A randomized trial of a new asthma medication in 400 patients shows no difference in FEV1 decline (primary endpoint: -120 mL/year in both groups; p=0.18). Investigators perform 40 unplanned subgroup analyses and report a significant benefit in left-handed women older than 70 years (n=12; p=0.04), with no prior mechanistic hypothesis. Baseline characteristics were balanced between groups. Which statistical problem most directly threatens validity of this subgroup finding?
- A)Data dredgingGABARITO
- B)Lead time bias
- C)Spectrum bias
- D)Confounding by indication
- E)Healthy worker effect
Explicação
Data dredging, also called p hacking, occurs when many unplanned analyses are performed until a statistically significant result appears by chance. The more comparisons that are made, the greater the false positive risk. Post hoc subgroup findings should there... Ver explicação completa e trilha adaptativa →