Longitudinal varying coefficient single-index model with censored covariates

Area of Science:

• Health Economics
• Biostatistics
• Cancer Research

Background:

• Estimating longitudinal medical costs from cancer diagnosis to death is crucial for health policy.
• Longitudinal cost data present statistical challenges: non-normality, skewness, zero-inflation, heteroscedasticity, nonlinear trajectories, and censoring.
• Existing models struggle with flexibility, parsimony, and interpretation when modeling patient characteristics' impact on varying cost trajectories.

Purpose of the Study:

• To develop a novel statistical model for estimating population-averaged longitudinal medical cost trajectories in cancer patients.
• To understand how patient characteristics influence these cost trajectories over time and survival.
• To address the statistical complexities of skewed, zero-inflated, and censored longitudinal cost data.

Main Methods:

• Proposed a novel longitudinal varying coefficient single-index model.
• Summarized multiple patient characteristics into a single index representing healthcare use propensity.
• Employed generalized estimating equations with an extended marginal mean structure to incorporate censored survival time.

Main Results:

• The proposed model effectively estimates nonlinear, time-varying associations between patient characteristics and longitudinal medical costs.
• Demonstrated the model's ability to handle complex cost data structures and censored survival times.
• Successfully applied the methodology to prostate cancer patient data from the SEER-Medicare-Linked Database.

Conclusions:

• The novel varying coefficient single-index model offers a flexible and interpretable approach for analyzing longitudinal medical costs in cancer patients.
• This methodology provides valuable insights for health policy research by elucidating the impact of patient characteristics on healthcare spending.
• The model's robust performance in simulations and real-world data application highlights its utility in health economics and outcomes research.