Dietary Patterns and Risk of a New Carious Lesion Postpartum: A Cohort Study

Abstract

Dental caries (cavities), one of the most common infectious diseases, is caused by a number of factors. Oral microbes, dietary practices, sociodemographic factors, and dental hygiene all inform caries risk. Assessing the impact of diet is complicated as individuals eat foods in combinations, and the interactions among the foods may alter caries risk. Our study aimed to prospectively assess the association between dietary patterns and caries risk in the postpartum period, a potentially sensitive period for caries development. We analyzed in-person dental assessments and telephone food frequency questionnaires (FFQs) from 879 Caucasian women participating in the Center for Oral Health Research in Appalachia Cohort 2 (COHRA2) that were collected biannually for up to 6 y. One-week recall of food intake frequency was assessed using a Likert scale. We used principal component analysis to summarize the FFQ data; the top 2 components described 15% and 12% of the variance in FFQ data. The first component was characterized by high consumption of fruits and vegetables, while the second component was heavily influenced by desserts and crackers. We used a modified Poisson model to predict the risk of an increase in the number of decayed, missing, and filled teeth in the postpartum period by 1) dietary patterns and 2) individual foods and beverages at the previous study visit, after controlling for other known risk factors, including history of carious lesions. Eating a dietary pattern high in desserts and crackers was associated with a 20% increase in the number of decayed, missing, and filled teeth in the postpartum period (95% confidence interval, 1.03–1.39). However, this effect was attenuated among those who also consumed a dietary pattern high in fruits and vegetables. Dietary patterns should be considered when devising interventions aimed at preventing dental caries.

Introduction

After the common cold, dental caries is the second most common infectious disease, affecting 2.4 billion people worldwide (Kassebaum et al. 2015). Dental decay can negatively affect social life, self-esteem (Kaur et al. 2017), and overall quality of life (Bukhari 2020), including increases in anxiety and depression compared to those with low or no caries incidence (Kastenbom et al. 2019). Many factors contribute to dental caries, including diet, sociodemographic factors, fluoride exposure, and dental hygiene practices (Aas et al. 2008; Costa et al. 2012; Pitts et al. 2017; Mosaddad et al. 2019). Diet has direct and indirect effects on risk of dental decay. Dietary acids such as citric acid found in carbonated and fruit beverages can erode enamel (Abou Neel et al. 2016), whereas a diet high in free sugars selects for acid-producing bacteria that cause decay (Conlon and Bird 2014).

While individual foods can modify risk of dental decay, most foods are consumed in combinations, not isolation; these combinations are often specific to culture, region, resources, and/or wealth. The idea that food combinations are more informative than individual food constituents (Jacobs and Tapsell 2013) suggests dietary patterns may be more important than individual foods in understanding the relationships between diet and health. Previous work by our group used the 2013 to 2014 National Health and Nutrition Examination Survey (NHANES) to assess the relationship between caries and diet patterns in 5,855 adolescents and adults. Results suggested prevalence of decay was associated with dietary patterns and not individual foods, although the effect sizes were small—possibly due to relatively low prevalence of caries at a nationally representative level (Blostein et al. 2020). Similar findings have been reported elsewhere, including a cross-sectional study of 14,517 Latino adults, linking higher diet quality scores to fewer decayed teeth (Sanders et al. 2020), and in a study of 7,751 adults who completed dental examinations and two 24-h NHANES dietary recalls in which odds of coronal caries were reduced among those who met fruits, greens and beans, and added sugars components of Dietary Guidelines for Americans (Kaye et al. 2020). Finally, in a prospective study of 533 older men, adherence to the Dietary Approaches to Stop Hypertension (DASH) diet was associated with a decreased adjusted lower root caries increment (Kaye et al. 2015).

Changes in the salivary factors associated with cariogenesis increase risk of dental decay during the postpartum period (Yousefi et al. 2020). Furthermore, changes in dietary patterns during pregnancy (Rio et al. 2020)—such as increased consumption of sweet snacks between meals—may increase risk of dental decay; these patterns may persist postpartum, especially when breastfeeding. Literature on mothers’ oral health during the postpartum period is sparse (Heimonen et al. 2008; Ramadugu et al. 2020). Therefore, understanding the role of dietary patterns on maternal risk of caries postpartum is sorely needed. Here, we investigate the relationships between diet and occurrence of a new dental lesion in postpartum mothers. Specifically, we prospectively compare associations between 1) dietary patterns and 2) individual foods and beverages reported by mothers at the previous study visit on the yearly risk of an increase in the number of decayed, missing, and filled teeth (DMFT).

