Models to predict future permanent tooth caries incidence in children using primary teeth caries experience.

Abstract

Objectives:

This paper predicts permanent tooth caries incidence (ΔDMFS) among a cohort of African American children using the presence of any caries experience in primary teeth (dmfs) and the presence of untreated primary tooth caries (ds) in two separate models.

Methods:

This is a secondary analysis of data from a prospective study conducted at the University of Alabama at Birmingham. Two models with different clinical indications were applied for predicting ΔDMFS from age 6 to ages 7, 8, 9, 10, 11 and 12 years, respectively. The first model used dmfs and the second model used ds as predictors (both at age 6). Sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were assessed.

Results:

The first model resulted in sensitivity from 81.8% to 100%, specificity from 35.6% to 42.6, PPV from 5.8% to 38.1%, and NPV from 90.0% to 100%. In the second model, the predictive values for ΔDMFS from ages 7 to 12 were from 33.3% to 55.6%, 62.7% to 72.7%, 3.4% to 45.5%, and 80.0% to 95.9%, respectively.

Conclusions:

The proposed models for ΔDMFS prediction are easy, not time consuming and clinically-applicable, and if validated, has potential to change the current paradigm for caries risk assessment.

Introduction:

Dental caries risk assessment has been of greater interest to many clinicians and researchers in recent decades after the decline in dental caries prevalence^1,2. In developed countries in the 21^st century, lower percentages of children, including African Americans, are experiencing dental caries in the permanent dentition. For example, among 6- to 11-year-old African American children in the United States, the prevalence of dental caries experience in primary and/or permanent teeth was 44.3%% in the National Health and Nutrition Examination Survey (NHANES) 2015-2016 data³. Identification of high caries risk children is important to better allocate the scarce resources available to prevent and/or manage dental caries in children^4,5.

Previous dental caries experience has been considered one of the contributing conditions in the caries risk assessment tools by the American Academy of Pediatric Dentistry⁶, the American Dental Association⁷, and Caries Management by Risk Assessment⁸. Studies from around the world have assessed dental caries in primary teeth as a predictor of dental caries in permanent teeth using sensitivity (SN), specificity (SP), positive predictive value (PPV), and negative predictive value (NPV). For example, Zhang et al.⁹ used the presence of primary tooth dental caries experience in Chinese children at ages 6 to 7 years as a diagnostic tool to predict positive caries increment in permanent teeth at two-year follow-up and the results were: SN (88%), SP (28%), PPV (32%), and NPV (85%). Raadal et al.¹⁰ used the presence of primary tooth caries experience at age 6 years to predict the presence of at least one site with early fissure caries in permanent first molars at age 8 years among Norwegian children and the results were SN (89%), SP (39%), PPV (42%), and NPV (88%).

Li and Wang¹¹ used the presence of primary molar tooth decay among 362 Chinese children aged 3 to 5 years at baseline (prevalence was 83.3%) to predict caries in permanent teeth after 8 years (age 11 to 13 years). They reported that the predictive values were SN (94%), SP (26%), PPV (53%), and NPV (85%). The main limitation in these studies was that either SN (true positive/ (true positive+ false negative)) or SP (true negative/ (true negative+ false positive)) values were low. Thus, researchers have been testing different cutoffs and definitions of the predicting tool to increase the predictor values. For example, when Li and Wang¹¹ used the presence of primary maxillary incisor tooth decay (prevalence at baseline was 56.4%) to predict caries in permanent teeth, the predictive values were SN (61%), SP (47%), PPV (59%), and NPV (56%). Further, ter Pelkwijk¹² used dmfs≥4 vs. dmfs>4 among 268 children at age 7 years to predict any future dental caries at the age of 11 years. They reported SN to be 69% and SP to be 72%. Similarly, in Europe, Steiner et al.¹³ reported that the SN was 75% and SP was 98%, using the absence of any sound primary molars among 7- to 8-year-old Swiss children as a predictor of having ≥4 DMFS as the outcome after 4 years (when children were 11 to 12 years old). The respective values were 81% and 77% when using ≥6 DMFS as the outcome (Helfenstein et al.¹⁴). Using the same sample, when changing the definition of the predictor variable to be the presence of any lingual or buccal smooth surfaces with white spots, the respective values were 81% and 60%, when using ≥4 DMFS as an outcome (Steiner el al.¹³).

In the U.S., several studies assessed the predictive values of previous dental caries experience as a diagnostic tool for predicting permanent dentition caries during the mixed dentition stage. However, results differed substantially across studies, probably because of differences in the definitions of the outcome variables (what these studies were designed to predict), the predictors and children’s ages. For example, Disney et al.¹⁵ reported SN of 60% and SP of 83% when using any caries experience (dmfs+DMFS) among 1,099 children at age 6 years in South Carolina as a predictor of having ≥4 DMFS at age 9 years. The respective values were 80% and 61% when using ≥1 DMS as the outcome¹⁶. Similarly, Disney et al.¹⁵ reported that SN and SP were 56% and 83% among 1,086 children at age 6 years in the state of Maine when using ≥2 DMFS at age 9 years as an outcome, while in the same study Beck et al.¹⁶ reported 66% and 78%, respectively, when using ≥1 DMFS as an outcome.

