EVALUATING COSTS WITH UNMEASURED CONFOUNDING: A SENSITIVITY ANALYSIS FOR THE TREATMENT EFFECT

Area of Science:

• Health Economics
• Biostatistics
• Cancer Treatment Analysis

Background:

• Observational studies estimating treatment cost effects are vulnerable to bias from unmeasured confounders.
• Assessing the potential magnitude of such biases is crucial for reliable cost-effectiveness analyses.

Purpose of the Study:

• To derive a general adjustment formula for loglinear models of mean cost to account for unmeasured confounders.
• To assess the performance and robustness of this adjustment method through simulation.
• To apply the method to real-world data for bladder cancer treatment cost evaluation.

Main Methods:

• Derivation of a general adjustment formula for loglinear models of mean cost.
• Exploration of special cases under various assumptions for unmeasured confounder distributions.
• Simulation studies to evaluate adjustment performance, focusing on robustness to conditional independence assumptions.
• Application to SEER-Medicare cost data for stage II/III muscle-invasive bladder cancer.

Main Results:

• The derived adjustment formula provides a method to quantify bias from unmeasured confounders in cost estimation.
• Simulation results demonstrate the method's performance and its robustness under specific assumptions.
• Analysis of bladder cancer data indicates that the significance of treatment effects (radical cystectomy vs. chemoradiation) is sensitive to potential unmeasured confounders.
• Sensitivity analysis revealed that plausible Bernoulli, Poisson, and Gamma confounders can alter the observed treatment effect significance.

Conclusions:

• The developed adjustment method is valuable for addressing unmeasured confounding in health economic evaluations.
• Findings highlight the importance of considering unmeasured confounders when interpreting cost-effectiveness of cancer treatments.
• The sensitivity of treatment effect significance to unmeasured confounders underscores the need for careful study design and analysis in observational cost research.