes painted with this high content of lead still exist, largely in inner cities, and without extensive abatement programs to remove lead from homes many children continue to be exposed through a source declared illegal over 25 years ago. Another type of ingestion, that of paint chips and/or the dust from lead paint, is also a significant cause of elevated blood lead levels (Lamphear et al 1996; Jacobs et a1 2002; Mielke 1999; Farmer 2001; Lamphear, Weitzman and Eberly 1996). Another contamination point from environmental sources is the ingestion of soil containing lead. There are multiple sources fi'om which soil has historically become contaminated with lead including peeling paint, industrial emissions, as well as leaded gasoline emissions. Living in areas near current or former industrial sites which contribute significant amount of lead emissions to the environment can also be a source of lead poisoning in children. As with the dust or chips from lead paint, the lead emitted into the air from industrial production settles in the soil in nearby areas. These industrial sites are usually in urban areas, and are often in close proximity to areas where minorities are concentrated (Weintraub 1997). Sociodemographic Correlactes of BLL In conjunction with these environmental sources of lead, the socioeconomic status and ascribed characteristics of children contribute significantly to the amount of exposure to lead that they incur. Manheirner and Sibergeld (1998) demonstrated that children from low-income families have considerably higher rates of EBLL than do children from high- income families. When you consider where it is that low-income families are likely to live this makes sense. There is considerable evidence showing that lead exposure from lead-based paint hazards in older housing is a common source of lead poisoning today (Bomschein et al. 1987; Clark et al. 1991; Jacobs 1995; Lanphear et al. 1995, 1998; Lanphear and Roghmann 1997; McElvaine et al. 1992; Rabinowitz et al. 1985; Shannon and Graef 1992). In a study by David Jacobs et al (2002) regarding the prevalence Of lead-based paint hazards, they found that “among low-income households, 35% of the units had lead-based paint hazards, compared with 19% of units among households with middle and upper incomes” (p 601). This puts children of low-income families at greater risk for exposure to lead, due in part to the prevalence of lead-based paint that exists in older homes. In addition to the socioeconomic status, the race of a child seems to matter when it comes to levels of blood lead poisoning. Evidence shows that Black children are more at risk than are White children for lead poisoning (Mahaffey et al. 1982; Brody et al. 1994). Additionally, in a study conducted by Lanphear, Weitzman and Eberly (1996), the mean blood lead level was nearly twice as high among Black children as it was for White children (p = .001), with a mean of 8.8 ug/dL for Blacks and a mean of only 4.7 ug/dL for White children. The authors of this study indicated that “housing condition and exposure to lead-contaminated house dust appear to be major contributors to the racial disparity in children’s blood lead levels” (Lanphear et al., p 1462). The authors concluded that the racial disparity in blood lead levels found could possibly be explained by White children’s homes being in ‘better condition’ as well as different mouthing behaviors of White and Black children (ibid, p 1462). The authors state that “Black children were more likely to put their mouths on window sills”, though they were unable “to determine if these behaviors were culturally or environmentally driven” (ibid, p 1462). Weintraub (1997) found that African-Americans were much more likely to live in pre-1950’s housing, and this supports the findings of Lanphear et al, in as much as the housing conditions contribute more to African-American children’s EBLL than it does for White children. Behavioral and/or cultural attributes contribute to lead poisoning risk as well. While ultimately all risk of lead poisoning can only stem from exposure to environmental risk, one’s behaviors and customs can increase some children’s risk. For instance, the absence of breast-feeding has been speculated to increase risk (Kaplowitz et al. 2000). This is likely due to the presence of lead in the water, which under these circumstances is often used to make formula in lieu of breast-feeding. Black children are less likely to be breast-fed (Kaplowitz et al. 2000), and this again increases their risk of lead poisoning. Similarly, the use of a pacifier could prevent a child from eating paint chips or exhibiting other hand-to-mouth activities, thereby limiting the ingestion of lead. Children can come into contact with lead in other manners also. One of the less frequent, but still relevant, contact points is through the use of certain cosmetics and home medical remedies that contain lead or lead by-products (MDCH 1999). Likewise, travel outside the United States to countries that use home remedies and cosmetics which contain lead, can increase a child’s exposure to lead. Additionally, many countries around the world have high levels of environmental risk factors. For those children that don’t travel out of the US. nor use any home remedies or cosmetics containing lead, there are still additional exposure risks. Often times lead can be brought into the home environment by other members of the family (parents, older siblings, etc.) who come into contact with lead in their jobs or lead-based hobbies (MDCH 1999; Kaplowitz et al. 2002; Farmer 2001). Needs for Research on Lead Poisoning Current Statewide ScreeningMeasures The Michigan Department of Community Health, Childhood Lead Poisoning Prevention Program (MDCH-CLPPP) formulated a set of statewide screening recommendations in June 1999. Screening, in this context, is defined as the non- invasive, pre—testing procedures for determining which children are more likely to have, or at least be at increased risk for, elevated blood lead levels. This allows for better targeted testing of those children at risk, without spending unnecessarily for blood testing those children not at risk. Following recommendations and guidelines issued by the CDC, as well as existing Federal and State government mandates; the state plan was created and disseminated to many parties. Health care professionals, “local health departments, managed care organizations, laboratory directors, rental property owners” (Kaplowitz et al. 2000), and others concerned with lead poisoning detection and prevention were made aware of the new guidelines and encouraged to implement the screening measures. The screening plan includes four criteria in descending priority order. They are Medicaid status, geography, questionnaire, and physician’s discretion. Medicaid requires testing of all one and two year old children that are enrolled in the program, as well as those children between the ages of 36 — 72 months who have not previously been tested. This is a Federal and State mandate. Since a large number Of these children are seen by local health departments, it has become a standard part of protocol at many County Health Departments to regularly test children for blood lead level as part of the Women, Infants, and Children (WIC) program. 10 Geography as a screening measure is clearly an area that needs to be refined. The CLPPP screening plan indicates mandatory testing for all non-Medicaid covered children that live within certain high-risk ZIP code areas to be tested. The definition of a “high- risk ZIP” is defined as follows: . ZIP codes with 12% or greater incidence of lead poisoning among children between the ages of twelve and thirty-six months. . ZIP codes with 27% or greater pre-1950 housing. . ZIP codes with a combination of a high percentage of pre-1950 housing, a high number of children under age six and a high percentage of children under age 6 living in poverty. Given this definition of “high-risk” nearly 50% of the 1170 ZIP codes in Michigan are classified as high-risk. The utility of ZIP code as a screening measure will be discussed later. The MDCH Childhood Lead Poisoning Questionnaire is the third component of the statewide screening plan. This questionnaire contains five questions (though several of these are somewhat double-barreled) and any single ‘yes’ or ‘don’t know’ response would warrant a physician or other health care worker classifying that child as ‘high—risk’ and administering a blood lead test. The five questions are as follows: a Does the child now or in the recent past live in or often visit a house built before 1950 with peeling or chipping paint? This could include day care, preschool, or home of a relative. 0 Does the child now or in the recent past live in or often visit a house built before 1978 that has been remodeled with the last year? 0 Does the child have a brother or sister (or playmate) with lead poisoning? 0 Does the child live with an adult whose job or hobby involves lead? (list provided) 11 0 Does the child’s family use any home remedies that may contain lead? (list provided) The last component of the screening plan is physician discretion. Even if a child meets none of the above criteria, the physician always has the option of determining that a particular child Should be tested if he or she believes it is warranted. This decision may stem from a child presenting symptoms consistent with lead poisoning or some other known exposure or risk factor. Even with all of these guidelines in place, especially that of over 50% of ZIP codes being classified as high-risk, only 11% of children were tested last year in Michigan. It is these components of the current screening plan, in addition to other factors deemed theoretically necessary to include, that will be tested in the data analysis section. Reasons for Continuing Research on Lead Poisoning The current literature on screening for detection of ‘high-risk’ children has yet to create a successful model whereby we can accurately predict which children are more at risk. According to Kaplowitz, et al (2001) Sensitivity (the ability to predict which individuals have EBLL) and Specificity (the ability to predict which individuals do not have EBLL) of specific questionnaires is the most frequent method of evaluating the existing screening measures. Sensitivity is computed by dividing the number of people for which the screening measure predicts to have EBLL by the number of people that actually have EBLL; in other words, the true positive rate. Specificity, on the other hand, is computed by dividing the number of people the screening measure predicts to not have EBLL by the number of people that actually do not have EBLL; the true negative rate. A recent article by Binns, LeBaily, Fangar, and Saunders (1999) reports on the Sensitivity 12 and Specificity of seven different screening studies where the median value for both Sensitivity (66%) and Specificity (67%) were not much better than chance. Though some studies