Evidence for clinical subtypes of early childhood caries

doi:10.21203/rs.3.rs-2629612/v1

Download PDF

Article

Evidence for clinical subtypes of early childhood caries

https://doi.org/10.21203/rs.3.rs-2629612/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

The current early childhood caries (ECC) case definition contains a substantial degree of clinical heterogeneity, and to address this, we sought to identify clinical subtypes of the disease. We used tooth surface-level dental caries experience from a discovery and 3 replication community-based cohorts of 3-to-5-year-old children (N=226,471). We identified five disease subtypes with distinct patterns of caries lesion intraoral distribution that largely replicated across cohorts. These subtypes were associated with established caries risk factors (e.g., history of nighttime bottle-feeding), showed familial concordance and microbiome differences, and predicted dental caries experience 7 years after subtype assignment. Notably, classification of children in these subgroups can be achieved by inspecting small sets of easily examinable tooth surfaces with reasonable accuracy. Collectively, our findings provide evidence for generalizable and clinically recognizable subtypes of ECC. Etiology, targeted prevention, and optimal management of these subtypes should be systematically investigated in future studies.

Health sciences/Diseases/Dental diseases/Dental caries

Health sciences/Health care/Dentistry/Paediatric dentistry

dental caries

early childhood caries

children

latent class analysis

precision medicine

Early childhood caries (ECC) is the most common non-communicable childhood disease. Despite advances in the science and practice of clinical dentistry and concerted public health efforts, it remains an important health problem affecting approximately 6 million preschool-age children in the US and 600 million children worldwide^1,2. The population distribution of ECC is marked by pronounced social and economic disparities³ and has substantial impact on children, their families, and the health system⁴. Consequently, considerable attention and efforts to prevent and manage ECC have been devoted by both the dental and the medical communities^5,6.

Dental caries is a tooth surface demineralization process caused by an imbalanced (i.e., dysbiotic) microbial biofilm influenced by health-promoting factors (e.g., oral hygiene) and disease-causing (e.g., high frequency of fermentable carbohydrate consumption) factors⁷. The current understanding of the etiology of ECC encompasses interactions between biological (e.g., individual susceptibility and microbiome), behavioral (e.g., diet and oral hygiene), environmental (e.g., fluoride exposure), and social (e.g., socioeconomic and health literacy) factors⁸. Importantly, the clinical manifestations of ECC are highly variable and this heterogeneity is arguably not captured in its current case definition^2,9. Furthermore, evidence from clinical^10,11 and biological¹² studies demonstrates clustering of caries experience within groups of teeth (e.g., anterior versus posterior teeth) or types of tooth surfaces (e.g., pits and fissures versus smooth surfaces).

Understanding individual variability in the causes and presentation of the disease is key for creating better disease taxonomies¹³ and optimizing disease prevention and management, the foundational components of precision medicine¹⁴. Besides cancers, the concept of disease subtyping has recently gained considerable attention in other common, complex, chronic conditions including obesity,¹⁵ type 2 diabetes,¹⁶ asthma,^17,18 neurological conditions¹⁹ and low birth weight infant phenotypes²⁰. In principle, disease subtyping has the potential to lead to both biological discoveries and pragmatic clinical benefits, such as individualized, person-centered care²¹. Leveraging informative clustering of caries experience at the individual level could therefore help improve ECC classification. Furthermore, the discovery of generalizable subtypes of the disease may help address the clinical or biological heterogeneity in ECC and thus enhance our ability to develop useful screening tools, prediction models, and potentially treatments. To this end, the overarching goal of this study was to identify and describe clinical subtypes of ECC. We hypothesized that 1) unsupervised clustering of tooth surface-level information on caries experience would identify subgroups of children defined by distinct and homogeneous patterns of caries experience and 2) these clinical subtypes of disease would be generalizable across populations. Additionally, we explored sociodemographic, behavioral, clinical, and biological differences (including heritability, familial concordance, and microbiome) between the identified clinical subtypes and examined whether small numbers of easy-to-screen tooth surfaces might be used to accurately classify children in these clinical groups.

The optimal solution in the discovery dataset included 5 clinical subtypes (Fig. 1, top panel). This model showed excellent classification diagnostics, i.e., > 0.90 entropy and posterior probabilities. The subtypes were characterized by symmetric caries experience, distinct patterns and prevalence of affection of posterior and anterior teeth, and a gradient of increasing number of affected tooth surfaces. The most common ECC subtype (type I, 49%) represents low-prevalence caries experience (mean number of decayed, missed [due to caries] or filled primary tooth surfaces, dmfs = 4) mostly affecting molar teeth. Type II is characterized by a pattern of marked affection of pits-and-fissures, type III is characterized by maxillary anterior disease, type IV involves posterior disease including molar smooth surfaces (e.g., proximal surfaces), and finally, type V represents generalized and severe involvement of most tooth surfaces (mean dmfs = 56). These patterns were similar when non-ECC cases (i.e., children who were cavity-free) were included in the LCA, wherein a large “health/mild disease” type I is formed, and types II-V remain stable (Fig. 1, middle panel).

There were important differences in demographic characteristics, and caries risk factors between the 5 ECC subtypes in ZOE 2.0. Hispanics and American Indians/Native Americans were more represented among those with the most severe subtypes (III, IV, and V), as were children whose parents/guardians had less than high school education (Table 1). There were small differences in mean age in the range of 3–4 months between subtypes I and IV/V. Children in subtypes involving maxillary anterior disease (i.e., III and V), were more likely to have a history of nighttime bottle feeding. Those with the greatest caries burden (i.e., type V) were the most likely to live in a household with sub-optimally fluoridated water (67% versus 48% among type I) and to report an episodic problem-driven pattern of dental care-seeking (18% versus 2% among type I). Reports of history of tooth pain increased from type I (4%) to types IV (37%) and V (38%).

