A Null Trial or an Undelivered Intervention? Check Fidelity Before Interpreting the Result
October 29, 2026 · 5 min read · Aqrab editorial team
A trial finds little difference between an intervention and usual care. Did the intervention fail, or did participants rarely receive it as intended? The result alone cannot distinguish those explanations. Start by reconstructing the intervention that was actually tested.
Follow the intervention from plan to practice
Imagine a hypothetical six-session rehabilitation program. The protocol describes six sessions, but the report says only that participants were “offered therapy.” That phrase does not tell us how many sessions therapists delivered, how many participants attended, or whether the content matched the protocol. The example is invented; it makes no claim about any real therapy.
| Layer | Evidence to look for |
|---|---|
| Planned | Components, provider training, schedule, dose, and allowed tailoring. |
| Delivered | Sessions and components actually provided; who assessed fidelity and how. |
| Received | Attendance or exposure, with a denominator and reasons for missed sessions. |
| Changed | When, why, and how the intervention was modified during the trial. |
The TIDieR reporting checklist separates how well fidelity was planned or assessed from how well the intervention was actually delivered. For nonpharmacologic trials, the CONSORT extension also asks authors to describe standardization, provider and participant adherence, and the treatments as implemented. These are requests for transparent reporting, not proof that fidelity caused the outcome.
The practical takeaway
When a trial result is null, ask for the planned dose, delivered dose, participant receipt, and comparator care before calling the intervention ineffective—or calling low fidelity the explanation.
Do not turn fidelity into a second randomized trial
Randomization assigns participants to a strategy; it does not randomize them to complete all sessions. People who attend more sessions may differ in health, motivation, access, or early response. Comparing “completers” with “non-completers” therefore does not, by itself, estimate the causal effect of receiving the full intervention. A low-fidelity report can limit interpretation and suggest a delivery problem worth studying, but it cannot rescue a null result with an unadjusted per-protocol comparison.
Also inspect usual care. If the comparator received similar components, the contrast may have been smaller than the protocol implied. If fidelity was not measured, label it unknown, not poor.
Four questions for a trial critique
- Could another team reproduce the intervention and comparator from the reported materials, providers, schedule, and tailoring rules?
- Was fidelity assessed with a stated method, and are actual delivery and receipt reported with denominators?
- Were modifications and departures from the protocol dated and explained?
- Does the conclusion describe the effect of assignment, or does it quietly switch to a post-randomization subgroup?
Where Aqrab fits
Use Aqrab to organize a methods critique around the gap between intervention plan and delivery, and to flag missing fidelity details for human review. It cannot infer unreported delivery or prove why a trial result was null.
Source anchors
The TIDieR checklist and guide defines intervention-description items, including modifications and planned versus actual fidelity. The 2017 CONSORT extension for nonpharmacologic treatments addresses standardization, provider and participant adherence, and reporting the treatment as implemented. Both are reporting guidance, not evidence that fidelity explains an individual trial result.
Keep reading
Don't stop at one method.
Good methods judgment comes from contrast. Read the neighboring guides, see where the assumptions diverge, and avoid treating every observational problem like it needs the same hammer.
The Trial Ended. Who Made It Into the Open-Label Extension?
A denominator-first audit of long-term extension findings: trace who enrolled, who stayed, and what an uncontrolled extension can actually establish.
Twenty Patients, Thirty Events: Read the Harms Denominator
A worked trial example separates the proportion of patients harmed from recurrent events per person-time, then shows why unequal follow-up needs more than a rate correction.
A 12-Month Benefit and a 30-Day Harm Are Not One Net-Benefit Number
A fictional trial timeline shows why benefit and harm counts measured over different windows cannot be silently collapsed into one net-benefit claim.