Scholarly article on topic 'Assessing causal effects of early life-course factors on early childhood caries in 5-year-old Ugandan children using directed acyclic graphs (DAGs): A prospective cohort study'

Assessing causal effects of early life-course factors on early childhood caries in 5-year-old Ugandan children using directed acyclic graphs (DAGs): A prospective cohort study Academic research paper on "Health sciences"

Share paper
Academic journal
Community Dent Oral Epidemiol
OECD Field of science

Academic research paper on topic "Assessing causal effects of early life-course factors on early childhood caries in 5-year-old Ugandan children using directed acyclic graphs (DAGs): A prospective cohort study"

Received: 22 July 2017 | Accepted: 15 May 2017 DOI: 10.1111/cdoe.12314


Assessing causal effects of early life-course factors on early childhood caries in 5-year-old Ugandan children using directed acyclic graphs (DAGs): A prospective cohort study

Nancy Birungi1'2© | Lars T. Fadnes1'2 | Arabat Kasangaki3 | Victoria Nankabirwa4'5 | Isaac Okullo3 | Stein A. Lie1 | James K. Tumwine6 | Anne N. Astr0m1 | for the PROMISE-EBF study group



department of Clinical Dentistry, University of Bergen, Bergen, Norway

2Centre for International Health, University of Bergen, Bergen, Norway

3Department of Dentistry, School of Health Sciences, College of Health Sciences, Makerere University, Kampala,Uganda

4Department of Epidemiology and Biostatics, School of Public Health, College of Health Sciences, Makerere University, Kampala, Uganda

5Centre for Intervention Science in Maternal and Child Health (CISMAC), Centre for International Health, University of Bergen, Bergen, Norway

6Department of Paediatrics and Child Health, School of Medicine, College of Health Sciences, Makerere University, Kampala,Uganda


Nancy Birungi, Department of Clinical Dentistry, University of Bergen, 5009 Bergen, Norway.


Funding information

This work was supported by European Union Sixth Framework International Cooperation-Developing Countries, Research Council of Norway, Swedish International Development Cooperation Agency, and Norwegian Programme for Development, Research and Education, South African National Research Foundation, and Rockefeller Brothers Foundation.


Objective: To estimate the effect of distal and proximal early life-course factors on early childhood caries (ECC) in 5-year-old Ugandan children, particularly focusing on the causal effect of exclusive breast feeding (EBF) on ECC using directed acyclic graphs (DAGs) for confounder selection.

Methods: This study had a nested prospective cohort design, focusing on 5 years of follow-ups of caregiver-children pairs from the PROMISE-EBF trial ( no: NCT00397150) conducted in 2011 in Eastern Uganda. Data were from recruitment interviews, 24-week, 2-year and 5-year follow-ups of a cohort of 417 mother-children pairs. Trained research assistants performed interviews with caregivers in the local language and ECC was recorded under field conditions using the World Health Organization's (WHO) decayed missing or filled teeth (dmft) index. Early life-course factors in terms of socio-demographic characteristics, EBF and other feeding habits were assessed at the various follow-ups. The outcome (ECC; dmft>0) was assessed at the 5-year follow-up. Causal diagrams as DAGs were constructed to guide the selection of confounding and collider variables to be included in or excluded from the final multivariable analysis. Negative binomial regression analyses were performed based on two comparative DAGs representing different causal models. Results: Model 1 based on DAG 1, showed EBF to be a protective factor against ECC, with an IRR and 95% CI of 0.62 (0.43-0.91). According to Model 2 based on DAG 2, EBF and having both parents living together had protective effects: the corresponding IRRs and 95% CI were 0.60 (0.41-0.88) and 0.48 (0.25-0.90), respectively. Conclusions: Both plausible models indicated that being exclusively breastfed for 24 weeks had a protective causal effect against ECC. Further research, examining the unmeasured variables included in the DAGs is necessary to strengthen the present finding and allow stronger causal claims.


ECC, epidemiology, paediatric dentistry, public health, statistics

This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made. © 2017 The Authors. Community Dentistry and Oral Epidemiology Published by John Wiley & Sons Ltd

Community Dent Oral Epidemiol. 2017;1-10. I 1


Conceptual models of early childhood caries (ECC, which is defined as presence of one or more noncavitated or cavitated decayed, missing or filled tooth surfaces in any primary tooth of children aged under 71 months) describe contextual and individual risk factors at different levels of proximity to ECC.1,2 According to these multilevel models, individual oral health-related behaviours and socio-demographic characteristics at the family and community levels are considered proximal and distal factors, respectively, related to ECC. Socio-demographic characteristics at the distal environmental level may affect ECC, directly or indirectly, through more proximal factors at the individual level. Moreover, risk factors may be conceptualized within a temporal framework while considering influences through time. Life-course epidemiology considers the long-term effects on health and disease of physical and social exposures during gestation, childhood, adolescence, young adulthood and later adult life.3 Accordingly, distal and proximal early life circumstances, such as socio-demographic characteristics, nutritional status, birth weight, and breastfeeding patterns may impact on current and future oral conditions, suggesting a gradual accumulation of exposures throughout the life-course. Alternatively, circumstances during a specific critical period in early life may provide subsequent and irreversible damage to health and oral health.4

Exclusive breastfeeding (EBF) is described as the practice of giving only breastmilk and no other liquids to a baby, with the exception of vitamins, mineral supplements or medicines.5 The benefits of EBF for growth during infancy and protection against early childhood mortality and morbidity have been extensively documented.6-8 By contrast, reviews investigating the influence of breastfeeding practices on ECC have been equivocal,9,10 although two meta-ana-lyses concluded that breastfeeding in early life may prevent ECC.11,12 In addition, three cross-sectional and one cohort study revealed that breastfed children were significantly less frequently affected by dental caries than bottle-fed children.11 A lack of internationally adopted definitions of breastfeeding and ECC as well as methodological flaws may have contributed to inconsistency in the conclusions of the existing research literature.9,13 The relationship between any early life-course factor and dental caries, as between EBF and ECC, is further complicated by bias from confounding, which is the mixing of effects from extraneous variables associated with both the exposure of interest and the outcome.13 If not adjusted for appropriately in multivariable regression models, spurious associations may result that subverts the goal of epidemiological research to produce unbiased, causal estimates of exposure with outcome.

In epidemiological studies, a distinction between prediction and causal approaches should be made to facilitate valid interpretation from the body of evidence.14 A prediction approach involves the use of statistical models to predict the outcome based on characteristics of the exposure variables and is independent of causal mechanisms.