The purpose of this study was to propose two models with different clinical indications to predict DMFS incidence and apply them to a sample of African American children.

Methods:

The study reported herein is a secondary analysis of data from a prospective study conducted at the University of Alabama at Birmingham. A cohort of low socioeconomic status, African-American school-aged children was recruited (n=98) from Perry County, Alabama. The inclusion criteria for the study were that children had to a) not have all first permanent molar teeth erupted, b) live with their biological mothers, c) plan to remain in the area for at least 3 years, and d) be free from systemic diseases.

The Institutional Review Board at the University of Alabama at Birmingham first approved the project in August 2006 and informed consent and waiver of assent were obtained from all parents prior to the study. Oral examinations were done at baseline (approximately age 6 years) and six annual follow-up examinations from ages ~7 to ~12 years by the same three dentists throughout the study. Oral examinations were done (cavitated decay vs. no decay) using portable equipment according to the World Health Organization (WHO) criteria¹⁷, without radiographs. All parents were encouraged to find dental homes for their children to have the required preventive and restorative treatments.

From the beginning of the study, calibration of the dental examinations of about ten children (~10%) was done annually. Both kappa and weighted kappa were used to assess the overall inter-examiner reliability (intra-examiner reliability was not assessed). There were semi-annual study visits for fluoride varnish application and oral hygiene instructions by the dental examiners, dental hygienists, and dental residents in the Department of Pediatric Dentistry at the University of Alabama at Birmingham.

Two models for predicting future caries incidence in permanent teeth (ΔDMFS) were applied: the first model used dental caries experience in primary teeth (dmfs) at age 6 as a predictor, and the second model used untreated dental caries in primary teeth (ds) at age 6 as a predictor. In the first model, children were assigned to four groups: 1) true positive (TP) where children had dmfs>0 at age 6 and ΔDMFS>0 at the follow-up examination, 2) false positive (FP) where children had dmfs>0 at age 6 and ΔDMFS=0 at the follow-up examination, 3) false negative (FN) where children had dmfs=0 at age 6 and ΔDMFS>0 at the follow-up examination and 4) true negative (TN) where children had dmfs=0 at age 6 and ΔDMFS=0 at the follow-up examination. In the second model, children were assigned to the same groups, except ds at age 6 was used as the predictor instead of dmfs.

Quantifying the predictive ability of primary tooth caries experience on future permanent tooth decay, we assessed 1) sensitivity¹⁸ (probability that dmfs>0 when ΔDMFS>0), 2) specificity¹⁸ (probability that dmfs=0 when ΔDMFS=0), 3) positive predictive value (PPV)¹⁹ (probability that ΔDMFS>0 when dmfs>0), 4) negative predictive value (NPV)¹⁹ (probability that ΔDMFS=0 when dmfs=0), and 5) percentage agreement (the probability that ΔDMFS>0 when dmfs>0 and ΔDMFS=0 when dmfs=0). All analyses were conducted using SAS 9.4 (SAS Institute Inc., Cary, NC, USA).

Results:

There were 78, 66, 66, 63, 62 and 62 children examined with both a) primary teeth at age 6 years and b) permanent teeth subsequently at ages 7, 8, 9, 10, 11 and 12 years, respectively. Calibration examinations resulted in average person-level weighted kappa scores of 0.98, 0.97 and 0.98 for dmfs count, DMFS count and dmfs plus DMFS count, respectively. For the average person-level simple kappa scores, the respective scores were 0.85, 0.90 and 0.83²⁰.

Table 1 summarizes the results of the first model predicting ΔDMFS using dmfs at age 6 years. The SN ranged from 81.8% to 100%, SP ranged from 35.6% to 42.6%, PPV ranged from 5.8% to 38.1%, NPV ranged from 90.0% to 100% and percent agreement ranged from 38.5% to 56.5%.

Table 1.

Sensitivity, specificity, positive predictive value, negative predictive value, and percentage agreement for primary tooth dental caries experience (dmfs) at age 6 as a diagnostic tool to predict dental caries incidence of permanent teeth at ages 7, 8, 9, 10, 11 and 12 years.