Methods

Study Population

We analyzed data from women participating in the Center for Oral Health Research in Appalachia Cohort 2 (COHRA2) (Neiswanger et al. 2015). Briefly, pregnant women were enrolled on a rolling basis between 2012 and 2016, and they and their child were followed regularly. We included everyone followed until 2018. As described previously (Neiswanger et al. 2015), participants were Caucasian Pennsylvanian or West Virginian pregnant women 18 y of age and older, fluent in English, and not immunocompromised. Mothers were recruited through clinics, community centers, hospitals, and outreach offices. Print, radio, and television advertisements were used to promote recruitment. The study protocol was approved by the institutional review boards at the Universities of Pittsburgh and West Virginia.

This analysis includes data that were cleaned in the database as of November 30, 2018, with additional corrections to the DMFT as of November 2020. Dietary data from 1,043 women (1,024 enrollments, 19 women enrolled twice) at the 10-wk telephone interview and all diet information collected at subsequent encounters were used to identify dietary patterns (5,553 interviews). Only women with an enrollment (prenatal) visit and a postnatal visit at 12 mo or more were included in the analysis, for a sample size of 879 for the imputed data set and 729 for the complete case analysis. The enrollment dental exam was used to determine baseline DMFT (Appendix Fig. 1). This study is reported using the Strengthening the Reporting of Observational Studies in Epidemiology guidelines (von Elm et al. 2007).

Exposure Data

Telephone interviews assessed eating habits, tooth care, demographics, and alcohol, drug, and tobacco use at 10 wk, 6 mo, and every 6 mo (±2 wk) thereafter for up to 6 y following the child’s birth. The 30- to 45-min calls were conducted by the University Center for Social & Urban Research (University of Pittsburgh). Food frequencies within the past 7 d were recorded using a 4-level Likert scale (never or once, every few days, once a day, several times a day). See Appendix Figure 2 and Appendix Table 1 for food categories and beverage items. All intake was scored, including consumption that occurred never or once. There were 6-mo intervals between diet interviews and dental examinations. For this analysis, we used diet data from the most recent interview preceding the dental exam of interest.

We grouped beverages by whether they were sugar-sweetened beverages (SSBs), artificially sweetened beverages, or unsweetened beverages (including plain water) and calculated intake by group. Cow’s milk, 100% juice, and alcohol were included as separate variables. To identify dietary patterns, 11 individual food variables and beverage groups from all enrollment and follow-up surveys completed by the 1,024 participants were used in an orthogonally rotated principal component analysis (PCA) (Appendix Fig. 1). Factors were retained using the scree plot “elbow rule” and eigenvalues greater than 2 (Nguyen and Holmes 2019).

Outcome Data

Registered dental hygienists and licensed dentists conducted dental examinations at 2 mo postpartum, yearly until 2 y postpartum, and then every other year up to 6 y postpartum. Calibration for oral health screenings was done regularly. The mean Cohen’s κ score for individual staff members ranged from 46.1 to 80.6, with a mean score of 70.8 across all staff (Neiswanger et al. 2015). Caries were scored using the protocol in the PhenX toolkit (https://www.phenxtoolkit.org/protocols/view/80301) based on the NHANES Decayed, Missing, or Filled Tooth Surfaces (DMFS) Index, 2003 to 2004 (wisdom teeth were excluded). Sealants were counted as sound. No X-rays were taken.

We defined incident decay as an increase in DMFT during the current dental exam over that noted during the prior dental exam. Two individuals had a DMFT count of 28 at enrollment. On occasion (n = 71), a participant’s DMFT decreased (e.g., 16 to 15). The dental exams for all visits for each of these were reviewed by 2 authors (ED and BF) and recorded for future reference.