Causal approaches, by contrast, may utilise directed acyclic graphs (DAGs), a graphical tool guided by qualitative assumptions, visualizing the hypothesized causal structure underlying the exposure and outcome under study, while at the same time considering potential confounders and mediators.15-17 Using DAGs has been proposed as a tool for confounder selection.18-20 A causal model based on DAGs can be designed and used as an aid to check for the sufficiency of confounder adjustment instead of relying entirely on P-values derived from statistical models or stratification.16 Using statistics to identify and adjust "for confounding" could induce bias unless built on well guided causal approach models.15 So far, conventional analytic approaches in the ECC literature have included adjustment for several risk factors in multiple variable statistical models without assumptions of causal relationships.21-23 The application of DAGs using scenarios from cross-sectional epidemiological studies of peri-odontitis and temporomandibular disorders has been demonstrated conceptually.17,24 However, few dental studies have so far employed DAGs in analyses of empirical data.25-28 In previous work, we showed that an intervention using peer-counsellors to promote EBF had no effect on ECC.29 In the present study, we investigate how the practice of exclusive breastfeeding itself impacts on ECC in the context of a cohort study design.

Using background knowledge from conceptual risk factor models of ECC (as well as the ECC risk factor literature),22,23,30,31 DAGs were constructed to guide selection of the variables to be adjusted for in statistical analyses to obtain unbiased estimates. This study aimed to estimate the effect of distal and proximal early life-course factors on ECC in 5-year-old Ugandan children, particularly focusing on the estimation of a causal effect of EBF on ECC using DAGs.


This study had a nested prospective cohort design, presenting the 5-year follow-up of caregiver-children pairs from the Ugandan site of the PROMISE-EBF trial ( no: NCT00397150) conducted in 2011 in Mbale district, Eastern Uganda.32 The PRO-MISE-EBF study was a community-based cluster-randomised trial assessing the effect of individual home-based peer counselling to promote exclusive breastfeeding. During the first 6 months postde-livery, peer counselling on EBF by community workers took place with around half of the cohort presented in this study. Detailed information about the PROMISE-EBF trial has been published pre-viously.32,33 A total of 863 pregnant women were recruited at 7 months gestational age into the Ugandan site of the PROMISE EBF trial. A total of 765 healthy mother-infant pairs were enrolled during the first 6 months after birth. Follow-ups were carried out at household level at 3, 6, 12, 24 weeks, 2 and 5 years after birth between 2006 and 2011 (Figure appendix in supplementary file). This study utilises information obtained from 417 caregiver-child pairs at the recruitment interview (marital status, socioeconomic

status, maternal education level), and at household visits at the child ages of 24 weeks (exclusive breastfeeding), 2 years (breastfeeding duration) and 5 years (oral health, anthropometric status). Ethical approval for the study was granted by Makerere University School of Medicine, Research and Ethics Committee (SOMREC), the Uganda National Council for Science and Technology and Regional Committees for Medical and Health Research Ethics, Western Norway (05/8197). Informed consent was given by the study participants before data collection.

2.1 | Interviews with caregivers at recruitment, 24-week, 2-year and 5-year follow-up interviews

Trained research assistants conducted interviews with caregivers in the local language, Lumasaba. Socio-demographic characteristics assessed at the recruitment interview were mothers' educational level, marital status and socioeconomic status. Caregivers' level of education was assessed by the question, "What is your highest level of education"? Responses included 18 alternatives from primary school level to above bachelor level at university. Marital status was assessed asking "Are you single, married, cohabitating, widowed, divorced or separated now?" Responses included single (i) married (ii) cohabitating (iii), widowed (iv), divorced/separated (v). For analysis, that variable was dichotomised into married/cohabitating (i) and single/widowed/divorced/separated (ii). Socioeconomic status was assessed by the questions "How many of the following items (chairs, foam mattress, lantern) do you have in your household? Do you have electricity in the home? Do you have the following (cupboard, bicycle, radio, TV, mobile phone, gas heater, refrigerator, motorcycle, car truck) in your household? "What (wood, charcoal, paraffin/ kerosene, gas, electricity, others) is used for cooking in your household"? Responses were yes (1) and no (2). Materials used for the construction of floor, roof, walls, windows, doors, status of the toilet, compound status of the household and the type of house were observed by the interviewer. A socioeconomic wealth index consisting of five quintiles was constructed based on the above attributes. For analysis, the wealth quintiles were dichotomised into (i) most poor (including the 1-3rd quintiles) and (ii) least poor (including 4-5th quintiles).34

Exclusive breastfeeding (EBF) was assessed at the 24-week follow-up using 24-hour and 7-day recalls. A baby who did not receive any food or liquids other than breastmilk and medicines was classified as exclusively breastfed. Breastfeeding duration was assessed at 2- and 5-year follow-ups for this study. Breastfeeding was assessed using the following questions, "Did you breastfeed?", "Are you still breastfeeding?" Responses were given as no (0) or yes (1). Breastfeeding duration was assessed at the 2nd and 5th year of follow-up visit by the question; "How long did you breastfeed?"—. Children's dental attendance was assessed at the 5-year follow-up by asking, "Before today have you taken your child to a dentist/dental therapist for a check-up?" Responses were given as no (0) or yes (1). Children's oral hygiene habits were assessed at the 5-year follow-up by the following questions; "What is used to clean your child's teeth? Responses

-I^NTCTRY™e -Wl I FY—-


ranged from 0-5; including tooth brush (i), chewing stick (ii), cloth (iii), finger (iv) and other (v). "Who brushed the child's teeth", with responses; child (i), parent (ii), someone else (iii) and teeth not brushed (iv). "How often are the child's teeth brushed" with responses less than once a day (i), once a day (ii), twice a day (iii) and more than twice a day (iv). Caregivers' oral hygiene was assessed at the 5-year follow-up by three questions "What device do you use to brush your teeth"? Responses included plastic tooth brush (i), chewing stick (ii), other (iii); the second was "How often do you brush?" with responses never (0), occasionally (1), once a day (2), more than once a day (3). "Do you use toothpaste when brushing your teeth?" Responses were no (0), yes (1), and do not know (9). The child's sugar consumption was assessed at the 5-year follow-up with the following questions: "How often does child drink fruit juice /other sweetened beverages?" responses were never (0), less than weekly (1), 1-3 times/week (2), 4-6 times/week (3), daily (4), do not know (9). "Has your child used sweetened cough syrup for a longer period" Responses were no (0) and yes (1). Caregivers' sugar consumption was assessed using: "Do you eat sugary snacks/drinks?" with responses no (0), yes (1), do not know (9). "How often do you eat/drink snacks/drinks?" with responses occasionally (1), once a day (2), more than once a day (3). To reduce the number of covariates under study, formative-summative (additive) indices were constructed for family oral hygiene and family sugar consumption. (see Table S1 in supplementary file for summary data of covariates comprising the indices). Family oral hygiene consisted of the following five variables: What is used to clean the child's teeth? How often are the child's teeth cleaned /brushed? What device do you (mother) use to brush your teeth? How often do you (mother) brush your teeth? Do you (mother) use tooth paste when brushing your teeth? Family sugar consumption consisted of the following variables: How often does child drink fruit juice? How often does child drink sweetened beverages? Has your child used sweetened cough syrup for a longer period? Do you take /eat sugared snacks? How often do you take sugared snacks? Do you take sugared drinks? How often do you take sugared drinks? These variables have been applied in previous sub-Saharan African studies and have been shown to associate with clinical and subjective measures of oral health.