Age (years)	Sample size	TP	FP	FN	TN	SE	SP	PPV	NPV	Percentage agreement
7	78	3	48	0	27	100%	36.0%	5.8%	100.0%	38.5%
8	66	7	38	0	21	100%	35.6%	15.6%	100.0%	42.4%
9	66	9	34	2	21	81.8%	38.2%	20.9%	91.3%	45.5%
10	63	13	29	1	20	92.9%	40.8%	31.0%	95.2%	52.4%
11	62	15	27	0	20	100%	42.6%	35.7%	100.0%	56.5%
12	62	16	26	2	18	88.9%	40.9%	38.1%	90.0%	54.8%

TP: True positive; FP: False positive; FN: False negative; TN: True negative; SE: Sensitivity; SP: Specificity; PPV: Positive predictive value; NPV: Negative predictive value

Table 2 summarizes the results of the second model predicting ΔDMFS using ds at age 6 years. The SN ranged from 33.3% to 55.6%, SP ranged from 62.7% to 72.7%, PPV ranged from 3.4% to 45.5%, NPV ranged from 80% to 95.9% and percent agreement ranged from 60.6% to 67.7%.

Table 2.

Sensitivity, specificity, positive predictive value, negative predictive value, and percentage agreement for untreated primary tooth caries (ds) at age 6 as a diagnostic tool to predict dental caries incidence of permanent teeth at ages 7, 8, 9, 10, 11 and 12 years.

Age (years)	Sample size	TP	FP	FN	TN	SE	SP	PPV	NPV	Percentage agreement
7	78	1	28	2	47	33.3%	62.7%	3.4%	95.9%	61.5%
8	66	3	21	4	38	42.9%	64.4%	12.5%	90.5%	62.1%
9	66	4	19	7	36	36.4%	65.5%	17.4%	84.7%	60.6%
10	63	7	15	7	34	50.0%	69.4%	31.8%	82.9%	65.1%
11	62	7	15	8	32	46.7%	68.1%	31.8%	80.0%	62.9%
12	62	10	12	8	32	55.6%	72.7%	45.5%	80.0%	67.7%

TP: True positive; FP: False positive; FN False negative; TN: True negative; SE: Sensitivity; SP: Specificity; PPV: Positive predictive value; NPV: Negative predictive value.

Discussion:

In this cohort of African American children, Ghazal et al.²⁰ reported that the prevalence rates for permanent tooth dental caries experience were 3.8% 10.5%, 16.4%, 22.2%, 24.2% and 29.0% at ages, 7, 8, 9, 10, 11, 12 years, respectively. This study proposed two models to predict ΔDMFS which are easy, clinically-applicable and time-efficient. In our cohort, the first model to predict ΔDMFS using dmfs at age 6 years resulted in high SN (SN values were 100% at ages 7, 8, and 11 years; false negative cases were very unlikely). However, the SP was low (35.6% to 40.9%). Therefore, a second model to predict ΔDMFS using ds at age 6 years was applied and had substantially higher SP values (62.7% to 72.7%, compared to 35.6% to 40.9% in the first model). In the second model, the false positive cases substantially decreased. In other words, this risk assessment model would be less likely to put a child in a high caries category and result in no decay at the follow-up visits.

Each of the two caries prediction models can be used with specific clinical scenarios in order to provide either an acceptable SN or SP²¹. In other published studies, caries predictive models were applied to all children in a similar manner regardless of their existing dental and general health status. For example, Zhang et al.⁹, Raadal et al.¹⁰, and Li and Wang¹¹ reported that SN values were 88% to 94%, however, the reported SP values were 26% to 39%.

Other published studies used different definitions and cutoffs to improve the predictive values. Generally speaking, using the same diagnostic test, but only changing the cutoff points of what is defined as disease, SN does not increase much without decreasing the SP (and vice versa)²². However, SN can increase without compromising SP only if a different “better” diagnostic test is developed, or a second test is added, such as baseline salivary levels of cariogenic bacteria²².

For example, when ter Pelkwijk¹² used dmfs≤4 vs. dmfs>4 to predict any future dental caries, the SN was 69% and the SP was 72%. Similarly, in Europe, Steiner et al.¹³ reported that the SN and SP using the absence of any sound primary molars to predict the presence of ≥4 DMFS were 75% and 98%, respectively, while the values were 81% and 77%, respectively, when using ≥6 DMFS as an outcome (Helfenstein et al.¹⁴). Although these studies were somewhat successful in improving the predictive values, the outcome definitions and the cutoffs were quite arbitrary and not based on any scientific bases. Also, using these random definitions can be very confusing and not clinically-applicable.

