Editor—We agree with Dickinson et al1 that larger and better designed randomised controlled trials are necessary to detect benefits of treatment in head injury.2 But increasing the sample size is not the only solution to show efficacy. The statistical power of a study can also be improved by randomising the same number of patients but taking prognostic factors, such as age or Glasgow coma scale, into account.
Firstly, one might limit the inclusion of patients to those with an intermediate prognosis—for example, between 20% and 80% probability of a favourable outcome.3 This leads to a focus on patients for whom treatment effects can be well determined. For the same power, a reduction in sample size of 30% might be achievable.3 After showing efficacy in the intermediate risk group, additional funding may be raised more easily to study patients with a poorer or better prognosis. Note that this reasoning assumes that the relative effect of a treatment is constant across risk groups. This is in contrast to the assumption of an absolute effect of 5% as discussed by Dickinson et al. Such an absolute effect is comparatively large at a baseline incidence of 20%, as indicated by an odds ratio of 0.71 for the comparison of an incidence of 15% versus 20%. In contrast, the odds ratio is 0.82 for the same absolute effect at 50% baseline incidence (45% v 50%). The absolute effect of 5% is more easily detected at a baseline incidence of 20%, while a relative effect such as an odds ratio of 0.71 is more easily detected at an incidence of 50%. So the assumption of an absolute or relative effect is crucial in reasoning about power and inclusion criteria.
Secondly, even if inclusion would be limited to patients at intermediate risk, heterogeneity will remain regarding the probability of a favourable outcome. Predictive characteristics which account for this heterogeneity can be adjusted for in the analysis, which will increase the statistical power to detect a treatment effect.4 In an analysis of patients with acute myocardial infarction, the potential reduction in sample size was 12% by adjustment for age.5
Besides dealing with heterogeneity, we may also consider restricting data collection to the essential variables, such that larger numbers of patients are accrued at the same costs. We hope that the strategies here proposed will be applied in the study of therapy for head injury, together with an increase in funding.
References
- 1.Dickinson K, Bunn F, Wentz R, Edwards P, Roberts I. Size and quality of randomised controlled trials in head injury: review of published studies. BMJ. 2000;320:1308–1311. doi: 10.1136/bmj.320.7245.1308. . (13 May.) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Maas AIR, Steyerberg EW, Murray GD, Bullock R, Baethmann A, Marshall LF, Teasdale GM. Why have recent trials of neuroprotective agents in head injury failed to show convincing efficacy? A pragmatic analysis and theoretical considerations. Neurosurgery. 1999;44:1286–1298. [PubMed] [Google Scholar]
- 3.Machado SG, Murray GD, Teasdale GM. Evaluation of designs for clinical trials of neuroprotective agents in head injury. J Neurotraum. 1999;16:1131–1138. doi: 10.1089/neu.1999.16.1131. [DOI] [PubMed] [Google Scholar]
- 4.Robinson LD, Jewell NP. Some surprising results about covariate adjustment in logistic-regression models. Int Stat Rev. 1991;59:227–240. [Google Scholar]
- 5.Steyerberg EW, Bossuyt PMM, Lee KL. Clinical trials in acute myocardial infarction: Should we adjust for baseline characteristics? Am Heart J. 2000;139:745–751. doi: 10.1016/s0002-8703(00)90001-2. [DOI] [PubMed] [Google Scholar]