Pre-specified vs post-hoc analysis
Whether an analysis was planned and written down before the data were seen (pre-specified) or invented afterwards (post-hoc). Post-hoc findings, especially in subgroups, are prone to be chance and count as hypotheses for a new trial, not as evidence.
Trial protocols and statistical analysis plans, registered in advance (ClinicalTrials.gov, EU CTR), list the endpoints, populations and tests that will count; regulators approve on pre-specified analyses and treat post-hoc results as exploratory with 'nominal' P values. Post-hoc subgroup analyses have produced both misleading claims (retrospective PD-L1 cut-offs) and genuine discoveries later confirmed (KRAS wild-type for cetuximab). Pre-registration is standard in clinical trials and the site's ideas argue for extending it to animal studies and to 'silent trial' evaluations of AI, where the lack of pre-registration is a known source of false positives.
Pages like this
not linked directly; found by shared links- TermFutility analysis (stopped for futility)
Shares Statistical significance (P values, alpha, multiplicity), Primary, secondary and co-primary endpoints.
- TermNon-inferiority trial
Shares Intention-to-treat (ITT) and per-protocol analysis, Statistical significance (P values, alpha, multiplicity).
- TermInterim analysis, readout and data cut-off
Shares Statistical significance (P values, alpha, multiplicity), Primary, secondary and co-primary endpoints.