Outcome risk model development for heterogeneity of treatment effect analyses: a comparison of non-parametric machine learning methods and semi-parametric statistical methods.

BACKGROUND: In randomized clinical trials, treatment effects may vary, and this possibility is referred to as heterogeneity of treatment effect (HTE). One way to quantify HTE is to partition participants into subgroups based on individual's risk of experiencing an outcome, then measuring treatment effect by subgroup. Given the limited availability of externally validated outcome risk prediction models, internal models (created using the same dataset in which heterogeneity of treatment analyses also will be performed) are commonly developed for subgroup identification. We aim to compare different methods for generating internally developed outcome risk prediction models for subject partitioning in HTE analysis. METHODS: Three approaches were selected for generating subgroups for the 2,441 participants from the United States enrolled in the ASPirin in Reducing Events in the Elderly (ASPREE) randomized controlled trial. An extant proportional hazards-based outcomes predictive risk model developed on the overall ASPREE cohort of 19,114 participants was identified and was used to partition United States' participants by risk of experiencing a composite outcome of death, dementia, or persistent physical disability. Next, two supervised non-parametric machine learning outcome classifiers, decision trees and random forests, were used to develop multivariable risk prediction models and partition participants into subgroups with varied risks of experiencing the composite outcome. Then, we assessed how the partitioning from the proportional hazard model compared to those generated by the machine learning models in an HTE analysis of the 5-year absolute risk reduction (ARR) and hazard ratio for aspirin vs. placebo in each subgroup. Cochran's Q test was used to detect if ARR varied significantly by subgroup. RESULTS: The proportional hazard model was used to generate 5 subgroups using the quintiles of the estimated risk scores; the decision tree model was used to generate 6 subgroups (6 automatically determined tree leaves); and the random forest model was used to generate 5 subgroups using the quintiles of the prediction probability as risk scores. Using the semi-parametric proportional hazards model, the ARR at 5 years was 15.1% (95% CI 4.0-26.3%) for participants with the highest 20% of predicted risk. Using the random forest model, the ARR at 5 years was 13.7% (95% CI 3.1-24.4%) for participants with the highest 20% of predicted risk. The highest outcome risk group in the decision tree model also exhibited a risk reduction, but the confidence interval was wider (5-year ARR = 17.0%, 95% CI= -5.4-39.4%). Cochran's Q test indicated ARR varied significantly only by subgroups created using the proportional hazards model. The hazard ratio for aspirin vs. placebo therapy did not significantly vary by subgroup in any of the models. The highest risk groups for the proportional hazards model and random forest model contained 230 participants each, while the highest risk group in the decision tree model contained 41 participants. CONCLUSIONS: The choice of technique for internally developed models for outcome risk subgroups influences HTE analyses. The rationale for the use of a particular subgroup determination model in HTE analyses needs to be explicitly defined based on desired levels of explainability (with features importance), uncertainty of prediction, chances of overfitting, and assumptions regarding the underlying data structure. Replication of these analyses using data from other mid-size clinical trials may help to establish guidance for selecting an outcomes risk prediction modelling technique for HTE analyses.

Delaying high school start times impacts depressed mood among students: evidence from a natural experiment.

PURPOSE: Delaying high school start times prolongs weekday sleep. However, it is not clear if longer sleep reduces depression symptoms and if the impact of such policy change is the same across groups of adolescents. METHODS: We examined how gains in weekday sleep impact depression symptoms in 2,134 high school students (mean age 15.16 ± 0.35 years) from the Minneapolis metropolitan area. Leveraging a natural experiment design, we used the policy change to delay school start times as an instrument to estimate the effect of a sustained gain in weekday sleep on repeatedly measured Kandel-Davies depression symptoms. We also evaluated whether allocating the policy change to subgroups with expected benefit could improve the impact of the policy. RESULTS: Over 2 years, a sustained half-hour gain in weekday sleep expected as a result of the policy change to delay start times decreased depression symptoms by 0.78 points, 95%CI (-1.32,-0.28), or 15.6% of a standard deviation. The benefit was driven by a decrease in fatigue and sleep-related symptoms. While symptoms of low mood, hopelessness, and worry were not affected by the policy on average, older students with greater daily screen use and higher BMI experienced greater improvements in mood symptoms than would be expected on average, signaling heterogeneity. Nevertheless, universal implementation outperformed prescriptive strategies. CONCLUSION: High school start time delays are likely to universally decrease fatigue and overall depression symptoms in adolescents. Students who benefit most with respect to mood are older, spend more time on screens and have higher BMI.

Using Power Analysis to Choose the Unit of Randomization, Outcome, and Approach for Subgroup Analysis for a Multilevel Randomized Controlled Clinical Trial to Reduce Disparities in Cardiovascular Health.