Table 1

Differences in demographics, caries risk factors, and guardian-reported child dental pain between the 5 identified ECC subtypes in the ZOE 2.0 study (N = 6,404).
	ECC subtype, n (col %)
	I	II	III	IV	V	χ², p¹
Race/ethnicity						< 0.005
African American (non-Hispanic black)	1,996 (50)	477 (46)	338 (46)	129 (38)	111 (38)
Hispanic	695 (17)	233 (22)	182 (25)	100 (29)	81 (28)
Non-Hispanic white	737 (18)	176 (17)	105 (14)	66 (19)	54 (19)
American Indian/Alaskan Native	68 (2)	23 (2)	26 (4)	12 (4)	19 (7)
more than 1 race	432 (11)	117 (11)	65 (9)	26 (8)	20 (7)
other	71 (2)	17 (2)	12 (2)	7 (2)	5 (2)
Age, months (SD)	53 (7)	56 (7)	53 (7)	57 (7)	56 (7)	< 0.005
Sex, girl n (%)	2,039 (51)	519 (50)	361 (50)	177 (52)	122 (42)	0.05
Guardian’s educational attainment						< 0.005
less than high school	620 (16)	224 (22)	173 (24)	90 (27)	89 (32)
high school or general educational development	1,465 (37)	369 (36)	292 (41)	125 (38)	106 (38)
some technical or community college or more	1,824 (47)	421 (42)	243 (34)	116 (35)	85 (30)
Lives in an optimally fluoridated area						< 0.005
yes	2,064 (52)	477 (47)	348 (49)	142 (43)	94 (33)
no	1,885 (48)	541 (53)	367 (51)	191 (57)	188 (67)
Brushes teeth at least twice a day						0.525
yes	2,491 (62)	632 (61)	448 (62)	216 (64)	192 (66)
no	1,496 (38)	406 (39)	275 (38)	122 (36)	98 (34)
Consumes 2 or more between-meal sugar-containing snacks and beverages daily						< 0.005
yes	2,773 (70)	763 (74)	551 (76)	262 (77)	222 (78)
no	1,216 (30)	273 (26)	170 (24)	77 (23)	64 (22)
Has even been put to bed with a bottle containing anything other than water						< 0.005
yes	876 (22)	237 (23)	251 (35)	101 (30)	107 (38)
no	3,047 (78)	781 (77)	461 (65)	235 (70)	175 (62)
Reason for visiting dentist						< 0.005
for check-ups and/or problems	3,465 (98)	883 (93)	588 (89)	285 (86)	236 (82)
only when problems	72 (2)	69 (7)	75 (11)	47 (14)	52 (18)
History of dental pain (not due to teething)						< 0.005
yes	165 (4)	165 (16)	193 (27)	124 (37)	109 (38)
no/do not know	3,794 (96)	864 (84)	531 (73)	212 (63)	179 (62)
¹The critical p value was 0.005 after applying a Bonferroni multiple testing statistical significance correction

When this all-population-inclusive 5-group classification was applied to 5-year-old SKaPa participants with 7-year prospective data, we found that it predicted permanent dentition caries experience. Specifically, classes II and III had approximately double mean caries experience at ages 12–13, class IV triple, and class V almost quadruple compared to the referent “healthy/mild disease” class I (Table 2). Additionally, these classes were robust to the change of diagnostic criteria and the inclusion of early-stage caries lesions (i.e., ICDAS ≥ 1, Fig. 1, bottom panel), with the only difference being the increase of apparent caries burden (i.e., dmfs) in almost all classes, due to the enumeration of early-stage caries lesions in estimating disease experience. A detailed breakdown of caries clusters (i.e., groups of caries-affected surfaces) by subtype is illustrated in Supplemental Figure S1.

Table 2

Permanent dentition caries experience (the number of decayed, restored and missing permanent tooth surfaces due to caries, DMFS index) among Swedish children in the SKaPa cohort at ages 12–13 according to ECC clinical subtype defined at the moderate/severe caries lesion detection threshold in the entire population when they were age 5.
	ECC subtype considering moderate/severe lesions¹ among all participants, n (column %)
	I	II	III	IV	V
SKaPa, n (row %)	36,317 (79)	4,697 (10)	2,143 (5)	2,246 (5)	585 (1)
Age, years (SD)	12.7 (0.5)	12.7 (0.5)	12.7 (0.5)	12.7 (0.5)	12.7 (0.4)
Sex, girl n (%)	18,161 (50)	2,306 (49)	995 (46)	963 (43)	229 (39)
Permanent dentition caries experience¹
Number of affected tooth surfaces, mean DMFS² (SD)	1.47 (2.27)	3.05 (3.62)	2.44 (3.23)	4.24 (4.72)	5.41 (6.39)
Mean ratio³ and 95% confidence interval⁴	reference	2.1 (2.0, 2.2)	1.7 (1.6, 1.8)	2.9 (2.7, 3.0)	3.7 (3.3, 4.0)
¹defined at the ICDAS ≥ 3 caries lesion detection threshold; ²DMFS, the number of decayed, missing due to caries, or filled (restored) permanent tooth surfaces; ³all pairwise differences in means are statistically significant at a conventional alpha of 0.05 after multiple testing correction based on Games-Howell test; ⁴obtained using Fieller’s theorem

The 5 ECC subtypes were generally consistent in the 3 independent cohorts, with NHANES more closely resembling the ZOE 2.0-derived solution (Fig. 2). Some variation was evident in the less common subtypes and where sample sizes were small (types IV-V among the Australian sample), but even in these instances, classes were differentiated by the patterns of disease among anterior and posterior teeth or pits-and-fissures versus smooth tooth surfaces. Most children in subtypes IV and V had received restorative dental treatment (Supplemental Table S1). The identified clinical subtypes were similar (i.e., differentiated by patterns of anterior versus posterior disease and increasing severity) when considering individuals without any evidence of prior dental treatment (Supplemental Figure S2).