have reported higher levels of Sensitivity (87%) by Tejada et al. (1994) and Specificity (80%) by Snyder et al. (1995) these results are based on small sample sizes and considered to be low in reliability and validity. Therefore it is clear that the existing screening measures are inadequate and can be improved. By improving the predictive capacity of instruments used to better target those most at risk, a higher rate of testing is assured, as well as a decrease in wasted resources through the elimination of unnecessary testing. There are other reasons for continuing research in this area as well. First, the continuing presence of affected children at consistent rates of 5 to 7% of those children tested is troubling and not indicative of a condition that is under control or on the decline. Secondly, while we as a society have effectively curtailed some of the risk factors for children, many environmental and behavioral issues which put children at risk continue to go unaddressed in large segments of our society. Proiect Goals This paper stems from a two-year research project performed by the MSU-BLL group (S Kaplowitz, P1) in collaboration with the Childhood Lead Poisoning Prevention Programs. The project was entitled Part C: Supplemental Studies “An Evaluation of the Predictiveness of the Elements of the Michigan Statewide Screening Plan and Other Factors”, funded by the Michigan Department of Community Health as a sub-grantee Of a CDC grant. For that reason the goals of that study are worth mentioning here. 13 The primary purpose of the project was to evaluate the predictiveness of the elements of the existing statewide screening program and other associated factors. Furthermore, the study proposed to examine the efficacy of assessing the risks associated with housing, poverty and the local environment by comparing the predictive value of ZIP codes versus census tracts and census block groups. The adequacy of brief questionnaires as screening devices to assess personal risk factors was also to be examined. Ultimately the project goal was to make recommendations regarding screening and testing children for lead poisoning. Through analysis of survey data and limited geographic information, the goal was to generate a ‘score’ that includes each child’s level of risk they are at for lead poisoning. This will allow physicians and other health care providers to allocate scarce resources toward testing those children at greatest risk. That then is the purpose of this paper; to identify non-invasive measures which can best predict who is at greatest risk for elevated blood lead levels (EBLL) as well as define which geographic/environmental, socioeconomic, and behavioral characteristics put some children at more risk than others. 14 Research Methods In order to test the predictiveness of the current statewide screening plan in Michigan a four-page survey was constructed. Survey construction was completed through a thorough examination of this screening plan and other factors determined through an extensive literature review. The Medicaid enrollment and “high risk” ZIP code were included based on self-reported data by the parent’s of children enrolled in the survey project. By selecting Medicaid as the child’s form of insurance coverage in survey question 16, the variable MDCH 6 was created with those children covered by Medicaid being given a value of one and all others being given a value of zero. The “high risk” ZIP code (ZIPRISK) was obtained by matching the address recorded by the parent with the self reported ZIP code, assuring it’s’ accuracy, and then matching with a list of “high risk” ZIP codes obtained from the MDCH. The five screening questions used in the Childhood Lead Poisoning Questionnaire were somewhat complex, so they were broken down into multiple survey items allowing for more precise measurement of exactly which part of those questions, if any, increased the variance explained. This consisted of using a matrix for housing characteristics of current and previous residences, separately asking about any day care facilities with peeling paint, asking separate questions about siblings and playmates with elevated blood lead levels, separate questions about adult jobs and hobbies, and a question about the use of home remedies or cosmetics which are known to contain lead. In addition to the questions discussed above, based on the MDCH questions, the survey construction was completed by considering other environmental, behavioral, and socioeconomic factors that have been linked to elevated blood lead levels by other 15 researchers. The screening questionnaire used by the Henry Ford Health System (HF HS) was analyzed and used in conjunction with the MDCH questions, as a starting point for the survey. By combining the HFHS and MDCH questionnaires, as well as including other relevant environmental, behavioral and socioeconomic factors, the survey presented in Appendix B was created. This survey consists of 18 questions, all of which were considered when designing regression models used in analysis and discussed later. 16 Research Hypotheses Thus far this researcher has elucidated not only the existing screening measures for lead poisoning, but other environmental, behavioral/cultural, and socioeconomic predictors and/or indicators which increase children’s risk of elevated blood lead levels. The discussion and literature review regarding lead exposure sources, behavioral characteristics which increase risk of lead poisoning and socioeconomic indicators of who is at risk for lead poisoning lead to a set of hypotheses around the ability to predict one’s risk for lead poisoning. These hypotheses can be divided into three components: environmental, behavioral/cultural, and socioeconomic. While these categories should not be considered mutually exclusive, for elucidation of the predictive ability of different variables, they must be considered separately from one another. It is clear that where a child lives can increase their risk of lead poisoning visa vi their exposure to environmental sources. This leads to the following hypotheses: 1. Children residing in areas defined as ‘high-risk’ ZIP codes by the MDCH will be more at risk for elevated blood lead levels and therefore, on average, have higher blood lead levels than children living in ZIP codes not classified as ‘high-risk’. 2. Children who live in or visit old houses with peeling paint are more likely to have EBLL than those that do not. 3. Children who live in housing where the water is supplied through lead pipes will have higher BLL than those reporting the absence of lead pipes. The next set of hypotheses proposes that certain behavioral and/or cultural practices increase risk for EBLL: 1. The absence of breast feeding increases a child’s risk for EBLL, therefore children whose parent’s report not breast feeding their child will have higher BLL. 2. Children exposed to certain cosmetics or home remedies that contain lead will have higher BLL. 3. Contact with adults who have hobbies or jobs in which they deal with lead, or lead by-products, will increase a child’s risk of EBLL. l7 The third set of hypotheses proposes that there are non-causal indicators which can predict who will be more at risk for elevated blood lead levels: 1. Any child having contact with a sibling, adult or playmate who has been diagnosed with EBLL will be more likely themselves to have EBLL. 2. Black children are more likely to be exposed to lead and, therefore, will have higher BLL, on average, than White children. The fourth and final set of hypotheses proposes that there are indicators of poverty, which either taken alone or when combined with old housing, increase risk for EBLL: 1. As the income level of a child’s family goes up the probability of a child having EBLL decreases. 2. Children who are covered by State or Federal insurance programs like Medicaid are more at risk to have EBLL than are children covered by private HMO or other insurance. 18 Sampling, Data Collection, And Descriptives Sample Selection and Data Collection Process The selection of whom to administer the questionnaire to was a purposive form of sampling. While not being a representative sample of children in the State of Michigan, the sample for this questionnaire was chosen carefully to increase the prevalence of four characteristics and thereby increase the ability to use characteristics and indicators that will predict better who is at risk for EBLL. Other project guidelines further shaped the make-up of the sample population. The desire to test the predictiveness of specific models for various segments of the population, such as rural versus urban populations or by race and ethnicity, was facilitated by recruiting certain types of medical clinics in certain locations to be included in our study. For example, many of the Upper Peninsula County Health Departments were contacted to seek their cooperation in the Lead Screening Study. By using WIC programs in primarily rural counties, the number of both rural and low income populations, a group deemed both theoretically and empirically to be more at risk, was increased. To ensure a large enough sample of African-American respondents many of the clinics within the Henry Ford Health System, located in and around Detroit, were enlisted. Since, according to the Census Bureau, the population of the Detroit Urbanized Area is comprised of nearly 26% Black or Afi'ican-American, well above the State total which comprises only 14.2% of the population (U .S. Census Bureau, Census Bureau Summary File 1); this area was over sampled to assure a larger proportion of Blacks or African-Americans in our study population. In addition, Detroit has one of the highest rates of EBLL of any part of the state (2002, CLPPP data reports). 