OnCo
ideasIdea

Run automated statistics and image checks on every cancer manuscript before review

Software can already spot impossible statistics, mismatched p-values and duplicated images in a paper. Journals should run these checks on every submission, as spell-check runs on every document.

Tools such as statcheck (recomputing p-values from reported test statistics), GRIM and SPRITE (checking whether means are possible given sample sizes) and image-duplication detectors identify errors and manipulation at scale. Some journals run image checks; almost none run statistical checks routinely. Integrating a pipeline at submission, with results shown to authors and reviewers, would catch a substantial fraction of errors before publication and deter fabrication.

Hypothesis
Routine checks flag actionable issues in at least 10% of oncology submissions, and journals using them see a lower subsequent correction and retraction rate than matched journals without them.
Rationale
statcheck found inconsistent p-values in about half of psychology papers, with a meaningful share changing significance; image duplication has been documented in about 4% of biomedical papers.
What would test it
Deploy the pipeline at two oncology journals for a year; report flag rates, author responses and downstream correction rates.
Maturity
early clinical
Who has to act
engineering
Cost to try
Small (under $1M)
Years to first evidence
1
Bottlenecks it attacks

Connected

5top