Concordance of the identified clinical subtypes was higher among closely related children (Supplemental Table S2) indicating that these capture a partially heritable underlying disease. The highest concordance among monozygotic (MZ) twins was found for the ICDAS ≥ 1 subtypes (κ = 0.45; 95% CI = 0.24–0.65), which was substantially higher compared to 1st degree (κ = 0.18) and 2nd /3rd degree relatives (κ = 0.04). As expected, random allocation to matched-size groups yielded no familial concordances. On the other hand, when we compared the new clinical subtypes classification against a simulated 5-group classification where participants were assigned according to rank-ordered caries severity (e.g., highest dmfs are assigned in class V and lowest in class I) we obtained similar estimates, indicating that the severity of caries experience (that is emergent and not by design in the ECC clinical subtypes) is likely underlying the observed heritability.

Participants in classes II, III and IV were more similar to each other as determined by lower β microbiome diversity compared to those in type I (Fig. 3), whereas the effective sample size of class V participants with microbiome information was prohibitively small. Relative abundances of S. mutans, L. wadei, and S. sputigena but not P. salivae increased across clinical subtypes compared to their health type I counterparts (Fig. 3). Finally, we found that small sets of easily-examinable tooth surfaces (i.e., maxillary anteriors and lower occlusal) could reproduce the ECC classification from a full examination with reasonable accuracy (Supplemental Figure S3). Except for Class IV for which AUC was high but sensitivity was low, information from 10 surfaces could classify children in ECC subtypes with > 0.72 sensitivity and > 0.88 specificity (validation data, Supplemental Table S3).

This study provides evidence for the existence of distinct clinical presentations of ECC that are evident in different populations and settings, robust to change in diagnostic criteria, and are predictive of caries experience. These clinical subtypes emerge in an unsupervised manner from diverse, population-based clinical data of preschool-age children, yet they are clinically recognizable. They are characterized by different disease patterns involving homogeneous clusters of tooth surfaces, and a naturally-emergent disease severity gradient. Key modifiable caries risk factors such as high frequency of sugar-containing snacks and beverages, episodic dental care, inadequate fluoride exposure, as well as known socioeconomic risk indicators were associated with the identified clinical subtypes. It is likely that these subtypes reflect different disease etiologies, susceptibilities, and perhaps distinct biological profiles and clinical trajectories. Future studies must investigate their utility in clinical, public health, and research settings.

Other oral disease subtypes have been developed recently. The 2017 World Workshop on the classification of periodontal and peri-implant diseases and conditions introduced a multi-dimensional staging and grading system²². This classification entails four stages of periodontal disease, with distinct clinical presentations, although it does not contain a ‘healthy’ category and it was based on expert opinion. In this study, we used an empirical, data-driven approach to identify subtypes that include one healthy or ‘mild disease’ category and four frank, pattern-like disease categories. While the currently accepted taxonomy of ECC is an all-encompassing case definition⁷, it is reasonable to consider the potential benefits of developing a formal ECC subtype taxonomy^8,9. There is a need to improve both risk assessment²³ and screening⁶ for ECC, and a formal disease taxonomy may help with this. We show that the 5 clinical subtypes can be identified by using caries experience information from 10 tooth surfaces that are relatively easy to access and examine—this finding must be validated in independent populations. ECC may be a collection of distinct diseases with overlapping clinical manifestations, a feature that could be leveraged in disease management and prevention. Although consensus preventive recommendations are available, it is possible that some population groups may benefit more from different interventions, for example, early counseling on feeding and dietary patterns for some and sealants or topical fluoride application for others—factors that are implicated with the presence, severity, and progression of dental caries in the primary dentition^24–26.

A clinical taxonomy of ECC subtypes would supplement rather than replace the consensus definition of ECC, which is useful and serves a purpose. However, disparate, heterogeneous clinical presentations are currently encompassed in the ECC case definition, and we believe that the expansion of taxonomy can have benefits for clinical documentation and surveillance and may offer more informative endpoints for biological investigations and clinical trials. Of note, while children’s mean ages differed by 3–4 months between the healthiest and the most diseased groups, we believe that the identified subtypes are largely stable among 3-5-year-olds—a 3-month period is arguably insufficient for progression from one subtype to another, however, it is understood that all children depart from clinical health and follow different longitudinal trajectories into the identified subtypes.

The frank differentiation of caries experience between symmetric clusters of tooth surfaces aligns with recent reports of biologically informed dental caries traits in the permanent dentition^27,28. In the primary dentition, at least 10 proposals of ‘caries patterns’ have been discussed in the literature¹⁰ but no effort has been undertaken to empirically identify and generalize groups of individuals with distinct disease patterns in population-based samples. Nevertheless, it is remarkable that observations about heterogeneous clinical patterns and distinct contributing etiologies in childhood dental disease put forward several decades ago appear to emerge in an unbiased manner from large-scale population-based data in this study^29,30. Similarly, we find that caries experience follows strongly symmetric patterns in the primary dentition, consistent with previous reports^31,32.

We found that this ECC subtype classification at age 5 was predictive of caries experience in permanent teeth at ages 12–13. Moreover, we recently showed that classification of ECC cases in the 5 subtypes perform well compared to a dentist-assigned caries risk score in a subset of Swedish children over a 7-year mean follow-up period³³. Importantly, these patterns appear to be robust to different caries lesion clinical detection thresholds, are relevant to the entire population of 3-5-year-olds and generalize across populations with different caries risk and access to care. Of note, in the Australian sample, subtypes IV and V were flipped in terms of prevalence and severity versus the other 3 cohorts, but this finding was based on a relatively very small number of 117 combined cases for both subtypes.

An important consideration when interpreting the identified ECC subtypes and their variation between the US, Swedish and Australian data is the impact of restorative dental care on what is being measured as caries experience. In theory, restorative dental treatment should follow the pattern of underlying caries lesions, however, it is known that full-coverage restorations (i.e., crowns) can inflate the measured caries experience. For instance, a two-surface caries lesions may receive a 5-surface restoration. This is reflected in subtypes IV and V, wherein almost all children have undergone at least some restorative dental treatment in the US and Swedish samples and their comparison with their unrestored counterparts. However, the caries lesion patterns remain the same.

