A leading healthcare company with a broad network of behavioral health facilities across the United States, focused on treating behavioral health and addiction disorders.

The client needed an accurate ML model to predict patient relapse rates effectively, allowing for more tailored treatment plans and efficient resource allocation.

The client, with support from Turing, implemented a comprehensive machine learning (ML) framework to tackle the challenge:

• Data labeling and feature engineering: Collaborated with Turing to clearly define relapse categories and develop labels. The team analyzed five years of patient data and identified 120 features that could impact prediction outcomes.
• Model training and optimization: The team trained and tested various classifier algorithms, selecting GBTClassifier for its accuracy and lesser tendency to overfit. Anti-overfitting techniques were employed, and the model was tuned to achieve the target ROC/AUC score of 0.8.
• Implementation and deployment: After refining the model, Turing deployed it on the client’s systems via Azure Cloud, ensuring real-time application and testing on live patients.

• 4% quarter-on-quarter revenue growth: Achieved through the introduction of customized treatment plans based on relapse predictions.
• Improved treatment customization: Additional data flags like "Discontinued" and "Treatment plan not followed" allowed the client to devise more tailored treatment plans.
• Enhanced predictive accuracy: The model provided accurate relapse predictions at the admission stage, supporting better patient outcomes and resource allocation.