3193 Improving Individual Clinical Outcomes in a Sequential Multiple Assignment Randomized Trial (SMART)

Abstract

OBJECTIVES/SPECIFIC AIMS: This work develops an algorithm that identifies patients in a Sequential Multiple Assignment Randomized Trial (SMART) who should switch treatments prior to the end of a stage because clinical effectiveness via their current intervention is unlikely. This algorithm uses as inputs patient baseline and interim measurements to assign a probability that a patient should switch or stay on their current intervention. First, the algorithm will be derived assuming both a linear and non-linear patient trajectory. Second, the performance of the algorithm will be assessed using trial data from the Establishing Moderators and Biosignatures of Antidepressant Response in Clinical Care† (EMBARC) study. The primary objective of the algorithm is to switch treatment in patients who will not reach clinical effectiveness by the end of the stage, and the secondary objective is to avoid accidentally switching treatment in patients who will reach clinical effectiveness by the end of the stage. †Trivedi et al. Journal of Psychiatric Research 78 (2016) 11-23 METHODS/STUDY POPULATION: First, the algorithm was derived assuming a linear or non-linear trajectory. Next, performance of the algorithm was assessed using data from the Establishing Moderators and Biosignatures of Antidepressant Response in Clinical Care† (EMBARC) study. This two-stage SMART design measured the effectiveness of sertraline in 242 patients with nonpsychotic Major Depressive Disorder (MDD). The algorithm was applied to baseline and interim measurements from the EMBARC study to predict end-stage Hamilton Depression (HAMD17) scores, the primary outcome of the study. True positive rate (TPR) and false positive rate (FPR) were used to measure respectively the primary study objective (switching treatment in patients who will not reach clinical effectiveness by the end of the stage), and the secondary study objective (avoiding accidentally switching treatment in patients who will reach clinical effectiveness by the end of the stage). TPR and FPR were calculated for the following prediction scenarios: (1) three separate two-point predictions: Baseline and Week 2, Baseline and Week 4, Baseline and Week 6, and (2) a single three-point prediction: Baseline and Weeks 2 and 6. †Trivedi et al. Journal of Psychiatric Research 78 (2016) 11-23 RESULTS/ANTICIPATED RESULTS: When using two-point prediction, we found TPR to increase and FPR to decrease as the interim measurements approached closer to the end of the stage. We also found TPR to increase when using a three-point prediction, but at the expense of FPR also increasing. Across these scenarios, TPR ranged between 70 and 90%, and FPR ranged between approximately 20 and 50%. DISCUSSION/SIGNIFICANCE OF IMPACT: Although SMART designs ultimately assign patients to more effective treatments, this process can take time and leave a patient (currently on an ineffective treatment) waiting until the end of a stage to try a potentially superior treatment. This disadvantage of the SMART design is currently addressed by this algorithm. By introducing a regression and likelihood approach to predict whether a patient should switch or stay on their current treatment, we move closer to the goal of designing rigorous, patient-centered studies. This work has the potential to improve individual clinical outcomes for patients enrolled in pragmatic clinical trials.