Predicting patient 'cost blooms' in Denmark: a longitudinal population-based study.

Area of Science:

• Health Informatics
• Predictive Analytics
• Healthcare Economics

Background:

• Identifying high-cost patients is crucial for healthcare resource allocation.
• Traditional models often struggle to predict future healthcare spending accurately.
• A subset of patients, termed 'cost bloomers,' experience a rapid increase in healthcare expenditures.

Purpose of the Study:

• To compare the predictive performance of standard versus enhanced models for identifying future high-cost patients.
• To specifically assess the ability of models to predict 'cost bloomers' – individuals transitioning to the upper decile of healthcare spending.
• To quantify the improvement in predictive power offered by enhanced models.

Main Methods:

• Developed six models, ranging from a standard 4-variable model to an enhanced model with 1053 non-traditional features.
• Utilized a large population dataset from Western Denmark (2004-2011).
• Evaluated prospective predictive performance using unseen data, measuring 'cost capture' (predicted vs. actual expenditures).

Main Results:

• The best enhanced model demonstrated a 21% improvement in cost capture for predicting overall high-cost patients.
• The enhanced model achieved a 30% improvement in cost capture for predicting cost bloomers.
• Enhanced models incorporating a diverse feature set and statistical learning methods showed superior performance.

Conclusions:

• Models enhanced with a large, diverse feature set and modern statistical learning methods outperform standard models.
• These enhanced models are particularly effective for predicting future high-cost patients and cost bloomers.
• Improved prediction of high-cost patients can lead to more efficient healthcare management and resource allocation.