Based on findings from heritability, familial concordance, and microbiome analyses, we support that the identified ECC clinical subtypes appear to contain biological information. The higher concordances among MZ compared to FS, and FS compared to HS/FC are reflective of biological underpinnings. ECC subtypes showed similar familial concordance with a simulated scenario where participants were ordered and grouped according to caries burden, suggesting that disease severity is a major driver of heritable variance. While more detailed investigations must be carried out in large sample sizes, this study’s microbiome analyses also support that some biological information is contained in ECC clinical subtypes—some were found to be less dissimilar than the health-associated class, and higher abundance of key species was found among the most severe ones.

The study’s findings are novel mainly due to the population-based nature of participating cohorts and the unbiased, data-driven analytical approach made possible from the large sample sizes. Still, our results must be viewed acknowledging some limitations. First, the study was limited to children ages 36–71 months at baseline. While the upper limit is by design the de facto upper age cut-off for the current definition of ECC, children may show early signs of dental disease much earlier than 36 months, soon after the eruption of the first primary teeth. The study is therefore unable to comment on caries patterns in 0-2-year-old children, which will have different clinical manifestation. Future studies that include this demographic may be able to determine subtypes of ECC that emerge in young ages and the trajectories that precede and follow the 3-5-year-old age group. In our studies, age 36 months was the lowest age limit, considering that children may not have a complete primary dentition before that age. Secondly, our study did not include radiographic examinations, and this may have led to underestimation of early-stage disease in proximal surfaces in cases with closed interdental contacts. We are of the view that this is not a major limitation because most of our analyses involved established/severe lesions (which would be detectable visually, even without radiographs), visual examinations in this demographic have been shown to yield high quality data³⁴ and the fact that posterior dental contacts may not close until later in life. Thirdly, it can be argued that the optimal number of ECC subtypes (i.e., latent classes) may be smaller or larger than our identified 5. While from a model fit and clinical interpretation standpoints, we prioritized a 5-class model as the optimal and replicated it existence in 3 external cohorts, however, more finely differentiated classes could be derived in large enough samples—at the extreme end, each person may be his or her own clinical case type; nevertheless, we believe that the presented classification is reasonable balance between parsimony and precision. Finally, the sample sizes for familial concordance and microbiome analyses were much smaller than that available for the main analysis of the 4 cohorts and thus these results should be interpreted with caution. In conclusion, we present evidence from four large population-based cohorts in support of generalizable clinical subtypes of early childhood caries. The stark variation in these clinical presentations likely reflects differences in underlying etiology. Implications for prevention, screening, and optimal management of these disease subtypes should be systematically investigated in future studies.

Study Design, Setting and Participants

We conducted a cross-sectional, multi-cohort study of dental caries experience among children ages 3-5-years-old from 4 community-based cohorts in the US, Sweden, and Australia (Supplemental Table S4). A well-characterized sample of 6,404 preschool-age enrolled in the ZOE 2.0 study (2016–2019) in North Carolina, US, served as the ‘discovery’ and main analysis cohort³⁵. Participating children attended public preschools (Head Start) in a state-wide sample in NC. Information on caries experience for all 88 primary tooth surfaces was collected by trained and calibrated clinical examiners in community locations using modified International Caries Detection and Assessment System (ICDAS) criteria³⁶. Using the established caries lesion detection threshold (ICDAS ≥ 3), 54% of participating children had ECC. A detailed description of the study protocols has been previously reported ^37–39.

The first replication cohort was the US-based, National Health and Nutrition Examination Survey (NHANES) oral health component⁴⁰. We combined 7 cycles of the NHANES including years 1999–2004 and 2011–2018, comprising a total of 3,958 3-5-year-old children, 33% of whom had ECC. The second replication cohort comprised 208,112 Swedish 3-5-year-old children (15% with ECC) participating in the Swedish Quality Registry for Caries and Periodontal Diseases (SKaPa, http://www.skapareg.se/). SKaPa included data from the Swedish dental public health system extracted from electronic health records of public and private dental clinics beginning in 2008⁴¹. The Australian cohort comprised 7,997 3-5-year-old preschool-age children who participated in a community-based outreach program “Wide Smiles” between 2013-19 in southwest Victoria. The program included on-site oral health screenings at preschool centers conducted by the local public dental service. Trained and calibrated clinical examiners documented dental caries experience at the tooth surface-level using International Caries Detection and Assessment System (ICDAS) criteria as previously reported⁴².

While ZOE 2.0 represents a low-income, high caries risk population (i.e., families are Medicaid-eligible and must meet additional socioeconomic vulnerability criteria to qualify for Head Start participation), NHANES served as the U.S. community-dwelling representative sample (although no survey weights were used prior to carrying out LCA). The Swedish population data served as another independent sample from a generally low caries-risk and high access-to-care population. The Australian data served as the fourth independent sample from an area of moderate caries risk and low access to care.

This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cross-sectional studies. Ethics approvals were obtained in all participating studies.

Data and Variables

Tooth surface-level information on caries experience for all 88 primary tooth surfaces (each coded as a binary variable) was available in all participating cohorts. ZOE 2.0, SKaPa and Wide Smiles recorded caries lesions using two different thresholds, i.e., one corresponding to cavitation (i.e., ICDAS ≥ 3, “established” or “advanced” caries lesions) and one representing the first visual changes or “white spots” in tooth enamel (i.e., ICDAS ≥ 1, all caries lesions including “early stage” ones). The clinical protocol of NHANES did not employ ICDAS criteria; however, it recorded caries lesions using a definition resembling ICDAS ≥ 3. Additional information is available for all cohorts included participants’ age (measured in in months), sex (boy/girl), and evidence of restorative or surgical dental procedures (i.e., fillings, crowns, extractions) (Supplemental Table S4).

