Bayesian Nonparametric Policy Search with Application to Periodontal Recall Intervals

Area of Science:

• Dentistry
• Public Health
• Data Science

Background:

• Periodontal disease causes significant tooth loss, impacting public health.
• Current six-month dental visit recommendations lack evidence and yield suboptimal outcomes.
• There is a need for improved, cost-effective dental care strategies.

Purpose of the Study:

• To develop a tailored approach for dental recall intervals.
• To optimize patient-specific visit recommendations to minimize periodontal disease progression.
• To ensure recommendations are interpretable for clinical integration.

Main Methods:

• Formalized a dynamic treatment regime for sequential decision-making.
• Combined non-parametric Bayesian dynamics modeling with policy-search algorithms.
• Utilized electronic dental records from HealthPartners HMO for analysis.

Main Results:

• The proposed method demonstrated improved dental health outcomes in simulations and real-world data.
• The tailored approach effectively minimized disease progression.
• No increase in average recommended recall time was observed compared to standard methods.

Conclusions:

• Personalized dental recall intervals offer a more effective and evidence-based approach to managing periodontal disease.
• Dynamic treatment regimes can be effectively modeled and implemented in clinical practice.
• This strategy improves patient outcomes while maintaining resource efficiency.