Understanding bias in probabilistic analysis in model-based health economic evaluation

Area of Science:

• Health Economics
• Biostatistics
• Computational Modeling

Background:

• Economic evaluations often use probabilistic analysis for less biased estimates due to non-linear relationships.
• However, other factors can influence bias in both probabilistic and deterministic analyses.

Purpose of the Study:

• To evaluate and compare the bias associated with probabilistic analysis versus deterministic analysis.
• To investigate how parameter characteristics, like wide confidence intervals, affect bias in non-linear models.

Main Methods:

• Conducted three simulation studies to assess bias in probabilistic and deterministic analyses.
• Examined bias in model inputs (risk ratios, cost estimates) and outputs (life-years).

Main Results:

• Probabilistic analyses can sometimes exhibit greater bias in inputs and outputs compared to deterministic analyses.
• Point estimates may reduce bias when parameters have wide, asymmetric confidence intervals.
• High parameter variance in probabilistic analyses can lead to extreme values biasing non-linear model results.

Conclusions:

• Neither probabilistic nor deterministic analysis guarantees less bias in non-linear models.
• Health economists must carefully consider potential biases and choose the most appropriate analytical approach.