Additional demographic and oral health-related information was collected in ZOE 2.0 via questionnaires administered to children’s guardians in the language of their preference (English or Spanish)⁴³. For the purposes of this study, we considered information on self-reported race/ethnicity (categorized as African American/non-Hispanic Black, Hispanic, non-Hispanic white, American Indian/Alaskan Native, more than 1 race, and other); guardian’s educational attainment (categorized as less than high school, high school or general educational development, and some technical or community college or more); tooth brushing frequency (twice a day or more versus less than twice a day); frequency of consuming between-meal sugar-containing snacks and beverages (categorized as 2 or more versus less than 2 per day); history of child having been put to bed with a bottle containing anything other than water (yes versus no); reason for visiting the dentist among those with a previous dental visit (for check-ups and/or problems versus only when problems arise); and history of the child having dental pain not from teething (yes versus no). Information about domestic water source fluoride content was measured directly in a subset of participants (n = 1,530) and was subsequently imputed to the remainder of the sample as optimal (i.e., ≥ 0.60ppm F) versus sub-optimal (i.e., < 0.60ppm F) using a machine learning-based imputation method leveraging geographic information systems⁴⁴.

Relatedness in ZOE 2.0 was determined from genome-wide genotyping data available for 96% (n = 6,144) of participants. Genotyping was performed using DNA from saliva samples at the Center for Inherited Disease Research (CIDR, Johns Hopkins University) using the Infinium Global Diversity Array-8 v1.0, offering 1,905,000 genotyped single nucleotide polymorphisms (SNPs). For this application, 183,287 autosomal SNPs were used to estimate identity-by-descent (IBD) probabilities using the KING-robust procedure⁴⁵. This subset of “high-quality” SNPs was selected by linkage disequilibrium (LD) pruning from an initial pool consisting of all autosomal SNPs with missing call rate < 2% and minor allele frequency > 5%, with all pairs of SNPs having r² < 0.1 in a sliding 10 Mb window. Subsequently, relatedness was estimated using the GENESIS package⁴⁶. Relationships between participant pairs were determined based on clustering of kinship coefficients around expected values, i.e., monozygotic twins (MZ): 0.50, full siblings (FS): 0.25, half siblings (HS): 0.125, and first cousins (FC): 0.0625.

Microbiome data in ZOE 2.0 were available for a subset of participants (n = 300). Specifically, supragingival biofilm samples from the first 150 ECC cases (using the ICDAS ≥ 3 threshold) and 150 non-cases using the same threshold was carried forward to whole genome shotgun sequencing (WGS, metagenomics-MTG) and RNA-seq (metatranscriptomics-MTX)³⁸. The resulting paired-end reads were trimmed of adapter sequences using Trim Galore (Babraham Institute, Cambridge, UK) and classified with Kraken2⁴⁷ and Bracken 2.5⁴⁸ using a custom database including human, fungal, bacterial, and the expanded Human Oral Microbiome Database (eHOMD) genomes⁴⁹ to produce an initial taxonomic composition profile. After eliminating reads identified as ‘host’, paired-end reads were joined with vsearch 1.10.2.⁵⁰ and again trimmed of adapter sequences using Trim Galore. We used microbiome MTG data to examine differences in measures of community diversity (i.e., Bray-Curtis estimates of beta diversity, a measure of similarity or dissimilarity) between clinical subtypes of ECC. Additionally, we examined differences in the relative abundance of 4 species, namely Streptococcus mutans, Selenomonas sputigena, Leptotrichia wadei, and Prevotella salivae, that our group recently identified as strongly associated with ECC experience⁵¹.

Latent Class Analysis

Discovery latent class analyses (LCA) included children with frank caries experience (i.e., only cases, defined at the established/severe caries lesion or cavitation ICDAS ≥ 3 threshold) in the ZOE 2.0 study (n = 3,465). After identifying an optimal LCA solution in that sample, we carried out two within-study sensitivity analyses and subsequently sought to replicate or generalize clinical subtypes in the 3 independent cohorts of similarly aged children. The first sensitivity analysis consisted of caries-free participants (i.e., the entire study population, n = 6,404) in the ZOE 2.0 LCA analysis to determine whether the solution was robust to the inclusion of non-cases. The second sensitivity analysis consisted of changing the diagnostic threshold so that all caries lesions (i.e., including early-stage, non-cavitated lesions) were included in the case-only analysis (n = 5,882), to determine whether the solution was robust to a different caries lesion diagnostic threshold.

LCA models were fitted using an expectation-maximization algorithm (EM) and robust maximum likelihood (MLR) considering each of the 88 primary tooth surfaces as a binary latent class indicator in a stepwise manner as in a previously reported application^43,52. Models were identified by sequentially increasing the number of classes (k) beginning with k = 1 until model non-identification was concluded. To identify the model with the optimal number of classes (i.e., ECC clinical subtypes), we assessed and compared absolute and relative measures of model fit of the two models with the closest fit (i.e., competing models). Measures of classification accuracy included overall entropy (i.e., a measure for each model) and average posterior class probabilities (i.e., a measure for each class). Additionally, we obtained frequencies and percentages of study participants assigned to each class or subtype (i.e., class proportions or ɣ-probabilities) and class-item-specific response probabilities (i.e., IRP, surface-specific caries prevalence).

Description and correlates of ECC clinical subtypes

We used three methods to describe differences in the clinical presentation of ECC between its clinical subtypes. First, we compared measures of caries experience in the different subtypes (e.g., number of caries-affected tooth surfaces). Next, to visualize differences in the severity and distribution of caries experience, we used a custom-build visualization pipeline (SculptorHD)⁵³ to produce annotated 3D representations of the primary dentition that are accessible via https://eccsubtypesdemo.web.app/ (demo site provided for peer-review purposes). Finally, because measures of total caries experience can mask differences in the pattern of disease distribution, we used spatially-informed measures of caries severity (termed ‘caries clusters’). These were derived using a hierarchical clustering analysis (HCA) of the 88 tooth surfaces using Ward’s minimum variance method⁵⁴. Upon inspection of the resulting dendrogram and clinical interpretability, we identified 7 right-left symmetric clusters of tooth surfaces, a grouping that parallels previous reports in the literature (e.g., separating pits-and-fissures versus smooth surfaces, as well as maxillary versus mandibular teeth) (Supplemental Figure S4).

