OBJECTIVE: Heterogeneity of patients is a common problem in randomized controlled trials (RCTs) in various fields of clinical research. We aimed to investigate the potential benefits of different approaches for dealing with heterogeneity in a case study on traumatic brain injury (TBI). DESIGN AND SETTING: Statistical modeling studies in three surveys and six randomized controlled trials. PATIENTS: Individual patient data (n = 8033) from the IMPACT database. INTERVENTIONS: We investigated the statistical power and efficiency of randomized controlled trials (RCTs) in relation to (1) selection according to baseline characteristics, (2) prognostic targeting (i.e., excluding those with a relatively extreme prognosis), and (3) covariate-adjusted analysis. Statistical power was expressed as the required sample size for obtaining 80% power and efficiency as the relative change in study duration, reflecting both gains in power and adverse effects on recruitment. Uniform and targeted treatment effects were simulated for 6 month unfavorable outcome. RESULTS: For a uniform treatment effect, selection resulted ina sample size reduction of 33% in the surveys and 5% in the RCTs, but decreased recruitment by 65% and 41%, respectively. Hence, the relative study duration was prolonged (surveys: 95%; RCTs: 60%). Prognostic targeting resulted in sample size reductions of 28% and 17%, and increased relative study duration by 5% in surveys and 11% in the RCTs. Covariate adjustment reduced sample sizes by 30% and 16%, respectively, and did not affect recruitment. For a targeted treatment effect, the sample size reductions by selection (surveys: 47%; RCTs: 20%) and prognostic targeting (surveys: 49%; RCTs: 41%) were larger and adverse effects on recruitment smaller. CONCLUSIONS: The benefits of selection and prognostic targeting in terms of statistical power are reversed by adverse effects on recruitment. Covariate adjusted analysis in a broadly selected group of patients is advisable if a uniform treatment effect is assumed, since there is no decrease in recruitment.
OBJECTIVE: Heterogeneity of patients is a common problem in randomized controlled trials (RCTs) in various fields of clinical research. We aimed to investigate the potential benefits of different approaches for dealing with heterogeneity in a case study on traumatic brain injury (TBI). DESIGN AND SETTING: Statistical modeling studies in three surveys and six randomized controlled trials. PATIENTS: Individual patient data (n = 8033) from the IMPACT database. INTERVENTIONS: We investigated the statistical power and efficiency of randomized controlled trials (RCTs) in relation to (1) selection according to baseline characteristics, (2) prognostic targeting (i.e., excluding those with a relatively extreme prognosis), and (3) covariate-adjusted analysis. Statistical power was expressed as the required sample size for obtaining 80% power and efficiency as the relative change in study duration, reflecting both gains in power and adverse effects on recruitment. Uniform and targeted treatment effects were simulated for 6 month unfavorable outcome. RESULTS: For a uniform treatment effect, selection resulted ina sample size reduction of 33% in the surveys and 5% in the RCTs, but decreased recruitment by 65% and 41%, respectively. Hence, the relative study duration was prolonged (surveys: 95%; RCTs: 60%). Prognostic targeting resulted in sample size reductions of 28% and 17%, and increased relative study duration by 5% in surveys and 11% in the RCTs. Covariate adjustment reduced sample sizes by 30% and 16%, respectively, and did not affect recruitment. For a targeted treatment effect, the sample size reductions by selection (surveys: 47%; RCTs: 20%) and prognostic targeting (surveys: 49%; RCTs: 41%) were larger and adverse effects on recruitment smaller. CONCLUSIONS: The benefits of selection and prognostic targeting in terms of statistical power are reversed by adverse effects on recruitment. Covariate adjusted analysis in a broadly selected group of patients is advisable if a uniform treatment effect is assumed, since there is no decrease in recruitment.
Authors: Elizabeth L Turner; Pablo Perel; Tim Clayton; Phil Edwards; Adrian V Hernández; Ian Roberts; Haleema Shakur; Ewout W Steyerberg Journal: J Clin Epidemiol Date: 2011-12-09 Impact factor: 6.437
Authors: Alexandros L Georgiadis; Yuko Y Palesch; David Zygun; J Claude Hemphill; Claudia S Robertson; Peter D Leroux; Jose I Suarez Journal: Neurocrit Care Date: 2015-06 Impact factor: 3.210
Authors: Michael O Harhay; Jason Wagner; Sarah J Ratcliffe; Rachel S Bronheim; Anand Gopal; Sydney Green; Elizabeth Cooney; Mark E Mikkelsen; Meeta Prasad Kerlin; Dylan S Small; Scott D Halpern Journal: Am J Respir Crit Care Med Date: 2014-06-15 Impact factor: 21.405
Authors: Andrew I R Maas; Ewout W Steyerberg; Anthony Marmarou; Gillian S McHugh; Hester F Lingsma; Isabella Butcher; Juan Lu; James Weir; Bob Roozenbeek; Gordon D Murray Journal: Neurotherapeutics Date: 2010-01 Impact factor: 7.620
Authors: Roelof Risselada; Hester F Lingsma; Andrew J Molyneux; Richard S C Kerr; Julia Yarnold; Mary Sneade; Ewout W Steyerberg; Miriam C J M Sturkenboom Journal: BMC Med Res Methodol Date: 2010-09-29 Impact factor: 4.615