Predicting over-the-counter antibiotic use in rural Pune, India, using machine learning methods

Area of Science:

• Public Health
• Infectious Disease Epidemiology
• Health Informatics

Background:

• Over-the-counter (OTC) antibiotic use contributes to antimicrobial resistance, a significant global health threat.
• Understanding the drivers of OTC antibiotic consumption is crucial for developing targeted interventions.
• Rural populations may exhibit unique patterns of antibiotic access and usage.

Purpose of the Study:

• To identify key predictors of over-the-counter antibiotic use in rural Pune, India.
• To develop and validate a machine learning model for predicting OTC antibiotic use.
• To inform public health strategies aimed at reducing inappropriate antibiotic consumption.

Main Methods:

• Machine learning algorithms including stepwise logistic regression, lasso, random forest, XGBoost, and Boruta were employed for feature selection.
• Regression and tree-based models were constructed using identified features to predict OTC antibiotic use.
• Model performance was evaluated using five-fold cross-validation, focusing on Area Under the Receiver Operating Characteristic Curve (AUROC) and log-loss.

Main Results:

• The prevalence of OTC antibiotic use in rural Pune was 35.9%.
• Significant predictors identified by the Boruta algorithm included the perception of direct pharmacy purchases being useful, antibiotic use for eye complaints, and increased household antibiotic consumption.
• The XGBoost+Boruta model, incorporating 7 predictors, achieved a high predictive accuracy (AUROC: 0.934).

Conclusions:

• The XGBoost+Boruta model demonstrated superior accuracy in predicting OTC antibiotic use in the study population.
• Perceived utility of direct pharmacy purchases, antibiotic use for eye conditions, and higher consumption levels are critical factors influencing OTC antibiotic use.
• These findings provide a basis for designing targeted interventions to promote responsible antibiotic stewardship in rural settings.