Subsequently, ECC clinical subtype memberships were used to investigate differences in terms of demographics (e.g., age, race/ethnicity), reported oral health-related behaviors (e.g., history of nighttime bottle feeding, history of tooth pain), and fluoride exposure (i.e., optimal versus sub-optimal). Comparisons between categorical variables were done using chi-square tests and comparisons involving quantitative measures (i.e., dmfs or age) were done with ANOVA including a Bonferroni multiple testing statistical significance correction.

Dental caries experience information at age 12–13 was available for 60% (45,998 out of 77,274) of SKaPa participants with ECC clinical subtype information at age 5. At ages 12–13, one would expect all primary teeth have exfoliated and all permanent teeth except 3rd molars (i.e., wisdom teeth) have emerged. This prospective clinical information was summarized by the number of caries-affected permanent tooth surfaces (i.e., the DMFS index) at ages 12–13 and was then compared between groups of the 5-class ECC classification that included all children, i.e., wherein class I includes caries-free and mild cases. For this purpose, we estimated mean ratios and 95% confidence intervals obtained based on Fieller’s theorem⁵⁵, then carried out multiple testing-adjusted pairwise comparisons using the Games-Howell test⁵⁶.

Familial concordance and microbiome differences

Estimates of ECC subtype concordance among related individuals in ZOE 2.0 were obtained via Cohen’s kappa (κ) with corresponding 95% confidence intervals and were compared between monozygotic twins (MZ), full siblings/1st degree relatives (FS), and half siblings/2nd degree relatives (HS) and first cousins/3rd degree relatives (FC). We compared concordances of the ECC subtypes estimates with two simulated control conditions of matched-size participant groups —a ‘positive’ control, wherein ZOE 2.0 participants were assigned to a similar number of groups according to their ordered disease severity (i.e., caries burden, as measured by the dmfs index) and a ‘negative’ control, where ZOE 2.0 participants were randomly assigned. We carried out these tests to determine to what degree ECC subtypes are associated with genetic variation. For instance, we expected concordance estimates to be larger among MZ twins compared to FS and those larger compared to HS/FC. Additionally, we hypothesized that concordance estimates would be larger for ECC clinical subtypes (i.e., along more than 1 axis of variation) compared to what could obtained from allocation of participants in caries severity rank-ordered groups (i.e., per the dmfs index, along 1 axis of variation, the “positive” control) and compared to random participant allocation (a “negative” control).

Additional between-subtype comparisons were completed among the subset of 300 ZOE 2.0 participants with supragingival microbiome information. We hypothesized that microbiome biomarkers would be similar within each clinical disease subtype but would diverge between different subtypes.

Assessment of the feasibility of ECC subtype screening using index teeth.

Given the tendency for clusters of teeth to share similar caries status, we hypothesized that examination of a small number of tooth surfaces might be sufficient to assign children to an ECC subtype. To determine whether information from examining small numbers of index teeth could help accurately categorize children in ECC subtypes, we employed machine learning using 3 sets of easily accessible tooth surfaces as inputs. The first set comprised 10 surfaces—the facial surfaces of the upper anterior 6 maxillary teeth (primary canines and incisors) and the occlusal surfaces of the lower 4 primary molars. The second set included 16 surfaces (i.e., the first set, plus the 6 proximal surfaces between the 4 maxillary primary incisors) and the third set included 20 surfaces (i.e., the second set, plus the 4 surfaces in the interproximal areas between maxillary lateral incisors (i.e., distal surfaces) and maxillary canines (i.e., mesial surfaces)]. We screened 8 candidate models (i.e., boosted neural network, generalized regression Lasso, nominal logistic, bootstrap forest, decision tree, support vector machine, k-nearest neighbors, and linear discriminant analysis), and selected the best performing model based on R², mean root average square error (RASE), and mean area under the receiver operator curve (ROC AUC) upon 50-fold validation using a random 60% of the data for model training, 15% for model validation, and 25% for model testing. For the selected model, we reported AUC values of a neural network (including 5-fold cross-validation and a total 100 boosted models with learning rate of 0.1) correctly identifying each ECC clinical subtype in training (80%) and validation (20%) data in the ZOE 2.0 study.

Software

For LCA, we used Mplus v.8.8 (Muthén & Muthén, Los Angeles, CA, US) and the poLCA package (v.1.4.1) in R v.4.1.2 (The R Foundation for Statistical Computing, Vienna, Austria). HCA and machine learning applications were implemented with JMP^® Pro 17.0 (SAS Institute Inc., Cary, NC, US). Stata v.17.0 (Stata Corp LLC, College Station, TX, US) was used for all additional analyses.

Funding: Miguel Simancas-Pallares, Poojan Shrestha, Hunyong Cho, Yu Gu, Jeannie Ginnis, Di Wu, Kari North, and Kimon Divaris were supported by NIH/NIDCR U01DE025046. Hunyong Cho, Di Wu, and Kimon Divaris were supported by NIH/NIDCR R03DE02898. Alexander Gormley and Simon Haworth were supported by the National Institute for Health and Social Care Research (NIHR) through the academic clinical fellowship and academic clinical lecturer schemes. David Burgner was supported by an NHMRC Investigator Grant (GTN1175744). Mihiri Silva was supported by a Melbourne Children’s Clinician-Scientist Fellowship. Ingegerd Johansson and Pernilla Lif Holgerson received support by the Swedish Patent Revenue Fund for Research in Preventive Odontology, TUA, the Västerbotten County council funding for dental research.

Acknowledgments

The authors thank the Swedish Quality Registry for Caries and Periodontal Diseases (SKaPa) for providing access to the Swedish registry data, and studies’ participants and their families for their contributions to the advancement of knowledge.