Additional Covariates

Assessment and interview protocol details have been previously described (Neiswanger et al. 2015). Sociodemographic factors (geographic location [by site], age, dental and medical insurance, family income, education level) were collected at enrollment and all phone interviews. Age was modeled as continuous, and education level (high school vs. higher) and family income (<$50,000/y vs. higher) were dichotomized. Dental hygiene practices and additional behaviors were assessed during the phone interview. Participants were asked if they ever brushed their teeth (yes/no) and ever flossed (yes/no) and the frequency of each. Breastfeeding or smoking at any time since the previous interview (the past 4–6 mo) was recorded as a yes for our binary variables. Household water fluoride levels were collected at enrollment. A range of 0.7 to 1.2 ppm was used as a reference as per the U.S. Department of Health and Human Services Federal Panel on Community Water Fluoridation (2015).

Statistical Modeling

We considered each visit as a separate unit of analysis and adjusted for correlation within an individual in our models. We fit modified Poisson models with autoregressive covariance matrices to account for correlation between repeated measures in our generalized estimating equations. The modified Poisson approach is appropriate for nonrare outcomes (Zou 2004; Naimi and Whitcomb 2020). We predict an increase in DMFT relative to the previous dental examination and include as a covariate history of DMFT increase at the prior dental exam (i.e., lagged any DMFT increase). We conducted the analysis using 1) complete cases (n = 729) and 2) a multiply imputed data set with 50 imputations, using the fully conditional specification (FCS) method (n = 879).

Our main predictors were quartiles of principal components (PCs) or the top-loading individual foods from the PCs plus SSBs and 100% juice. PC scores were categorized into quartiles for interpretability and used as ordinal variables for statistical modeling. The distributions of the PC scores are included in Appendix Figure 3 and Figure 1A. As a sensitivity analysis, we used nonlagged dietary data. We calculated the percentage of visits with an increase in DMFT for each combination of PCs (Fig. 1D) and adjusted ratios using an “estimate” statement in SAS for the diet quartiles of interest. All analyses were conducted using SAS software, Version 9.4. Copyright © 2013 SAS Institute Inc. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc., Cary, NC, USA.

Figure.

Dietary patterns over time and associations with an increase in decayed, missing, and filled teeth (DMFT) data from 5,553 phone interviews. (A) Distribution of principal component (PC) scores by site for individuals included in the models. Black square indicates mean PC score value. (B) PC food categories with factor loadings greater than 0.30. (C) Cross-tabulation of quartile scores for both PCs at each visit included in the models. (D) Three-dimensional bar graph showing percentage of visits with an increase in DMFT for selected combinations of PC quartile scores. Figure colors solely provided for ease of comparison across panels. All available dietary data used, including those from second pregnancy or incomplete records. Center for Oral Health Research in Appalachia Cohort 2, 2012 to 2018.

Results

Demographics

The 879 participants completed an average of 2.95 ± 1.01 phone interviews with a paired dental exam (range, 1–6). Participants averaged 30 y of age (SD, 5.12 y; range, 19–44) and DMFT of 6.6 (SD, 5.53; range, 0–28) at enrollment (Appendix Fig. 4). At the 12-mo visit, 42% of participants had an increase in DMFT compared to enrollment. Age, education, income, smoking status, flossing, marital status, and DMFT at enrollment differed significantly among those with and without an increase in DMFT (Table 1).

Table 1.

Sociodemographic and Behavioral Characteristics of 879 Women Included in Models at Enrollment, by Increase in the Number of Decayed, Missing, or Filled Teeth in the Postpartum Period.

Characteristic	No Increase in DMFT during Postpartum Period		Increase in DMFT during Postpartum Period		P a
	n	%	n	%
Total	396	45	483	54
Age					<0.0001
18–25	67	17	146	30
26–29	103	26	126	26
30–32	116	29	98	20
+33	110	28	113	23
Site					0.12
Pittsburgh	232	59	258	53
West Virginia	164	41	225	47
Education					<0.0001
Greater than high school	353	89	358	74
Marital status					<0.0001
Single	72	18	158	33
Married	297	75	281	58
Domestic partner	19	4.81	26	5
Widowed	0	0	1	0.2
Divorced	7	2	17	4
Missing	1
Income					<0.0001
<25,000	59	15	128	28
25,000–49,999	64	17	96	21
50,000–74,999	147	38	143	31
75,000+	113	30	88	19
Missing	41
Insurance
Has health insurance	388	98	476	99	0.52
Has dental insurance	287	73	336	72	0.57
Missing	19
No smoking	370	93	389	82	<0.0001
Missing	10
Brush teeth	394	99.8	482	99.8	0.89
Missing	1
Brush frequency					0.05
Never to a few times a week	7	2	16	3
Once a day	83	21	121	25
Twice a day	278	71	300	62
3 or more times a day	26	7	45	9
Missing	3
Floss	325	82	339	70	<0.0001
Floss frequency					0.002
Never to a few times a week	209	64	193	57
Once a day	99	30	104	31
Twice a day	16	5	29	9
3 or more times a day	1	0.3	13	4
Missing	215
No alcohol	247	87	277	89	0.37
Missing	283
DMFT at enrollment					<0.0001
0	70	18	30	6
1–2	60	15	76	16
3–7	134	34	186	39
8 or more	132	33	191	40
Fluoride in toothpaste					0.80
Yes	322	91	364	91
No	33	9	35	9
Missing	138
Fluoride in water (ppm)					0.12
0.00–0.69	118	30	158	33
0.70–1.20	261	67	290	61
1.21–1.84	13	3	27	6
Missing	12

