Primary Investigator (PI) Name
Christina Scherrer
Department
SPCEET – Industrial and Systems Engineering
Abstract
Dental caries remains one of the most common chronic diseases in American children and young adults. Dental sealants can prevent up to 80% of cavities, yet there is little guidance on when to apply or reseal them. Existing economic evaluations typically compare static strategies or sweep through fixed predictive-model thresholds using finite-horizon markov models or discrete-event simulation, but they do not adapt large-scale data-driven decisions for finding high-risk individuals from a large population. We propose a two-stage framework that integrates machine-learning (ML) risk prediction, microsimulation, and model-based reinforcement learning (RL) to identify cost-effective, state-dependent sealant policies. In stage one, we will train an ML model on NHANES 2011–2016 child data (~6,000 children, ~37% with caries) using demographic, behavioral, laboratory, and dietary predictors to estimate each individual’s baseline risk; these individualized risk profiles both initialize and parameterize a microsimulation of tooth-state transitions (healthy, sealed, sealant lost, decayed, filled) using published estimates of sealant effectiveness, retention, caries incidence hazards, service use, and costs. In stage two, a finite-horizon, model-based dynamic-programming RL algorithm (e.g., value iteration) will learn—at six‑month decision epoch over a 10‑year horizon—whether to seal, reseal, filling or defer in order to maximize discounted net monetary benefit. We will benchmark the learned policy against conventional static and threshold-based comparators, reporting cost per averted decay, cost per disability-adjusted life year (DALY) averted, incremental net monetary benefit, and subgroup results (age and race/ethnicity). By explicitly modeling risk heterogeneity and allowing preventive actions to adapt over time, the study aims to deliver high-performance, interpretable decision rules for preventive dental care that are practical for providers and policymakers and balance costs, health benefits, and equity considerations.
Reinforcement Learning Driven Cost-Effectiveness Analysis of Dental Caries Prevention Policies Under Uncertainty
Dental caries remains one of the most common chronic diseases in American children and young adults. Dental sealants can prevent up to 80% of cavities, yet there is little guidance on when to apply or reseal them. Existing economic evaluations typically compare static strategies or sweep through fixed predictive-model thresholds using finite-horizon markov models or discrete-event simulation, but they do not adapt large-scale data-driven decisions for finding high-risk individuals from a large population. We propose a two-stage framework that integrates machine-learning (ML) risk prediction, microsimulation, and model-based reinforcement learning (RL) to identify cost-effective, state-dependent sealant policies. In stage one, we will train an ML model on NHANES 2011–2016 child data (~6,000 children, ~37% with caries) using demographic, behavioral, laboratory, and dietary predictors to estimate each individual’s baseline risk; these individualized risk profiles both initialize and parameterize a microsimulation of tooth-state transitions (healthy, sealed, sealant lost, decayed, filled) using published estimates of sealant effectiveness, retention, caries incidence hazards, service use, and costs. In stage two, a finite-horizon, model-based dynamic-programming RL algorithm (e.g., value iteration) will learn—at six‑month decision epoch over a 10‑year horizon—whether to seal, reseal, filling or defer in order to maximize discounted net monetary benefit. We will benchmark the learned policy against conventional static and threshold-based comparators, reporting cost per averted decay, cost per disability-adjusted life year (DALY) averted, incremental net monetary benefit, and subgroup results (age and race/ethnicity). By explicitly modeling risk heterogeneity and allowing preventive actions to adapt over time, the study aims to deliver high-performance, interpretable decision rules for preventive dental care that are practical for providers and policymakers and balance costs, health benefits, and equity considerations.