Estimating incremental costs with skew: a cautionary note.

Area of Science:

• Health Economics
• Biostatistics
• Econometrics

Background:

• Healthcare cost data frequently exhibit skewness, prompting the use of log transformations or generalized linear models with log links.
• These non-linear modeling approaches can result in non-linear incremental effects, where treatment costs vary significantly across different covariate levels.
• This variability can lead to substantial distortions in predicted costs.

Purpose of the Study:

• To demonstrate that log link functions and log transformations are not essential for modeling skewed data.
• To highlight the unintended and potentially misleading effects of these non-linear approaches on cost estimations.
• To advocate for alternative modeling strategies that avoid these distortions.

Main Methods:

• Simulated skewed cost data using a linear model incorporating a treatment variable, a covariate, and excessive cost observations.
• Utilized actual cost data from a hip-replacement patient pain-relief intervention study.
• Estimated cost models using various methods designed to handle data skewness and calculated incremental treatment costs across different covariate levels.

Main Results:

• All tested methods yielded unbiased estimates of incremental treatment costs at the mean covariate level.
• Log-based models exhibited a doubling of implied incremental treatment costs between extreme low and high covariate values.
• This pattern in log-based models was inconsistent with the underlying linear model's structure.

Conclusions:

• The potential for log-based specifications to produce misleading incremental cost estimates is often overlooked.
• Specification checks are crucial when employing any cost modeling technique.
• Researchers and policymakers must recognize the limitations of log-based models for patient populations differing from the sample mean, especially in cost-containment contexts.