I
339‘] —- —-~
LIBRARY
Michigan Sftute a
UniveISIW

 

 

This is to certify that the
thesis entitled

CARIES PATTERNS IN 7 YEAR OLD C§ILDREN IN

BELGIUM: THE SIGNAL TANDMOBIEL PROJECT.

presented by

OBIANUJU HELEN NNAMA

has been accepted towards fulfillment
of the requirements for the

  

MS. EPIDEMIOLOGY

 

 

 

 

8/28/09

 

Date

MSU is an Affinnative Action/Equal Opportunity Employer

 

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

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/08 K:IProj/Aoc&Pres/ClRCIDateDue.indd

CARIES PATTERNS IN 7 YEAR OLD CHILDREN IN BELGIUM: THE SIGNAL
®
TANDMOBIEL PROJECT.
By

Obianuju Helen Nnama

A THESIS

Submitted to
Michigan State University
In partial fiIlfillment of the requirements
for the degree of

MASTER OF SCIENCE
Epidemiology

2009

ABSTRACT

CARIES PATTERNS IN 7 YEAR OLD CHILDREN IN BELGIUM: THE SIGNAL
®
TANDMOBIEL PROJECT.

By
Obianuju Helen Nnama

Dental caries or tooth decay is the most common oral health problem. Although it
is quite preventable, it still remains the most common chronic disease of children aged 5
to 17 years, and has been linked with poor academic performance, low self confidence
and poor peer interactions.

Despite the many epidemiologic studies conducted to assess the intra-oral pattern
of dental caries, there are still some unanswered questions. There are four quadrants in
the mouth believed to be inter-correlated and it is widely believed that dental caries
develop symmetrically within the mouth but much statistical results are still needed. Most
of the previous studies have relied on the usage of the decayed, missing, and filled (DMF)
index developed in the 19303, to analyze dental caries outcomes. It has its limitations in
explaining the development of caries and is unable to measure tooth-specific problems
such as caries patterns. It is speculated that a more effective manner to study the spatial
distribution of dental caries may be to analyze using adequate statistical models such as
generalizing estimated equations (GEE), and Alternating Logistic Regression (ALR).
Such models can be used to determine tooth surface susceptibility to caries formation and
to answer questions regarding dental caries spatial distribution that may exist in the
mouth. These findings should enable dentists/hygienists to detect dental caries at earlier

stages and provide more subject specific treatment plans.

Cepyright by
OBIANUJU HELEN NNAMA

2009

ACKNOWLEDGMENTS

I would like to express my sincere gratitude to my advisor Dr. David Todem for
all his assistance and encouragement. His dedication to the research, as well as all his
generous help with the statistical models allowed me the opportunity to accomplish this

task, and for this I am truly grateful.

®
My thanks go to everyone associated with The Signal Tandmobiel Project

globally for allowing me to use this data set. Appreciation is extended to Dr. Zhang for
his initial contribution to the research.

I would also like to extend my appreciation to Dr. Fu and Dr. Chung for being my
thesis committee members, and lending their support to this project. Appreciation is also
extended to the entire Department of Epidemiology faculty and staff who have
contributed in many ways throughout my time here.

I extend my deepest gratitude to my parents and siblings for their unending love
and great support in helping me with my academic pursuits and with life. Especially for

instilling in me a strong foundation of faith in God, from which I draw my strength in life.

iv

TABLE OF CONTENTS

LIST OF TABLES ................................................................................. vi
LIST OF FIGURES ............................................................................... vii
CHAPTER 1 ...................................................................................................................... 1
Introduction ....................................................................................................................... 1
1.1 Background and objectives ................................................................................. 1
1.2 Literature review ................................................................................................. 2
CHAPTER 2 ...................................................................................................................... 6
Population and Methods .................................................................................................. 6
. . . ® .

2.1 Study Population: The Signal Tandmobiel Pr0ject ........................................ 6

2. 1 . 1 Introduction ................................................................................................. 6
2.1.2 Study Design ............................................................................................... 6
2.1.3 Project Conduct ........................................................................................... 7
2.1.4 Data ............................................................................................................. 8

2.2 Statistical Methods and Design ........................................................................... 9
2.2.1 Research Questions/Purpose of the study ................................................... 9
2.2.2 Modeling correlated binary data using the alternating logistic regression 11
CHAPTER 3 .................................................................................................................... 18
Analysis and Results ....................................................................................................... 18
Descriptive analysis ...................................................................................................... 18
CHAPTER 4 .................................................................................................................... 26
Discussion ......................................................................................................................... 26
4.1 Research Study .................................................................................................... 26

4.2 Implications of findings. ..................................................................................... 28
BIBLIOGRAPHY ........................................................................................................... 3O

LIST OF TABLES

2. 1 Parameter 7jk based on tooth position in the quadrant ....................................... .13

2. 2. Description of the log odds ratio Parameters au(jk,j*k‘) .................................. 15

3. 1. Prevalence of caries experience (% affected) in the primary dentition of 7-year-old
children n=4,351. .............................................................................................................. 18

3. 2 Analysis of dental caries experience: results from using ALR model for estimates of
the log odds of dental caries for each tooth (standard errors) and corresponding 95%
Confidence Interval in 7-year-old children ............................................................... 20

3. 3 Analysis of dental caries experience: results of comparison of prevalence of dental
caries in upper and lower jaws in 7-year-old children. ............................................. 21

3. 4 ALR Results: log odds ratio estimates, standard error, 95% confidence interval and

Z score for dental caries for all teeth in 7-year-old children ..................................... 22

3. 5 Hypotheses Test Results ........................................................................................... 23

vi

 

 

m: '

 

 

LIST OF FIGURES

2. 1 The location of the deciduous teeth in the mouth (adapted from Bogaerts et a1.
2002)... .. .................................................................................................................. 10

2. 2 Parameters representing the quadrant-level association ........................................... 16

vii

CHAPTER 1

Introduction

1.1 Background and objectives

Dental caries is a leading cause of tooth loss, which can affect chewing, and food
intake selection. In the children population especially where nutrients are needed for
growth, this can lead to an increased risk of systemic diseases such as cardiovascular
conditions at an adult age.