2.2 | Clinical oral examination and anthropometry at the 5-year follow-up

A full-mouth clinical oral examination was carried out at household level by two experienced and calibrated dentists (NB and AK). Early childhood caries (ECC) was assessed on fully erupted primary teeth using the decayed, missing, and filled teeth index (dmft) in accordance with the World Health Organization (WHO) guidelines for field conditions.35 A tooth was recorded as decayed if it was visually cavitated using a disposable mirror and dental explorer (Double ended No.23). A missing tooth was qualified as missing if extracted due to caries, as confirmed by the caregiver. In the present analysis, dmft was used as a count variable. The count variable was a sum score of decayed, missed or filled teeth. Details of the inter-rater

and intrarater agreement between and within dental examiners have been reported previously.29 Double data entry was carried out.

Anthropometry measurement for height to the nearest 0.1 cm was carried out using a "Shorr" height and length measuring board provided from the UNICEF supplies in Uganda. The weight was measured to the nearest 0.5 kg using the Seca weight model 762. Prevalence of stunting was defined as the proportion of children who fell below 2 standard deviations from the WHO child growth standard median.36 The degree of stunting from minus 1 standard deviation was constructed as a measure better reflecting the nutrition status of the children.

2.3 | Directed acyclic graphs (DAGs)

Following Hernan and Robins,37 the underlying causal relationships between early life-course factors and ECC were visualised using DAGs. Based on causal diagram theory, DAGs are considered sets of arrows that characterise causal associations between exposures and outcomes and also specify relationships among other variables that influence the exposure or outcome.18 Generally, the variables in the DAGs are referred to as nodes that are connected by head to tail arrows called vertices or edges that make up paths. Paths between exposure and outcome are any noncrossing, nonrepeating series of arrows connecting them, starting at the exposure and ending with the outcome, irrespective of the direction of the arrows. Front door paths have arrowheads pointing from exposure towards the outcome representing the presence but not the strength of causal effects. The assumption is that the variables connected directly to the outcome have a causal effect. Front door paths can be open with all arrows pointing in the same direction or they can be closed with a change in direction of one or more arrows. Backdoor paths are alternative noncausal biasing paths between exposure and outcome, characterized by at least one arrowhead pointing towards the exposure and one arrow head pointing towards the outcome and all other arrows pointing in the same direction. Closed backdoor paths include collider variables, which are common effects of two other variables on the path. Front- or backdoor paths including collider variables are closed and there is no flow of association from exposure to outcome along those collider paths. Backdoor paths may confound a direct effect between exposure and outcome when left open.

In this study, two alternative DAGs, depicted in Figures 1-2, were used to illustrate assumed relationships between early life-course factors and ECC. Before being included in the DAGs, possible direct paths between all variables considered were evaluated for plausibility based on theory and previous empirical evidence. The wealth asset index was assumed to have direct effects on dental attendance,38 family sugar consumption,38 anthropometric sta-tus,39,40 family oral hygiene,38 breastfeeding duration41 and breastfeeding exclusivity.42 Maternal education was assumed to have direct effects on dental attendance,42family sugar consump-tion,38 marital status, anthropometric status,43 family oral hygiene,38 breastfeeding duration42 and EBF.42,43 Marital status was assumed

to have direct effects on the wealth assets index,23 dental atten-dance,23family oral hygiene,44-46 breastfeeding duration42 and EBF.47 Breastfeeding duration was assumed to have direct effects on anthropometric status,48 family sugar consumption,49enamel hypoplasia,50 and cariogenic bacteria.51,52 Family oral hygiene was assumed to have direct effects on cariogenic bacteria53 and ECC,54 whereas family sugar consumption was assumed to have direct effects on family oral hygiene,55 EBF,49 anthropometric status56,57 and cariogenic bacteria. Anthropometric status was assumed to have direct effects on breastfeeding exclusivity58 and enamel hypoplasia.59 Enamel hypoplasia was assumed to affect cariogenic bacteria60 and ECC.61 Cariogenic bacteria were assumed to have direct effects on ECC.62 Early childhood caries was assumed to have a direct effect on dental attendance.63,64 A direct effect of EBF on ECC was hypothesized; hence causal paths between EBF and ECC, mediated through cariogenic bacteria and enamel hypoplasia, were included in the DAGs.

In summary, proximal individual- and family-level covariates; anthropometric status (individual level), breastfeeding duration (individual), sugar consumption (family level) and oral hygiene (family level) were assumed to have direct effects on ECC, whereas the effect of more distal family-level covariates; socioeconomic status, maternal education, and marital status were assumed to be mediated through the proximal variables. Dental attendance was assumed to be a collider variable in some paths (model Figure 1) and a mediator in other paths (model 2, Figure 2). Unmeasured variables in terms of enamel hypoplasia and cariogenic bacteria were included in the DAGs for the purpose of providing a clearer picture of the assumed causal mechanisms.

2.4 | Statistical analysis

The statistical package, Stata Intercooled version 13.1 (Stata Corporation, College Station, TX, USA) was used for data analysis. The DAGitty ( online tool was used to check the postulated DAGs for consistency and validity of the minimum adjustment sets. Comparison of retention rates after 5 years of follow-up was undertaken for levels of selected baseline characteristics. Confidence intervals were used to evaluate whether or not any differences between levels of these covariates were likely to be of importance. A likelihood ratio test was performed for overdispersion. Following the use of ECC as a count variable, negative binomial regression for count data with overdispersion of mean from the variance was used in regression analyses. Incidence rate ratios, IRR and 95% confidence intervals (95% CI) were calculated. To avoid overes-timation of the precision of the estimates due to clustered data, analyses were adjusted for the cluster effect by including the cluster variable in the regression models.65 To adjust for potential differences in loss to follow-up, an inverse probability weighted method was applied using probit- regression for binary outcomes to predict risk of loss to follow-up based on background factors (socioeconomic status, level of education and residence in rural/urban area). The median of the weights was 1.8.

