Accurate EVSI Estimation for Nonlinear Models Using the Gaussian Approximation Method.

Area of Science:

• Health economics
• Decision analysis
• Biostatistics

Background:

• The expected value of sample information (EVSI) quantifies the benefit of additional data collection.
• Traditional nested Monte Carlo methods for EVSI estimation are computationally intensive.
• The Gaussian approximation (GA) offers efficiency but can be biased for nonlinear models, potentially leading to suboptimal study designs.

Purpose of the Study:

• To extend the Gaussian approximation (GA) approach for accurate EVSI estimation in the presence of nonlinear decision models.
• To address the bias introduced by conventional GA in complex, nonlinear health economic models.
• To enhance the reliability of EVSI estimates for optimizing study designs.

Main Methods:

• A novel method combining Taylor series approximation with splines to model conditional expectations.
• Approximation of conditional moments using GA and Fisher information.
• Validation through data collection exercises with non-Gaussian parameters and nonlinear decision models, comparing against nested Monte Carlo, conventional GA, and nonparametric regression.

Main Results:

• The proposed spline-based Taylor series GA method yields accurate EVSI estimates for non-Gaussian parameters and nonlinear decision models across various sample sizes.
• The computational efficiency of the new approach is comparable to existing advanced methods.
• Demonstrated improved accuracy over conventional GA for nonlinear models.

Conclusions:

• The developed approach accurately and efficiently estimates EVSI across different sample sizes, even with nonlinear decision models.
• This method supports researchers in designing economically optimal studies by providing reliable EVSI calculations.
• Enhances resource allocation strategies in health economics through precise EVSI estimation.