Dental caries is defined as “the dynamic de- and re—mineralizing processes
resulting from microbial metabolism on the tooth surface. Overtime this may result in the
net loss of mineral and subsequently possibly, but not always, will lead to cavitations”
(Fejerskov, 1997). Dental caries occurs as a result of the interplay of three main factors
over time: dietary carbohydrates, bacteria within dental plaque, and susceptible hard
tooth surfaces. The hard tissues of the crown (coronal, enamel, and dentine) and root
(cementum, dentine) surfaces of teeth are affected in the demineralization process, which
is caused by acids produced by bacteria. There are a number of bacteria linked to dental
caries namely mutans Streptococci and possibly Lactobacilli. Each carious process
begins with the formation of biofilms on the surface of the tooth involved. The biofilm is
able to form on any surface where there is adequate supply of water and nutrients such as
the oral cavity. The pellicle of the tooth surface present on enamel, dentine and
cementum is the key location where the biofilm initiates (Kidd and F ejerskov, 2004). The
formation of biofilm and plaque is a stage wise process, where each stage shows a more
complex colonization of the bacteria, increased selectiveness of the biofilm regarding

bacteria, and an advanced level of organization in the biofilm. Over time, the film

hardens with the deposition of minerals and the secretions produced by the bacteria. The
constant metabolic activity of these bacteria ensures constant multiplication and thereby
increased thickness of the plaque, which in turn requires constant nutritional support
(Kidd and Fejerskov, 2004). Most of this support is met by the food and liquid intake via
the oral cavity into the mouth, with saliva being the medium of nutrition transport. The
bacterial secretions produced drop the pH to around 5.3, and causes demineralization of
the tooth surfaces. This leads to “softening” of the tooth structure, which gives way to
cavity formation (Kidd and Fejerskov, 2004).

The plaque and saliva are saturated with calcium and phosphate ions, so if the pH
returns fairly rapidly above the 5.3 levels, ions will re-enter the enamel and re-crystallize.
The process of re-mineralization takes longer in an acidic environment, but is rapid if the
fluid next to the enamel is neutral or even alkaline. It is therefore uncommon to find
caries in those parts of the mouth near the outflow of salivary glands, like the lower
incisors, where the environment is filled with buffers and concentrated calcium ions of
saliva (Manji et a1, 1991). Thus saliva flow rate and composition can influence the
progression rate of caries lesions (Melberg, 1986). If the total outflow of saliva is
increased, there is a lower risk of caries development. Caries develop when the process of

re-mineralization is slower than the process of demineralization.

1.2 Literature review
Previous studies have shown that certain teeth are more susceptible to dental

caries than others. The maxillary incisors are more prone to developing caries than the

mandibular incisors. Posterior teeth which have pit-and-fissure (occlusal) surfaces are

more prone to dental caries formation than anterior teeth with smooth (labial and lingual)
surfaces (Reid and Grainger, 1955). A recent study of caries susceptibility has replicated
previous findings that teeth can be placed in the order of susceptibility from greatest to
least, as follows: 1) mandibular (lower) second molars, 2) mandibular first molars, 3)
maxillary (upper) second premolars, 4) maxillary first premolars, 5) maxillary central and
lateral incisors, 6) maxillary and mandibular canines and 7) mandibular central and
lateral incisors (Macek et al., 2003)

While understanding the caries susceptibility of teeth is informative, it does not
provide information regarding the progression of the disease. Dental caries has unique
characteristics in that it represents both past and current disease processes. When a person
goes to the dental clinic for a routine examination, it would be beneficial for the dentist to
have an understanding regarding caries patterns to help suggest subject specific
preventive measures. The traditional DMFT/S “Decayed/ Missing/ Filled Teeth/Surface”
index is a well known method for analyzing the prevalence of caries. It was introduced by
Klein et a1. (1938) and provides aggregated scores which provide a summary of mouth-
level caries information for each individual at tooth level or tooth surface level. It counts
the number of decayed teeth within the individual’s mouth. It is beneficial in the
evaluation and comparison of dental caries risks among different population groups but it
does not give detailed information among individuals regarding which specific teeth is
decayed (Todem, 2007). For a dentist, counting the number of decayed, missing and
filled teeth alone does not help address intra-oral caries incidence pattern. When a patient
comes for a routine dental examination and a caries lesion is found on a specific tooth in

a quadrant, a dentist with knowledge of intra-oral caries distribution patterns will be able

to predict the next tooth with a high risk of developing caries. It has been suggested that
the use of tooth-level models should enhance the interpretation of complex dependent
data generated from studies in dental research.

The initial attempt at understanding the caries experience was to establish first
whether the patterns observed are symmetric or asymmetric in nature. Recent studies
have helped identify three more categories of carious lesions based on their location. The
“random” caries series is one where the lesions have equal opportunity of developing in
the contra-lateral sides of the mouth (Vanobbergen et al., 2007). The “aggregated”
pattern of caries shows the concentration of carious lesions on one side of the mouth,
while the “regular” pattern shows a more symmetrical distribution of caries on both sides
of the mouth (V anobbergen et al., 2007). Using the data from the National Survey of Oral
Health (1985-1986), the Hujoel et al (1994) study investigated three different caries
patterns (random, aggregated, and regular) seen in 12,776 subjects. From the study they
determined that the distribution of caries lesions on the teeth and surfaces were not
random but rather aggregated or clustered (P<0.0001) across the different subgroups. A
two-sided test was performed, which resulted in a large calculated z-score rejecting the
null hypothesis of a random pattern, in favor of an aggregated pattern. This result
confirmed previous findings that adjacent pairs have a higher risk of developing caries.
The study by Hujoel et. al., (1994), provided possible explanations that might explain this
aggregated caries pattern, one of which suggests that the neighboring teeth are infected
by the proximity of the caries lesions regardless of it being on the right or left side

(Hujoel et a], 1994).