19 In addition to the above considerations for sample composition sampling was also limited by other factors as well. A necessary condition for being a source of data was that the clinic staff be willing to participate in the study. The decision to include specific centers and clinics in patient recruitment was mediated by finding those that were willing to do so. Those willing were mostly those clinics already doing universal BLL testing. These were typically clinics serving low-income populations and other clinics in Detroit. While there was little financial cost associated with participating, participation did involve extra work for the staff in these locations. This work involved determining which patients to administer the survey to, explaining the protocol of the study, and convincing parents to allow their children to participate. The recruitment of centers required multiple contacts with the staffs and much convincing these staffs of the importance of the study (this often included pressure or inducement from the MDCH on the WIC offices in the County Health Depts. to cooperate). Furthermore, multiple visits were required to introduce the study, devise research protocols for each specific center, and follow up with the data collection process. This included making revisions to the survey instrument (by eliminating non-predictive questions, we reduced the amount of time parents spent filling out the survey), stimulating an increase in patient recruitment and other such details. The data collection process, overall, was both labor-intensive and time- consuming, though the goal of obtaining over 4000 surveys and corresponding blood samples was nearly achieved. In addition to the multiple visits, an incentive program was implemented for both respondents and for health care providers for recruitment of subjects at the WIC clinics and physicians’ offices. Parents were given long-distance 20 calling cards, varying in length from twenty to forty-five minutes, as an incentive to enroll their child in the study. Furthermore, clinics and pediatric offices were provided with incentives to actively recruit patients for the study. These incentives came in the form of cookies, chocolates and other snacks for clinic staff for every forty completed surveys they returned. Through this arduous process, a sample of 4177 surveys was collected, for which the pertinent demographic information is discussed below in detail. Sample Population Descriptive Characteristics The sample for this study was compiled by surveying the parents or guardians of 4177 children in Michigan that were having blood drawn for the purposes of being tested for blood lead levels. In order to better understand the results of the analyses, we should explore some of the characteristics of our sample population. Of these 4177 children for whom survey data was received, a corresponding blood lead level (BLL) was received for 3875, or 92.8%, of the children. As demonstrated in Table 2, our sample found that 3.6% of the children tested for lead poisoning were found with elevated blood lead levels of _>_ 10 ug/dL. This measure falls just below the national average of 4.4% that had been identified by the third National Health and Nutrition Examination Survey 1988-94 (NHANES 111). Furthermore, our sample with 3.6% having elevated blood lead level (EBLL) is below the statewide findings from 2001, when 7.4% of young children tested for lead poisoning were confirmed at or above 10 ug/dL. However, our survey was not implemented statewide and therefore could not be expected to replicate those findings. Table 2: Blood Lead Level Results — Survey Respondents N = 3875 BLL RESULTS % OF RESPONDENTS BLL < 5 pgdL 79.8 BLL 5 to 9 Eg/dL 16.6 BLL 10 to 19 ug/dL 2.9 BLL 20 or morglg/dL 0.7 21 The decision to over-sample low-income, urban communities provided a higher proportion of minority group members than would have been obtained from a simple random sample of all children in Michigan. As presented in Table 3 the racial/ethnic makeup of the sample population consisted primarily of Whites and Afi'ican-Americans. The ability for respondents to check the boxes Of all racial and ethnic groups that applied to them allowed for the totals to exceed 100%, with the majority of the multiple selections containing Hispanic/Latino as the secondary descriptive. The sample population, reported below, included a rate of Hispanic or Latino respondents five times the population proportion for the state of Michigan as provided by the Census Bureau. Additionally, even with including health care providers in predominantly Arab communities in our sampling frame, only 0.7% of respondents reported being Arab or Chaldean. Therefore our sample is under-representative of the Arab community in Southeastern Michigan. Inclusion of the migrant population provided by the Telarnon Head Start program led to a higher rate of American Indians than is found in the population of Michigan. According to the Census Bureau, 0.6% of all Michigan residents reported their race/ethnicity as Native American while our sample included a rate over 10 times this at 6.3%. By these measures it is clear that the project’s goal of over-sampling minority group members was successful. As stated, in spite of oversampling “at risk” populations this did not yield a higher proportion of children with elevated BLL 2 10 ug/dL, as was anticipated (see Table 2). 22 Table 3: Race/Ethnicity of Respondents N = 4102 RACE VALID % I(E’IEiZNSUS % FOR White 59.7 80.2 Black or African-American 31.3 14.2 Hispanic/Latino 1 8.8 3 .3 American Indian or Native Alaskan 6.3 0.6 All other race/ethnicity categories“ 2.5 3.1 *includes Asian, Arab/Chaldean, and Hawaiian/Pacific Islander Both the Centers for Disease Control (CDC) and the current statewide screening recommendations for Michigan, set forth the by the Michigan Department of Community Health (MDCH), include specific age groups of children that should be tested for elevated blood lead levels. The CDC recommends testing children ages one and two years of age (or those between 36 and 72 months if not previously screened) who reside in high risk ZIP codes or who indicate a “yes” or “don’t know” to questions in a basic personal risk questionnaire (discussed earlier). Given these guidelines we established a research protocol whereby all children at the participating health care facilities and WIC offices that were between 9 — 36 months would be offered the Lead Screening Questionnaire as well as receiving a BLL test. Given that older children previously not screened (36 — 72 months) are also included in the recommendations for screening, we did not strictly limit patient recruitment to the 9 — 36 month age group, though we did urge providers to universally screen those in that age group. The ages of the children whom received blood lead tests and whose parents completed questionnaires for our study are provided in Table 4 below. 23 Table 4: Age of Respondents N = 3788 AGE GROUP VALID % Less than 9 months 2.2 9 to less than 25 months 64.0 25 to 36 months 18.0 36.1 to 72 months 15.4 Older than 72 months 0.4 As is clear from this table, 82% of the children in the analyses to follow were in our target age range of 9 to 36 months. Furthermore, given the expanded guidelines provided by CDC and MDCH, all but 2.6% (those children less than 9 months of age and those children older than 72 months of age) were in an acceptable age range to be included in our study. Furthermore, given that the objective was to test the predictiveness Of the elements of the current statewide screening recommendations, we are still able to use the data provided by the parents of children outside the 9 — 72 month age group. The last two descriptive characteristics of the sample population worth mentioning here are the rates at which the children’s health insurance was provided by Medicaid and the prevalence of children who reside in high risk ZIP codes. Federal regulations require the testing of all one and two year old children on Medicaid, as well as those on Medicaid between 36 and 72 months not previously tested. Our sample population (N = 4085) contained 65.2% of children being covered by Medicaid. This is not surprising due to the over-sampling of low-income respondents. The Census Bureau reported that in 1999 14.1% of households in Michigan had an income of less than $14,999. Compared to that, 47.8% of the respondents to our survey reported a total household income of less than $14,999. While the predictiveness of ZIP code has been called into question it is still currently used as a screening measure. In our sample (N = 4190) 80.3% of the children 24 were reported to live in high-risk ZIP codes. However, the substantial variation in housing age and poverty levels that exists within ZIP codes suggests the need for a more precise geographic predictor. This measure will be improved upon by a more precise measurement of old housing and poverty through the use of Census data. By geocoding each residence to the Census Block Group level and incorporating the housing and income characteristics of each Census Block Group into the analysis we will better be able to predict, geographically, the risk level of all children. Though it is beyond the scope of this paper, future publications will address these additional geographic predictors for blood lead levels. 25 Data Analysis QataScreening Techniques Before beginning any data analysis techniques for discerning the predictability of the independent variables on blood lead level it is imperative that the data is explored for any violations of regression assumptions. Furthermore, testing the assumptions of normality, linearity, and homoscedasticity for our dependent and independent, or predictor, variables will assure that the results of later regression and residual analyses will be valid. Later analysis of the regression residuals will be used to examine these three assmnptions simultaneously, as well as checking for multicollinearity. As discussed earlier, the fact that all of the independent variables being used in the analyses to follow are discrete allows for a check of univariate outliers by simply examining frequencies of response categories. In the case of all of the independent variables being considered for inclusion in the regression analyses, there are no univariate outliers present in the data set. Furthermore, the dependent variable is a continuous, ratio-level measurement represented by the number of ug/dL of lead found in the child’s blood upon screening. By examining a histogram of BLL it is evident that there is no disconnect between the extreme cases and the mean value of 3.36 ug/dL nor the median value of 3.00 ug/dL (see Appendix A). The range for this variable was fifty-three with a minimum of one and a maximum of fifty-four, with the higher, extreme cases being vital for testing the predictiveness of the independent variables. The need to check for multivariate outliers is not present due to the lack of univariate outliers in our discrete predictor variables, as well as in the dependent variable. 