Calculating the Expected Value of Sample Information in Practice: Considerations from 3 Case Studies.

Area of Science:

• Health economics
• Decision analysis
• Biostatistics

Background:

• Efficient research investment is vital for policy decisions.
• Expected value of sample information (EVSI) aids study design and sample size selection.
• Traditional EVSI estimation methods are computationally intensive.

Purpose of the Study:

• Compare the performance of recently developed EVSI approximation methods.
• Evaluate computational speed and accuracy across diverse health economic models.
• Identify optimal EVSI methods for various research contexts.

Main Methods:

• Compared four EVSI approximation methods.
• Utilized three distinct, previously published health economic models.
• Assessed performance with varying study outcomes, missing data, and observational data.

Main Results:

• Approximation methods achieved accurate EVSI estimates in minutes/hours, versus weeks for traditional Monte Carlo.
• Specific methods excel for large sample sizes, multiple design comparisons, or computationally expensive models.
• All evaluated methods yielded comparable results to traditional Monte Carlo.

Conclusions:

• EVSI can now be computed efficiently and reliably in realistic scenarios.
• The choice of EVSI method depends on the specific model, data, and user expertise.
• No single EVSI computation method is universally superior.