1. Interpretation of subgroup results in clinical trial publications: insights from a survey of medical specialists in Ontario, Canada
- Author
-
Andrea B. Parker and C. David Naylor
- Subjects
Male ,Pediatrics ,medicine.medical_specialty ,Cross-sectional study ,Research methodology ,Decision Making ,MEDLINE ,Cardiology ,law.invention ,Randomized controlled trial ,law ,Surveys and Questionnaires ,Medicine ,Humans ,Practice Patterns, Physicians' ,Randomized Controlled Trials as Topic ,Ontario ,business.industry ,Interpretation (philosophy) ,Patient Selection ,Clinical trial ,Harm ,Cross-Sectional Studies ,Family medicine ,Female ,Cardiology and Cardiovascular Medicine ,business ,Ontario canada - Abstract
Clinicians routinely apply randomized trial evidence to their patients who meet study selection criteria. However, little is known about how clinicians interpret conflicting subgroup data.We mailed a self-administered survey to all practicing cardiologists (n = 309) and 695 randomly chosen other specialists in Ontario, Canada. The survey presented 6 hypothetical trials where a subgroup result deviated from the overall result. We also elicited responses to some general statements about clinical evidence and subgroups.Completed surveys were received from 435 physicians (44%). Faced with overall benefit but no apparent treatment effect in a subgroup, almost 44% would exclude subgroup-type patients, notwithstanding the hazard of beta error. Given overall harm but significant benefit for a subgroup, responses were split approximately 60:40 between continuing conventional therapy for all and treating subgroup-type patients with the new drug. For an overall null result with a positive treatment-subgroup interaction term, 25% of respondents would continue conventional therapy, whereas 69% would adopt the new drug for subgroup-type patients. Physicians with an academic appointment, devoting more time to research, or with formal training in research methodology were more likely to ignore subgroups unless a treatment-subgroup interaction term was significant (P values ranging from .018 to.0001). Asked if in general they paid special attention to individual subgroup results, respondents were again divided with 37.5% agreeing, 39.5% disagreeing, and the rest undecided.Clinicians disagree sharply in interpretation of clinical trials when the overall and subgroup results diverge. Clearer guidelines are needed for undertaking, reporting, and interpreting subgroup analyses.
- Published
- 2005