Predicting costs over time using Bayesian Markov chain Monte Carlo methods: an application to early inflammatory...

Area of Science:

• Health Economics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Modeling healthcare costs over time presents challenges due to data skewness and within-subject correlation.
• Accurate cost prediction is crucial for effective healthcare service planning and budgeting.

Purpose of the Study:

• To compare the predictive performance of four different multilevel/hierarchical models for longitudinal healthcare cost data.
• To inform the selection of appropriate statistical models for analyzing and predicting disease-related costs over time.

Main Methods:

• Analysis of healthcare costs in a cohort of individuals with early inflammatory polyarthritis (IP) over a 5-year period.
• Comparison of linear regression, two-part log-transformed cost models, and two-part gamma regression models using Bayesian Markov chain Monte Carlo (MCMC) simulation.
• Model performance assessed using a learning and test sample split, with retransformation factors applied for original cost scale prediction.

Main Results:

• The study evaluated the performance of various hierarchical models in predicting healthcare costs.
• Different models showed varying degrees of accuracy in predicting costs on the original scale.
• Bayesian MCMC methods were employed for all model fitting and prediction evaluations.

Conclusions:

• The choice of hierarchical model impacts the accuracy of longitudinal healthcare cost predictions.
• Understanding model performance is essential for reliable healthcare budget and service planning.
• Multilevel/hierarchical models offer a framework for handling correlated cost data over time.