Center for Oral Health Research in Appalachia Cohort 2 (COHRA2), 2012 to 2018. Totals vary because of missing data.

DMFT, decayed, missing, and filled teeth.

^aP value from χ² test and Fisher’s test when indicated.

PCs Were Influenced by Fruits & Vegetables and Desserts & Crackers

Data from all dietary interviews (n = 5,553 interviews) were used in the PCA. PCs were named “fruit & vegetable” and “dessert & cracker” after the variables, which had >|0.30| loadings (Figure 1B). PC1 “fruit & vegetable” loaded positively on fruits (canned or fresh, including applesauce), vegetables (excluding potatoes), unsweetened beverages and water, nuts (peanut butter, beans, seeds, soy products, veggie meat, protein bars, or supplements), cheese (or yogurt), and meats (poultry, fish, and eggs) and negatively on SSBs. PC2 “dessert & cracker” loaded positively on desserts, crackers (crackers, breads, and biscuits), candies, chips, potatoes, SSBs, cereal, and cheese.

Dietary Patterns Were Stable during Follow-up

The 879 participants included in the models tended to stay in the same PC or adjacent quartile throughout follow-up. Sixty-seven percent of participants who started in the 2 highest quartiles of the fruit & vegetable PC remained in the 2 highest quartiles. Sixty-four percent of those starting in the lowest 2 quartiles remained in the lowest quartiles. Quartile membership in the dessert & cracker PC was slightly more variable: 48% of participants who started in the top 2 quartiles remained in the top 2 quartiles, and 57% of those who started in the bottom 2 quartiles remained there.

Percentage of Visits with Increase in DMFT Changed with PC Quartile Combination

To further examine how the PC quartiles interact, we first classified each visit by their food PC quartiles (Fig. 1C). (Each visit has a score for both PCs.) We then determined the proportion in each group that had an increase in DMFT (Fig. 1D). For example, an increase in DMFT was detected in 42% of the 126 participant visits in quartile 1 for the fruit & vegetable PC and quartile 4 (dessert & cracker), compared to 30% of the 113 participant visits in quartile 4 for PC1 and PC2.

Associations between Diet and Increase in DMFT Using a Modified Poisson Model

We predicted risk of an increase in DMFT at each visit and included as a covariate history of DMFT increase at the prior visit (lagged) for the complete case and imputed data (Table 2). While the fruit & vegetable PC was associated with lower risk of an increase in DMFT in a dose–response manner, after adjustment, the effect was no longer seen. However, the positive association between an increase in DMFT and the fourth desserts & crackers PC quartile remained (imputed risk ratio [RR] = 1.20; 95% confidence interval [CI], 1.03–1.39). However, the strongest predictor of decay was an increase in DMFT at the previous visit (imputed RR = 3.57; 95% CI, 3.16–4.05). In addition, low income and dental insurance were significantly associated with an increase in DMFT, while medical insurance was associated with a decrease in risk. Complete case and imputed results were similar.

Table 2.

Risk Ratios from Modified Poisson Models, Modeling the Association of Principal Components (PCs) with an Increase in the Number of Decayed, Missing, and Filled Teeth in the Postpartum Period by Quartile Increase in PC Score from the Previous Visit (Complete Case Analysis, n = 729; Imputed, n = 879).

