Assessing and comparing costs: how robust are the bootstrap and methods based on asymptotic normality?

Area of Science:

• Health Economics
• Biostatistics
• Statistical Modeling

Background:

• Cost data in health economics often present highly skewed distributions.
• Common statistical methods like bootstrapping and asymptotic normality may be technically valid but can yield inefficient or misleading inferences.
• Existing approaches may not adequately address the inherent skewness in cost data.

Purpose of the Study:

• To challenge common perceptions regarding the statistical assessment of costs and cost-effectiveness.
• To advocate for statistical methods that appropriately handle skewed cost distributions.
• To emphasize the importance of incorporating prior information in Bayesian analyses for health economic evaluations.

Main Methods:

• Critique of standard statistical techniques (bootstrap, asymptotic normality) for skewed cost data.
• Demonstration of the limitations of these methods in health economic contexts.
• Exploration of Bayesian analysis incorporating prior information.

Main Results:

• Standard methods, while technically valid, can lead to inefficient and misleading inferences for skewed cost data.
• Methods specifically designed to recognize and address data skewness are crucial for accurate cost-effectiveness analysis.
• Bayesian approaches that incorporate relevant prior information can enhance the reliability of health economic assessments.

Conclusions:

• Statistical assessment of health economic costs requires methods that acknowledge and manage skewed distributions.
• Relying solely on methods assuming normality or employing standard bootstrapping may be inadequate.
• Integrating prior information within a Bayesian framework offers a more robust approach to health economic evaluations.