Predicting obesity and smoking using medication data: A machine-learning approach.

Area of Science:

• Public Health
• Health Informatics
• Machine Learning in Healthcare

Background:

• Administrative health datasets are crucial for public health research but often lack key confounding variables.
• Predicting health outcomes like obesity and smoking is vital for targeted interventions.

Purpose of the Study:

• To develop and validate machine learning (ML) models to predict obesity and smoking using medication data.
• To assess the utility of Australia's Pharmaceutical Benefits Scheme (PBS) data for predicting these health outcomes.

Main Methods:

• Utilized data from the D-Health Trial and QSkin Study, linking to PBS medication records.
• Employed gradient-boosted ML models trained on age, sex, and medication data (Anatomical Therapeutic Classification codes).
• Validated models using a hold-out dataset and compared performance using 3 vs. 5 years of PBS data.

Main Results:

• Models achieved an area under the receiver operating characteristic curve of 0.70 for obesity and 0.71 for smoking prediction.
• Performance was notably better in women than in men.
• Using 5 years of PBS data offered only a marginal improvement over 3 years.

Conclusions:

• Medication data, combined with age and sex, can effectively predict obesity and smoking.
• These ML models offer valuable predictive capabilities for researchers utilizing administrative health data.