The goal of the Ananth and Kantor (2004) study was to assess if caries aggregated
within a subject. This complex dataset utilized a radiographic efficacy National Institutes
of Health (NIH) data with multiple levels of nesting. In this study, they investigated the
clustering of caries on each of the 16 tooth surfaces per subject. There was a complex
multiple levels of nesting: tooth surfaces within an interproximal (IP) regions, IP rejoins
within a jaw, and jaws within a subject. It was reported that caries lesions appear to
aggregate strongly within subjects with a spatially distributed risk (p<0.05).

Using data from two Belgium surveys: the Signal Tandmobiel® (ST) project and

the Tandje de Voorste-Smile for Life (TDV) project, the Vannobergen et a1 (2007) study
investigated the distribution and spatial correlation of caries lesions. This sample covered
1,291 3-year old and 1,315 5-year old children. from the TDV project and 4,468 7-year
old Flemish school children from the ST project. A symmetrical pattern was noted at the
population level, left-right caries distribution patterns appearing strongest in the mandible.
At an individual level, randomness of caries patterns could be rejected, as caries tended to
cluster on one side of the mouth (p=0.0005).

The prevailing hypothesis remains that there is indeed a spatial distribution of
dental caries, and a need for statistical models that allow for such evaluations. To answer
this question, there is a need for data collected at the tooth or tooth-surface level (Hujoel

et al., 1994).

CHAPTER 2
Population and Methods

®
2.1 Study Population: The Signal T andmobiel Project

2.1.1 Introduction

The Signal Tandmobiel project is an extensive longitudinal and cross-sectional

study carried out from 1996 till 2001, and conducted with the support of three Flemish
universities (Catholic University at Leuven, University of Gent, Free University at
Brussels), the Association of Flemish Dentists (VV T), the Flemish scientific Association
for Youth Health (VWVJ) and the industry (LeverElida). The Signal Tandmobiel®
project’s objectives were to carefully examine the oral hygiene of the school children
ages 7 to 12 year old over six years, and assess the effect of yearly oral hygiene education
conducted at the random schools. The three parts of the study were: (1) oral health
condition screening, (2) obtaining health information and education, and (3) a scientific

contribution.

2.1.2 Study Design
The study included 4468 Flemish children in Belgium, who were selected by a

stratified clustered random sample. The children were born in 1989, and followed
throughout elementary school. At enrollment, children were about 7 years of age, and by
the end of the study, they were 12 to 13 years of age. An equal number of boys and girls
were examined. The selected schools were divided into three groups known as samples A,

B and C. Group A consists of a cohort of children who were followed longitudinally

during their 6-year period in elementary school from 1996 to 2001 and were examined
yearly and also subjected to a yearly educational program. The cohort of children in
Group B, were examined in the first and last year of the study. Those in Group C were
sampled yearly, starting from the second year, to allow a cross-sectional comparison with
group A. Group B was used as a control in assessing the effect of the examinations and
educational program in group A. Group B is identical to Group C in year 1. Schools,
instead of children, were selected randomly with a probability proportional to the number
of children in school. The sampling was done after stratification for educational system
and province. This stratification allows that the group of children is representative of the
population. Whenever a school was selected, all children in the first grade of the selected
school were allowed to participate. Selecting individual children instead of schools would
not be feasible for practical and economical reasons and it would be unfair to allow only

some of the children from the same school to participate.

2.1.3 Project Conduct
The children were examined by trained dentists in a mobile dental clinic

(Tandmobiel), using the standardized and well established diagnostic criteria of the
World Health Organization (WHO) and the British Association for the Study of
Community Dentistry (BASCD). There were 16 dentists in the “Tandmobiel” who had
been trained and evaluated on a yearly basis to conduct the dental examination. The
examination was divided into three main parts; the evaluation of the dental hygiene, the
evaluation of the gingival health status and the examination of hard dental tissue for the

presence of caries and fillings. In addition to the examinations, questionnaires were also

administered; there was a parental questionnaire that addressed specific questions
regarding the child’s oral hygiene, fluoride exposure, visits to a dentist, dietary habits and
dental trauma. A medical center questionnaire was completed for each child by a
physician at the medical center that provided name, date of birth, gender, nationality,
school, class, socioeconomic status, region and medical precedents. A preventive health
education program was an important component of the project, and was adapted each
year to fit the age of the children. The message was conveyed through the use of slides,
tapes, brochures, and a reward. Study results were not directly reported to the child but in
considerable details to the medical center and by means of a letter to the parents for

possible preventive instructions, advise, treatment or referral to their dentist.

2.1.4 Data
The collected outcome data comprise information on (1) the caries experience of

all deciduous and permanent teeth, including the tooth surfaces affected by caries, (2) the
plaque index measured at six locations in the mouth, (3) the gingival condition, etc.
Questionnaires on oral hygiene behaviors are distributed yearly to the children, filled out
by the parents and returned on the day of the examination in the Tandmobiel. Risk factor
information such as the consumption of candy and dry biscuits, the utilization of
toothpaste and with or without fluoride, the brushing habit of the child as well as that of
her parent, the visit of the child to the dentist, etc, were also collected. Other aggregated
information such as the nationality, the province, the urbanization status of the place of
residence and the educational system of the child were also collected. More description of

the Signal-Tandmobiel® can be found elsewhere (Vanobbergen et al., 2000).

 

2.2 Statistical Methods and Design

2.2.1 Research Questions/Purpose of the study
This study aims to investigate the spatial distribution of dental caries on tooth

surfaces in the primary dentition of 7 year olds, using statistical models for correlated

data. The data used are from the Signal Tandmobiel® study. Our study sample was

restricted to 4351 school children who attained their seventh birthday anniversary during
the course of the study. Specifically, in this study, we are interested in the following
questions:

0 Which tooth surfaces are more susceptible to caries experience?

0 With respect to dental caries, is there any spatial association of dental caries at the

quadrant level?

An answer to these questions requires the analysis to be performed at the tooth or tooth-
surface level. This then necessitates the use of methods for correlated data. In dental
research, intra-mouth data present a unique set of challenges to statistical analysis
(Leroux et al., 2006). These challenges include, but are not limited to, large cluster sizes
(large number of teeth within a mouth); informative cluster sizes and multilevel data
structures which generate a very complex correlation structure.