Variable	Risk Ratio $e^{\beta }$ (95% CI)
	Unadjusted Complete Case Model	Unadjusted Imputed Model	Adjusted Complete Case Model	Adjusted Imputed Model
Vegetable and fruit PC a
Second quartile	0.88 (0.72–1.06)	0.81 (0.68–0.96)	0.98 (0.83–1.16)	0.96 (0.83–1.11)
Third quartile	0.86 (0.70–1.04)	0.78 (0.66–0.93)	1.04 (0.88–1.23)	1.03 (0.88–1.20)
Fourth quartile	0.62 (0.50–0.78)	0.57 (0.46–0.69)	0.93 (0.76–1.13)	0.90 (0.76–1.07)
Dessert and cracker PC a
Second quartile	1.08 (0.86–1.35)	1.16 (0.95–1.41)	1.10 (0.92–1.32)	1.15 (0.98–1.36)
Third quartile	1.13 (0.91–1.41)	1.17 (0.96–1.43)	1.17 (0.98–1.40)	1.21 (1.03–1.43)
Fourth quartile	1.26 (1.02–1.56)	1.26 (1.05–1.53)	1.18 (1.00–1.39)	1.20 (1.03–1.39)
Covariates
Site: West Virginia			0.94 (0.83–1.07)	0.91 (0.81–1.03)
Mother’s age			0.99 (0.97–1.00)	0.98 (0.97–1.00)
Medical insurance			0.71 (0.57–0.88)	0.78 (0.64–0.96)
Dental insurance			1.37 (1.14–1.64)	1.27 (1.08–1.49)
Low income b			1.20 (1.05–1.38)	1.21 (1.06–1.38)
Flossing			1.01 (0.90–1.14)	0.98 (0.88–1.09)
Breastfed c			1.05 (0.93–1.19)	1.03 (0.91–1.16)
Cigarettes d			1.14 (1.00–1.31)	1.21 (1.07–1.35)
Months postpartum			1.00 (0.99–1.00)	1.00 (0.99–1.00)
Low fluoride levels e			1.05 (0.93–1.19)	1.05 (0.94–1.17)
High fluoride levels f			1.14 (0.90–1.44)	1.08 (0.85–1.37)
Increase in DMFT at most recent visit			3.80 (3.31–4.37)	3.57 (3.16–4.05)

Center for Oral Health Research in Appalachia Cohort 2, 2012 to 2018. Bold text indicates P value < 0.05.

CI, confidence interval; DMFT, decayed, missing, and filled teeth.

^aReference group: first quartile.

^bHousehold annual income less than $50,000.

^cBreastfed in the past 4 to 6 mo.

^dSmoked cigarettes since last interview.

^eLow fluoride defined as household water levels below 0.70 ppm.

^fHigh fluoride defined as household water levels above 1.20 ppm. Imputed models include 50 imputations using the fully conditional specification method.

We next modeled risk of an increase in DMFT at each visit by the top-loading foods for each PC plus SSBs and 100% juice using complete case and imputed data (Table 3). After adjustment, only the association with eating dessert every few days or once a day was significantly associated with an increase in DMFT (dessert once a day, imputed RR = 1.22; 95% CI, 1.05–1.43). As in the models using PC quartiles, the covariate with the strongest association with an increase in DMFT was an increase in DMFT at the previous dental exam (Table 3). Complete case and imputed results were similar.

Table 3.

Risk Ratios from Modified Poisson Models, Modeling the Association of Top-Loading Cariogenic Food Groups with an Increase in the Number of Decayed, Missing, and Filled Teeth in the Postpartum Period by Frequency of Food Consumption at the Previous Visit (Complete Case Analysis, n = 729; Imputed, n = 879).