-I^NTCTRYane -Wl I FY"


Dental attendance

FIGURE 1 Directed acyclic graphs (DAG)-1 includes the following variables: Family sugar consumption, family oral hygiene, breastfeeding duration, exclusive breast feeding (EBF), anthropometric status. *Unmeasured variables. Numbers along arrows indicate citation for the assumed direct effect between variables

FIGURE 2 Directed acyclic graphs (DAG) 2 includes the following variables exclusive breast feeding (EBF), breastfeeding duration, family oral hygiene, marital status, education level, wealth asset index, anthropometric status and family sugar consumption. *Unmeasured variables. Numbers along arrows indicate citation for the assumed direct effect between variables


Comparison of baseline socio-demographic characteristics between the children followed up and those lost to follow-up showed that socioeconomic status and parity assessed at the recruitment interview at baseline differed slightly between the two groups.

Most were married or cohabiting, attended antenatal clinic at least once, had a pit latrine but did not have electricity, and had more than one child (Table 1). The median age (25%-75% percentile) of the participating children was 4.5 years (4.1-5.1). The median maternal age (25%-75% percentile) was 30 years (26-36).

6_, , ,, , „ , Community

■Wl I FY—Dentistry«

t t 1 1 qralgpidemiology

BIRUNGI et al.

TABLE 1 Socio-demographic characteristics at baseline by follow-up status at 5-year follow-up

Variable Category Followed up n (%) Lost to follow- up n (%) Odds ratio (95% CI)

Socioeconomic status Poor .east poor 270 (64.7) 147 (35.3) 190 (54.8) 157 (45.2) 0.66 (0.49-0.88)

Marital status Married/cohabiting 390 (94) 311 (90.7) 0.62 (0.36-1.07)

Single /widowed/ divorced,/separated 25 (6.0) 32 (9)

Yrs of education .ess than 6 yrs More than 6 yrs 229 (54.9) 188 (45.1) 168 (48.4) 179 (51.6) 0.77 (0.56-1.02)

Antenatal visit Yes 299 (74.8) 247 (75.3) 0.97 (0.69-1.36)

No 101 (25.2) 81 (24.7)

Place of delivery Out of facility Facility 199 (49.4) 204 (50.6) 155 (47.1) 174 (52.7) 0.91 (0.68-1.22)

Parity Primipara 77 (18.7) 96 (28.3) 1.72 (1.22-2.42)

Multipara 335 (81.3) 243 (71.8)

Electricity in house Yes 60 (14.6) 63 (18.5) 0.75 (0.51-1.10)

No 352 (85.4) 277 (81.5)

Toilet None 87 (23.8) 56 (18.5) 0.73 (0.50-1.06)

Pit/ventilated improvec pit latrine 278 (76.2) 246 (81.5)


TABLE 2 Summary data % (n) of early life-course factors in the total study group followed up for 5 years

Variable Category % (n)

Sex Boys 50 (208)

Girls 50 (209)

Exclusive breastfeeding (assessed at No 36 (145)

24 wks follow-up) Yes 64 (254)

Breastfeeding duration for >2 yrs (assessed at 2 and 5 yrs follow-up) No Yes 51 (196) 49 (188)

mothers reported having breastfed for at least 2 years. Some, 36% of the children were stunted and 20% were underweight. A total of 83% reported children's general health condition as good, and 88% had never visited a dentist in their lifetime. The prevalence of ECC (dmft>0) and mean dmft was 39% and 1.6, respectively (for complete data on all variables included, see Table S1).

The likelihood ratio test comparing negative binomial regression and Poisson models indicated that the negative binomial model was more appropriate than the Poisson, with a P<.001 (data not shown).

Stunting (assessed at 5 yrs follow-up) No 64 (266)

Yes 36 (151)

Underweight (assessed at 5 yrs follow- No 80 (333)

up) Yes 20 (84)

General oral health (assessed at 5 yrs Neither good 17 (70)

follow-up) or bad/ bad

Good 83 (342)

Family sugar consumption index Less 16 (456)

More 84 (2463)

Family oral hygiene index Bad 24 (508)

Good 76 (1577)

Dental attendance (assessed at 5 yrs No 88 (360)

follow-up) Yes 12 (51)

Table 2 presents summary data of various covariates among participants followed up for 5 years. Two hundred and eight (50%) of 417 children studied were boys. Exclusive breastfeeding for 24 weeks was confirmed for 63% of the children while 49% of the

3.1 | Adjustment sets and regression analysis

The minimum adjustment set based on DAG 1 included family sugar consumption, breastfeeding duration, anthropometric status and family oral hygiene as confounding factors to be adjusted in the multivariable model regressing ECC on EBF. According to Model 1 based on DAG 1 (Table 3), EBF for 24 weeks had a protective causal effect on ECC with an IRR and (95% CI) of 0.62 (0.43-0.91). No other significant effects were observed. The minimum adjustment set based on DAG 2 included, family sugar consumption, breastfeeding duration, anthropometric status and family oral hygiene, dental attendance, marital status, maternal education and the wealth assets index. According to Model 2 based on DAG 2, EBF for 24 weeks had a protective direct effect on ECC with an IRR of 0.60 (95% CI 0.41-0.88). Marital status had a secondary effect (effect of a covari-ate not of primary interest in the model, for example a confounder or effect measure modifier), with IRR of 0.48 (95% CI 0.25-0.90). Detailed justifications for selection of variables to be included in the

TABLE 3 Negative binomial regression of early childhood caries (ECC) on exclusive breast feeding (EBF) adjusted for confounding variables as identified by directed acyclic graphs (DAG) 1 (Model 1), DAG 2 (Model 2). Incidence rate ratios (IRR) and 95% confidence Intervals (CI)

Covariates Model 1 IRR (95 % CI) Model 2 IRR (95 % CI)

Exclusive breastfeeding 0.62 (0.43-0.91)a 0.60 (0.41-0.88)a

Breastfeeding duration 1.01 (0.98-1.05)b 1.01 (0.97-1.04)b

Sugar consumption index 1.04 (0.87-1.24)b 1.06 (0.88-1.27)b

Oral hygiene index 1.00 (0.80-1.26)b 0.96 (0.77-1.20)b

Anthropometric status 0.94 (0.77-1.14)b 0.94 (0.75-1.17)b

Marital status N/A 0.48 (0.25-0.90)b

Education level N/A 0.93 (0.71-1.23)b

Wealth asset index N/A 0.96 (0.62-1.48)b

Dental attendance N/A 1.47 (0.87-2.50)b

aEffect of primary exposure of interest.

bEffect of covariate not of primary interest in the model, for example a confounder or effect measure modifier.

minimum adjustment sets have been described in a supplementary file.