26 The only independent variable that would normally be considered continuous, household income is being treated as a discrete, grouped variable. This warrants a brief discussion regarding the linearity of household income. By creating eight dummy variables for the nine response categories for household income tests for non-linearity were conducted. NO significant non-linearity was found in the effect of income on BLL. The issue of normality needs to be addressed for the continuous dependent variable of blood lead level. After doing so it will be appropriate to investigate the correlations amongst the independent variables as well as with each of these predictors and the dependent variable. TO check if BLL is normally distributed, a graphical approach, along with an examination of the skewness and kurtosis of this variable, presents the clearest understanding of normality. A normal probability plot is a strong graphical technique for assessing normality. In these plots, all the cases are ranked and sorted, and then an expected 2 score is calculated and compared with the normal value. If the variable is distributed normally the plotth points for the cases will line up along the diagonal line from lower left to upper right. By examining both the normal probability plot and histogram of BLL we can graphically see that the distribution of BLL is not normal (see Appendix A for histograms and normal p-plots depicting normality). At low values of BLL, there are too many cases above the diagonal, and at high values of BLL there are too many cases below the diagonal, reflecting the patterns of skewness and kurtosis. Along with this graphical information an examination of the skewness and kurtosis values clearly indicate a need to transform the dependent variable. In this case, both the graphical and statistical information indicate a substantial positive skew. This type of non-normality tends to 27 indicate a need to perform a logarithmic transformation though for thoroughness and accuracy an examination of all reasonable transformation techniques will be presented. The table below provides the skewness and kurtosis values for our dependent variable in its’ non-transformed value as well as for square root, logarithmic, and inverse transformations. Table 5: Skewness and Kurtosis of non-transformed and transformed BLL || BLL LOG OF BLL SORT OF BLL INV OF BLL l l Skewness 5.32 .418 1.893 .687 || || Kurtosis 50.057 -.005 7.377 -953 I] As is evident from Table 5, all three types of transformations improve the normality of BLL. Both square root and inverse transformations of BLL do reduce the skewness and kurtosis, more so through the inverse transformation. However, by decreasing both the skew and kurtosis of BLL to as near zero as possible, incorporating a logarithmic transformation of BLL is necessary to assure that the regression analyses will approach multivariate normality. Data Analysis I Returning now to the research hypotheses stated earlier, the effectiveness of the various predictor variables can begin to be tested. In many of the previous studies on elevated blood lead levels in children, the utility of certain predictor or indicator variables was determined by conducting an analysis of the sensitivity and specificity of each item (Binns, LeBaily, Fangar and Saunders, 1999, p. 105). Using the conventionally agreed upon definition within the medical community of EBLL being at or above 10 ug/dL, the sensitivity and specificity of each item will be examined. AS discussed earlier, sensitivity is computed by dividing the number of people for which the screening measure predicts to have EBLL by the number of people that actually have EBLL. Specificity, discussed 28 earlier as well, is computed by dividing the number of people the screening measure predicts to not have EBLL by the number of people that actually do not have EBLL. These measures will provide a good understanding how well a particular item can predict a child’s risk for EBLL. Our dependent variable for purposes of this statistical analysis is dichotomous. Either a child is classified as ‘low BLL’ at less than 10 ug/dL, or they are considered ‘elevated BLL’ at 10 ug/dL or higher. Therefore, as is common in the medical community, we are testing each item’s ability to predict whether a child is ‘diseased’ or ‘not diseased’ as is defined by the threshold mentioned. The only predictor variable not dichotomous is income. In order to test the sensitivity and specificity all independent variables need to have only two response categories. Therefore, a recoding of income is required and a logical threshold is the income level used to determine poverty levels by the US. Census Bureau. This is used to classify respondents as either ‘low income’ or ‘non low income’. ‘Low income’ can be used as a predictor for EBLL and its’ sensitivity and specificity can be examined. Below are the sensitivity and specificity, as well as the positive predictive value, for each variable discussed in the research hypotheses. The positive predictive value (PPV) is “the probability that a person with a positive test is a true positive (i.e., does have the disease)” (Last, 129). In this case “the disease” would be a BLL of >= 10 ug/dL. This statistic is computed by dividing the number of children with positive screening results and elevated blood lead levels by the actual number of children with elevated blood lead levels. From the table below we get some idea as to which items will be supported as good indicators or predictors for EBLL. 29 Table 6: Sensitivity and Specifici ofpredictor variables for BLL >= lO_u_flL SENSITIVITY SPECIFICITY POSITIVE PREDICTIVE VALUE High Risk ZIP Code 96.4 19.5 4.3 Exposure to peeling paint 13.0 93.7 6.3 Water through lead pipes 22.2 94.8 12.3 Absence of breast feeding 40.9 48.1 2.9 Use of certain. cosmetics/home 0.8 99.5 5.6 remedies contalnlng lead Adult job Involves contact wrth 12.6 80.4 23 lead Adult hobby involves contact 5.6 85.7 1.4 With lead Sibling with EBLL 22.9 96.5 20.5 Adult household member with EBLL 9.7 98.3 17.6 Playmate with EBLL 12.0 97.3 14.4 Black or African-American 65.9 69.5 7.5 Low Income (below $20,000) 75.6 44.0 4.5 Child covered by Medicaid 79.1 36.5 4.5 The following items perform best on sensitivity: ZIP Code, Medicaid, Low Income, and respondent being Black or African-American. However, they have low specificity. For example, ZIP code, as a measure of the poverty rate and proportion of old housing, is a very sensitive predictor in that almost all who EBLL are in high risk ZIP codes. However, with so many of Michigan’s ZIP codes being classified as ‘high risk’ it has very low specificity and a weak positive predictive value. The following variables perform best on specificity: exposure to peeling paint, adult household member with EBLL, playmate with EBLL, sibling with EBLL, adult hobby involving contact with lead, adult job involving contact with lead, and child being Black or Afiican-American. However, these items have low sensitivity. The following items do best on positive predictive value: sibling with EBLL, adult household member with EBLL, sibling with 30 EBLL, and water through lead pipes. The hypothesis concerning a child’s proximity to others with EBLL increasing their risk of being lead poisoned appears to be supported, with the strongest positive predictive value in our study being that of ‘sibling with EBLL’ at 20.5%. However, since this does not take into account the multivariate relationship it is a somewhat limiting level of analysis. For a more complete analysis, which will allow decisions to be made regarding the research hypotheses, a multiple regression analysis will be presented. The use of ordinary-least-squares regression allows for the significance of regression coefficients as being different from zero to be tested. It also estimates the quantitative effect of each variable on BLL. Other multiple regression models will follow including those variables needed to test the effectiveness of the current screening measures. For the purposes of hypothesis testing, an OLS regression was conducted with Ln (BLL), where Ln is the logarithm to base e, as the dependent variable. Brody et al. (1994) found that the distribution of BLL is log normal. Therefore, to meet the assumptions of regression, we must use a logarithmic transformation on BLL as is documented by Kaplowitz et al. (2000). (This will be demonstrated in the next section.) “Thus, the regression coefficient is the effect of a one-unit change in a predictor on the logarithm of BLL. For a dichotomous independent variable, the coefficient is the difference between the means of the logarithm of BLL levels for groups with different values of that predictor” (Kaplowitz et al. 2000). The amount of variance in BLL explained by this set of predictor variables is 13.3%, with an adjusted R2 of 12.9% of variance explained. While this is a fairly small R2, an analysis of the significance of the individual regression coefficients, presented 31 below, will allow for decisions to be made regarding the research hypotheses. In cases where the regression coefficient is small but significant, the hypothesis in question will qualify as supported. A glance at Table 7 easily and quickly distinguishes which independent variables have a significant effect on BLL and therefore allows us to make decisions regarding the stated research hypotheses. The variables that have effects, and therefore should be used to predict a child’s risk for EBLL, are: high risk ZIP code, water through lead pipes, Sibling with EBLL, Adult with EBLL, child is Black or African-American, and the child’s household income. Given the large number of variables whose effects are not significant, it is fair to say that many of the research hypotheses will not be supported. Table 7: OLS Regression - Independent variables in Research Hypotheses UNSTANDARDIZED SIGNIFICANCE VARIABLE NAME COEFFICIENT High Risk ZIP Code .212 .000'" Exposure to peeling paint (any house 074 166 child has lived in, built prior to 1950) ' ' Water through lead pipes .220 .000'" Absence of breast feeding .009 .410 Use of certain cosmetlcs/home .05 5 .760 remedles contalmng lead Adult job involves contact with lead .028 .424 Adult hobby involves contact with lead .043 .275 Sibling with EBLL .480 .000'" Adult household member with EBLL .322 .003" Playmate with EBLL .107 .195 Black or African-American .318 .000." Low Income (below $20,000) -.O46 .000'" ‘ Child covered by Medicaid .032 .318 p<.05 , p<.01m,p<.001 a. Predictors: (Constant), high risk ZIP code, Any house built before 1950 with peeling paint inside, Any house with lead pipes, Absence of breastfeeding, home remedies or cosmetics used that may contain lead, Adult job involves lead, Adult hobby involves lead, Sibling with elevated blood lead level, Adult household member with EBLL, Playmate with EBLL, Respondent is Black or African- American, Total Household Income, Child covered by Medicaid, 32 The hypotheses in the environmental risk section are mixed. High risk ZIP codes increase a child’s mean BLL and this result is significant at p < .001. We can be 99.9% confident that whether a child resides in an area defined by ZIP code (with a certain percentage of poverty and old housing) has some value for predicting EBLL and therefore should be used in any prediction equation. The question of whether exposure to peeling paint in old houses (those built prior to 1950) seems to have limited utility for predicting EBLL in children. This variable as a predictor is not significant, even at a more conservative level of p < .10, has low sensitivity and positive predictive value, and therefore we should not support the hypothesis that children whose parents report that their child was exposed to peeling paint in old houses are more at risk for EBLL than those that are not exposed. The last hypothesis in the environmental risk section is the question of whether knowing that water is supplied through lead pipes increases a child’s propensity for EBLL. Given the significance of the regression equation, the moderate sensitivity and decent positive predictive value, this hypothesis