Visualizing the value of diagnostic tests and prediction models, part II. Net benefit graphs: net benefit as a function of the exchange rate.

Abstract

Background and objective: In this second of a 3-part series, we move from expected gain in utility (EGU) graphs to net benefit (NB) graphs, which show how NB depends on w= C/B, the treatment threshold odds, equal to the harm of treating unnecessarily (C) divided by the benefit of treating appropriately (B).

Method: For NB graphs, we shift from the perspective of testing individual patients with varying pretest probabilities of disease to the perspective of applying a test or risk model to an entire population with a given prevalence of disease, P₀. As with EGU graphs, we subtract the harm of testing and the expected harm of treating according to the results of a test or model when it is wrong from the expected benefit of treating when it is right. The difference is that for NB graphs, the prevalence is fixed at P₀ , and the x-axis is w. NB graphs show the NB of 3 strategies: 1) "Treat None"; 2) "Test" and treat those with predicted risk greater than the treatment threshold; and 3) "Treat All" in the population regardless of predicted risk.

Results: The "Treat All" line intersects the y-axis at NB = P₀ and the x-axis at w = P₀/(1 - P₀). The "Test" line intersects the "Treat All" line at the Treat-Test threshold value of w; it intersects the x-axis at the Test-No Treat value of w.

Conclusion: When NB is plotted as a function of w, NB graphs can be drawn as straight lines from easily calculated intercepts.