Figure 2.1 is an illustration of the primary dentition. In Europe, these deciduous
teeth are numbered with two digits where the first digit represents the quadrant and the
second digit the position of the tooth within a quadrant. As an example, tooth 55
represents the last molar of the fiflh quadrant. Quadrants V and VI form the maxilla or
the upper jaw whereas quadrants VII and VIII form the mandible or the lower jaw. This

pictorial representation of the primary dentition exhibits a two-level spatial association

structure. The first-level Spatial association structure is that among quadrants (V)-(VIII),
whereas the second-level spatial association structure is that of teeth nested within a

quadrant.

 

 

 

.59

6
4a
81

84

33
VIII 82

 

 

Figure 2.1 The location of the deciduous teeth in the mouth (adapted from
Bogaerts et al. 2002).

Vanobbergen et al. (2007) have proposed a GEE—based model to answer some of the
questions related to the spatial association of dental caries in the mouth. Specifically,
these authors have used so-called alternating logistic regression (ALR) to model
simultaneously the marginal expectation of each binary outcome and the pair-wise
associations between outcomes (Liang, Zeger and Qaqish, 1992). This model was used to
circumvent some of the numerical problems related to the second order moments-GEE2
model due to larger cluster sizes. Although the ALR model is known to be numerically
stable, its convergence for unstructured second order moments remains an issue for
moderately larger cluster Sizes. Indeed, when the cluster size is relatively large, one may

encounter a numerical difficulty when an unstructured odds ratio model is imposed. This

10

 

 

 

 

limitation was the reason for these authors to perform their analysis on a small subset of
teeth (8 per mouth). When the ALR model with unstructured odds ratios is attempted on
the full dataset (20 teeth per mouth), the algorithm did not converge. To circumvent
some of the computational difficulties of the ALR model for larger cluster sizes, we
imposed some structure on the odds ratio model. These constraints are carefirlly chosen,
as overly simplistic models may limit our ability and flexibility to study the spatial
distribution of dental caries in the mouth. We discuss this in the section below.

2.2.2 Modeling correlated binary data using the alternating logistic
regression

Generalized Estimating Equations (GEEs), developed for the analysis of
correlated data, provide a flexible method for modeling the marginal mean response
while adjusting inferences for correlation. The estimates of the mean response that this
approach yields are therefore population-averaged estimates, analogous to those obtained
from cross-sectional data. GEE uses a “working correlation” to treat the within-subject
correlation, and produces standard error estimates that take into account the correlation of
responses within subjects (Liang and Zeger, 1986; and Zeger and Liang, 1986). As a
result, the within-subject correlation is not modeled explicitly but adjusted for using the
so-called sandwich-based estimator of the variance covariance matrix of the parameter
estimates. The classical GEE model treats the intra-oral correlation as a nuisance
parameter. This is troublesome since the intra-oral association terms are important in
understanding the spatial distribution of dental caries in the mouth.

The so-called Alternating Logistic Regression (ALR) model developed by Carey,

Zeger, and Diggle (1993), overcomes this problem by explicitly modeling the association

ll

 

between the intra-oral caries outcomes. The algorithm switches back and forth between

the GEE and the logistic regression, hence the name ALR. Another alternative to
modeling association using odd ratios is the use of Pearson correlation coefficient. But as
pointed by Diggle, Liang, and Zeger (1994) modeling association among binary
responses with correlation has a disadvantage, and they propose using the odds ratio

instead. We illustrate the application of the ALR model to data on intra-oral caries

outcomes. For tooth k in quadrant j of child i, we denote by yijk a binary response

variable taking value 1 if caries is present and 0 if otherwise. The first moment

P(yijk = 1) is modeled as:

0g P(yijk =1) _ 7
1—P(y,-,. =1) 1"

 

where 711. represents the log odds of caries for tooth k in quadrant j. These parameters

help estimate the caries prevalence for any tooth at a specific location in the mouth. In
this logistic regression model, the position of the tooth in the mouth is treated as the
independent variable. Details are given in the Table 2.1.

Note that teeth 71, 72, 81, and 82 were removed from our analyses as these teeth were

primarily unaffected by caries. To compare the frequency of dental caries between two

teeth, the parameter 7}]; is used. As an example, to compare teeth 85 and 75, we

Consider the following null hypothesis,

785 =775°

12

 

Table 2.1 Parameter yjk based on tooth position in the quadrant

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

QUADRANT POSITION TOOTH m PARAMETER
5 5 55

755
5 4 54

754
5 3 53

753
5 2 52

752
5 1 51

751
6 1 61

761
6 2 62

762
6 3 63

763
6 4 64

764
6 5 65

765
8 5 85

785
8 4 84

784
8 3 83

783
7 3 73

773
7 4 74

774
7 5 75

775

 

In addition to these first moments, we model the association parameters as,

P(yijk =1)/(1—P(yijk =1))\

 

P(yij.k. = 1)/(1— P(yij.k. = 1))

13

 

utjkfk‘)’

"' T - at“:
" ' -' . 33A~2"-§= In"... ..
-"' rm?) :...;-°—rmw -. w

 

Here the parameter au(jk,j‘k‘ ) measures the log odds of dental caries for tooth k in

quadrant j relative to that of tooth k* in quadrant j*. A full description of au(jk,j'k')

is given in Table 2.2. As an example, the parameter a, captures the lag l association in
quadrant 5. This lag l association is captured by the pairs of teeth (55,54), (54,53),
(53,52), and (52,51) of the fifth quadrant. This model obviously assumes that the odds
ratio of dental caries for the pair of teeth (55,54), is the same as that of the pairs (54,53),
(53,52) and (52,51). The parameter a22 captures the constant log odds ratio between any
tooth in quadrant 5 to any tooth in quadrant 8. These constraints may be viewed as
restrictive, but as stated in the section above, we have imposed these conditions to ensure

convergence of the ALR model. Other restrictions are shown in the Table 2.2.

