In randomized clinical trials, analyses of time-to-event data without risk stratification, or with stratification based on pre-selected factors revealed at the end of the trial to be at most weakly ...associated with risk, are quite common. We caution that such analyses are likely delivering hazard ratio estimates that unwittingly dilute the evidence of benefit for the test relative to the control treatment. To make our case, first, we use a hypothetical scenario to contrast risk-unstratified and risk-stratified hazard ratios. Thereafter, we draw attention to the previously published 5-step stratified testing and amalgamation routine (5-STAR) approach in which a pre-specified treatment-blinded algorithm is applied to survival times from the trial to partition patients into well-separated risk strata using baseline covariates determined to be jointly strongly prognostic for event risk. After treatment unblinding, a treatment comparison is done within each risk stratum and stratum-level results are averaged for overall inference. For illustration, we use 5-STAR to reanalyze data for the primary and key secondary time-to-event endpoints from three published cardiovascular outcomes trials. The results show that the 5-STAR estimate is typically smaller (i.e. more in favor of the test treatment) than the originally reported (traditional) estimate. This is not surprising because 5-STAR mitigates the presumed dilution bias in the traditional hazard ratio estimate caused by no or inadequate risk stratification, as evidenced by two detailed examples. Pre-selection of stratification factors at the trial design stage to achieve adequate risk stratification for the analysis will often be challenging. In such settings, an objective risk stratification approach such as 5-STAR, which is partly aligned with guidance from the US Food and Drug Administration on covariate-adjustment in clinical trials, is worthy of consideration.
Personalized medicine, a paradigm of medicine tailored to a patient's characteristics, is an increasingly attractive field in health care. An important goal of personalized medicine is to identify a ...subgroup of patients, based on baseline covariates, that benefits more from the targeted treatment than other comparative treatments. Most of the current subgroup identification methods only focus on obtaining a subgroup with an enhanced treatment effect without paying attention to subgroup size. Yet, a clinically meaningful subgroup learning approach should identify the maximum number of patients who can benefit from the better treatment. In this paper, we present an optimal subgroup selection rule (SSR) that maximizes the number of selected patients, and in the meantime, achieves the pre-specified clinically meaningful mean outcome, such as the average treatment effect. We derive two equivalent theoretical forms of the optimal SSR based on the contrast function that describes the treatment-covariates interaction in the outcome. We further propose a ConstrAined PolIcy Tree seArch aLgorithm (CAPITAL) to find the optimal SSR within the interpretable decision tree class. The proposed method is flexible to handle multiple constraints that penalize the inclusion of patients with negative treatment effects, and to address time to event data using the restricted mean survival time as the clinically interesting mean outcome. Extensive simulations, comparison studies, and real data applications are conducted to demonstrate the validity and utility of our method.
Randomized clinical trials are often designed to assess whether a test treatment prolongs survival relative to a control treatment. Increased patient heterogeneity, while desirable for ...generalizability of results, can weaken the ability of common statistical approaches to detect treatment differences, potentially hampering the regulatory approval of safe and efficacious therapies. A novel solution to this problem is proposed. A list of baseline covariates that have the potential to be prognostic for survival under either treatment is pre-specified in the analysis plan. At the analysis stage, using all observed survival times but blinded to patient-level treatment assignment, 'noise' covariates are removed with elastic net Cox regression. The shortened covariate list is used by a conditional inference tree algorithm to segment the heterogeneous trial population into subpopulations of prognostically homogeneous patients (risk strata). After patient-level treatment unblinding, a treatment comparison is done within each formed risk stratum and stratum-level results are combined for overall statistical inference. The impressive power-boosting performance of our proposed 5-step stratified testing and amalgamation routine (5-STAR), relative to that of the logrank test and other common approaches that do not leverage inherently structured patient heterogeneity, is illustrated using a hypothetical and two real datasets along with simulation results. Furthermore, the importance of reporting stratum-level comparative treatment effects (time ratios from accelerated failure time model fits in conjunction with model averaging and, as needed, hazard ratios from Cox proportional hazard model fits) is highlighted as a potential enabler of personalized medicine. A fiveSTAR R package is available at https://github.com/rmarceauwest/fiveSTAR.
Rare variant associations are essential to fully understanding and trying to effectively treat complex diseases. Rare variants are often hypothesized to jointly explain much of the missing ...heritability from common variant genome-wide association studies, and are believed to be more likely than their common counterparts to have direct functional effect on disease etiology. However, rare variant associations are still quite difficult to detect with existing statistical methodology, especially when the question of interest goes beyond simple genetic main effect testing. Kernel machine models provide a flexible framework to facilitate complex rare variant studies.We consider three such interesting problems: (1) multi-factor analysis using multiple kernels to jointly model different variable sets, testing for the effect of one set while adjusting for nuisance effects; (2) crossdisorder gene variable selection, using kernels to leverage information on the co-occurrences of and genetic correlations between diseases; and (3) rare variant prioritization using protein structure, forming variant-level tests from local kernels that borrow information from variants that are nearby on the protein tertiary space. In project 1, we demonstrate use of a low-rank decomposition of the nuisance effect terms to improve computational efficiency over traditional estimation (i.e., using an ExpectationMaximization algorithm or penalization).We extend this idea in project 2, showing how this decomposition can be used to facilitate gene-level variable selection using group lasso when sample size and number of variants is large. Further, in project 2 we demonstrate a significant power gain from borrowing information from correlated diseases. In project 3, we continue looking at variable selection but try to prioritize rare variants at the variant level and demonstrate how protein folding structure can be incorporated into variant-level association testing to improve power and generate more biologically interesting hypotheses. We establish the validity and power of our methods using simulation studies with continuous, binary, and survival traits.We further apply our methods to data from the Vitamin Intervention for Stroke Prevention (VISP), CoLaus, and Action to Control Cardiovascular Risk in Diabetes (ACCORD) clinical trails, finding promising associations for follow-up analyses.
PDF