is supported and should be considered a valued screening question. The hypotheses regarding certain behavioral and/or cultural practices increasing risk for EBLL will be addressed now. With an unstandardized regression coefficient of .009 and a p-value of .410 it is clear that the breast feeding variable is not statistically significant as a predictor for EBLL. Furthermore, the moderate sensitivity (40.9%) and the low positive predictive value (2.9%) it is clear that the absence of breast feeding makes no perceptible difference. Therefore the hypothesis that the lack of breast feeding places children at greater risk for EBLL is not supported. The second behavioral or cultural hypothesis regards the use of certain cosmetics or home remedies which contain 33 lead. This hypothesis is not supported, due to the high p-value of the regression coefficient (p = .760), as well as the extremely low sensitivity (0.8%). Note that only .5% of the respondents answered in the affirmative for this question. The last behavioral hypothesis is regarding children’s exposure to adults who have hobbies or jobs in which they deal with lead or lead by-products. The hypothesis stated that this exposure would increase a child’s risk for EBLL. Neither the job nor hobby of any adult having contact with the child significantly increases their risk of EBLL. This is evident through the high p-values of the regression coefficients for job and hobby respectively (p = .424, p = .275) in addition to the extremely low sensitivity (12.6%, 5.6%) and poor positive predictive value (2.3%, 1.4%). The third set of hypotheses proposes that there are non-causal indicators which predict who will be more at risk for elevated blood lead levels are much more strongly supported. The first of these hypotheses is that any child having contact with a sibling, adult or playmate that has EBLL will have an increased risk of EBLL. This statement actually contains three different variables which were measured separately from each other. Knowing whether a child has a sibling with EBLL is one of the strongest predictors for EBLL. With a p-value < .001, as well as having the strongest predictive value, this part of the hypothesis is supported. A child having contact with an adult with EBLL is supported as well, with a p-value < .01 and strong predictive value. Having a playmate with EBLL is not a statistically significant predictor for BLL. The second hypothesis related to non-causal indicators for EBLL stated that, due to increased exposure to lead, Black children will have higher BLL, on average, than White children. 34 With a p-value < .001 this variable is a statistically significant indicator for BLL, and, therefore, this research hypothesis is supported. The items concerning the socioeconomic status of the child’s family (total household income and Medicaid coverage) are the last two hypotheses presented for analysis. The first hypothesis states that as the income level of a child’s family goes up the probability of a child having EBLL decreases. With a significant regression coefficient (p-value < .001), and good sensitivity, this item predicts fairly well which children are at increased risk. Therefore, this hypothesis is supported and income level should be considered a good predictor for EBLL. In tandem with income level is the type of insurance coverage the child has. In this sample of 3875 children in Michigan, Medicaid coverage is not a good predictor as is demonstrated by the low significance of the regression coefficient (p-value = .318). A major reason why Medicaid comes out so poorly in this regression is that household income is being controlled for. Medicaid is highly correlated with household income; therefore once you already know household income, Medicaid status adds little additional predictive value. While this clarifies the accuracy of the research hypotheses, to gain a clearer picture of how best to predict if children are at risk for EBLL, more data analysis is needed. Lead Poisoning: Predictive Model Building This section will continue to clarify how best to predict which children are at risk for EBLL. By presenting several regression models the analysis will show which variables are good predictors for EBLL and therefore should be included in a revised screening plan. As the environmental justice framework discussed earlier indicates, the risk factors for elevated blood lead levels (EBLLs) lead us to include certain variables as 35 predictors in regression analyses. These risks, as is pointed out by Kaplowitz et a1 (2000), include, among other things, those children living in older housing (those which were built pre-l950). This is due to the presence of lead-based paint. Close contact with adults who have jobs and/or hobbies that involve lead creates a risk for children, as does water contamination from lead water mains (usually found in older houses). The rates of elevated blood lead levels vary tremendously among ethnic and income groups. “Twenty-one percent of Afiican-American children have EBLLs as compared to 1.5% of non-Hispanic white children (NHANES III, Phase 11 1991-94).” Two models of ordinary-least-squares regression with the logarithm of BLL as the dependent variable will be presented. The first model will be done using the MDCH screening criteria as the independent variables; the second model will include some of the MDCH criteria, but will be more reflective of the strongest survey questions as predictors. Again, the current screening procedure includes Medicaid participation, high- risk ZIP codes, and screening questions, which have been set forth by the MDCH-CLPPP (see Current Statewide Screening Measures section for a detailed listing and discussion of these questions). The MDCH screening criteria includes five screening questions that were designed to capture a child’s possible exposure to lead contaminants. These questions are to be used for all children not included in the Statewide Screening Plan by Medicaid enrollment or residence in a “high risk” ZIP code. These three components together comprise the current MDCH screening guidelines which I will first test the predictiveness of through regression analysis. In all, the MDCH predictors (Model 1) consisted of 11 independent variables created from our survey. I will present the results of this regression analysis momentarily. 36 The second model presented was arrived at by a combination of the theoretical inclusion of certain variables as well as some statistical screening techniques for the decisions about whether or not to include certain variables in the regression. Besides the aforementioned survey questions created to duplicate the MDCH screening questions, a number of survey questions were included to further understand any other possible predictors for blood lead levels. As a starting point for the analysis of the survey instruments’ ability to predict BLL, all of the items from the questionnaire were analyzed through many combinations and statistical techniques. This involved looking at correlations, regression runs, as well as specificity and sensitivity of each variable in relation to it’s’ ability to predict elevated blood lead level. As an additional check on which variables to include in the final prediction model (Model 2 below) a stepwise regression was conducted using all possible survey items as independent variables and the log of blood lead level as the dependent variable. Once this was completed, the variables remaining through stepwise regression and some variables that theoretically deemed inclusion in the model were run through an OLS. This model (Model 2), though still far from being considered strongly predictive, did show a significant improvement from the MDCH model. Table 8: Comparison of regression models R R2 ADJ R2 F-TEST DEGREES SIG STAT OF FREEDOM Model 1 .265a .070 .067 23.210 11 .000a Model 2 .388b .150 .148 59.589 10 .0001) a. Predictors: (Constant), Child covered by Medicaid, Any home remedies used that may contain lead, high risk ZIP code, Any playmate with elevated blood lead level, Adult hobby involves lead, Any house built before 1950 with peeling paint outside where child plays outside, Sibling with elevated blood lead level, Any house with peeling paint that child regularly visits, Any house built from 1950 — 1980 that has been remodeled, Adult job involves lead, Any house built before 1950 with peeling paint inside 37 b. Predictors: (Constant), Pacifier use, Lead pipes for water in any house, Total household income, Playmates with elevated blood lead level, High risk ZIP code, Adults with elevated blood lead level, Sibling with elevated blood lead level, Respondent is Black or Hispanic, Any house child has lived in with peeling paint inside, Child’s age at time of specimen While neither of these models does an outstanding job of predicting blood lead levels, it is obvious that the survey model, Model 2, does a much better job than the current MDCH screening guidelines. By more than doubling the R2 and increasing the F score by nearly three times, we can see that Model 2 is a much better predictor than Model 1. Looking at the value for R2, it is clear that the MDCH screening measures can only explain 7% of the variance in the log of blood lead level. This is improved to a modest 15% of variance explained when using the survey items as predictors. Furthermore, in Model 2 the F value (10, 3371) = 59.589, easily exceeds the critical value of 2.96, p < .001, as did the F value in Model 1. As mentioned, the 11 MDCH variables in Model I explained only 7% (6.7% adjusted) of the variance in the logarithm of blood lead level (log BLL). Of those 11 predictors, five proved to significantly explain a small portion of the variance. Sibling with EBLL had an unstandardized coefficient of .575 and a significance of pg .001. Sibling with EBLL (St,2 = .026)1 explained the largest portion of the variance in Model 1 and is, by far, the strongest predictor that the MDCH has in place. The direction of this coefficient indicates that for those children having a sibling with EBLL, the log of BLL increases by .575 which means that BLL was multiplied by 1.78. Ziprisk (dichotomized ZIP code as either ‘high risk’ or not) had an unstandardized coefficient of .232, was significant at the .001 level, and explained 1.7% of the variance (sri2 = .017). Again, the ' This is a measure of the unique variance explained by the independent variable. This is the amount by which R2 is reduced if the independent variable were deleted from the regression equation. The sum of the measures of sr2 are usually less than R2 with this difference being explained by shared variance of two or more independent variables. 