The manuscripts

Over the past year we edited manuscripts in neurointerventional surgery, critical care, respiratory medicine and developmental psychology. The fields, journals and evidence standards differed. The same three structural failures kept appearing.

1. The conclusion went beyond what the data could support

A single-centre study (n = 43) concluded by recommending a specific device oversizing threshold. “Our cohort supports X” and “clinicians should adopt X” are separated by the whole evidence hierarchy.

A paediatric study of sleep and executive function (n = 204) had a sound cross-sectional design. It was written as if it had established causal relationships. It had not.

The fix in both cases was to match each claim to the study design. Randomised controlled trials recommend. Retrospective cohorts suggest. Single-centre case series observe. These verbs are not a matter of style. Each one states how much evidential weight the design can carry.

2. Internal contradictions were left unresolved

A postpartum case report on hereditary haemorrhagic telangiectasia (HHT) showed that inhaled iloprost produced rapid haemodynamic improvement, while conventional vasodilators had limited effect. The discussion listed each treatment “failure” but did not explain why patients with an ALK1 mutation might respond differently.

A WEB device manuscript argued for a stent-free technique. Its own data showed that some cases needed rescue stenting. Left unaddressed, the data appeared to weaken the paper’s own argument.

Both manuscripts needed the contradiction built into the argument. Rescue stenting was presented as a planned, bounded fallback within the risk-benefit balance. The treatment failures in the HHT case were presented as mechanistic information that supports the conclusion.

Reviewers do not expect clean data. They expect consistent reasoning. An unresolved contradiction is a problem of argument structure, not of data.

3. The statistics were reported but did not support a claim

A retrospective ARDS study (n = 38) used end-expiratory lung volume as a prognostic indicator. Its methods were sound: linear mixed models, repeated measures and appropriate handling of missing data. The interpretation claimed prognostic value without citing effect sizes. In the WEB manuscript, odds ratios from a multivariable regression were reported without explaining why that analysis was appropriate.

Two questions resolve this. Which clinical claim does this result support? Does the effect size appear where that claim is made? The statistics do their job only after both questions are answered.

What this means for manuscript support

These failures recurred across four unrelated fields because they are structural. None can be seen by reading for fluency. All become visible when the manuscript is read as a reviewer reads it: checking whether the argument holds from evidence to conclusion.

High-quality manuscript support is therefore a logic review, not only translation or proofreading. It should be done by an editor who knows both the field and the structure of an argument, acting as the most demanding first reviewer before the manuscript reaches peer review.