Sequential Multiple Assignment Randomized Trials for COMparing Personalized Antibiotic StrategieS (SMART-COMPASS): Design and Software
Open AccessPatient management is dynamic, a sequence of decisions with therapeutic adjustments made over time. Adjustments are personalized, tailored to individuals as new information becomes available. Strategies allowing for such adjustments are infrequently studied. Two major treatment decisions occur during the treatment of serious bacterial infections: empiric and definitive therapies. Empiric therapy selection is based on immediately available and often limited information upon recognition of the clinical syndrome. Definitive therapy is selected once organism identification and antibiotic susceptibility testing (AST) results are known, frequently 48-72 hours later than empiric therapy selection. The COMparing Personalized Antibiotic StrategieS (COMPASS) trial design aims to compare strategies consistent with clinical practice, decision-rules that guide empiric and definitive therapy decisions, and provides the opportunity to create and compare new strategies in a randomized setting using Sequential Multiple Assignment Randomized (SMART) COMPASS. SMART-COMPASS is a pragmatic trial design that mirrors clinical treatment decision-making and addresses the most relevant issue for treating patients: identifying the strategy that optimizes ultimate patient outcomes.However, several statistical challenges arise when designing a SMART-COMPASS for both continuous and binary endpoints including how to: appropriately estimate strategy means or proportions and associated standard errors within the context of sequential randomization, and identify the best strategy while controlling trial-wise error and related sample size and power calculations. Estimators of strategy means or proportions weighted in both individual and path levels, along with their expectations and variance-covariance, are discussed. The sequentially rejective test procedure, a data-driven multiple testing procedure to identify the best strategy, is necessary to control error probability as sequential randomization implies at least three strategies. Conjunctive power and type I error probability can be numerically calculated by using multivariate normal integrals and the overall sample size can then be determined through a grid-search. The behaviors of sample size, powers and type I error probability are evaluated with examples under several scenarios.Planning the sample size of a SMART-COMPASS is challenging as it requires an additional assumption on the AST results i.e., determination of resistance vs. susceptibility to initial regimen selection in addition to strategies’ sizes. The proportions of AST susceptible are associated with differences in strategy means or proportions and their variance-covariance and thus power and sample size. It is therefore natural to monitor the proportions of AST susceptible or resistant during a SMART-COMPASS. We discuss approaches to this issue, particularly in a blinded manner. We illustrate the approaches with hypothetical examples and provide guidance on their use.SMART-COMPASS R package and a user-friendly interactive Shiny application are developed. Basic features and implementation of the functions are described. Dynamic report output is customized based on input parameters.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.