Interpretation of the Expected Value of Perfect Information and Research Recommendations: A Systematic Review and Empirical Investigation.

Abstract

Expected value of perfect information (EVPI) calculations are increasingly performed to guide and underpin research recommendations. An EVPI value that exceeds the estimated cost of research forms a necessary (although not sufficient) condition for further research to be considered worthwhile. However, it is unclear what factors affect researchers' recommendations and whether there is a notional threshold of positive returns below which research is not recommended. The objectives of this study were to explore whether EVPI and other factors have a bearing on research recommendations and to assess whether there exists a threshold EVPI below which research is typically not recommended. A systematic literature review was undertaken to identify applied EVPI calculations in the health care field. Study characteristics were extracted, including funder, location, disease group, publication year, primary language, and outcome measure. Population EVPI values and willingness-to-pay thresholds were also extracted alongside verbatim text excerpts describing the authors' research recommendations. Recommendations were classified according to whether further research was recommended (a positive recommendation) or not (negative). Factors affecting the likelihood of a positive recommendation were examined statistically using logistic regression and visually by plotting the results in graphs. Eighty-six articles were included, of which 13 suggested no further research, 66 recommended further research, and 7 gave no recommendation. EVPI appears to be a key driver of researchers' recommendations for further research. Disease area, funder, study location, publication year, and outcome may have a bearing on recommendations, although none of these factors reached statistical significance. A threshold EVPI value below which research is typically not recommended was found at around £ 1.48 million.