The existing caries incidence prediction models are applied in a similar way to all children without taking into consideration the differences in their dental and/or health conditions. In some clinical scenarios, a highly sensitive tool is needed for prediction of ΔDMFS, while in others a highly specific tool is needed. For example, a six-year-old patient who had the last recall examination more than 2 years ago requires a sensitive predictive tool, compared to a patient who has been coming for recall exams every 6 months during the past 3 years (the latter requires a specific tool). For the first patient, high sensitivity is needed in order to avoid the situation where the patient has a cavity that is missed and does not return for several years for an exam until the cavity is bigger and then it needs to have a large restoration. However, low specificity is less of a concern (high probability of determining she would have caries, but ends up caries-free). For the second patient, a predictive tool with high specificity is needed (low probability of determining she has would have caries but ends up caries-free). Having low sensitivity is less of a concern (high probability of predicting she would be without caries, but ends up with caries), because it is likely that she would come in for her next recall exam and the dentist probably would be able to diagnose the lesion when it is still small and manage it with non-invasive or minimally-invasive techniques.

Thus, our study was designed to overcome the limitations of the previous studies by introducing two models for predicting ΔDMFS that can be used with different clinical scenarios. These two models are easy, applicable, and not time-consuming. If these two models were validated in other samples and populations, they could have the potential to change the current paradigm for caries risk assessment and caries prediction models which have always been applied to all children in the same way, which could result in low SN or SP results, respectively, when high SN or SP values are needed.

The results of this study showed that both models provided high NPVs (80%-100%), but low PPVs (3.4%-45.5%). In other words, both models showed that, when a child was considered at high risk for getting dental caries, there was a high probability that this child would be caries-free (in this scenario, clinicians would be unnecessarily aggressive in the diagnostic and preventive approaches when they were not necessary). However, both models showed that when a child was considered low-risk for getting dental caries, there is a high probability that this child will be caries-free (in this scenario, clinicians could be fine with being less aggressive in the diagnostic and preventive approaches, such as obtaining radiographs every 18 months, etc.).

To demonstrate clinical applicability of this project that can be used at an individual level, we discuss three different scenarios. Scenario #1: A six-year-old new patient presented initially to the dental provider and the mother did not remember when the last dental examination was, but it was more than 2 years ago. For this patient, we want a sensitive predictive tool (low probability of determining she is caries-free, but really has cavities). High sensitivity is needed because we want to avoid the patient having a cavity and, consistent with their history, not returning for an exam for several years until the cavity has progressed sufficiently that it then needs to have a large restoration. However, clinicians do not worry too much about having low specificity (high probability of determining the child would have cavities, but ends up caries-free). Therefore, we need to be aggressive with our treatment and preventive approaches. Based on the results of this research, we recommend using dmfs to predict ΔDMFS (high sensitivity).

Scenario #2: A five-year-old patient on the waiting list for a bone marrow transplant. For this patient, we recommend the same thing as in Scenario #1. A highly sensitive tool is needed in order to minimize the chance of having unpredicted cavities that might need restorative treatment when there are contraindications because of the transplant, such as low platelet count, etc.

Scenario #3: A six-year-old patient has been coming for recall examination every 6 months during the past 3 years. For this patient, the clinician needs a predictive tool with high specificity (low probability of determining she has cavities, but is really caries-free). Low sensitivity (high probability of determining the child is without caries, but results in caries) might not be a major issue, because it is likely that she would come for her next recall exam and the dentist probably would be able to diagnose the lesion when it is small and manage it with sealants or a small preventive resin restoration. Thus, with Scenario #3, we do not need to be as aggressive on treatment of questionable/incipient lesions and can rely on future active surveillance and re-evaluation for possible remineralization of these lesions. Based on the results of this research, we recommend using ds to predict ΔDMFS (high specificity).

The current study has several strengths, including the homogenous cohort of African American children and the determination of SN, SP, PPV and NPV at all follow-up dental visits using two separate models. Since previous dental caries experience is the sum of the cumulative effects of different, complex etiological and preventive factors, including microbiological, demographic, dietary and oral hygiene ones^23–28, we focused on dental caries in primary teeth as a predictor for having dental caries in permanent teeth among African-American children.

The main limitations of this study are the relatively small convenience sample (difficult to generalize the findings), lack of data on non-cavitated lesions, and reliance solely on clinical examination without radiographs. Another limitation is that some children at baseline had already exfoliated primary teeth that could have had a history of caries before the dmfs score was recorded. Although the best tool to predict future caries is past caries experience, it is more useful to determine caries risk before the disease is manifested. Also, we need a strong predictor for dental caries which is based on health, not disease.

In the future, additional research studies are needed to show which teeth/surfaces can best predict future caries incidence, as emphasized by Divaris²⁹. Also, additional studies are needed to assess the caries risk at mid to late adolescence.

Conclusion:

1. This study presented two separate models to predict future caries of the permanent teeth which are easy, clinically-applicable, and time-efficient. When selecting the appropriate model based on the clinical history and scenario, clinically-useful enhanced sensitivity and/or specificity can be obtained.

2. If these two models were validated by additional studies, they have potential to add to the current paradigm for caries risk assessment and caries predication..