++
CLINICAL SCENARIO
You are a physician working at a regional trauma center. Your unit's committee, which is responsible for standardization of care, is considering using tranexamic acid to treat trauma patients who arrive 3 hours after injury. Almost all of the information on this topic is derived from a single blinded trial that randomized trauma patients to tranexamic acid or placebo.1
The original publication reported that 99% of the enrolled patients were followed up and there was a reduction in all-cause mortality (relative risk [RR], 0.91; 95% confidence interval [CI], 0.85-0.97) with no apparent subgroup effect.1 A subsequent publication2 focused on an additional analysis that addressed death from bleeding and reported a powerful subgroup effect with a large benefit for patients treated within 3 hours of injury and possible harm if treated 3 or more hours after injury. The committee's mandate is to decide whether tranexamic acid should not be given to patients 3 hours or more after injury. The credibility you place on the subgroup analysis will determine your decision.
+++
The Challenge of Subgroup Analysis
++
Clinicians making treatment decisions use evidence applying most closely to the individual patient and treatment under consideration. To address this issue, clinical trialists and authors of systematic reviews with meta-analyses frequently conduct subgroup analyses to identify groups of patients (ie, sicker patients) who may respond differently to treatment than other groups (ie, less sick patients) or find more and less effective ways of administering treatment (eg, intravenous vs oral).3,4 Although subgroup analyses may help individualize treatment, they may also mislead clinicians.
++
For example, the Second International Study of Infarct Survival (ISIS-2) investigators reported an apparent subgroup effect: patients presenting with myocardial infarction born under the zodiac signs of Gemini or Libra did not experience the same reduction in vascular mortality attributable to aspirin that patients with other zodiac signs had (Table 25.2-1).5 Despite the statistical significance of the test for interaction (the probability that the difference in the effect of aspirin in Gemini and Libra patients vs those born under other zodiac signs was 3/1000), the investigators did not believe the subgroup effect, and they reported the results to indicate the dangers of subgroup analysis. Table 25.2-2 lists 19 examples in which other randomized clinical trial (RCT) authors have, when faced with biologically more plausible effects, claimed subgroup effects unsupported by subsequent evidence.
+++...