EVPI
Expected Value of Perfect Information — the maximum amount one should be willing to pay to eliminate all uncertainty.
voiage
Cross-domain Value of Information analysis for health economics, finance, engineering, and public policy
voiage is a Value of Information (VOI) analysis library with a broad Python workflow, selected Rust calculations, and narrower R and Julia source bindings.
VOI methods help quantify the economic value of acquiring additional information to reduce uncertainty in decision-making, particularly in fields like health economics, risk analysis, and decision sciences.
Core VOI Metrics
Section titled “Core VOI Metrics”EVPPI
Expected Value of Partial Perfect Information — identifies key drivers of decision uncertainty by subset of parameters.
EVSI
Expected Value of Sample Information — the expected benefit of collecting new data through a specific study design.
ENBS
Expected Net Benefit of Sampling — EVSI minus research cost, indicating whether research is economically worthwhile.
Additional Methods
Section titled “Additional Methods”- Structural VOI — value of reducing uncertainty about model structure
- Network Meta-Analysis VOI — multi-treatment comparison through evidence networks
- Adaptive Trial EVSI — value of adaptive study designs
- COSS and EVSI/EVPI efficiency — experimental finite-design optimization using signed ENBS and an explicitly commensurate information-efficiency ratio
- Portfolio VOI — optimal allocation across a research portfolio
- Sequential VOI — value of sequential decision-making
- Value of Heterogeneity — estimating returns to individualized care
- Calibration VOI — value of model calibration studies
- Observational VOI — value of observational data collection
Design
Section titled “Design”- Explicit contracts — documented schemas, units, assumptions, provenance, and deterministic fixtures.
- Shared reference calculations — Python, R, and Julia test the same Rust EVPI kernel against common numerical cases.
- Evidence-labelled methods — stable, approximate, and experimental methods are distinguished in the API and documentation.
- Python analysis workflow — NumPy, SciPy, pandas, xarray, and optional backends support model preparation, analysis, and reporting.
Python currently provides the complete workflow and the provenance-bearing
result envelope. R and Julia expose scalar EVPI and require a compatible
voiage-ffi native library. Mojo remains an external, unreleased boundary.
Target Audience
Section titled “Target Audience”- Health economists and HTA practitioners
- Decision analysts and operations researchers
- Clinical trial designers and Bayesian statisticians
- Researchers in environmental, financial, and engineering domains
- Students learning VOI methods