Significant interactions were identified between EBF and the covariates anthropometric status, dental attendance and level of education. The corresponding IRRs and (95% CI) for the interaction terms were 0.54 (95% CI 0.35-0.81), 2.2 (95% CI 1.01-4.91) and 1.86 (95% CI 1.05-3.31), respectively (see Table S2).

Sensitivity analysis, adjusting for patterns in loss to follow-up with inverse probability sample weights taking socioeconomic status, years of education of the caregiver and site of residence into account, did not change the results (except second decimal changes) based on unweighted analyses. The likelihood ratio test comparing negative binomial regression and Poisson models indicated that the negative binomial model was more appropriate than the Poisson, with a chi-squared value of 684.77 with one degree of freedom (data not shown).


The causal inference approach based on DAGs utilized in this study is useful both when it comes to select an appropriate set of confounding variables as well as identifying collider variables to be left unadjusted in the final statistical model. Only a few previous studies in dental public health have applied DAGs to guide the analyses of empirical data.25-28 In this study, the regression models based on DAG 1 and DAG 2 suggested a protective causal effect of exclusive breastfeeding for 24 weeks on ECC. No other direct effect from distal and proximal early life-course factors was observed. Based on DAG 1, back door paths with respect to the effect of EBF on ECC through dental attendance and anthropometric status were blocked or closed due to the collider status of those variables along the paths. According to the causal graph theory, variables along back

-I^NTCTRY™e -Wl I FY—-


door paths including collider variables should be left unadjusted to avoid unblocking or opening of paths and thus creating instead of limiting confounding bias. If collider variables are adjusted for, as discussed by various authors, this may lead to invalid estimates.15,66,67 In the alternative DAG 2, the path from dental attendance to cario-genic bacteria opened previously closed backdoor paths for flow of association, EBF^ marital status ^ maternal education ?wealth index ? dental attendance ?cariogenic bacteria ?ECC. This justifies inclusion of the distal socio-demographic characteristics in the minimum adjustment set of Model 2.

Although traditional methods for confounder selection, such as stratification and multivariable regression models, could be argued from an empirical point of view, causal diagrams such as DAGs enhance the appreciation of confounding paths and provide information about colliders.24 Notably, however, given the complex multifactorial nature of ECC, it is uncertain whether the optimal DAGs have been created.68 Thus, it has been suggested to include comparative DAGs with different adjustment sets of variables to indicate how findings are influenced by differences in causal assumptions. The path from dental attendance to ECC through cariogenic bacteria as illustrated by DAG 2 was based on the assumption that the ECC situation is a consequence of dental attendance patterns typically seen in high-income settings, where utilization of dental care may be more preventively oriented.69 The path from ECC to dental attendance included in DAG 1 (Figure 1) illustrates a situation typically seen in low-income countries, where dental attendance may be problem oriented and occurs as a consequence of having oral problems.63,64 Notably, the validity of the analysis used in the causal approach depends on the correctness of the postulated DAG; the causal diagram is as useful as the substantive knowledge used to create it. Omission of key variables or relationships, inclusion of nonexisting causal paths and errors in path direction, may lead to an inappropriate analytical model.

A major strength of the present study is the utilization of DAGs to identify appropriate confounders to be adjusted for in the multiple variable analyses. However, as with any observational study, residual confounding cannot be ruled out. Although we adjusted for sugar consumption and breastfeeding duration, not all aspects of early dietary habits were recorded. A second strength of this study is the use of a prospective cohort design being important when investigating risk factors and contributing higher level of evidence as compared to cross-sectional studies. Reliance on the memory of the participants is a problem with retrospective studies and memory problems are worse for data pertaining to events in the past compared to recent events that have occurred just before the inter-view.31 Moreover, we tried to limit the possibility of self-report data being subject to recall or social desirability bias by training the research assistants and by use of pretested questionnaires developed by experts in the field. According to previous evidence, the accuracy of recalled information against historical records has provided good levels of agreement.70 As in most longitudinal studies, attrition represents a limitation in terms of potential for differential selection in loss to follow-up (Table 1). However, the IPW analyses did not change the findings revealed by the unweighted analyses.

Thus, it is unlikely that loss to follow-up biased the present findings substantially, although IPW is indicated for data that is missing at random and not for nonrandom missing.71 The present study was conducted under standardized field conditions according to the WHO criteria, whereas an accurate diagnosis of ECC requires more optimal dental equipment, such as artificial lightning and x-ray units.35 This may have resulted in misclassification and under reporting of the ECC prevalence by overlooking incipient carious lesions and approximal lesions.72,73 A satisfactory agreement (as indicated by the test-retest reliability scores obtained) suggests that misclassification and biased estimates do not constitute an important problem.74 Although key variables such as fluoride consumption, earlier ECC experience, and bacteria load were not directly measured, fluoride consumption was reflected by the summative indices derived from variables measuring family oral hygiene.75

The present findings suggesting a protective effect of EBF on ECC in preschool children support those of a recent systematic review and meta-analysis.11 A previous cross-sectional study of Ugandan infants based on dietary recalls since birth- and 24-hours, showed a positive association between late cessation of any breastfeeding and late first-time exposure to sugared snacks and drinks.49 Moreover, children breastfed for longer than 12 months had less dental caries than those exposed to shorter time of breastfeeding.11 If late first time exposure to sugared snacks and drinks is a consequence of late breastfeeding cessation, this could explain a protective effect of EBF on ECC.76 Evidence from developing countries such as Uganda has suggested that urbanization and unhealthy dietary advertisements through radio and television influence caregivers to feed their children with sweets, add sugar to food and to spend more money on sweets.77 The mediating role of developmental enamel defects that was assumed in this study is supported by previous evidence suggesting a negative association between EBF and enamel defects and a positive association between enamel defects and ECC.50 As depicted in Table 3, living with both parents was independently associated with ECC in Model 2, thus being in accordance with previous studies that have confirmed single parent status, a marker for disadvantaged social status, to be a risk factor for developing ECC.23 A common finding in the ECC literature is that having married or cohabiting parents is negatively associated with ECC development23,78 No effect of breastfeeding duration on ECC was observed in this study. By contrast, previous studies included in a literature review have shown that breastfeeding for longer than 1 year and night-time breastfeeding beyond eruption of the primary dentition are associated with some form of ECC.9 It is important to note that prolonged breastfeeding may include other factors linked to nocturnal feeding that was not investigated in this study. Moreover, the three studies reported in the literature review were cross-sectional studies and relied on data collected retrospectively.79-81 It is difficult to compare the findings of the present study with those previously reported in the literature review because the interpretation should be consistent with the performed analyses. Most previous studies have utilized statistical models only to identify and adjust for