Variable	Risk Ratio $e^{\beta }$ (95% CI)
	Unadjusted Complete Case Model	Unadjusted Imputed Model	Adjusted Complete Case Model	Adjusted Imputed Model
Vegetable a
Every few days	0.77 (0.55–1.07)	0.74 (0.57–0.95)	0.88 (0.66–1.18)	0.85 (0.68–1.07)
Once a day	0.71 (0.51–0.98)	0.70 (0.54–0.90)	0.81 (0.61–1.08)	0.81 (0.64–1.02)
Several times a day	0.70 (0.50–0.98)	0.63 (0.48–0.83)	0.83 (0.62–1.13)	0.81 (0.63–1.04)
Fruit a
Every few days	0.83 (0.61–1.12)	0.85 (0.66–1.09)	1.05 (0.84–1.31)	0.98 (0.81–1.19)
Once a day	0.83 (0.61–1.12)	0.81 (0.63–1.04)	1.12 (0.89–1.40)	1.00 (0.82–1.22)
Several times a day	0.77 (0.56–1.05)	0.76 (0.59–0.99)	1.11 (0.88–1.41)	1.02 (0.83–1.26)
Dessert a
Every few days	1.09 (0.90–1.32)	1.06 (0.90–1.26)	1.22 (1.04–1.43)	1.16 (1.01–1.33)
Once a day	1.19 (0.97–1.47)	1.11 (0.92–1.34)	1.27 (1.08–1.51)	1.22 (1.05–1.43)
Several times a day	1.26 (0.92–1.73)	1.17 (0.89–1.54)	1.23 (0.93–1.61)	1.13 (0.89–1.44)
Cracker a
Every few days	1.03 (0.81–1.31)	1.11 (0.90–1.38)	1.04 (0.85–1.29)	1.13 (0.93–1.37)
Once a day	0.89 (0.70–1.13)	1.00 (0.80–1.25)	1.00 (0.81–1.23)	1.10 (0.91–1.34)
Several times a day	1.00 (0.76–1.32)	1.03 (0.80–1.34)	1.12 (0.88–1.43)	1.13 (0.90–1.41)
Sugar-sweetened beverages b
Not specified	0.81 (0.66–1.00)	0.82 (0.68–1.00)	0.93 (0.78–1.11)	0.91 (0.77–1.08)
Once a day	0.96 (0.78–1.19)	1.00 (0.82–1.21)	0.92 (0.76–1.11)	0.95 (0.80–1.12)
Several times a day	0.98 (0.79–1.22)	1.02 (0.84–1.23)	0.96 (0.79–1.15)	0.94 (0.80–1.11)
100% juice a
Every few days	0.99 (0.83–1.20)	1.03 (0.87–1.21)	0.91 (0.78–1.07)	0.96 (0.84–1.10)
Once a day	1.15 (0.92–1.44)	1.20 (0.98–1.46)	0.97 (0.80–1.17)	1.01 (0.85–1.20)
Several times a day	1.62 (1.25–2.10)	1.55 (1.21–1.98)	1.04 (0.83–1.30)	1.09 (0.89–1.34)
Covariates
Site: West Virginia			0.95 (0.84–1.08)	0.92 (0.82–1.04)
Mother’s age			0.99 (0.97–1.00)	0.98 (0.97–1.00)
Medical insurance			0.69 (0.55–0.87)	0.77 (0.63–0.96)
Dental insurance			1.41 (1.18–1.68)	1.29 (1.10–1.52)
Low income c			1.22 (1.07–1.40)	1.22 (1.07–1.39)
Flossing			1.00 (0.88–1.14)	0.97 (0.87–1.08)
Breastfed d			1.05 (0.93–1.20)	1.03 (0.92–1.16)
Cigarettes e			1.17 (1.01–1.34)	1.22 (1.08–1.38)
Months postpartum			0.99 (0.99–1.00)	1.00 (0.99–1.00)
Low fluoride levels f			1.06 (0.94–1.20)	1.05 (0.94–1.17)
High fluoride levels g			1.17 (0.92–1.50)	1.11 (0.87–1.41)
Increase in DMFT at most recent visit			3.85 (3.35–4.42)	3.58 (3.16–4.05)

Center for Oral Health Research in Appalachia Cohort 2, 2012 to 2018. Bold text indicates P < 0.05.

CI, confidence interval; DMFT, decayed, missing, and filled teeth.

^aReference group: Never or once.

^bReference group: Every few days.

^cHousehold annual income less than $50,000.

^dBreastfed in the past 4 to 6 mo.

^eSmoked cigarettes since last interview.

^fLow fluoride defined as household water levels below 0.70 ppm.

^gHigh fluoride defined as household water levels above 1.20 ppm. Imputed models include 50 imputations using the fully conditional specification method.

