PhD defence F. (Fang) Li

More precision in depression care

Optimizing Cost-Effectiveness and Personalized Treatment with Real-World Data

This thesis investigates strategies to optimize depression treatment by analyzing real-world data to match patients with the most effective interventions. Through various studies, including health economic evaluations and machine learning analyses, the research reveals that treatment outcomes depend significantly on factors like duration and intensity, with non-linear relationships suggesting optimal levels vary among patient subgroups. The findings indicate that algorithm-guided treatment programs, while potentially beneficial for long-term outcomes, may not be more cost-effective than predefined duration programs in the short term. Early symptom changes, particularly in mood and anxiety by week five of treatment, emerge as strong predictors of overall treatment success. The research introduces a mathematical model (POMDP and MDP) that shows promise in identifying non-responders early and optimizing treatment decisions. The thesis concludes that while extending treatment beyond optimal durations may not provide additional benefits, personalizing treatment approaches based on individual patient characteristics could improve outcomes while making more efficient use of healthcare resources.