confounding variables. Although the associations reported in those studies are not causal, they could still be useful in identifying children with a particular risk of ECC and thus inform the inclusion of groups to be prioritized for oral health care programmes. It may be argued that the EBF intervention could have had an effect on the ECC incidence and that the RCT groups should have been included in the models of this present study. Even though the primary outcome of the intervention was to evaluate effects on EBF and diarrhoea prevalence, the confidence intervals for the effect of the intervention on ECC described in another article 0.91 (95% CI 0.65-1.2),29 suggests that if any effect at all, it was neither of statistical or clinical significance. Moreover, the intervention study was powered to evaluate effects on EBF and diarrhoea prevalence and not ECC and this could have led to type II errors, thus the nonsignificant findings of the intervention on ECC.29

From a public health perspective, this study highlights the importance of early oral health intervention with caregivers both before and immediately after delivery, suggesting that those interventions should be included in primary health care in Uganda. Scaling up optimal early feeding practices and strengthening the policies to support their implementation could prevent ECC in Ugandan preschool children.


The use of DAGs seemed to facilitate the estimation of a causal effect of EBF on ECC in 5-year-old Uganda preschool children. Both plausible models applied indicated that being exclusively breastfed for 24 weeks had a protective causal effect on ECC. Further research, particularly including some of the unmeasured variables included in the DAGs, is necessary to strengthen the present finding and allow stronger causal claims.


The authors thank the members of the PROMISE-EBF Study Group led by the steering committee comprising of Thorkild Tylleskar, Philippe van de Perre, Eva-Charlotte Ekstrom, Nicolas Meda, James KTumwine, Chi-pepo Kankasa, and Debra Jackson. The authors would also like to acknowledge the investigators from the different participating countries with the country PI listed first and others in alphabetical order of surname (Norway: Thorkild Tylleskar, Ingunn MS Engebretsen, Lars T Fad-nes, Eli Fjeld Falnes, Knut Fylkesnes, J0rn Klungs0yr, Anne Nordrehaug Astr0m, 0ystein Evjen Olsen, Bjarne Robberstad, Halvor Sommerfelt, France: Philippe Van de Perre, Sweden: Eva-Charlotte Ekstrom, Barni Nor, Burkina Faso: Nicolas Meda, Hama Diallo, Thomas Ouedrago, Jeremi Rouamba, Bernadette Traore Germain Traore, Emmanuel Zabsonre, Uganda: James K. Tumwine, Caleb Bwengye, Charles Karamagi, Victoria Nankabirwa, Jolly Nankunda, Grace Ndeezi, Margaret Wandera, Zambia: Chipepo Kankasa, Mary Katepa-Bwalya, Chafye Siuluta, Seter Siziya, South Africa: Debra Jackson, Mickey Chopra, Mark Colvin, Tanya Doh-erty, Carl Lombard, Ameena E Googa, Lyness Matizirofa, Lungiswa Nkonki, David Sanders, Sonja Swanevelder, Wanga Zembe).


None to declare.


1. Fisher-Owens S, Gansky S, Platt L, et al. Influences on children's oral health: a conceptual model. Pediatrics. 2007;120:e510-e520.

2. Petersen PE. Sociobehavioural risk factors in dental caries - international perspectives. Community Dent Oral Epidemiol. 2005;33:274-279.

3. Ben-Shlomo Y, Kuh D. A life course approach to chronic disease epidemiology: conceptual models, empirical challenges and interdisciplinary perspectives. Int J Epidemiol. 2002;31:285-293.

4. Barker D. Mothers, Babies and Health in Adult Life. Edinburgh: Churchill Livingstone; 1998.

5. Nutrition World Health Organisation. 2016. tion/topics/exclusive_breastfeeding/en/. Accessed November 6, 2016.

6. León-Cava N, Lutter C, Ross J, Martin L. Quantifying the Benefits of Breastfeeding: a Summary of the Evidence. Washington DC: Pan American Health Organization; 2002.

7. Kramer MS, Kakuma R. The optimal duration of exclusive breastfeeding: a systematic review. Adv Exp Med Biol. 2004;554:63-77.

8. Salone LR, Vann WF, Dee DL. Breastfeeding: an overview of oral and general health benefits. J Am Dent Assoc. 2013;144:143-151.

9. Valaitis R, Hesch R, Passarelli C, Sheehan D, Sinton J. A systematic review of the relationship between breastfeeding and early childhood caries. Can J Public Health. 2000;91:411-417.

10. White V. Breastfeeding and the risk of early childhood caries. Evid Based Dent. 2008;9:86-88.

11. Avila WM, Pordeus IA, Paiva SM, Martins CC. Breast and bottle feeding as risk factors for dental caries: a systematic review and meta-analysis. PLoS ONE. 2015;10:e0142922.

12. Tham R, Bowatte G, Dharmage S, et al. Breastfeeding and the risk of dental caries: a systematic review and meta-analysis. Acta Paedi-atr. 2015;104:62-84.

13. Ribeiro NME, Ribeiro MAS. Breastfeeding and early childhood caries: a critical review. J Pediatr (Rio J). 2004;80:s199-s210.

14. Shmueli G. To explain or to predict? Stat Sci. 2010;25:289-310.

15. Suttorp MM, Siegerink B, Jager KJ, Zoccali C, Dekker FW. Graphical presentation of confounding in directed acyclic graphs. Nephrol Dial Transplant. 2015;19:1418-1423.

16. Williamson EJ, Aitken Z, Lawrie J, Dharmage SC, Burgess JA, Forbes AB. Introduction to causal diagrams for confounder selection. Respirology. 2014;19:303-311.

17. Akinkugbe AA, Sharma S, Ohrbach R, Slade GD, Poole C. Directed acyclic graphs for oral disease research. J Dent Res. 2016;95:853-859.

