Expected value of sample information calculations in medical decision modeling

Area of Science:

• Decision Analysis
• Health Economics
• Medical Informatics

Background:

• Increasing interest in expected value of information (EVI) theory for medical decision-making.
• Expected value of sample information (EVSI) used for optimizing clinical trial design.
• Challenges in applying EVSI due to multiple uncertainty sources and data heterogeneity.

Purpose of the Study:

• To derive simple Monte Carlo methods for EVSI calculations in medical decision applications.
• To extend EVSI to models with multiple uncertainty sources and complex data structures.
• To provide methods for calculating EVSI with relative efficacy measures and heterogeneous data.

Main Methods:

• Development of simple and nested Monte Carlo procedures for EVSI calculation.
• Application to multi-parameter decision models with probability, rate, or continuous variables.
• Approximate methods for relative measures (risk differences, odds ratios, risk ratios, hazard ratios).
• Specific EVSI calculations for random effects meta-analyses (individual vs. population decisions).

Main Results:

• Demonstration of Monte Carlo procedures for calculating EVSI with various data types and structures.
• Adaptation of EVSI calculations for relative efficacy measures and heterogeneous literature estimates.
• Distinction between EVSI for specific patient groups versus entire populations based on meta-analysis.
• Consideration of EVSI for studies updating both baseline and relative treatment efficacy.

Conclusions:

• The proposed Monte Carlo methods effectively extend EVSI calculations to complex medical decision models.
• These methods facilitate research prioritization and sample size determination in the presence of multiple uncertainties.
• The approach is applicable to most probabilistic decision models, aiding evidence-based healthcare decisions.