38 positive unstandardized coefficient means the increase in the log of BLL by .232 for residence in a ‘high risk’ ZIP code. A child being covered by Medicaid had an unstandardized coefficient of .130, a significance of .000, and explained .73% of the variance (sri2 = .007). With a positive sign the regression coefficient indicates that insurance coverage through Medicaid increases the log of BLL by .087. The last two independent variables, any house child visits with peeling paint (sri2 = .002) and playmate with EBLL (St;2 = .005) have unstandardized coefficients of .117 and .298, respectively. They are both significant at the .05 level and explain a very small portion of the variance (any house with peeling paint explains .24% and playmate with EBLL explains .47%). As demonstrated through the analysis of sriz, these variables uniquely explain 5.7% of the variance in the log of BLL. The remaining 1.3% of variance is explained through a combination of two or more of these variables. Hoping to improve on the understanding of how best to predict those at risk for elevated blood lead levels, the analysis was refined and included the survey variables as mentioned previously. All ten variables proved to contribute significantly to the variance explained in log of BLL. Table 9: Regression Model 2: Coefficients, Significance, and Uni ue variance explained Variable Name Unstd Significance Unique variance Coefficient (sriz) Lead pipe in any house .119 .016 .001 Household Income -.042 .000 .018 Playmates with EBLL .212 .002 .002 High risk ZIP code .224 .000 .016 Adults with EBLL .234 .003 .002 Sibling with EBLL .387 .000 .011 Respondent Black or Hispanic .246 .000 .028 Any house with peeling paint inside .174 .000 .008 Pacifier Use -.060 .000 .004 _A_ge in months at time of specimen .005 .000 .007 39 Table 9 demonstrates the size and direction of the unstandardized coefficient, the significance level, and the unique variability explained by each. A comprehensive discussion of these predictors for the log of BLL would be timely and cumbersome. The explanation of Table 9 is fairly straightforward and can be discussed in less specific terms. All of the variables, with the exception of household income and age in months, are dichotomous variables (yes-no responses). For all of the above, except household income and pacifier use, a slight increase in the dependent variable is associated with an increase in each of these variables. That is, if the behavior or condition is experienced and reported, the predicted log of BLL increases. For example, if the child is either Black or Hispanic, the log of BLL increases by .246 ug/dL. If the child lives in a ‘high risk’ ZIP code the log of BLL increases by .224 ug/dL. When looking at the amount of variance explained in the log of BLL the semipartial correlations for these ten variables explain .097 of the R2. In other words, .097 of R2 is attributed to unique sources, leaving .053 which can be explained by the contribution of all ten variables jointly. Overall, it is clear that the amount of variance in the log of BLL is far from adequately explained by these variables and further analysis is needed. Qgistic Regression An_alysis of Lead Poisoning: The use of dichotomized dependent variables is a common practice in the medical community. In most cases those in the medical community want to be able to predict or determine the likelihood that someone is diseased or not diseased. Therefore, predicting whether or not someone has the disease is often the primary goal of research in the medical field. For instance, in our case we would want to be able to predict whether or not someone is diseased as defined by exceeding a conventionally agreed upon amount of 40 lead in the blood. While we are able to use OLS to determine the amount to which our dependent variable will increase with the presence of indicator variables, there is another component to screening that is important. While it is good to know that children that live in ‘high risk’ ZIP codes increase the log of BLL by .224 ug/dL, for example, being able to predict if someone has an elevated blood lead level is vital. This information can allow for more effective use Of the resources available to test only those children that are at risk of being elevated, while not missing any children that should be tested. For the case of BLL there has been a good discussion of what constitutes elevated levels. While it is not cut and dry most of the literature in the field indicates that the use of 10 ug/dL as the threshold for elevated blood lead level. Other threshold levels discussed range from a low of 2.5 (Canfield et a1, 2003) to as high as 75 ug/dL (though this amount was recommended by a representative for the paint industry). While a recent article in the American Journal of Public Health calls for a threshold of 5 ug/dL (Bernard, 2003), for purposes of this analysis, a 10 ug/dL threshold for the Split point between elevated blood lead levels and non-elevated blood lead levels will be used. The decisions as to which independent variables to include in the logistic regression were fairly simple. Given the knowledge gained from previous regression analyses, it was determined that ability to predict elevated blood lead level could best be accomplished by including the same variables as were used in the ordinary-least-squares regression. These are age (in months), presence of lead pipes, any inside paint peeling, adult with EBLL, child being black or Hispanic, a Sibling with EBLL, high risk ZIP code, playmate with EBLL, and income (dichotomized at 20K). The decision to dichotomize income at 20K came from analyzing the bivariate relationship between BLL and income. 41 From this it was clear that this threshold was significant for increased risk of lead poisoning, whereby those below 20K had a much higher incidence of elevated levels. Lastly, the decision to not include pacifier use was made based on statistical inference discovered during the logistic regression analysis. By entering the independent variables using stepwise selection and using a likelihood-ratio test, it was determined that pacifier use did not improve the goodness-of-fit of the model and therefore eliminated the variable from the analysis. A test of the full model with all predictor variables against a model with the constant only was found to be statistically significant. This was determined by examining x2 (with 9 degrees of freedom and a sample size of 2822) = 199.78, p < .001. The goodness of fit of this model, however, remains small. Two techniques were employed to measure this. First, McFadden’s p2, calculated by a transformation of the likelihood ratio statistic, is intended to mimic an R2 in being between 0 and 1. The results were p2 = .234. A similar statistic is Nagelkerke R2, which for the complete model was .262. The prediction success was not very good for this model, however, with 99.7% of low BLL being correctly predicted but only 10.2% of elevated BLL cases being correctly predicted. This resulted in an overall prediction rate of 96.6%. Table 10 shows the regression coefficients, Wald test statistic, and Odds ratios (with 95% confidence intervals) for each of the nine predictors. As is evidenced by the significance of the Wald statistic, all of the predictors except ‘adult with EBLL’ significantly predicted status on our dependent variable. 42 Table 10: Logistic regression coefficients, Wald, and Odds-ratio with 95% CI Variable Name B We“? Odds ratio 95% CI for Odds Rat”) Test Upper Lower Lead pipes in any 1.011 11.290*** 2.749 4.958 1.524 house Any house with 1.070 20.855*"‘* 2.916 4.617 1.842 peeling paint High risk ZIP code 2.206 9.355" 9.076 37.303 2.208 Respondent is Black 1.272 22.334"‘** 3.569 6.048 2.106 or Hispanic Playmate with EBLL .879 5.311* 2.409 5.088 1.141 Sibling with EBLL 1.106 12.564*** 3.023 5.573 1.640 Child’s age (in .019 9.052* * 1.019 1.031 1.007 months) Income .749 8.965" 2.1 16 3.455 1.295 Adult with EBLL .849 3.753 2.337 5.516 .990 (constant) -7.767 100.308 3 Significance levels for Wald statistic: * p<.05, **p<.01, ***p<.001. The interpretation of the regression coefficients in a logistic regression model is best done by discussing the odds ratios for the predictor variables. The odds ratio (if greater than 1) is the increase in odds of a respondent having an elevated blood lead level with a one-unit increase in the predictor variable. For instance, a child that is reported to have lived in any house with lead pipes has an Odds of EBLL which is 2.7 times that of a child that has not lived in any house with lead pipes. In our sample, the lowest odds ratio present in this model is the child’s age. The reason for this is that the unit of measurement (age in months) is small, therefore an increase in the child’s age by only 1 month produces only a small increase in the odds of having elevated levels. If age was recoded and measured by years, instead of months, the odds ratio would be Significantly higher. The high Odds ratio from ZIP code is over 9. SO a child living in a high risk ZIP code has 9 times the Odds Of having elevated blood lead level than a child not living in a high risk ZIP code. This warrants some explanation. The best explanation for this high 43 odds ratio revolves around the prevalence of high risk ZIP codes in our sample. As defined by the Michigan Department of Community Health, nearly 50% of all ZIP codes in the state are deemed as high risk. Combined with the sampling done in this study to over represent those living in areas deemed ‘high risk’, the number of children living in non-high risk areas is fairly small and constituted less than 20% of the sample population. In addition to this, the relatively low rate of elevated blood lead levels found in the sample confounds our ability to predict for all variables, though seemingly more so in the case of ZIP code. While only 5 respondents having elevated blood lead levels reported living in non-high risk areas, Ziprisk does appear to be a good predictor. As indicated earlier, filrther analysis of more precise geographic indicators, such as Census Block Groups, as well as the housing and demographic characteristics of those Census Block Groups, should provide better understanding of the variation that exists in BLL. Some of the other odds ratios also indicate a strong ability to predict elevated blood lead level. Being Black or Hispanic increases the likelihood of having elevated BLL by over 3.5, having a sibling with elevated BLL increases odds by over 3, and being exposed to peeling paint increases odds by nearly 3. The other predictor variables considered in this model all increase a child’s odds of having EBLL, though to a lesser degree than those discussed in detail. Conclusion Discussion A number of statistical analyses have been presented in order to gain a better understanding of not only the effectiveness of the current screening procedures in place in Michigan, but which items can better predict which children are at greater risk for EBLL. Variables were selected and used according to various categories. These included environmental (including geographic predictors), behavioral/cultural, non-causal indicators and measures of poverty all of which were considered as possible sources for explaining the variance in BLL. Some of the findings presented earlier warrant further discussion. The use of ZIP code as a predictor for EBLL is somewhat problematic. While high-risk ZIP code is a significant predictor of EBLL and is, therefore, in the final predictive equation, this variable warrants closer examination. While this variable seems to provide an explanation of a small portion of the variance in BLL, the low specificity of this item (as discussed earlier) raises questions as to the usefulness of this as a predictor. The heterogeneity within ZIP codes needs to be a consideration when understanding the weak predictive capacity of this variable. While ZIP codes are fairly large geographic areas with diverse living conditions, the idea of predicting risk based on place of residence is still a valid research agenda. Though beyond the scope of this paper, the inclusion of more precise geographic indicators and Census block group aggregate data values will yield to a regression analysis whereby we are able to explain, satisfactorily, a much larger part of the variance. 