18. Greenland S, Pearl J, Robins JM. Causal diagrams for epidemiologic research. Epidemiology. 1999;10:37-48.

19. Hernón MA, Hernóndez-Díaz S, Robins JM. A structural approach to selection bias. Epidemiology. 2004;15:615-625.

20. Greenland S, Neutra R. Control of confounding in the assessment of medical technology. Int J Epidemiol. 1980;9:361-367.

21. Congiu G, Campus G, Luglie PF. Early Childhood Caries (ECC) prevalence and background factors: a review. Oral Health Prev Dent. 2014;12:71-76.

22. Horowitz HS. Research issues in early childhood caries. Community Dent Oral Epidemiol. 1998;26:67-81.

23. Hooley M, Skouteris H, Boganin C, Satur J, Kilpatrick N. Parental influence and the development of dental caries in children aged 0-6 years: a systematic review of the literature. J Dent. 2012;40:873-885.

24. Merchant AT, Pitiphat W. Directed acyclic graphs (DAGs): an aid to assess confounding in dental research. Community Dent Oral Epidemiol. 2002;30:399-404.

25. Chaffee BW, Feldens CA, Vitolo MR. Association of long-duration breastfeeding and dental caries estimated with marginal structural models. Ann Epidemiol. 2014;24:448-454.

-I^NTCTRY™e -Wl I FY—-


26. Palkovicova L, Ursinyova M, Masanova V, Yu Z, Hertz-Picciotto I. Maternal amalgam dental fillings as the source of mercury exposure in developing fetus and newborn. J Expo Sci Environ Epidemiol. 2008;18:326-331.

27. Nieri M, Rotundo R, Franceschi D, Cairo F, Cortellini P. Pini Prato G. Factors affecting the outcome of the coronally advanced flap procedure: a Bayesian network analysis. J Periodontol. 2009;80:405-410.

28. Nieri M, Crescini A, Rotundo R, Baccetti T, Cortellini P, Pini Prato GP. Factors affecting the clinical approach to impacted maxillary canines: a Bayesian network analysis. Am J Orthod Dentofacial Orthop. 2010;137:755-762.

29. Birungi N, Fadnes LT, Okullo I, et al. Effect of breastfeeding promotion on early childhood caries and breastfeeding duration among 5 year old children in Eastern Uganda: a cluster randomized trial. PLoS ONE. 2015;10:e0125352.

30. Leong PM, Gussy MG, Barrow S-YL, de Silva-Sanigorski A, Waters E. A systematic review of risk factors during first year of life for early childhood caries. Int J Paediatr Dent. 2013;23:235-250.

31. Harris R, Nicoll A, Adair P, Pine C. Risk factors for dental caries in young children: a systematic review of the literature. Community Dent Health. 2004;21:71-85.

32. Tylleskär T, Jackson D, Meda N, et al. Exclusive breastfeeding promotion by peer counsellors in sub-Saharan Africa (PROMISE-EBF): a cluster-randomised trial. Lancet. 2011;378:420-427.

33. Engebretsen IM, Jackson D, Fadnes LT, et al. Growth effects of exclusive breastfeeding promotion by peer counsellors in sub-Saharan Africa: the cluster-randomised PROMISE EBF trial. BMC Public Health. 2014;14:633.

34. Schellenberg JA, Victora CG, Mushi A, et al. Inequities among the very poor: health care for children in rural southern Tanzania. Lancet 2003;361:561-566.

35. WHO. Oral Health Surveys:basic Methods, 5th edn. Albany, NY, USA: World Health Organization; 2013.

36. De Onis M, Blässner M. The World Health Organization global database on child growth and malnutrition: methodology and applications. Int J Epidemiol. 2003;32:518-526.

37. Hernan MA, Robins JM. Causal Inference. Boca Raton, Florida: Chapman & Hall/CRC; 2017.

38. Mashoto KO, Astrom AN, Skeie MS, Masalu JR. Socio-demographic disparity in oral health among the poor: a cross sectional study of early adolescents in Kilwa district, Tanzania. BMC Oral Health. 2010;10:7.

39. Wamani H, Tylleskar T, Astrom AN, Tumwine JK, Peterson S. Mothers' education but not fathers' education, household assets or land ownership is the best predictor of child health inequalities in rural Uganda. Int J Equity Health. 2004;3:9.

40. Kikafunda JK, Walker AF, Collett D, Tumwine JK. Risk factors for early childhood malnutrition in Uganda. Pediatrics 1998;102:E45.

41. Fadnes LT, Engebretsen IM, Wamani H, Wangisi J, Tumwine JK, Tylleskaär T. Need to optimise infant feeding counselling: a cross-sectional survey among HIV-positive mothers in Eastern Uganda. BMC Pediatr. 2009;9:1-11.

42. Bbaale E. Determinants of early initiation, exclusiveness, and duration of breastfeeding in Uganda. J Health Popul Nutr. 2014;32:249-260.

43. Wamani H, Astram AN, Peterson S, Tylleskär T, Tumwine JK. Infant and young child feeding in western Uganda: knowledge, practices and socio-economic correlates. J Trop Pediatr. 2005;51:356-361.

44. Weitoft GR, Hjern A, Haglund B, Rosén M. Mortality, severe morbidity, and injury in children living with single parents in Sweden: a population-based study. Lancet. 2003;361:289-295.

45. Mattila M-L, Rautava P, Sillanpaa M, Paunio P. Caries in Five-year-old Children and Associations with family-related factors. J Dent Res. 2000;79:875-881.

46. Wigen TI, Wang NJ. Caries and background factors in Norwegian and immigrant 5-year-old children. Community Dent Oral Epidemiol. 2010;38:19-28.

47. Engebretsen IM, Wamani H, Karamagi C, Semiyaga N, Tumwine J, Tylleskar T. Low adherence to exclusive breastfeeding in Eastern Uganda: a community-based cross-sectional study comparing dietary recall since birth with 24-hour recall. BMC Pediatr. 2007;7:10.

48. Lamberti LM, Walker CLF, Noiman A, Victora C, Black RE. Breastfeeding and the risk for diarrhea morbidity and mortality. BMC Public Health. 2011;11:S15.

49. Masumo R, Bardsen A, Mashoto K, Astr0m AN. Feeding practice among 6-36 months old in Tanzania and Uganda: reliability and relationship with early childhood caries, ECC. Acta Odontol Scand. 2013;71:1309-1318.

50. Masumo R, Bardsen A, Astr0m AN. Developmental defects of enamel in primary teeth and association with early life course events: a study of 6-36 month old children in Manyara, Tanzania. BMC Oral Health. 2013;13:21.

