Clinical Trials 3
Presenter: Ling Li
When: Thursday, July 14, 2016 Time: 4:30 PM - 6:00 PM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
CHALLENGES IN THE DESIGN AND ANALYSIS OF STEPPED-WEDGE CLUSTER RANDOMIZED CONTROLLED TRIALS TO EVALUATE THE EFFECTIVENESS AND SAFETY OF ELECTRONIC MEDICAL RECORD SYSTEMS
Cluster randomized controlled trials (CRCTs) are increasingly being used in the health care context to evaluate interventions when it is inappropriate or impossible to use individual randomization. CRCTs commonly use a parallel group design, in which the clusters are randomized to either the intervention or control arm of the study. It is often regarded as unethical to withhold an intervention from a proportion of participants if it is believed that the intervention will do more good than harm. Stepped wedge trial designs, where the intervention is delivered sequentially to all trial clusters over a number of time periods, is an alternative to the traditional parallel groups design. Electronic medical record systems, containing patients medical and clinical data, are expected to improve quality of care delivery in hospitals, e.g. reduce medication errors and increase the appropriate prescribing of antibiotics. However, rigorous evidence demonstrating these effects is limited. The objectives of this paper are: 1) to describe the design and analysis of stepped-wedge cluster randomized controlled trials (SW-CRCTs) for evaluating the effectiveness of electronic medical record systems, and 2) to discuss the practical and methodological challenges of SW-CRCTs, drawing on a specific case study in pediatrics. In SW-CRCTs, the order in which the clusters, e.g. wards or hospitals, receive the electronic systems is randomized, and by the end of the study all clusters will have the systems implemented. Sample size calculations take into account the estimated between-cluster variance and the design effect associated with the stepped-wedge design. The main outcome of interest may include changes in specific outcomes such as medication error rate or repeat test rate following implementation of the systems. For each outcome of interest, data collected across all measurement periods and all study steps are used in the analyses comparing intervention status (pre versus post intervention). The analyses include multiple time points for both pre and post intervention. The study design potentially also allows determination of temporal changes in system effectiveness, e.g. to determine if error rates continue to decline over time.
Clinical Trials 3
Presenter: Sumithra Mandrekar
When: Thursday, July 14, 2016 Time: 4:30 PM - 6:00 PM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
Resampling based methods to assess the clinical Utility of Phase II endpoints based on tumor measurements to predict overall survival outcomes in Phase III Trials
Phase II (P2) clinical trials inform go/no-go decisions for proceeding to phase III (P3) trials, and appropriate end points in P2 trials are critical for facilitating this decision. P2 solid tumor trials have traditionally used end points defined by the Response Evaluation Criteria for Solid Tumors (RECIST). We previously reported that absolute and relative changes in tumor measurements demonstrated potential, but not convincing, improvement over RECIST to predict overall survival (OS). In this work, we used resampling methods to assess the clinical utility of metrics to predict phase III outcomes from simulated phase II trials. 2,000 phase II trials were simulated from four actual phase III trials (two positive for OS and two negative for OS). Cox models for absolute and relative change, and RECIST tumor response were fit for each phase II trial. Clinical utility was assessed by positive and negative predictive negative predictive value, by prediction error, and by concordance index. Absolute and relative change metrics had higher positive and negative predictive values than RECIST in five of the six treatment comparisons and lower prediction error curves in all six. However, the differences were negligible. No statistically significant difference in c-index across metrics was found. Despite work in the literature suggesting that continuous metrics have better promise than categorical metrics in predicting OS, our analyses demonstrated that the absolute and relative change metrics are not meaningfully better than RECIST based categorical metrics in predicting OS.
Clinical Trials 3
Presenter: Timothy NeCamp
When: Thursday, July 14, 2016 Time: 4:30 PM - 6:00 PM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
Cluster-level adaptive interventions and sequential, multiple assignment, randomized trials: Estimation and sample size considerations
Individual-level adaptive interventions, also known as dynamic treatment regimens, are used to guide the sequencing of intervention decisions based on an individuals changing health information, including responsiveness to prior interventions. Sequential, multiple assignment, randomized trials (SMARTs) are a particular type of multi-stage trial design used to develop high-quality adaptive interventions. This talk introduces cluster-level adaptive interventions in which sequential intervention decisions are, for example, made at the clinic or classroom level. We present cluster-level SMART designs in which randomization occurs at the cluster level and outcomes are at the individual-level (e.g. entire schools randomized to receive a school-wide intervention, while student level outcomes are the outcomes of interest). We develop an easy-to-use weighted least squares regression approach for comparing cluster-level adaptive interventions embedded in a SMART. In addition, we develop a sample size calculator to be used by researchers designing clustered randomized SMARTs. We develop the sample size formulae for two common types of two-stage SMART designs. The sample size is expressed as a function of the effect size, Type-I error, statistical power, rate of response to first-stage treatment, within cluster correlation, and correlation between baseline cluster-level covariates and individual outcomes. The validity and robustness of the calculator under various settings, including when working assumptions are violated, are evaluated through simulation. To illustrate our methods we utilize Adaptive Implementation of Effective Programs Trial (ADEPT), a cluster randomized SMART currently being conducted in which community-based mental health clinics are assigned to one of three embedded cluster-level adaptive interventions. The interventions aim to increase a clinics adoption of an evidence-based practice for mood disorders and, in turn, improve patient-level mental health outcomes.
