Cost function estimation: the choice of a model to apply to dementia

Area of Science:

• Health Economics
• Biostatistics
• Econometrics

Background:

• Statistical analysis of healthcare cost data is challenging due to high variability caused by a few patients with extreme costs.
• Predicting individual patient costs necessitates regression models that prioritize prediction accuracy.

Purpose of the Study:

• To estimate the annual healthcare cost of dementia by developing a robust cost function.
• To explore econometric issues in modeling skewed health care cost data.

Main Methods:

• Regression models were employed to predict health care costs based on dementia severity, sex, age, marital status, and comorbidities.
• Models included log-transformed dependent variables with re-transformation methods to estimate costs on the original scale.
• Model performance was evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Theil U-statistic.

Main Results:

• A two-part model, comprising a logistic regression for cost probability and a second stage for cost level, was selected as the best predictor.
• The choice of model significantly influenced predicted annual healthcare costs, with estimates varying widely based on the model used.
• The selected two-part model provided cost estimates ranging from DKK 44,714 for very mild dementia to DKK 197,840 for severe dementia.

Conclusions:

• The two-part model offers a superior approach for estimating dementia-related healthcare costs compared to other regression techniques.
• Accurate prediction of healthcare costs is crucial for resource allocation and policy-making in dementia care.
• The study highlights the substantial economic burden of dementia, varying significantly with disease severity.