14

Table 2. 2 Description of the log odds ratio Parameters au( jk, j ‘k‘)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

PARAMETERS QUADRANT 4. a.
a POSITION TOOTH PAIRS (jk. j k )
uum‘k‘)
(11 Q5 lagl (55,54),(54,53),(53,52),(52,51)
(12 Q5 lag2 (55,53),(54,52),(53,51)
(13 Q5 lag3 (55,52)(54,51)
(14 Q5 lag4 (55,51)
(15 Q6 lagl (61,62)(62,63)(63,64)(64,65)
(16 Q6 lag2 (61,63)(62,64)(61L65)
(17 Q6 lag3 (61,64)(62,65)
(18 Q6 lag4 (62,65)
(19 Q8 lagl (85,84)(84,83)
(110 Q8 lag2 (85,83)
(1111 Q7 lagl (73,74)(74,75L
(112 Qj lag2 (73,75)
(113 Q5Q6lagl (51,61)
(114 Q5Q6lag2 (52,61)(5L62)
(115 Q5Q6lag3 (53,61)(52,62)(51,63)
(116 Q5Q6lag4 (54,61)(53,62)(52,63)(5 1,64)
(117 Q5Q6la$5 (55,61)(54,62)(53,63)(52,64)(51,65)
(118 Q5Q6lag6 (55,62)(54,63)(53,64)(52,65)
(119 Q5Q6lag7 (55,63)(54,64)((53,65)
(120 Q5Q6lag8 (55,64)(54,65)
(121 Q5Q6lag9 (55,65)
(122 Q5Q8 (55,85)(55,84)(55,83)(54,85)
(54,84)(54,83)((53,85)(53,84)
(53,83)(52,85)(52,84)(52,83)
(51,85)(51,84)(51,83)
(123 Q5Q7 (55,73)(55,74)(55,75)(54,73)
(54,74)(54,75)(53,73)(53,74)
(53,75)(52,73)(52,74)(52,75)
(51,73)(51,74)(51,75)
(124 Q6Q8 (61,85)(61,84)(61,83)(62,85)
(62,84)(62,83)(63,85)(63,84)
(63,83)(64,85)(64,84)(64,83)
\ (65,85)(65,84)(65,83)
(125 Q6Q7 (61 ,73)(61 ,74)(61 ,75)(62,73)
(62,74)(62,75)(63,73)(63,74)
(63,75)(64,73)(64,74)(64,75)
\ (65,73J(65,7fl(65,75)
\ (126 Q8Q7lag5 (83,73)
\ (127 Q8Q7lag6 (84,73)(83,74)
-\ (128 Q8Q7lag7 (85,73)(84,74)(83,75)
(129 Q8Q7lag8 (85,74)(84,75L
(130 Q8Q7lag9 (85,75)

 

15

 

 

® (113- (121 @

u23 a24
u22 a25

 

 

 

@ ®
(126- (130

Figure 2. 2 Parameters representing the quadrant-level association

In Figure 2.2, we present the parameters used to assess the spatial association of

caries outcomes at the quadrant level. As an example, the set of parameters 6213 to 0:21
denoted a'I3 - a2, captures the left right association for the upper jaw and the parameters

(126 —a30 that of the lower jaw. Our model assumes that “across” and the up-down

associations are captured by a single parameter. As an example, the association between
quadrant 5 and quadrant 8 is captured by 6222. This model assumes that the log odds ratio
Of dental caries of any tooth in quadrant 5 relative to any tooth in quadrant 8 is constant.
S imilar assumptions are made for quadrant 5 and quadrant 7, quadrant 6 and quadrant 7,
quadrant 6 and quadrant 8. This structured odds ratio model was necessary to achieve

II‘Odel convergence. Any attempt to leave these association terms unstructured, lead to

c QIrrputational challenges.

16

Hence, to assess the left-right quadrant-level association as compared to other

quadrant-level association, we consider the following null hypothesis,

“13:21+a26:30 _ “22 + a25 +5123 +5124

14 4

 

 

In the formula above, we define (11m... = au + au+1 +A + duh] + a“:

foru < u'. To evaluate whether the association between quadrants 5 and 6 is the same as

that of quadrants 7 and 8, we consider the following null hypothesis,

a13:21 _ “26:30

9 5

Other hypotheses tests about model parameters, which are not of primary importance, can
be evaluated to reduce the number of parameters in the working model. This will be
discussed in the results section. These hypotheses can be evaluated using the classical
Wald test statistic with its chi-squared limiting distribution. For the specific null

hypotheses examples above, a 1 degree of freedom Wald test can be used to conduct the

test

17

CHAPTER 3

Analysis and Results

Descriptive analysis

From table 3.1, it can be seen that the highest prevalence of dental caries is found

in the molars of the mandible (lower jaw). This is consistent with the findings of previous

studies (see for example Zhang et al., 2009). There is also as expected, an extremely low

prevalence of dental caries in the incisors of the lower jaw teeth (82, 81, 71, 72 ).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 3.1 Prevalence of caries experience (% affected) in the primary dentition
of 7-year-old children n=4,351.
TOOTH 55 54 53 52 51 61 62 63 64 65 I
Prevalence 8.92 5.20 0.74 3.72 7.81 7.06 2.23 1.86 5.20 8.55 I
TOOTH 85 84 83 82 81 71 72 73 74 75 1
Prevalence 10.78 13.75 1.12 0.74 0.37 0.37 0.37 0.37 11.15 9.67 I

 

 

 

 

 

 

 

 

 

 

 

These descriptive statistics suggest a left-right spatial symmetrical pattern in

terms of caries frequencies. Specifically, paired teeth 5k-6k, k =1,2,. . .,5 have roughly the

Same caries observed fiequencies. Similarly findings are observed for paired teeth 8k-7k,

k =3,4,5. One, however, should be careful in interpreting these results as they only

1' e[Dresent population averaged frequencies, which may not be relevant in understanding

the spatial distribution in the mouth. Indeed, at the mouth level, molars 55 and 65, as an

e)‘i-Elmple, may still achieve the same caries prevalence estimates by being negatively

c C>rrelated.