45 Another environmental variable which is currently used in the statewide screening program is the presence of lead pipes through which water is delivered to the home in which the child lives. The utility of this item remains problematic largely due to parents’ lack of awareness to the types of pipes at their residence. This is demonstrated by the low response rate and large number of ‘Don’t Know’ responses that this question received in the survey. However, a more systematic analysis determining which areas of Michigan still use lead pipes could improve the predictive capacity of this variable by eliminating occupants’ knowledge of the types of pipes in their homes from consideration. The variables used to represent ‘non-causal indicators’ for BLL warrant some discussion as well. While the relationship between a child having a sibling, or contact with an adult diagnosed, with EBLL was established, the hypothesis concerning exposure to a playmate with EBLL increasing risk for a child was not supported. This could simply be due to parents not having knowledge of their young children’s playmates’ medical conditions. Household income is another variable that requires a brief discussion. This variable was found to have negative regression coefficient. However, this is logical due to the fact that as income level increases the likelihood of exposure to environmental lead risk factors (such as peeling paint) are minimized. A closer examination of the interaction between income and age of housing could shed some light on this. It is likely, as local conditions reflect, that some old housing which would possibly contain lead paint, is associated with higher income levels and therefore reduced risk. 46 One cultural/behavioral variable worth discussing in a bit of detail is pacifier use. The direction of this regression coefficient is negative and with the coding of pacifier use ranging from 0 = never used to 2 = use frequently, it appears that as the use of a pacifier increases the Ln(BLL) will decrease by .06 contradicting the research hypothesis presented earlier. While this variable explains very little of the variance, it is worth noting that this inverse relationship can be explained. It is likely that the use of a pacifier prevents children from putting other objects in their mouths. This behavior, therefore, might limit the ingestion of paint chips and lead dust reducing a child’s risk for EBLL. Lastly, the use of the age in months at time of specimen variable needs a bit of explanation. The reason for including this variable in the analysis is fairly simple. By including it in the regression equation we can control for age when examining the other variables, such as pacifier use. In addition, it is worth noting that age is related to BLL in two possible ways. First, younger children’s physiology is such that they absorb lead more readily and, therefore, tend to have higher BLL. This creates a negative relationship between age and BLL. On the other hand, those older children tested for EBLL tend, disproportionately, to be those that have already exhibited EBLL. Thus, they are selected from the high risk group. This creates a positive statistical relationship between age and BLL. We could not know, in advance, whether the net relationship of age and BLL would be positive or negative. Future Research Needs The prevalence of children in the 21St century still combating the health risks associated with lead poisoning is troubling. The rates of EBL have decreased somewhat in the past twenty years or SO, thereby creating a false sense that the problem of lead 47 poisoning is one of the past. There is currently not enough being done to ensure the welfare of the future of our society, by protecting those that will be the leaders, decision makers and parents of the future. As much of the current literature indicates, the reduction in IQ associated with even low levels of exposure to lead (Raloff 2001; Needleman 1994) could theoretically prevent great advances in our future society. According to Joel Schwartz, of the Harvard School of Public Health, the reduction in IQ is directly linked to a reduction in economic productivity. “[Prior research] showed that if you lower the mean IQ of the US. population by one point, you lower the productivity of the economy by about one percent” (Schwartz, in Raloff 2001). Certainly a society driven by the all-mighty bottom line should be influenced, if not by the damaging health consequences of individual children, by the economic impacts of letting this problem continue. While this research has presented numerous techniques for screening children to determine their level of risk for lead poisoning, there is still much room for improvement. The best analysis of the survey data presented here can, at best, explain around 15% of the variance in blood lead levels. We will need to do better if we are going to more accurately target those children most at risk. The most effective technique of preventing even one child from suffering with this preventable malady is to institute universal testing for lead poisoning, with all children being tested regardless of place of residence, race, income level or insurance type. However, the current economic and political climate in the United States does not lend itself to throwing money at the problem, at least not before generating enough public interest to warrant it. Additional data collection and more detailed analyses done during this research project have provided a more cost 48 effective screening method. By making use of Census Block Group data such as age of housing, income levels, and other aggregate level data, the variance explained has been increased to upwards of 35%. Short of creating a public health environment where universal testing for EBLL is conducted, we must develop strategies for creating awareness of the health problems associated with EBLL, as well as promoting the use of these new screening measures. Lastly, by developing campaigns aimed at parents, perhaps in conjunction with corporate wellness programs in the workplace, the importance Of having children tested for BLL can put this problem in the public discourse. AS indicated, the problem of lead poisoning is far from fully examined. One area which needs to be examined is the biological component of lead poisoning. The data Show that minorities are more at risk, though it is often assumed that this is due to increased exposure caused by lower socioeconomic status and therefore higher proportions of children living in poverty as well as inadequate housing. By examining more closely the biological components of lead poisoning we can begin to better understand if there is a difference in the absorption into the blood of lead, which may account for the higher levels of African-American children having EBLL. Again, this will help focus finite resources so that money may be spent where it is most needed. Future research will also need to better refine the analysis of geographic predictors, beyond that of ZIP code and even Census Block Group data. While these measures do capture the proportion of old housing and poverty, which seem to be excellent predictors of increased risk for EBLL, they are far from perfect. By better identifying the characteristics of areas where high rates of EBLL exist, we can use these 49 data to more accurately predict who is at risk and thereby spend precious resources testing those children. Additionally, once these areas of increased risk are mapped and testing is done to guarantee children’s welfare, legislation and more resources need to be earmarked for the abatement of lead from existing contaminated sites. Furthermore, we need to increase the accountability to the public for removal or abatement of possible exposure sources. Only through public awareness, aggressive education campaigns about the dangers associated with lead poisoning, and more stringent guidelines for mandatory testing (and of course providing resources for doing so) can we truly begin to solve this problem. 50 Appendix A: Histograms and P-Plots Depicting Normality of BLL 51 Expected Cum Prob Expected Cum Prob Normal P-P Plot of Final BLL results 1.00 n .751 .50- .25 " a 0.00 I 0.00 .25 .50 .75 1.00 Observed Cum Prob Normal P-P Plot of Log of BLL 1.00 .75“ .50- .25 . 0.00 0.00 .25 .50 .75 1.00 Observed Cum Prob 52 nor-transformed BLL 2000 1000 6 ac, Std. Dev = 3.35 agr- Mean = 3.4 L: 0 N = 3808.00 Final BLL results Log Transformed BLL 1200 1000 800 600 400 5‘ C 200 Std. Dev = .71 d) 8. k Mean = .94 1.9:) 0 N = 3808.00 0.00 .50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 Log of BLL Appendix B: Lead Screening Questionnaire LEAD SCREENING QUESTIONNAIRE Parent or Guardian’s Name: Child’s Name: Date of Birth:(mo/date/yr) Address: [:1 St El Ave 1:] Rd El Dr [3 Cir C] Blvd Cl Sq U Tr City: State: Zip Code: + Please circle, check or fill in the answer that applies. If you are not absolutely certain, please indicate your best guess. Ifyou are veg uncertain, please leave the item blank. 1. How long was this child breast-fed? CI never El six to twelve months [:1 less than three months D more than twelve months El three to six months 2. How often does the child use a pacifier? El never or almost never El sometimes [:1 daily or almost daily 3. Has your child been outside the United States for more than three months? El yes C] no If “Yes,” which countries? 