We give examples of three features in the design of randomized controlled clinical trials which can increase power and thus decrease sample size and costs. We consider an example multilevel trial with several levels of clustering. For a fixed number of independent sampling units, we show that power can vary widely with the choice of the level of randomization. We demonstrate that power and interpretability can improve by testing a multivariate outcome rather than an unweighted composite outcome. Finally, we show that using a pooled analytic approach, which analyzes data for all subgroups in a single model, improves power for testing the intervention effect compared to a stratified analysis, which analyzes data for each subgroup in a separate model. The power results are computed for a proposed prevention research study. The trial plans to randomize adults to either telehealth (intervention) or in-person treatment (control) to reduce cardiovascular risk factors. The trial outcomes will be measures of the Essential Eight, a set of scores for cardiovascular health developed by the American Heart Association which can be combined into a single composite score. The proposed trial is a multilevel study, with outcomes measured on participants, participants treated by the same provider, providers nested within clinics, and clinics nested within hospitals. Investigators suspect that the intervention effect will be greater in rural participants, who live farther from clinics than urban participants. The results use published, exact analytic methods for power calculations with continuous outcomes. We provide example code for power analyses using validated software.

Accounting for expected attrition in the planning of cluster randomized trials for assessing treatment effect heterogeneity.

BACKGROUND: Detecting treatment effect heterogeneity is an important objective in cluster randomized trials and implementation research. While sample size procedures for testing the average treatment effect accounting for participant attrition assuming missing completely at random or missing at random have been previously developed, the impact of attrition on the power for detecting heterogeneous treatment effects in cluster randomized trials remains unknown. METHODS: We provide a sample size formula for testing for a heterogeneous treatment effect assuming the outcome is missing completely at random. We also propose an efficient Monte Carlo sample size procedure for assessing heterogeneous treatment effect assuming covariate-dependent outcome missingness (missing at random). We compare our sample size methods with the direct inflation method that divides the estimated sample size by the mean follow-up rate. We also evaluate our methods through simulation studies and illustrate them with a real-world example. RESULTS: Simulation results show that our proposed sample size methods under both missing completely at random and missing at random provide sufficient power for assessing heterogeneous treatment effect. The proposed sample size methods lead to more accurate sample size estimates than the direct inflation method when the missingness rate is high (e.g., ≥ 30%). Moreover, sample size estimation under both missing completely at random and missing at random is sensitive to the missingness rate, but not sensitive to the intracluster correlation coefficient among the missingness indicators. CONCLUSION: Our new sample size methods can assist in planning cluster randomized trials that plan to assess a heterogeneous treatment effect and participant attrition is expected to occur.

Costs of fall injuries in the STRIDE study: an economic evaluation of healthcare system heterogeneity and heterogeneity of treatment effect.

OBJECTIVES: The Strategies to Reduce Injuries and Develop Confidence in Elders (STRIDE) Study cluster-randomized 86 primary care practices in 10 healthcare systems to a patient-centered multifactorial fall injury prevention intervention or enhanced usual care, enrolling 5451 participants. We estimated total healthcare costs from participant-reported fall injuries receiving medical attention (FIMA) that were averted by the STRIDE intervention and tested for healthcare-system-level heterogeneity and heterogeneity of treatment effect (HTE). METHODS: Participants were community-dwelling adults age ≥ 70 at increased fall injury risk. We estimated practice-level total costs per person-year of follow-up (PYF), assigning unit costs to FIMA with and without an overnight hospital stay. Using independent variables for treatment arm, healthcare system, and their interaction, we fit a generalized linear model with log link, log follow-up time offset, and Tweedie error distribution. RESULTS: Unadjusted total costs per PYF were $2,034 (intervention) and $2,289 (control). The adjusted (intervention minus control) cost difference per PYF was -$167 (95% confidence interval (CI), -$491, $216). Cost heterogeneity by healthcare system was present (p = 0.035), as well as HTE (p = 0.090). Adjusted total costs per PYF in control practices varied from $1,529 to $3,684 for individual healthcare systems; one system with mean intervention minus control costs of -$2092 (95% CI, -$3,686 to -$944) per PYF accounted for HTE, but not healthcare system cost heterogeneity. CONCLUSIONS: We observed substantial heterogeneity of healthcare system costs in the STRIDE study, with small reductions in healthcare costs for FIMA in the STRIDE intervention accounted for by a single healthcare system. TRIAL REGISTRATION: Clinicaltrials.gov (NCT02475850).

Regression-based estimation of heterogeneous treatment effects when extending inferences from a randomized trial to a target population.