Clinical Trials 3
Presenter: Christian Ritz
When: Thursday, July 14, 2016 Time: 4:30 PM - 6:00 PM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
Simultaneous inference from separate mixed model fits - an alternative to joint modelling
In recent years joint modelling of various types of mixed outcomes has received a lot of attention. The resulting models are complex in order to suitably describe correlations between outcome variables of intrinsically different types, e.g., a binary and a continuous outcome. Similarly, non-standard and complex joint modelling approaches are needed for joint modelling of continuous outcomes recorded on different scales. We propose a novel and versatile framework for simultaneous inference on parameters that are estimated from linear and generalized linear mixed models, fitted separately to a number of outcomes within the same study (one model fit per outcome). By combining asymptotic representations of parameter estimates from separate model fits we derive the joint asymptotic normal distribution for all parameter estimates of interest for all outcomes considered. Intrinsic features of the estimation procedure for mixed models are exploited. This result enables the construction of simultaneous confidence intervals and calculation of adjusted p-values, but only using the separate model fits. Simultaneous coverage is demonstrated through simulation. Comparisons to a joint modelling approach are also shown. The work is a generalization of the approach proposed by Pipper et al. (2012) and Pallmann et al. (2016). Finally, we discuss the proposed methodology, which solely relies on standard models, as an attractive and flexible alternative to various joint modelling approaches in general. References: Pallmann, P., Pretorius, M., Ritz, C. (2016). Simultaneous comparisons of treatments at multiple time points for small samples: combined marginal models versus joint modeling. To appear in Statistical Methods in Medical Research Pipper, C. B., Ritz, C., Bisgaard, H. (2012). A versatile method for confirmatory testing of the effects of a covariate in multiple models. Journal of the Royal Statistical Society, Series C (Applied Statistics) 61, 315--326
Clinical Trials 3
Presenter: Annette Kopp-Schneider
When: Thursday, July 14, 2016 Time: 4:30 PM - 6:00 PM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
Outcome-adaptive interim monitoring in Phase II trials: Criteria for early stopping
A phase II trial is typically a small-scale study to determine whether one or more experimental treatments should continue further clinical evaluation. In this setting, interim analyses are commonly performed to allow for early stopping, either for futility and/or efficacy. The use of Bayesian posterior and predictive probabilities as decision rule for early stopping has been suggested in the last decade (e.g., Berry et al 2013, Saville et al 2014), especially for the context of biomarker-targeted therapies with small numbers of patients. We investigate the operating characteristics of a number of stopping criteria for a one-arm trial with dichotomous endpoint. Criteria are based on the Bayesian posterior probability that response probability exceeds a prespecified threshold, either the response rate under standard of care or a response rate that is perceived as interestingly efficacious. In addition, Bayesian predictive probabilities will be evaluated. The influence of the choice of the Bayesian model and the prior distribution on the performance of the design will be discussed. The final choice of stopping criterion will depend on the principal investigators preference on the basis of the designs operating characteristics. For illustration, the trial design for the INFORM2 phase I/II trial series addressing individualized therapy for relapsed malignancies in childhood will be presented. References: Berry SM et al (2013), Clin Trials 10: 720-34 Saville BR et al (2014), Clin Trials 11: 485-93
Clinical Trials 3
Presenter: Svenja Schueler
When: Thursday, July 14, 2016 Time: 4:30 PM - 6:00 PM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
Choice of futility boundaries for group sequential designs with two endpoints
In clinical trials, the opportunity for an early stop during an interim analysis (either for efficacy or for futility) may relevantly save time and financial resources. This is especially important, if the planning assumptions required for power calculation are based on a low level of evidence. For example, when including two primary endpoints in the confirmatory analysis, the power of the trial depends on the effects of both endpoints and on their correlation. Assessing the feasibility of such a trial is therefore difficult, as the number of parameter assumptions to be correctly specfied is large. For this reason, so-called 'group sequential designs' are of particular importance in this setting. Whereas the choice of adequate boundaries to stop a trial early for efficacy has been broadly discussed in the literature, the choice of optimal futility boundaries has not been investigated so far, although this may have serious consequences with respect to performance characteristics. In this work, we propose a general method to construct 'optimal' futility boundaries according to predefined criteria. Further, we present three different group sequential designs for two (co-primary) endpoints applying these futility boundaries. Our methods are illustrated by a real clinical trial example and by Monte-Carlo simulations.
