Moving beyond threshold-based dichotomous classification to improve the accuracy in classifying non-responders

Abstract

We examined maximal oxygen consumption responses following exercise training to demonstrate the limitations associated with threshold-based dichotomous classification of responders and non-responders and proposed alternative methods for classification. Specifically, we: 1) calculated individual probabilities of response, and 2) classified individuals using response confidence intervals (CI) and reference points of zero and a smallest worthwhile change of 0.5 METs. Our findings support the use of individual probabilities and individual CIs to improve the accuracy in non-response classification.

Keywords: Individual responses; VO2max; interindividual variability; non-responder; typical error.

Figure 1

(A) Observed individual VO ₂max responses and 2x TE threshold‐based dichotomous classification of responders and non‐responders. (B) Observed VO ₂max responses and probabilities of a positive response (>0 mL/kg/min) for participants 3–4.

Figure 2

Observed VO ₂max responses, response 50% CIs (black) and 90% CIs (gray), and probabilities of a positive response (percentages) relative to a zero‐based threshold (A–C) and a SWC‐based threshold (D–F) for three representative participants. Participant numbers correspond to Figure 1A. Red, black, and green circles represent non/adverse, uncertain, and positive responders, respectively.