The Kidney‐TIME Dose‐Response Rests on a Covariate the Trial Cannot Pin Down

To the Editor:

The as‐treated contrast that Kayler and colleagues carry forward‐ higher donor inquiry among optional‐video viewers than non‐viewers (22.9% vs 9.8%, SHR 2.52) [1]‐ turns on the one baseline variable the authors flag as imbalanced and then adjusts away the computer ownership. That variable gates the dose, and the trial cannot say by how much the arms differ on it. The dose‐response reading fails on both counts.

Consider what delivered the dose. The optional videos were delivered via web link; a working computer is a prerequisite for viewing them. Splitting the intervention arm into viewers and non‐viewers therefore largely sorts by device access. The viewer group is enriched for the connected; the non‐viewer group for those without a computer and, by the authors' own moderator result, those reporting low social support. The comparison recovers the deprivation gradient the intervention meant to close and reports it as a treatment gain.

Now the numbers. Table 1 gives computer ownership as 97% (203) in controls and 63% (132) in the intervention arm, with p = 0.033. Those counts and that p‐value cannot both be correct: a 203‐versus‐132 split at these denominators implies a p‐value far below 0.001, not 0.033. The reader cannot tell whether the arms differ trivially or by more than thirty points on the single covariate that determines who could receive the dose. This is not a footnote. The Methods state that computer ownership was entered as a covariate in every adjusted model, primary and as‐treated, so the discrepancy propagates through all of them.

The general reason as‐treated fails is older than this trial. In the Coronary Drug Project, placebo patients who adhered well had much lower five‐year mortality than poor adherers (15.1% vs 28.3%), an effect adjustment could not remove [2]; the pooled odds ratio for mortality with good placebo adherence is 0.56 [3]. Engagement marks prognosis rather than causing it, and baseline covariate adjustment does not recover a valid per‐protocol effect from a self‐selected subgroup [4].

The intention‐to‐treat null (SHR 0.84, 95% CI 0.52 to 1.36) [1] is the interpretable result, though a weak test: attrition left the study underpowered against its 15% target, the comparator delivered its own 13‐minute video, open materials leaked across arms, and outreach behaviors sat near ceiling in both groups. That patient education can raise LDKT pursuit is established where the contrast is real and intensive [5].

The remedy is to randomize the dose itself, reminder intensity, or added human coaching, and analyse by that assignment. Before the as‐treated result travels further, the computer‐ownership entry needs correction, because every adjusted estimate in the paper depends on it.

Funding

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

AI Usage

Grammarly for Grammar, and clarity

Data Availability Statement

Data sharing not applicable to this article as no datasets were generated or analysed during the current study