E-Newsletter - November 2019
Message from the Group Statistician


Leveraging Statistical, Methodological Research to Advance Alliance Initiatives: Alliance statistics and Data Management Center

At Alliance, the Statistics and Data Management Center (SDMC) continues on its journey in cutting edge statistical and methodological research, publishing extensively on novel statistical and bioinformatics methods, analyses and software, and providing statistical support and leadership on publications which leverage individual patient data from multiple Alliance studies. Take a glimpse into three such research endeavors that have a direct impact on Alliance research.

A sequential multiple assignment randomized trial (SMART) design and analysis is appropriate when comparing adaptive interventions defined by an individual’s sequence of treatment decisions that are guided by intermediate outcomes, according to Amy Ruppert Stark, PhD, Faculty Statistician for the Alliance Leukemia Committee and Alliance Lymphoma Committee. At the completion of a SMART, a rich data source exists to identify an optimal treatment strategy that, if followed by all patients in the population, would lead to the most beneficial outcome on average. SMART designs and analyses are often applied in social and behavioral science studies, but not prospectively used in oncology studies, particularly with time-to-event endpoints. The recently activated randomized phase III trial in older patients with CLL (Alliance A041702; NCT03737981) can be used to illustrate how a SMART design can be applied in the oncology setting with progression-free survival (PFS) as the primary endpoint. Alliance study A041702 initially randomizes patients to a standard doublet or more intensive triplet therapy as first-stage treatment. [1] Complete remission without minimal residual disease (CR MRD negative) is an intermediate outcome that can be utilized to make subsequent treatment decisions. In patients randomized to triplet therapy, those achieving CR MRD negative status discontinue therapy whereas those without CR MRD negative status receive maintenance therapy. All patients randomized to doublet therapy continue with maintenance. The primary objective is to determine whether more intensive therapy followed by response-driven discontinuation has superior PFS compared to doublet therapy plus maintenance. A secondary objective determines whether CR MRD negative rates are significantly higher with triplet versus doublet therapy. A slight modification of the current design, via re-randomizing patients achieving CR MRD negative status with triplet therapy to either maintenance or observation, results in the general structure of a SMART design. [2] Through this modification and use of a SMART design, not only can the primary and secondary objective of the current design be addressed, but we can also answer whether triplet therapy with maintenance has superior PFS versus doublet therapy with maintenance and whether CR MRD negative status is associated with PFS when adjusting for initial therapy. The SMART design provides the data required to answer a variety of questions, some of which cannot be addressed when using a randomized two-arm treatment strategy design, but at the cost of a modest increase in sample size. In this example, sample size increased by 13 percent when using a SMART design with a re-randomization component. This alternative design strategy has since been published in the context of Alliance A041702; SMART designs are a new concept in oncology that should be considered when designing a trial with adaptive interventions. [2]

[1] Woyach JA, Ruppert AS, Mandrekar S, Perez G, Booth A, Feldman D, Dib E, Jatoi A, Le-Rademacher J, Heerema N, Lozanski G, Little R, Ding W, Stone R, Byrd JC. Alliance A041702: A Randomized Phase III Study of Ibrutinib Plus Obinutuzumab Versus Ibrutinib Plus Venetoclax and Obinutuzumab in Untreated Older Patients (≥ 70 Years of Age) with Chronic Lymphocytic Leukemia (CLL). Abstract submitted Aug 2019.
[2] Ruppert AS, Yin J, Davidian M, Tsiatis AA, Byrd JC, Woyach JA, Mandrekar SJ. Application of a sequential multiple assignment randomized trial (SMART) design in older patients with chronic lymphocytic leukemia. Ann Oncol 30(4):542-550, 2019.

Within healthcare delivery research (HDR), evidence that can be used to improve clinical practice patterns is sought and frequently generated through standard, parallel-arms cluster randomized trial (CRT) designs that test interventions implemented at the clinical practice (center)-level and are characterized by the randomization of centers to interventions. Although the primary endpoint of these trials is often a center-level outcome, patient-level factors may vary between centers and, consequently, may influence the center-level outcome, according to David Zahrieh, PhD, Faculty Statistician for the Alliance Cancer Care Delivery Research Committee. Furthermore, there may be important factors that predict the variation in the center-level outcome and this knowledge can help contextualize the trial results and inform practice patterns. The symbolic two-step method that applies symbolic data analysis to account for patient-level factors when estimating and testing a center-level effect on both the average center-level outcome and its variation was developed for such settings. [1, 2] This method was compared to the single-stage data analysis approaches and sought to extend it to prospectively size a CRT so that the application of our method in data analysis is consistent with the design. The method performed well compared with the single-stage models studied, but unlike those models, the method allowed the data analyst to model the within-center variation to identify predictive factors of that variation. The proposed formulaic approach to sample size planning within the cluster-sampling framework incorporated this knowledge and accounted for patient-level characteristics; furthermore, the sample size approximation performed well in many different settings [3, submitted manuscript].  The symbolic two-step method provides an alternate approach in both the design and analysis of CRTs for HDR.

[1]  Le-Rademacher J, Billard L. Likelihood functions and some maximum likelihood estimators for symbolic data. Journal of Statistical Planning and Inference. 2011;141:1593-602.
[2]  Le-Rademacher J. A center-level approach to estimating the effect of center characteristics on center outcomes. Advances in the Mathematical Sciences:  Research from the 2015 Association for Women in Mathematics Symposium. 2016:301-21.
[3] Zahrieh D, Le-Rademacher J.  Sample size planning in the design and analysis of cluster randomized trials in healthcare delivery research using the symbolic data framework.  Manuscript submitted Aug 2019.

For clinical trials utilizing a time-to-event variable as the primary endpoint, monitoring the number of events is crucial since the interim analysis and final primary endpoint analysis timing depends on the number of events observed in a clinical trial, according to Fang-Shu Ou, PhD, Faculty Statistician for the Alliance Gastrointestinal Committee. Methods [1-3] have been developed previously for predicting the time to observing a predefined number of events (referred to as a “milestone” hereafter). These methods require careful consideration and comparisons between various models, thus requiring subjective statistical input for each prediction and can be difficult to be bundled into a more automated process. To aid clinical trial monitoring, combining the predictions from several models was suggested to come up with a better prediction by using prediction synthesis. The amalgamation of simple individual models leads to an easily implemented and largely automated process without the need for extensive modeling. The implementation of prediction synthesis for milestone prediction and its performance can be found in the published manuscript. [4]  An R shiny app was also developed to carry out the prediction and is available to the public at https://rtools.mayo.edu/a_milestone_prediction/

[1] Rubinstein LV, Gail MH, Santner TJ. Planning the duration of a comparative clinical trial with loss to follow-up and a period of continued observation Journal of chronic diseases. 1981;34:469–479.
[2] Bagiella E, Heitjan DF. Predicting analysis times in randomized clinical trials Statistics in medicine. 2001;20:2055–2063.
[3] Chen TT. Predicting analysis times in randomized clinical trials with cancer immunotherapy BMC medical research methodology. 2016;16:12.
[4] Fang‐Shu Ou, Martin Heller, Qian Shi; Milestone prediction for time‐to‐event endpoint monitoring in clinical trials.  Pharmaceutical statistics. 2019; 18:433-446.
 




Sumithra J. Mandrekar
Alliance Group Statistician

 

For other articles in this issue of the Alliance E-News newsletter, see below.