- Published
- Updated
Converting a continuous measurement into success or failure may simplify presentation, but can discard information and exaggerate small differences near a threshold. The decision needs a scientific rationale beyond making a table easier to read.
Define the question and assumptions
First explain the question answered by the continuous measurement. If a binary definition is proposed, assess the origin of the threshold, sensitivity to measurement error and implications for sample size. Show that mean change and response proportion summarise different information; neither should automatically be assumed to represent the other.
Hypothetical study example
In a hypothetical trial, mean changes are similar between arms, but response rates appear different because many participants sit near the chosen threshold. The team reviews the distribution and threshold rationale together. It does not select the threshold producing the largest difference after examining the results.
Design review checks
- Document the clinical rationale and independent evidence supporting threshold selection.
- Examine how measurement variability near the cutoff changes participant classification.
- Compare information and sample size requirements for continuous and binary summaries.
Record the decision and its limits
The accepted analysis plan identifies the primary summary and the supportive role of the other. Avoid presenting an exact cutoff as inherently meaningful without justification. If alternative thresholds are investigated, state their purposes prospectively; exploratory evaluations should not be presented with confirmatory certainty, and selection effects should be discussed.
Frequently asked questions
- Is a binary endpoint always easier to interpret?
- Presentation may be simpler, but it does not explain the meaning of differences around the cutoff.
- Does it make continuous analysis unnecessary?
- No. Distributions and effect sizes can reveal information that response rates omit.