Author Contributions:

MSP: data acquisition, curation, analysis and interpretation, manuscript draft; AG: data curation, analysis, manuscript revision; PS: data acquisition, analysis, curation, manuscript revision; YG: data analysis and manuscript revision; HC: data analysis and manuscript revision; HDS: website creation, application for 3D visualizations and manuscript revision; ZB: website creation and manuscript revision; MS: data acquisition, curation and manuscript revision; SD: data acquisition, curation and manuscript revision; DB: data acquisition, curation and manuscript revision; AFZ: study design, data acquisition, interpretation and manuscript revision; JG: study design, data acquisition, curation and manuscript revision; WFV: manuscript revision; AE: data acquisition, curation, and manuscript revision; JR: data curation, analysis, interpretation and manuscript revision; AAR: data interpretation and manuscript revision; DW: data analysis, interpretation and manuscript revision; MJS: data acquisition, curation, manuscript revision; PLH: data interpretation, manuscript revision; SH: data curation, analysis, interpretation and manuscript revision; IJ: data acquisition and manuscript revision; KEN: study design, data interpretation and manuscript revision; KD: study design, data acquisition, curation, analysis, interpretation, and manuscript draft.

Competing Interests:

The authors declare no competing interests.

Data Availability:

Clinical data of the ZOE 2.0 study are available via an institutional data repository: https://cdr.lib.unc.edu/concern/data_sets/kk91fv385 Clinical data of the NHANES are available via: https://wwwn.cdc.gov/nchs/nhanes/ Clinical data of the Swedish and Australian cohorts are not publicly available and cannot be shared internationally; investigators interested in collaborating with these cohorts are encouraged to contact the study authors. Molecular data (human genotypes and microbiome sequencing files) are available via the dbGaP repository under the umbrella "Trans-Omics for Precision Dentistry and Early Childhood Caries-TOP-DECC" and accession: phs002232.v1.p1.