4. Please look at the list of cosmetics and home remedies below. (Each of these is used by some ethnic groups. However, many people have never heard of them.) Has the child ever been given any of them? El yes [:1 no If “Yes,” please check those which the child has been given. El Albayalde D Alkohl CI Ayurvedoc El Azarcon“ El Ba Bow Sen Cl Bali goli C1 Cebagin CI Coryceps Cl Ghasard Cl Greta El Hai ge fen Cl Kandu El Kohl Cl Kushta CI Mai ge fen El Pay-loo-ah El Poying tan El Surma Cl X-yoo-Fa (*chon is also called Alarcon, Coral, Luiga, Maria Luisa and Rueda) 5. Please think about any house, day care center or preschool that the child has regularly visited. Did any of them have peeling or chipping paint? El yes [:1 no 5a) Of all these places that the child has regularly visited, please think about the place where the child has spent the greatest number of hours. Please give the address on the next page. 54 Please provide some information about this place the child has visited below City: State: Zip Code: ElSt CIAve DRd EIDr UCir UBlvd ElSq ElTr Street Name and number USt DAve 1:.le CIDr UCir EIBlvd DSq UTr name Of nearest cross-street 6A. Please look at the list of jobs below. Has the child had regular contact with an adult with any of the following jobs? El no (If no, please skip to Question 7). El yes (If yes, please check all jobs which apply ). El Abrasive Blasting of Metal Structures 1:] Glass Manufacturing Cl Automobile Radiator Repair Cl Migrant Farm Worker El Battery Repair and Production El Plastics Manufacturing El Brass/Bronze/ and Copper Processing D Plumbing or Pipe Fitting 13 Construction Work 13 Rubber Products Manufacturing (remodeling older homes, replacing older windows, etc.) El Firing Range Instructor 68. How many days per week does the child spend with this adult? 7A. none E] one 121 two or three El four or five 1:] six or seven Please look at the list of hobbies below. Has the child had regular contact with an adult with one of the hobbies below? Cl no (If “no,” please skip to Question 8). [:1 yes (If “yes,” please check all hobbies that apply). El Automobile or Boat Radiator Repair 13 Making Lead Shot, Fishing Sinkers or Bullets [:1 Casting Lead Figures El Painting (exterior, marine or highway (toy soldiers, etc.) Paint) [:1 Floor, Woodwork, or Furniture i El Pottery/Ceramic Ware El Jewelry Making Cl Stained Glass Making [:1 Lead Crystal Making El Target Shooting with Guns Cl Lead Soldering (for electronics, etc.) 55 78. How many days per week does the child spend with this adult? El none B one El two or three 13 four or five Cl six or seven 8. To which racial and ethnic groups does the child belong? Please check all that apply. El White [:1 American Indian or Alaskan Native [:1 Black/Afi'ican American 13 Native Hawaiian or Pacific Islander 13 Asian Cl Hispanic or Latino [:1 Arab or Chaldean C] Other: 9. In how many places has the child lived since birth? 10. Please think about the current place that the child lives and any previous place that the child has lived. If the child has lived in more than one place before the current residence, please use as “previous place” the previous place that the child lived in longest. Please circle, or fill in, the response which fits each place. Current Previous Place Place that the Child Lived in Longest Before 1950 Before 1950 _ . 1950 to 1980 1950 to 1980 When was '1 3m"? After 1980 After 1980 Don’t Know Don’t Know Child Moved In? / / (month/year) (month/year) Child Moved Our? / / (month/year) (month/year) Was it Owned or Rented? Owned Rented Owned Rented Did the Drinking Water Come From Lead Pipes? Yes No Yes No Don’t Know Don’t Know Was it Remodeled or Renovated While You Lived There? Yes No Yes No Don’t Know Don’t Know Did it Have Peeling or Chipping Paint on the Inside? Yes No Yes No Don’t Know Don’t Know Did it Have Peeling or Chipping Paint on the Outside? Yes No Yes No Don’t Know Don’t Know Did your Child Play Outside (Within Three Feet of the Yes No Yes No building)? Don’t Know Don’t Know 10A. What is the address of the previous place that the child lived in longest? [:1 St El Ave El Rd E] DrCI CirClBlvdElSq Cl Tr Address: City: State: Zip Code: DSt DAve Ele ClDr ClCir UBlvdElSquTr name Of nearest cross-street -56- + ll. 12. l3. 14. 15. l6. l7. 18. Has the child ever lived within a mile of a battery plant, lead smelter, or other industrial facility where lead may be used? El yes [:1 no El don’t know Which best describes how you keep your windows during the summer? [:1 almost always open D almost always closed [:1 they are opened and closed each day Think about all adults who live with the child. Have any of them ever been told that they had lead poisoning or high blood lead levels? El yes C] no D If yes, relationship to child: Think about the child’s brothers or sisters. Have any of them ever been told they had lead poisoning or high blood lead levels? El yes [:1 no El never tested Think about the child’s playmates. Have any of them ever been told that they had lead poisoning or high blood lead levels? [:1 yes Cl no What kind of health insurance covers this child? (Check one) D Medicaid El Individually Purchased Insurance 1:] MICHILD Cl No Insurance El Other Government Program [:1 Don’t Know [:1 Insurance Provided by Job Where do you usually take your child for health care? (Check one) D Doctor’s Office El Hospital Out-patient Clinic [3 Community Health Center Cl Urgent Care El Public Health Clinic El Don’t Know Cl Hospital Emergency Room Last year, how much was your total household income? Please count all sources, including your job. (Check one) El Less than $5,000 Cl $30,000 to $39,999 [:1 $5,000 to $9,999 Cl $40,000 to $49,999 [:1 $10,000 to $14,999 Cl $50,000 to $59,999 Cl $15,000 to $19,999 [:1 More than $60,000 Cl $20,000 to $29,999 -57- References Agresti, Alan and Barbara Finlay. 1997. Statistical Methods for the Social Sciences. 3rd edition. Upper Saddle River, NJ: Prentice Hall. Bernard, Susan M. 2003. Should the Centers for Disease Control and Prevention Childhood Lead Poisoning Intervention Level Be Lowered? American Journal of Public Health. Vol. 93, No. 8: 1253 -- 60. Binns, Helen, LeBaily, Susan A., Fingar, Ann R., and Saunders, Stephen. 1999. Evaluation of Risk Assessment Questions Used to Target Blood Lead Screening in Illinois. Pediatrics 103: 100-106. Bomschein RL, Succop P, Kraft KM, Clark CS, Peace B, Hammond PB. 1987. Exterior surface dust lead, interior house dust lead and childhood lead exposure in an urban environment. In “Trace Substances in Environmental Health” (Hemphill DD, ed). Proceedings of University of Missouri’s 20th Annual Conference, Vol. 20. Columbia, MO: University of Missouri, 322- 332. Brody DJ, Pirkle JL, Kramer RA, et al. 1994. Blood Lead Levels in the US Population: Phase I of the Third National Health Examination Survey (NHANES III, 1988 to 1991). JAMA. 272: 277-283. Canfield, Richard L, Henderson Jr., Charles R., Cory-Slechta, Deborah A., Cox, Christopher, Jusko, Todd A., and Bruce P. Lanphear. 2003. Intellectual Impairments in Children with Blood Lead Concentrations Below 10 pg per Deciliter. NEJM. 348(16): 1517—26. CDC. 1991. Preventing Lead Poisoning in Young Children: A Statement by the Centers for Disease Control. Report No. 99-2230. Atlanta, GA: Centers for Disease Control and Prevention. Clark, CS, Bomschein R, Succop P, Roda S, Peace B. 1991. Urban Lead Exposures of Children. J Chem Speciation Bioavail Vol 3: 163-171. Jacobs DE. 1995. Lead-based paint as a major source of childhood lead poisoning: A review of the evidence. In “Lead in Paint, Soil, and Dust: Health Risks, Exposure Studies, Control Measures, and Aquality Assurance (Beard ME, Iske SDA, eds). ASTM STP 1226. Philadelphia: American Society for Testing and Materials, 175-187. Kaplowitz, S. H. Perlstadt, L. A. Post, Y Bouraoui, and M. Barrat. 2001. “Testing the Predictiveness of the Elements of the Michigan Statewide Screening Plan -53- and other measures” Part C: Supplemental Studies Grant to Michigan Dept of Community Health, by Centers for Disease Control. Kazal, Louis A. MD. 1997. The Failure of CDC Screening Questionnaire to Efliciently Detect Elevated Lead Levels in a Rural Population of Children. The Journal of Family Practice. Vol 45, No. 6: 515-518. Lanphear, BP, Weitzman, M, and Eberly, S. 1996. Racial Dijfkrences in Urban Children’s Environmental Exposures to Lead. American Journal of Public Health. Vol. 86, No. 10: 1460-63. Lanphear, BP, Emond M, Jacobs DE, Weitzman M, Tanner M, Winter N, et al. 1995. A Side by Side Comparison of Dust Collection Methods for Sampling Lead- Contaminated House Dust. Envion Res 86: 114-123. Lanphear BP, Matte TD, Rogers J, Clickner RP, Dietz B, Bomschein RL, et al. 1998. The Contribution of Lead Contaminated House Dust and Residential Soil to Children ’s Blood Lead Levels: A Pooled Analysis of 12 Epidemiological Studies. Environ Res 79: 51-68. Lanphear B P, Roghmann KJ. 1997. Pathways of Lead Exposure in Urban Children. Env Res 74(1): 67-73. Last, John M. Ed. 1995. A Dictionary of Epidemiology. Assoc Eds. J .H. Abramson et al. New York, Oxford Press. Lee, Charles. 2002. Environmental Justice: Building a Unified Vision of Health and the Environment. Environmental Health Perspectives. 110 (suppl 2): 141 — 144. Mahaffey KR, Annest J L, Roberts J, Murphy RS. 1982. National Estimates of Blood Lead Levels: United States, 1976-1980. Association with Selected Demographic and Socioeconomic Factors. N Engl J Med. 307: 573-579. Markowitz, Gerald E., and Rosner, David. 2002. Deceit and Denial: The Deadly Politics of Industrial Pollution. Berkeley: University of California Press. McElvaine MD, DeUngria EG, Matte TD, Copley CG, Binder S. 1992. Prevalence of Radiographic Evidence of Point Chip Ingestion among Children with Moderate to Severe Lead Poisoning, St. Louis Missouri, 1989-1990. Pediatrics 89: 740-742. Michigan Department of Community Health. 2002. Childhood Lead Poisoning Data Facts — All Michigan Counties (I 998 - 2001). Mielke, Howard W. 1999. Lead in the Inner Cities. American Scientist. Vol. 87: 62-73. -59- Needleman, H.L. 1994. Preventing Childhood Lead Poisoning. Preventative Medicine, Vol. 23. Norusis, Marija J. 1993. SPSS® for WindowsTM Advanced Statistics Release 6.0. Chicago: SPSS Inc. Phelps, LeAdelle. 1999. Low-Level Lead Exposure: Implications for Research and Practice. School Psychology Review. Vol.28, No. 3: 477-492. Rabinowitz M, Leviton A, Bellinger D. 1985. Home Refinishing: Lead-Based Paint and Infant Blood Lead Levels. Am J Public Health 75(4): 403-404. Raloff, Janet. 2001. Even Low Lead in Kids has a High IQ Cost. Science News. Vol. 159, No. 18:277. Shannon MW, Graef JW. 1992. Lead Intoxication in Infancy. Pediatrics 89(1): 87- 90. Snyder et al. 1995. Dev elopement of a Population Specific Risk Assessment to Predict Elevated Blood Lead Levels in Santa Clara County CA. Pediatrics 93: 643-648. Tabachnick, Barbara G. and Linda S. Fidel]. 2001. Using Multivariate Statistics. 4‘h edition. Boston: Allyn and Bacon. Tejeda D. et al. 1994. Do Questions about Lead Exposure Predict Elevated Lead Levels? Pediatrics 93: 92-94. US. Census Bureau, Current Population Survey, 2000, 2001, and 2002. Annual Demographic Supplements. Weintraub, Max. 1997. A Childhood Lead Poisoning: An Environmental Justice Success Story? A Commissioned paper for Conference on Healthy and Sustainable Communities: Building Model Partnerships for the 21st Century (March 20, 1997) Atlanta, GA., Clark Atlanta University, Environmental Justice Resource Center (www.cjrgauedu) -60- will lG twilightTilt" 1 5.1..- '02-.“