Background: Standard randomized trials are designed to detect the average treatment effect for a population. They may fail to detect important differences in benefits and harms for subpopulations. The impact is that treatment recommendations based on the results of standard trial designs may be suboptimal, leading to poor patient outcomes and wasting healthcare resources. This problem affects virtually all disease areas, since it stems from how randomized trials, the gold standard for evaluating treatments, are currently designed and analyzed.
Objectives: Randomized trial designs that adaptively change enrollment criteria during a trial, called adaptive enrichment designs, have potential to provide improved information about which subpopulations benefit from new treatments. According to the PCORI draft Methodology Report (Chapter 8, 2012), "Adaptive designs are particularly appealing for PCOR because they have the potential to maintain many of the advantages of randomized clinical trials while minimizing some of the disadvantages" and "researchers are looking for guidance on how to design and conduct PCOR adaptive trials." We will develop new statistical methods and a freely available software tool enabling investigators to construct new adaptive designs tailored to answer their specific research questions. The software will automatically compare a wide variety of designs and recommend those with the best performance. The resulting improved designs will generate crucial information directly relevant to decisions of patients, clinicians, and regulators. Specifically, our designs aim to determine treatment benefits and harms for subpopulations defined by a risk factor such as age, sex, or disease severity measured at baseline.
Methods: Our designs will be implemented in user-friendly, freely available, trial-planning software, in order to make the methods widely available to clinical investigators. We will conduct extensive simulation studies to determine how our new designs perform as we vary different design features. We will also conduct simulations based on data from randomized trials of interventions for stroke and HIV. Anticipated Impact: Our new designs will generate crucial information directly relevant to patient and clinician decisions about benefits and harms of new treatments. This is because our designs aim to draw finer-grained conclusions about which patients, based on their personal characteristics such as age, sex, or disease severity, benefit from different treatments.
Díaz I, Colantuoni E, Rosenblum M. Enhanced precision in the analysis of randomized trials with ordinal outcomes. Biometrics. 2015 Nov 17. doi: 10.1111/biom.12450. [Epub ahead of print] PubMed PMID: 26576013. (Abstract only available)
Díaz I, Rosenblum M. Targeted Maximum Likelihood Estimation using Exponential Families. Int J Biostat. 2015 Nov 1;11(2):233-51. doi: 10.1515/ijb-2014-0039. PubMed PMID: 26197469. (Abstract only available)
Webb AJ, Ullman NL, Morgan TC, Muschelli J, Kornbluth J, Awad IA, Mayo S, Rosenblum M, Ziai W, Zuccarrello M, Aldrich F, John S, Harnof S, Lopez G, Broaddus WC, Wijman C, Vespa P, Bullock R, Haines SJ, Cruz-Flores S, Tuhrim S, Hill MD, Narayan R, Hanley DF; MISTIE and CLEAR Investigators. Accuracy of the ABC/2 Score for Intracerebral Hemorrhage: Systematic Review and Analysis of MISTIE, CLEAR-IVH, and CLEAR III. Stroke. 2015 Sep;46(9):2470-6. doi: 10.1161/STROKEAHA.114.007343. Epub 2015 Aug 4. PubMed PMID: 26243227; PubMed Central PMCID: PMC4550520. (Abstract only available)
Colantuoni E, Rosenblum M. Leveraging prognostic baseline variables to gain precision in randomized trials. Stat Med. 2015 Aug 15;34(18):2602-17. doi: 10.1002/sim.6507. Epub 2015 Apr 14. PubMed PMID: 25872751. (Abstract only available)