Fleming, E. & Afful, J. Prevalence of total and untreated dental caries among youth: United States, 2015–2016. NCHS Data Brief 307, 1–8 (2018).
Pitts, N.B. et al. Early childhood caries: IAPD Bangkok declaration. J Dent Child (Chic) 86, 72 (2019).
Satcher, D. & Nottingham, J.H. Revisiting oral health in america: a report of the surgeon general. Am J Public Health. 107, S32-S33 (2017).
Casamassimo, P.S., Thikkurissy. S., Edelstein, B.L., Maiorini, E. Beyond the dmft: the human and economic cost of early childhood caries. J Am Dent Assoc. 140, 650–657 (2009).
Garcia, R. et al. Progress in early childhood caries and opportunities in research, policy, and clinical management. Pediatr Dent. 37, 294–299 (2015).
Clark, M.B. & Braun, P.A. Promotion of oral health and prevention of dental caries among children in primary care. JAMA. 326, 2139–2140 (2021).
Pitts, N.B., et al. Dental caries. Nat Rev Dis Primers. 3, 17030 (2017).
Divaris, K. Predicting dental caries outcomes in children: a "risky" concept. J Dent Res. 95, 248–254 (2016).
Crystal, Y.O., Keels, M.A., Silva, M.J., Divaris, K. Clinical need and scientific evidence base for a precise early childhood caries classification. Int J Paediatr Dent. 32, 773–775 (2022).
Psoter, W.J., Morse, D.E., Pendrys, D.G., Zhang, H., Mayne, S.T. Historical evolution of primary dentition caries pattern definitions. Pediatr Dent. 26, 508–511 (2004).
Psoter, W.J, Pendrys, D.G., Morse, D.E, Zhang, H.P, Mayne, S.T. Caries patterns in the primary dentition: cluster analysis of a sample of 5169 Arizona children 5–59 months of age. Int J Oral Sci. 1, 185–195 (2009).
Zeng, Z., et al. Genome-wide association study of primary dentition pit-and-fissure and smooth surface caries. Caries Res. 48, 330–338 (2014).
National Research Council. Toward precision medicine: building a knowledge network for biomedical research and a new taxonomy of disease. (2011). at < http://www.nap.edu/catalog/13284/toward-precision-medicine-building-a-knowledge-network-for-biomedical-research>
Collins, F.S. & Varmus, H. A new initiative on precision medicine. N Engl J Med. 372, 793–795 (2015).
Field, A.E., Camargo, C.A. Jr, Ogino, S. The merits of subtyping obesity: one size does not fit all. JAMA. 310, 2147–2148 (2013).
McCarthy, M.I. Genomics, type 2 diabetes, and obesity. N Engl J Med. 363, 2339–2350 (2010).
Deliu, M., et al. Features of asthma which provide meaningful insights for understanding the disease heterogeneity. Clin Exp Allergy. 48, 39–47 (2018).
Ross, M.K., Eckel, S,P., Bui, A.A.T., Gilliland, F.D. Asthma clustering methods: a literature-informed application to the children's health study data. J Asthma. 59, 1305–1318 (2022).
Espay, A.J., Brundin, P., Lang, A.E. Precision medicine for disease modification in Parkinson disease. Nat Rev Neurol. 13, 119–126 (2017).
Matsushita, F.Y., Krebs, V.L.J., de Carvalho, W.B. Identifying clinical phenotypes in extremely low birth weight infants-an unsupervised machine learning approach. Eur J Pediatr. 181, 1085–1097 (2022).
Dahl, A., Zaitlen, N. Genetic influences on disease subtypes. Annu Rev Genomics Hum Genet. 21, 413–435 (2020).
Papapanou, P.N., et al. Periodontitis: consensus report of workgroup 2 of the 2017 World Workshop on the Classification of Periodontal and Peri-Implant Diseases and Conditions. J Periodontol. 89, S173-S182 (2018).
Halasa-Rappel, Y.A., Ng, M.W., Gaumer, G., Banks, D.A. How useful are current caries risk assessment tools in informing the oral health care decision-making process?. J Am Dent Assoc. 150, 91–102 (2019).
Slade, G.D., Grider, W.B., Maa,s W.R., Sanders, A.E. Water fluoridation and dental caries in U.S. children and adolescents. J Dent Res. 97, 1122–1128 (2018).
Feldens, C.A., Rodrigues, P.H., de Anastácio, G., Vítolo, M.R, Chaffee, B.W. Feeding frequency in infancy and dental caries in childhood: a prospective cohort study. Int Dent J. 68, 113–121 (2018).
Lee, H.H., Faundez, L., LoSasso, A.T. A cross-sectional analysis of community water fluoridation and prevalence of pediatric dental surgery among medicaid enrollees. JAMA Netw Open. 3, e205882 (2020).
Shaffer, J.R., et al. Clustering tooth surfaces into biologically informative caries outcomes. J Dent Res. 92, 32–37 (2013).
Haworth, S., et al. Heritability of caries scores, trajectories, and disease subtypes. J Dent Res. 99, 264–270 (2020).
Johnsen, D.C. Dental caries patterns in preschool children. Dent Clin North Am. 28, 3–20 (1984).
Johnsen, D.C., Schultz, D.W., Schubot, D.B., Easley, M.W. Caries patterns in Head Start children in a fluoridated community. J Public Health Dent. 44, 61–66 (1984).
Vanobbergen, J., Lesaffre, E., García-Zattera, M.J., Jara, A., Martens, L., Declerck, D. Caries patterns in primary dentition in 3-, 5- and 7-year-old children: spatial correlation and preventive consequences. Caries Res. 41, 16–25 (2007).
Zhang, Y., Todem, D., Kim, K., Lesaffre, E. Bayesian latent variable models for spatially correlated tooth-level binary data in caries research. Stat Modelling. 11, 25–47 (2011).
Gormley, A., et al. Subtypes of early childhood caries predict future caries experience. Community Dent Oral Epidemiol. 2022;10.1111/cdoe.12795.
Pontes, L.R.A., et al. Impact of visual inspection and radiographs for caries detection in children through a 2-year randomized clinical trial: The Caries Detection in Children-1 study. J Am Dent Assoc. 151, 407–415 (2020).
Divaris, K., et al. Cohort profile: ZOE 2.0-a community-based genetic epidemiologic study of early childhood oral health. Int J Environ Res Public Health. 17, 8056 (2020).
Young, D.A., et al. The American Dental Association caries classification system for clinical practice: a report of the American Dental Association council on scientific affairs. J Am Dent Assoc. 146, 79–86 (2015).
Ginnis, J., et al. Measurement of early childhood oral health for research purposes: dental caries experience and developmental defects of the enamel in the primary dentition. Methods Mol Biol. 1922, 511–523 (2019).
Divaris, K., et al. The supragingival biofilm in early childhood caries: clinical and laboratory protocols and bioinformatics pipelines supporting metagenomics, metatranscriptomics, and metabolomics studies of the oral microbiome. Methods Mol Biol. 1922, 525–548 (2019).
Agler, C.S., et al. Protocols, methods, and tools for genome-wide association studies (GWAS) of dental traits. Methods Mol Biol. 1922, 493–509 (2019).
Dye, B.A., Afful, J., Thornton-Evans, G., Iafolla, T. Overview and quality assurance for the oral health component of the National Health and Nutrition Examination Survey (NHANES), 2011–2014. BMC Oral Health. 19, 95 (2019).
Mensah, T., Tranæus, S., Cederlund, A., Naimi-Akbar, A., Klingberg, G. Swedish quality registry for caries and periodontal diseases (SKaPa): validation of data on dental caries in 6- and 12-year-old children. BMC Oral Health. 21, 373 (2021).
Mason, A., et al. A preventative approach to oral health for children in a regional/rural community in South-West Victoria, Australia. Dentistry. 6, 313–319 (2015).
Simancas-Pallares, M.A., et al. Children's oral health-related behaviours and early childhood caries: a latent class analysis. Community Dent Oral Epidemiol. 50, 147–155 (2022).
Gu, Y., et al. Partitioning around medoids clustering and random forest classification for GIS-informed imputation of fluoride concentration data. Ann Appl Stat. 16, 551–572 (2022).
Manichaikul, A., Mychaleckyj, J.C., Rich, S.S., Daly, K., Sale, M., Chen, W.M. Robust relationship inference in genome-wide association studies. Bioinformatics. 26, 2867–2873 (2010).
Conomos, M.P., Thornton, T., Gogarten, S.M. GENetic EStimation and inference in structured samples (GENESIS): statistical methods for analyzing genetic data from samples with population structure and/or relatedness. R package. 2, 1–58 (2016).
Wood, D.E., Lu, J., Langmead, B. Improved metagenomic analysis with Kraken 2. Genome Biol. 20, 257 (2019).
Lu, J., et al. Bracken: estimating species abundance in metagenomics data. PeerJ Computer Sci. 3, e104 (2017).
Dewhirst, F.E., et al. The human oral microbiome. J Bacteriol. 192, 5002–5017 (2010).
Rognes, T., Flouri, T., Nichols, B., Quince, C., Mahé, F. VSEARCH: a versatile open source tool for metagenomics. PeerJ. 4, e2584 (2016).
Alaluusua, S., & Renkonen, O.V. Streptococcus mutans establishment and dental caries experience in children from 2 to 4 years old. Scand J Dent Res. 6, 453–457 (1983).
Masyn, K.E. Latent Class Analysis and Finite Mixture Modeling Ch. 25 (Oxford Univ. Press, New York, 2013).
Spangler, H.D., et al. A web-based rendering application for communicating dental conditions. Healthcare (Basel). 9, 960 (2021).
Ward, Jr, J.H. Hierarchical grouping to optimize an objective function. J Am Stat Assoc. 58, 236–244 (1963).;
Fieller, E.C. The biological standardization of insulin. Supplement to the Journal of the Royal Statistical Society. 7, 1–64 (1940).
Games, P.A. & Howell, J.F. Pairwise multiple comparison procedures with unequal n’s and/or variances: a Monte Carlo study. J Educ Stat. 1, 113–125.

There is NO Competing Interest.

ECCClinicalLCANatCommsSuppl230218.docx
NCOMMS2305934RS.pdf
Reporting Summary

Download PDF

Version 1

posted

You are reading this latest preprint version

Evidence for clinical subtypes of early childhood caries

Status:

Version 1

Abstract

Figures

Introduction

Results

Discussion

Methods

Study Design, Setting and Participants

Data and Variables

Latent Class Analysis

Description and correlates of ECC clinical subtypes

Familial concordance and microbiome differences

Software

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1