51. Holgerson PL, Vestman NR, Claesson R, et al. Oral microbial profile discriminates breast-fed from formula-fed infants. J Pediatr Gastroen-terol Nutr. 2013;56:127-136.

52. Danielsson Niemi L, Hernell O, Johansson I. Human milk compounds inhibiting adhesion of mutans streptococci to host ligand-coated hydroxyapatite in vitro. Caries Res 2009;43:171-178.

53. Seow WK, Cheng E, Wan V. Effects of oral health education and tooth-brushing on mutans streptococci infection in young children. Pediatr Dent. 2003;25:223-228.

54. Gibson S, Williams S. Dental caries in pre-school children: associations with social class, toothbrushing habit and consumption of sugars and sugar-containing foods. Further analysis of data from the National Diet and Nutrition Survey of children aged 1.5-4.5 years. Caries Res. 1999;33:101-113.

55. Touger-Decker R, Van Loveren C. Sugars and dental caries. Am J Clin Nutr. 2003;78:881S-892S.

56. Dubois L, Farmer A, Girard M, Peterson K. Regular sugar-sweetened beverage consumption between meals increases risk of overweight among preschool-aged children. J Am Diet Assoc. 2007;107:924-934.

57. Fadnes LT, Nankabirwa V, Engebretsen IM, et al. Effects of an exclusive breastfeeding intervention for six months on growth patterns of 4-5 year old children in Uganda: the cluster-randomised PROMISE EBF trial. BMC Public Health. 2016;16:555.

58. Perera PJ, Ranathunga N, Fernando MP, Sampath W, Samaranayake GB. Actual exclusive breastfeeding rates and determinants among a cohort of children living in Gampaha district Sri Lanka: a prospective observational study. Int Breastfeed J. 2012;7:21.

59. Alvarez JO. Nutrition, tooth development, and dental caries. Am J Clin Nutr. 1995;61:410s-416s.

60. Pascoe L, Kim Seow W. Enamel hypoplasia and dental caries in Australian aboriginal children: prevalence and correlation between the two diseases. Pediatr Dent. 1994;16:193-199.

61. Sheetal A, Hiremath VK, Patil AG, Sajjansetty S, Kumar SR. Malnutrition and its oral outcome - a review. J Clin Diagn Res. 2013;7:178-180.

62. Fejerskov O, Kidd E, Nyvad B. Dental caries: the disease and its clinical management. Oxford: Blackwell Munksgaard; 2015.

63. Okullo I, Astr0m A, Haugejorden O. Social inequalities in oral health and in use of oral health care services among adolescents in Uganda. Int J Paediatr Dent. 2004;14:326-335.

64. Nasir EF, Astr0m AN, David J, Ali RW. Utilization of dental health care services in context of the HIV epidemic-a cross-sectional study of dental patients in the Sudan. BMC Oral Health. 2009;9:30.

65. Froot KA. Consistent covariance matrix estimation with cross-sectional dependence and heteroskedasticity in financial data. J Financ Quant Anal. 1989;24:333-355.

66. Hernan MA, Hernandez-Diaz S, Werler MM, Mitchell AA. Causal knowledge as a prerequisite for confounding evaluation: an application to birth defects epidemiology. Am J Epidemiol. 2002;155:176-184.

67. Cole SR, Platt RW, Schisterman EF, et al. Illustrating bias due to conditioning on a collider. Int J Epidemiol. 2010;39:417-420.

68. Annerosa B, Maik W, Susanne K. Early Childhood Caries: a multifactorial disease. OHDMBSC. 2010;9:32-38.

69. Chen M, Andersen R, Barmes D, Leclerq M, Lyttle S. Comparing Oral Health Systems. A Second International Collaborative Study. Geneva: World Health Organization; 1997.

70. Berney LR, Blane DB. Collecting retrospective data: accuracy of recall after 50 years judged against historical records. Soc Sci Med. 1997;45:1519-1525.

71. Seaman SR, White IR. Review of inverse probability weighting for dealing with missing data. Stat Methods Med Res. 2013;22:278-295.

72. Ismail AI, Brodeur J-M, Gagnon P, et al. Prevalence of non-cavitated and cavitated carious lesions in a random sample of 7-9-year-old schoolchildren in Montreal, Quebec. Community Dent Oral Epidemiol. 1992;20:250-255.

73. Amarante E, Raadal M, Espelid I. Impact of diagnostic criteria on the prevalence of dental caries in Norwegian children aged 5, 12 and 18 years. Community Dent Oral Epidemiol. 1998;26:87-94.

74. Landis JR, Koch GG. An application of hierarchical kappa-type statistics in the assessment of majority agreement among multiple observers. Biometrics. 1977;33:363-374.

75. Wandera M, Engebretsen IM, Okullo I, Tumwine JK, Astrom AN, Group P-ES. Socio-demographic factors related to periodontal status and tooth loss of pregnant women in Mbale district, Uganda. BMC Oral Health. 2009;9:18.

76. Johansson I, Lif Holgerson P, Kressin NR, Nunn ME, Tanner AC. Snack-ing habits and caries in young children. Caries Res. 2010;44:421-430.

77. Hawkes C. Uneven dietary development: linking the policies and processes of globalization with the nutrition transition, obesity and diet-related chronic diseases. Global Health. 2006;2:4.

78. Wigen TI, Espelid I, Skaare AB, Wang NJ. Family characteristics and caries experience in preschool children. A longitudinal study from pregnancy to 5 years of age. Community Dent Oral Epidemiol. 2011;39:311-317.

79. Williams SA, Hargreaves JA. An inquiry into the effects of health related behaviour on dental health among young Asian children resident in a fluoridated city in Canada. Community Dent Health. 1990;7:413-420.

80. Al-Dashti A, Williams S, Curzon M. Breast feeding, bottle feeding and dental caries in Kuwait, a country with low-fluoride levels in the water supply. Community Dent Health. 1995;12:42-47.

81. Derkson G, Ponti P. Nursing bottle syndrome; prevalence and etiology in a non-fluoridated city. J Can Dent Assoc. 1982;48:389-393.


Additional Supporting Information may be found online in the supporting information tab for this article.

How to cite this article: Birungi N, Fadnes LT, Kasangaki A, et al. ; For the PROMISE-EBF study group. Assessing causal effects of early life-course factors on early childhood caries in 5-year-old Ugandan children using directed acyclic graphs (DAGs): A prospective cohort study. Community Dent Oral Epidemiol. 2017;00:1-10. cdoe.12314