18

 

An appropriate method to study the spatial distribution of dental caries in the

mouth needs additional parameters that capture the within-mouth associations. This is
achieved by explicitly modeling the odds ratios using the ALR model. To allow this ALR
model to converge reliably, we imposed some structure on the odds ratio model as
described in the previous chapter. The model estimates for the ALR model are given in

Tables 3.2 and 3.4.

19

Table 3.2

 

Analysis of dental caries experience: ALR model estimates of 7 jk

(log odds of dental caries for each tooth) (standard errors) and corresponding 95%
confidence intervals in 7-year-old children

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TOOTH 1" ‘R‘ METER 71'" ESTIMATES 95% CI
11149 (0.0523) (-1.2173, -1.0124)
55 755
—1.3094 (0.0551) (-1.4174, -1.2015)
54 754
-3.5345 (0.1368) (-3.8027, -3.2663)
53 7’53
-3.0670 (0.1282) (-3.3182, -2.8158)
52 752
51 -1.9837 (0.1167) (22125, -1.7550)
751
61 -2.0377 (0.1212) (-2.2753, -1.8001)
761 .
62 -3.2256 (0.1400) (-3.4999, -2.9512)
762
63 -3.7970 (0.1538) (4.0985, -3.4955)
763
64 -1.2136 (0.0537) (-1.3188, -1.1084)
764
65 -1.0238 (0.0512) (11242, -09234)
765 '
85 -0.9002 (0.0499) (09979, -0.8024)
785
84 07349 (0.0483) (-0.8296, -O.6401)
784
83 -3.2840 (0.1205) (-3.5201, -3.0479)
783
73 -3.5176 (0.1345) (-3.7812, -3.2540)
7 73
74 -0.8005 (0.0490) (0.0490, -O.8966)
7 74
75 -O.9603 (0.0504) (-1.0592, -O.8615)
775

 

 

 

 

20

 

 

Table 3.3 Analysis of dental caries experience: results of comparison of
prevalence of dental caries in upper and lower jaws in 7-year-old children.

 

 

tooth 54 = tooth 64

TEETH INVOLVED WALD STATISTIC (P-
VALUE)
Molars in upper jaw
tooth 55 = tooth 65 6.005 (0.0497)

 

Molars in lower jaw
tooth 85 = tooth 75

tooth 84 = tooth 74

3.857 (0.1454)

 

Canines in upper jaw
tooth 53 = tooth 63

2.536 (0.1113)

 

Canines in lower jaw
tooth 83 = tooth 73

3.752 (0.0527)

 

Incisors in upper jaw

tooth 4= tooth 7
tooth 5= tooth 6

 

 

1.237 (0.5387)

 

We used parameters 7 fit from the ALR logistic regression model to assess

whether the frequencies of caries for each type of teeth in the lefi quadrants are the same
as those of the corresponding teeth in the right quadrants. Results of these analyses are
presented in Table 3.3. All teeth, except for the lefi and right molars in the upper jaw,
have the same caries prevalence with the corresponding teeth across the midline. Even

the result for the left and right molars in the upper jaw was marginally significant at 5%

level (p-value=0.0497).

21

 

Table 3.4 ALR Results: log odds ratio estimates (standard errors), 95%
confidence interval and Z statistic values for dental caries for all teeth in 7-year-old
children

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

PARAMETERS LOG ODDS RATIO 95% CI Z
ESTIMATES (STD STATISTIC
ER.)
(11 2.5716 (0.1139) 2.3483-2.7949 22.57
(12 2.0906 (0.1902) 1.7178-2.4634 10.99
(13 2.0510 (0.1468) 1.7633-2.3387 13.97
(14 2.0141 (0.1596) 1.7013-2.3269 12.62
(15 2.3822 (0.0869) 22118-25526 27.40
(16 1.8365 (0.1026) 16354-20375 17.90
(17 1.9646 (0.1364) l.6972-2.2320 14.40
(18 1.7960 (0.2167) l.3712-2.2208 8.29
(19 2.2510 (0.1397) 1.9771-2.5249 16.11
(110 2.8227 (0.4282) l.9834-3.6620 6.59
(111 2.1807 (0.2174) 1.7546-2.6068 10.03
(112 0.4988 (0.5913) -0.6601-l.6577 0.84
(113 2.8487 (0.1873) 2.4815-3.2159 15.21
(114 1.9026 (0.1608) 15875-22178 11.83
(115 1.9292 (0.1592) 16257-22328 12.46
(116 1.8720 (0.1207) 1.6354-2.1085 15.51
(117 2.1789 (0.0993) 1.9843-2.3735 21.95
0.18 2.1477 (0.0967) 1.9581-2.3373 22.20
(119 2.1708 (0.1014) 1.9721-2.3694 21.42
(120 2.0753 (0.1090) l.8617-2.2890 19.04
(121 2.1450 (0.1464) 1.8580-2.4320 14.65
(122 1.9406 (0.0827) 1.7786-2.1027 23.48
(123 2.0953 (0.1094) l.8807-2.3098 19.14
(124 2.1764 (0.0732) 2.0330-2.3197 29.75
(125 2.2554 (0.1013) 2.0567-2.4540 22.25
(126 2.7768 (0.2313) 23234-32302 12.00
(127 2.1216 (0.4104) 13172-2925 5.17
(128 2.6272 (0.2066) 2.2223-3.0321 12.72
(129 1.8707 (0.1828) 1.5124-2.2291 10.23
(130 2.7120 (0.2726) 2.1777-3.2464 9.95

 

 

 

 

For meaning of (1 values, please refer to Table 2.2.

