Fairness of Machine Learning Algorithms for Predicting Foregone Preventive Dental Care for Adults

Area of Science:

• Public Health
• Health Informatics
• Dental Public Health

Background:

• Preventive dental care is crucial for overall health and preventing advanced dental disease.
• Identifying at-risk populations is key for targeted public health interventions.

Purpose of the Study:

• To develop and assess machine learning models for predicting foregone preventive dental care in adults.
• To evaluate the algorithmic fairness of these models across sociodemographic subgroups.

Main Methods:

• Secondary analysis of longitudinal data from the US Medical Expenditure Panel Survey (MEPS) (2016-2019).
• Utilized tree-based ensemble machine learning models with 50 predictors.
• Assessed model performance using area under the receiver operating characteristic curve (AUC).

Main Results:

• Models accurately predicted foregone preventive dental care (overall AUC=0.84).
• Performance varied across sociodemographic subgroups, with lower accuracy for underrepresented groups.
• Previous dental visit patterns, healthcare utilization, and dental benefits were key predictors.

Conclusions:

• Machine learning models show promise for identifying adults at risk of missing preventive dental care.
• Algorithmic bias against underrepresented groups was observed, emphasizing the need for fairness assessments.
• Ensuring model fairness is critical to avoid exacerbating existing health disparities.