Sensitivity analysis for healthcare models fitted to data by statistical methods

Area of Science:

• Statistics
• Biostatistics
• Health Economics

Background:

• Complex statistical models are essential for data analysis but can generate uncertainty in predictions.
• Identifying sources of uncertainty is crucial for effective model refinement and decision-making.
• Existing methods for sensitivity analysis can be time-consuming or overly complex.

Purpose of the Study:

• To describe a simple, quick methodology for performing sensitivity analysis on complex statistical models.
• To provide a decision-support tool for modelers to guide further model development or data acquisition.
• To illustrate the application of this sensitivity analysis method using a breast cancer screening model.

Main Methods:

• A straightforward sensitivity analysis technique is presented.
• The methodology is applied to a previously published breast cancer screening model.
• A simulation study evaluates the method's error as a function of sample size using a simplified 3-parameter model.

Main Results:

• The described method provides rapid insight into model output uncertainty.
• The sensitivity analysis effectively highlights which model parameters contribute most to prediction variability.
• The simulation study demonstrates the method's performance across different sample sizes.

Conclusions:

• This simple sensitivity analysis is a valuable, time-efficient tool for modelers.
• It aids in prioritizing efforts for model improvement and targeted data collection.
• The method is applicable to various complex models, including those in health and screening research.