22

 

Table 3.5 Hypothesis test results on odds ratio parameters

 

 

 

 

 

 

HYPOTHESIS PARAMETERS UNDER NULL WALD
HYPOTHESIS STATISTIC
(P-VALUE)
(Q51agl)= (Q6lag1) (11= (15
(Q51ag2) = (Q6lag2) 92= a6
= . 1 0.507
(Q51ag3) = (Q6lag3) “3 “7 3 3 3 ( )
4:
(Q5984)=(Q6lag4) ‘1 “8
(Q81agl) = (Q7lag1) (19= (111
(110: (112 11.935 (0.003)
(Q81ag2) = (Q71ag2)
(QSIagl) = (Q71ag1) (19= (111 0.092 (0.762)
(Q81ag2) = (Q71ag2) a10= (112 11.809 (0.001)

 

Overall horizontal left-right

 

 

quadrant compared with all a _ + a . a + a + a + a, 4.835 (0.028)
other quadrant associations ”‘21 26'” = 22 25 23 24
14 4
Left-right in Lower and ‘
Upper Jaw 943$ = m 3.092 (0.079)
9 5

 

(Q5Q8) = (Q5Q7) = (Q6Q8)
=(Q6Q7)

(122= (123= (124= (125

13.370 (0.004)

 

(Q5Q8) = (Q6Q7)

(122= (125

7.295 (0.007)

 

 

(Q5Q7) = (Q6Q3)

 

(123= (124

 

0.461 (0.497)

 

Log odds ratios estimates with associated 95% confidence intervals (CI) are

presented in Table 3.2. These estimates and the associated variance covariance matrix are

used to evaluate hypotheses related to odds ratios. A Wald statistic with its chi-squared

limiting distribution is used to conduct the test. Following the results from the logistic

regression model of caries prevalence, we evaluate whether the left and right quadrant

have the same second moments. We found that the association of any lag in quadrant 5 is

the same as the corresponding association in quadrant 6 (p-value=0.507). For quadrants 7

23

 

 

 

and 8, only the lag l association is found to be the same (p-value=0.762). The lag 2
association for these quadrants was found to be different at 5% level (p-value=0.001).

The hypotheses comparing the left-right association compared to other
associations at the quadrant level were also evaluated using a traditional Wald test. The
lefi-right quadrant-level association was significantly higher than to other quadrant-level
associations (across, up and down) (p-value=0.028). This finding is consistent with that
of Zhang et al. (2009) in their work on modeling spatially correlated tooth-level binary in
caries research. Moreover this finding suggests that caries experience might Show
approximately a symmetric pattern across the midline in the deciduous dentition. Dentists
believe there is a stronger left right association at the quadrant level, but it has never been
evaluated using modern statistical tools. From a public health perspective, this finding
suggests that if a 7-year-old presents with caries in any tooth in quadrant 5, he or she has
an increased chance of presenting caries experience in quadrant 6 compared to other
quadrants within the mouth. This also applies when a 7-year-old presents with caries in
any tooth in quadrant 8, they have an increased chance of presenting caries experience in
quadrant 7 compared to other quadrants within the mouth.

Further tests were to evaluate whether the left right association at the quadrant
level was the same for both the upper and the lower jaw. There was no statistical
difference between these jaws in terms of lefi-right association at the quadrant level (p-
value=0.079). Therefore, a 7-year-old who presents with caries in any tooth in quadrant 5
has the same chance of presenting with caries experience in any tooth in quadrant 6, as a

7-year-old with caries in quadrant 8 has in presenting caries experience in quadrant 7.

24

 

Other quadrant level associations were also evaluated to better understand the
spatial distribution of dental caries in the population of 7 year old school children. We
tested the hypothesis (Q5Q8) = (QSQ7) = (Q6Q8) = (Q6Q7) of whether the association
between quadrants 5 and 8, was the same as that of quadrants 5 and 7, quadrants 6 and 8
and quadrants 6 and 7. The result of the test was significant at 5% level (p-value=0.004).
Further analysis test was performed to identify the difference. The “across” associations
represented by the odds ratio between quadrants 5 and 7 and odds ratio between
quadrants 6 and 8 were found to be the same (p-value=0.497). The “up and down”
associations represented by the odds ratio between quadrants 5 and 8 and odds ratio

between quadrants 6 and 7 were not the same (p-value=0.007).

25

 

CHAPTER 4

Discussion

4.1 Research Study

The purpose of this study was to investigate the spatial distribution of dental
caries on tooth surfaces in the primary dentition of 7 year olds, using statistical models
for correlated data. The main objective was to determine which tooth surfaces are more
susceptible to caries experience, and with respect to dental caries, if any spatial
association of dental caries exists at the quadrant level in the mouth. This study
demonstrates an application of a statistical model, where the association structure
between teeth is the scientific focus of research. It further extends that done of Hujoel et
al. (1994), and that of Vanobbergen et al. (2007) by using data from the entire mouth (all
20 teeth found in 7 year old children), and not just selected teeth.

Given a caries lesion on a tooth, this structured GEE-based alternating logistic
regression (ALR) model enables us to determine the odds of caries lesions formation on
corresponding teeth in a different quadrant. The ALR models focus on marginal
expectation of the response and the second order association through pair-wise odds
ratios. This class of models provides consistent estimates, and is easy to implement using
commercial software. To avoid any misspecifications of the association structure at the
expense of efficiency, unstructured odds ratio model, as seen in the Vanobbergen et al.
(2007) study, are usually advocated. In caries research, a GEE-based ALR model with an
unstructured design matrix for the odds ratio model requires an overwhelming number of
parameters for each quadrant. For 7 year old children, it requires 190 = 20 * (20 — 1) / 2

pair-wise odds ratios to be estimated. In this study we restricted the ALR model by

26

 

adding structure to the Odds ratio model, to produce fewer parameters (30 parameters for
the whole mouth), and allow for convergence.

The spatial association in this analysis refers to the association of dental caries
outcomes at quadrant levels. Lesaffre et al. (2005) and Vanobbergen et a1. (2007) have
shown that ALR can be used to evaluate the left-right association at the quadrant level.
These authors with few selected teeth have shown that a tooth in the left quadrant has the
strongest association with a corresponding tooth across the midline, in the right quadrant.
Our analysis goes beyond those findings to investigate whether complete spatial
association at the quadrant level exists when all teeth (20 teeth) are used. Also within
quadrants, we observed the associations (within quadrant) seen in adjacent teeth pairs,
with corresponding teeth pairs in other quadrants. In addition to the left-right quadrant
level associations, we also looked at “up and down” and “across” quadrant associations,
in order to fully capture the within-mouth caries associations.

