Modeling Mental Health Case-Mix for Quality Improvement-A Comparison of Statistical and AI Models

Area of Science:

• Health Services Research
• Artificial Intelligence in Healthcare
• Mental Health Informatics

Background:

• Rising mental health (MH) disorder prevalence necessitates improved care quality and effectiveness.
• Accurate staffing needs assessment and outcome benchmarking are crucial for MH care improvement.
• Lack of robust case-mix adjustment systems hinders accurate evaluation of staffing and patient outcomes.

Purpose of the Study:

• To develop a robust case-mix adjustment system for mental health (MH) care.
• To leverage advanced modeling techniques for enhanced predictive accuracy in MH patient classification.
• To identify optimal models for risk adjustment to improve staffing and benchmark patient outcomes.

Main Methods:

• Retrospective population-based study of over two million mental health (MH) patients (n = 2,088,174).
• Patients grouped into 162 clinically homogeneous categories using Clinical Classifications Software Refined (CCSR).
• Evaluated four statistical models and four artificial intelligence (AI) models for predictive performance.

Main Results:

• Box-Cox regression showed highest predictive power among statistical models (R² = 0.42).
• CatBoost AI model demonstrated superior performance (R² = 0.458).
• AI models offered modest improvements over traditional statistical models; robustness confirmed by sensitivity analyses.

Conclusions:

• Both Box-Cox and CatBoost models exhibit superior predictive performance for MH case-mix adjustment.
• Developed models can support risk adjustment for optimizing staffing levels in mental healthcare.
• Findings facilitate benchmarking patient outcomes to drive quality improvement initiatives in mental health services.