This paper develops a methodology and defines a measure that can be used to separate subjects that received an experimental therapy into those that benefitted from those that did not in recent‐onset type 1 diabetes. Benefit means a slowing (or arresting) the decline in beta‐cell function over time. The measure can be applied to comparing treatment arms from a clinical trial or to response at the individual level.

An analysis of covariance model was fitted to the 12‐month area under the curve C‐peptide following a 2‐hour mixed meal tolerance test from 492 individuals enrolled on five TrialNet studies of recent‐onset type 1 diabetes. Significant predictors in the model were age and C‐peptide at study entry. The observed minus the model‐based expected C‐peptide value (quantitative response, QR) is defined to reflect the effect of the therapy.

A comparison of the primary hypothesis test for each study included and a t test of the QR value by treatment group were comparable. The results were also confirmed for a new TrialNet study, independent of the set of studies used to derive the model. With our proposed analytical method and using QR as the end‐point, we conducted simulation studies, to estimate statistical power in detecting a biomarker that expresses differential treatment effect. The QR in its continuous form provided the greatest statistical power when compared to several ways of defining responder/non‐responder using various QR thresholds.

This paper illustrates the use of the QR, as a measure of the magnitude of treatment effect at the aggregate and subject‐level. We show that the QR distribution by treatment group provides a better sense of the treatment effect than simply giving the mean estimates. Using the QR in its continuous form is shown to have higher statistical power in comparison with dichotomized categorization.