The structured ALR model revealed that caries experience does follow a spatial
distribution pattern. We observed that the overall left to right quadrant associations of
caries experience appeared to have the strongest association. This is supported with the
similar caries frequencies seen especially in the molars of quadrants 5 and 6. The molars
of quadrant 5 (54 and 55) are 5.20 and 8.92, while that of quadrant 6 (64 and 65) are 5.20
and 8.55 respectively. It is also important to note the association between teeth 51 and 61,
with almost identical caries frequencies of 7.81 and 7.06 respectively. Here we can see
that the left-right teeth pairs have the strongest association which confirms the findings of

the previous study by Vannobergen et al. (2007). The highest caries frequencies are seen

27

in the molars of the lower jaw (mandible), which are consistent with previous findings
(Macek et al., 2003).

While the results of this study confirmed the findings of Vannobergen et al.
(2007), it conflicted with the findings of Hujoel et al. (1994), which showed caries
lesions tend to aggregate or cluster on one side of the mouth. In our study, clustering was
only observed in the left molars in the upper jaws. Given the young age of the population,
a plausible hypothesis more so than chewing patterns, would be tooth brushing
style/patterns, and a change in the current oral health education might influence this left
Sided caries clustering. This will suggest a need for more community dental health
programs for the parents and children, and within the school systems to encourage more

effective ways of tooth brushing.

4.2 Implications of findings.

Dentists and dental hygienists should look towards incorporating left to right
quadrant level associations to provide more effective treatment plans for patients with
dental caries. If a patient presents with caries on the molars in quadrant 5, the dentist
should first also look for the early development of caries on the molars in quadrant 6, as
these have an increased chance of developing caries. If a patient presents with caries on
any tooth in quadrant 5, firstly, care should be taken to examine the corresponding tooth
in quadrant 6. They should also look towards using other association patterns such as the
“across” and “up and down” quadrant level associations. Similarly, if a patient presents
with caries on any tooth in quadrant 6, the dentist needs to be educated to check the teeth
in quadrant 8 for caries first before looking at the other quadrants. These association

methods should be implemented in dental health training. In the long term, this

28

implementation should help decrease the prevalence of dental caries and further
debilitating periodontal diseases, as well as the financial burden on dental public health,
as preventive measures are undertaken. It should help encourage better quality of living,

as people will be able get their teeth restored early, without the need for drastic measures

such as root canals, to remove teeth.

29

 

BIBLIOGRAPHY

Ananth CV, Kantor ML: Modeling multivariate binary responses with multiple levels of
nesting based on alternating logistic regressions: an application to caries aggregation. J
Dental Res 2004; 83:776-781.

Beck JD, Weintraub JA, Disney JA, Graves RC, Stamm JW, Kaste LM, Bohannan HM:
University of North Carolina caries risk assessment model: comparisons of high risk
prediction, any risk prediction and risk etiologic models. Community Dent Oral
Epidemiol 1992; 20:313-321.

Carey V, Zeger SL., Diggle P: Modelling multivariate binary data with alternating
logistic regressions. Biometrika 1993; 80: 517-526.

Fejerskov 0. Concepts of dental caries and their consequences for understanding the
disease. Community Dent Oral Epidemiol 1997; 25:5-12.

Hujoel PP, Lamont RJ, DeRouen TA, Davis S, Leroux BG: Within-subject coronal caries
distribution patterns: an evaluation of randomness with respect to the midline. J Dental
Res 1994, 73:1575-1580.

Kidd EA, F ejerskov 0. What constitutes dental caries? Histopathology of carious enamel
and dentin related to the action of cariogenic biofilms. J Dent Res 2004;83zC35—C38.

Leroy R, Bogaerts K, Lesaffre E, Declerck D. Effect of caries experience in primary
molars on cavity formation in the adjacent permanent first molar. Caries Res. 2005 Sep-
Oct;39(5):342-9.

Liang K-Y, Zeger SL: Longitudinal data analysis using generalized estimating equations.
Biometrika 1986; 73:13-22.

Liang, K.-Y., Zeger SL,, and Qaqish B. Multivariate regression analyses for categorical
data. J. R. Stat. Soc. 1992 B 54(1):3-40

LiSH, Kingman A, Forthofer R, Swango P: Comparison of tooth surface-specific dental
caries attack patterns in US schoolchildren from two national surveys. J Dent Res 1993;
72:1398-1405.

Macek MD, Beltran-Aguilar ED, Lockwood SA, Malvitz DM. Updated comparison of
the caries susceptibility of various morphological types of permanent teeth. J Public
Health Dent. 2003 Summer;63(3): 174-82.

Manji F, Fejerskov O, Nagelkerke NJD, Baelum V. A random effects model for some

epidemiological features of dental caries. Community Dent Oral Epidemiol 1991;19:324-
328.

30

Melberg JR. Demineralization and remineralization of root surface caries. Gerodontology
1986 Spring; 5(1):25-31.

Reid DB, Graingcr RM. Variations in the caries susceptibility of children’s teeth. Hum
Biol. 1955; 27:1.

Todem D., Kim K., Lesaffre E: Latent-variable models for longitudinal data with
bivariate ordinal outcomes. Stat Med 2007; 26(5):1034-54.

Vanobbergen, J., Lesaffre, E., Garca-Zattera, M.,J., Jara, A., Martens, L., Declerck, D.
Caries Patterns in Primary Dentition in 3-, 5- and —year-old Children: Spatial Correlation
and Preventive Consequences. Caries Res 2007, 41:16-25.

Vanobbergen, J ., Martens, L., Lesaffre, E., Declerck, D: The Signal Tandmobiel® project,
a longitudinal intervention health promotion study in Flanders (Belgium): baseline and
first year results. Eur J Paediatr Dent 2000;2: 87-96.

Zeger SL, Liang K-Y. Longitudinal data analysis for discrete and continuous outcomes.
Biometrics 1986;42:121-130.

Zhang Y., Todem D., Kim K., Lesaffre E : Bayesian Latent Variable Models for Spatially
Correlated Tooth-level Binary Data in Caries Research. Stat Modelling 2009 (in press)

31