As a sensitivity analysis, we calculated risk ratios using the dietary PCs and individual foods from the current visit instead of the previous visit (Appendix Tables 2 and 3). Although dietary variables were no longer statistically significant, the directions of the associations are consistent with lagged dietary results. Use of the imputed data set did not change observed associations. We also compared the associations with current diet to those with lagged diet using a combined ratio from quartile rank in both PCs (Table 4). Unadjusted risk ratios associated with diet combinations that included the highest quartiles of the fruit & vegetable PC were in the protective direction, including combinations with a high dessert & cracker quartile ranking regardless of whether current or lagged diet was used. However, after adjustment, there was no apparent association. Results were similar using the imputed data (Appendix Table 4).

Table 4.

Unadjusted and Adjusted Risk Ratios and 95% Confidence Intervals from Complete Case Analysis (n = 729).

Dessert & cracker PC quartiles
Characteristic		4	3	2	1	Model Type
Previous diet
Fruit & vegetable PC quartiles	4	0.79 (0.58–1.07)	0.71 (0.52–0.96)	0.67 (0.49–0.92)	0.62 (0.50–0.78)	Unadjusted
	3	1.08 (0.81–1.44)	0.97 (0.73–1.30)	0.92 (0.69–1.23)	0.86 (0.70–1.04)
	2	1.11 (0.84–1.47)	1.26 (0.82–1.92)	0.95 (0.70–1.28)	0.88 (0.73–1.07)
	1	1.26 (1.02–1.56)	1.13 (0.91–1.41)	1.08 (0.86–1.35)	Reference
Fruit & vegetable PC quartiles	4	1.10 (0.86–1.40)	1.09 (0.84–1.41)	1.03 (0.79–1.33)	0.93 (0.76–1.13)	Adjusted
	3	1.23 (0.97–1.56)	1.22 (0.96–1.56)	1.15 (0.91–1.47)	1.04 (0.88–1.23)
	2	1.16 (0.92–1.46)	1.37 (0.98–1.92)	1.09 (0.86–1.38)	0.98 (0.83–1.16)
	1	1.18 (1.00–1.40)	1.18 (0.98–1.41)	1.11 (0.92–1.33)	Reference
Current diet
Fruit & vegetable PC quartiles	4	0.78 (0.58–1.04)	0.60 (0.44–0.81)	0.74 (0.56–0.98)	0.69 (0.56–0.85)	Unadjusted
	3	0.83 (0.63–1.10)	0.64 (0.48–0.86)	0.79 (0.60–1.05)	0.74 (0.60–0.90)
	2	1.00 (0.76–1.31)	0.77 (0.58–1.02)	0.95 (0.73–1.25)	0.89 (0.74–1.07)
	1	1.13 (0.92–1.38)	0.87 (0.70–1.08)	1.08 (0.88–1.31)	Reference
Fruit & vegetable PC quartiles	4	1.01 (0.79–1.29)	0.88 (0.68–1.14)	1.01 (0.79–1.30)	0.99 (0.82–1.20)	Adjusted
	3	0.90 (0.72–1.14)	0.79 (0.61–1.01)	0.90 (0.71–1.15)	0.89 (0.75–1.05)
	2	1.03 (0.82–1.30)	0.90 (0.71–1.14)	1.03 (0.82–1.30)	1.01 (0.87–1.19)
	1	1.02 (0.87–1.20)	0.89 (0.74–1.06)	1.02 (0.86–1.21)	Reference

Center for Oral Health Research in Appalachia Cohort 2, 2012 to 2018. Bold text indicates P < 0.05. Modified Poisson models for an increase in the number of decayed, missing, and filled teeth in the postpartum period by principal component (PC) quartile. Results shown using the dietary quartiles from the same interview (current diet) and the most recent interview (previous diet). Adjusted for geographical site, mother’s age, medical insurance, dental insurance, increase in decayed, missing, and filled teeth (DMFT) at previous visit, annual household income less than $50,000, months postpartum, household water fluoride level, breastfeeding, and cigarette smoking since last interview.

Discussion

In this prospective cohort study of 879 postpartum Appalachian women at high risk of dental caries, we show that eating a dietary pattern high in fruits and vegetables lowers the impact of eating a diet high in desserts and crackers on risk of caries. Associations between diet and caries risk were similar when diet information was collected at the time of the dental exam or the year prior. However, other factors, particularly an increase in DMFT at the previous dental exam and socioeconomic factors, attenuated this association. This result is consistent with the hypothesis that the effects of cariogenic foods may be modified by the consumption of other foods and with our previous cross-sectional analysis of NHANES (Blostein et al. 2020).

