Measuring Responsiveness to Preventative Treatments for Dental Caries: Towards a Predictive Model Using Machine Learning

Objectives: Our short-term objective is to assess the comparative effectiveness of caries prevention agents and measure the re-occurrence rate of dental caries in children receiving preventative therapies for cavities. Our long-term objective is to understand the clinical and microbial determinants of re-occurrence, leading to the development of a predictive model for the “responsiveness” to care.
Methods: Analyses for the longitudinal impact of prevention used binary Markov chains. Model outcomes were defined as the prevalence of untreated decay, stratified by baseline decay status (no decay versus any decay). Participants were treated with either sealants and interim therapeutic restorations or a 38% silver diamine fluoride solution on all teeth pits, fissures, and symptomatic carious lesions. Salivary samples and supragingival plaque were collected from 15 children with and without caries and analyzed using 16S sequencing.
Results: Approximately 32% of participants had untreated decay (deciduous or permanent teeth) at baseline. Model results indicate a significant increase in the odds of untreated decay over time for children without decay at baseline with a large negative quadratic effect. For children with baseline decay, there was a significant reduction of decay prevalence at subsequent observations but an increasing quadratic effect over time. Further, results indicate that approximately 15% of study participants receiving care experience a re-occurrence of dental caries regardless of baseline caries prevalence within two years.
Conclusions: We demonstrate a disparity in the effects of caries prevention. Further research in this area will consist of linking data from sequencing analyses to clinical and epidemiologic data in support of a machine learning algorithm to predict responsiveness to care in order to improve the targeting of caries prevention programs.