Our findings support and elaborate on previously reported associations between an overall high-quality diet and lower risk of caries (Kaye et al. 2020; Sanders et al. 2020). Several previous studies have emphasized the role of free sugar consumption, specifically SSBs, and other highly processed foods and beverages on risk of caries (Burt et al. 2006; Moynihan and Kelly 2014). A systematic review by Moynihan and Kelly (2014) found that in 42 of 50 children studies and 5 of 5 adult studies, there was at least 1 positive association between sugars and caries. By contrast, here and in our previous study using NHANES (Blostein et al. 2020), SSBs alone did not always explain associations with caries risk after adjustment for other foods and known risk factors, underscoring the importance of dietary patterns.

Decreasing caries incidence cannot be achieved without addressing the multiple individual and societal factors that increase risk. History of caries is a marker of cariogenic oral microbiome composition, cariogenic diet, and other factors increasing caries risk, and it is perhaps unsurprising that a history of new caries increased risk of a new lesion more than 3-fold, overwhelming associations observed with diet. Obtaining a nutritiously adequate diet is difficult for low-income families due to the higher cost of minimally processed, healthy foods and inaccessibility to merchants that provide these foods (French et al. 2019; Karpyn et al. 2020). Lack of a secure, nutritious diet occurs even when individuals are eligible for government-based assistance programs (Leung et al. 2012). This suggests there are additional difficulties experienced by low-income families that are not solved by an increase in funds alone. If dietary patterns are to be used as an intervention point for oral health, interventions will need a holistic approach that increases access to nutritious food. In our study, those with an annual household income below $50,000 had a 20% increase in risk of an additional caries lesion in the postpartum period.

A particular strength of our study is the longitudinal nature of our design, which allowed us to evaluate diet at multiple time points and establish temporality of exposure and outcome. In our analysis, we used diet information from the visit prior to the detection of a dental lesion, as the cariogenic process necessarily took place before the visit when a new lesion was detected. Furthermore, although diets were assessed by measuring intake over the previous week, we found diet was relatively stable across visits. Other strengths include assessment of dental health by trained, calibrated personnel and a large and relatively homogeneous sample of women, limiting the possibility of confounding. However, our study does have limitations. The dental examinations did not include X-rays. The diet questionnaire asked about food frequency over the past 7 d and targeted cariogenic foods; frequency categories are less granular than included in NHANES. We did not adjust for participant body mass or validate food intake.

When generalizing our results, it should be noted that all participants were Caucasian with similar geographical location and cultural practices. Loadings of foods found here may differ from other populations. In the previously analyzed NHANES population, we identified 3 PCs that loaded heavily on different foods: breads and fats, SSBs and sandwiches, and milk and cereal (Blostein et al. 2020). Also, the body undergoes many changes during the pregnancy and postpregnancy period, including the oral microbiome. This should not affect our conclusions regarding the importance of dietary patterns on caries risk, as previous studies suggest the oral microbiome returns to a healthy state during the postpartum period (Balan et al. 2018), and the foods identified are consistent with our understanding of cariogenesis. While our study population is one at high risk of caries and included only postpartum women, we believe our results are robust, biologically plausible, and therefore likely more broadly generalizable.

In conclusion, our study highlights the importance of considering dietary patterns, in that eating healthy foods such as vegetables can balance the adverse effects of eating sugary foods, such as desserts. Risk of new caries was high in the postpartum period, and multiple factors in addition to diet were associated with caries risk. Interventions to prevent caries should take a multifactorial approach.

Author Contributions

E. Davis, contributed to data analysis and interpretation, drafted and critically revised the manuscript; G. Martinez, contributed to conception, design, data analysis, and interpretation, drafted and critically revised the manuscript; F. Blostein, T. Marshall, A.D. Jones, E. Jansen, contributed to data analysis and interpretation, critically revised the manuscript; D.W. McNeil, K. Neiswanger, M.L. Marazita, contributed to data acquisition, analysis, and interpretation, critically revised the manuscript; B. Foxman, contributed to conception, design, data acquisition, analysis, and interpretation, drafted and critically revised the manuscript. All authors gave final approval and agree to be accountable for all aspects of the work.