Most work on extending (generalizing or transporting) inferences from a randomized trial to a target population has focused on estimating average treatment effects (i.e., averaged over the target population's covariate distribution). Yet, in the presence of strong effect modification by baseline covariates, the average treatment effect in the target population may be less relevant for guiding treatment decisions. Instead, the conditional average treatment effect (CATE) as a function of key effect modifiers may be a more useful estimand. Recent work on estimating target population CATEs using baseline covariate, treatment, and outcome data from the trial and covariate data from the target population only allows for the examination of heterogeneity over distinct subgroups. We describe flexible pseudo-outcome regression modeling methods for estimating target population CATEs conditional on discrete or continuous baseline covariates when the trial is embedded in a sample from the target population (i.e., in nested trial designs). We construct pointwise confidence intervals for the CATE at a specific value of the effect modifiers and uniform confidence bands for the CATE function. Last, we illustrate the methods using data from the Coronary Artery Surgery Study (CASS) to estimate CATEs given history of myocardial infarction and baseline ejection fraction value in the target population of all trial-eligible patients with stable ischemic heart disease.

Overall average treatment effects from clinical trials, one-variable-at-a-time subgroup analyses and predictive approaches to heterogeneous treatment effects: Toward a more patient-centered evidence-based medicine.

Despite the predominance of the evidence-based medicine paradigm, a fundamental incongruity remains: Evidence is derived from groups of people, yet medical decisions are made by and for individuals. Randomization ensures the comparability of treatment groups within a clinical trial, which allows for unbiased estimation of average treatment effects. If we treated groups of patients instead of individuals, or if patients with the same disease were identical to one another in all factors that determined the harms and the benefits of therapy, then these group-level averages would make a perfectly sound foundation for medical decision-making. But patients differ from one another in many ways that determine the likelihood of an outcome, both with and without a treatment. Nevertheless, popular approaches to evidence-based medicine have encouraged a reliance on the average treatment effects estimated from clinical trials and meta-analysis as guides to decision-making for individuals. Here, we discuss the limitations of this approach as well as limitations of conventional, one-variable-at-a-time subgroup analysis; finally, we discuss the rationale for "predictive" approaches to heterogeneous treatment effects. Predictive approaches to heterogeneous treatment effects combine methods for causal inference (e.g. randomization) with methods for prediction that permit inferences about which patients are likely to benefit and which are not, taking into account multiple relevant variables simultaneously to yield "personalized" estimates of benefit-harm trade-offs. We focus on risk modeling approaches, which rely on the mathematical dependence of the absolute treatment effect with the baseline risk, which varies substantially "across patients" in most trials. While there are a number of examples of risk modeling approaches that have been practice-changing, risk modeling does not provide ideal estimates of individual treatment effects, since risk modeling does not account for how individual variables might modify the effects of therapy. In "effect modeling," prediction models are developed directly on clinical trial data, including terms for treatment and treatment effect interactions. These more flexible approaches may better uncover individualized treatment effects, but are also prone to overfitting when dimensionality is high, power is low, and there is limited prior knowledge about effect modifiers.

Sample Size Requirements to Test Subgroup-Specific Treatment Effects in Cluster-Randomized Trials.

Cluster-randomized trials (CRTs) often allocate intact clusters of participants to treatment or control conditions and are increasingly used to evaluate healthcare delivery interventions. While previous studies have developed sample size methods for testing confirmatory hypotheses of treatment effect heterogeneity in CRTs (i.e., targeting the difference between subgroup-specific treatment effects), sample size methods for testing the subgroup-specific treatment effects themselves have not received adequate attention-despite a rising interest in health equity considerations in CRTs. In this article, we develop formal methods for sample size and power analyses for testing subgroup-specific treatment effects in parallel-arm CRTs with a continuous outcome and a binary subgroup variable. We point out that the variances of the subgroup-specific treatment effect estimators and their covariance are given by weighted averages of the variance of the overall average treatment effect estimator and the variance of the heterogeneous treatment effect estimator. This analytical insight facilitates an explicit characterization of the requirements for both the omnibus test and the intersection-union test to achieve the desired level of power. Generalizations to allow for subgroup-specific variance structures are also discussed. We report on a simulation study to validate the proposed sample size methods and demonstrate that the empirical power corresponds well with the predicted power for both tests. The design and setting of the Umea Dementia and Exercise (UMDEX) CRT in older adults are used to illustrate our sample size methods.

Evaluating effectiveness of public health intervention strategies for mitigating COVID-19 pandemic.

Coronavirus disease 2019 (COVID-19) pandemic is an unprecedented global public health challenge. In the United States (US), state governments have implemented various non-pharmaceutical interventions (NPIs), such as physical distance closure (lockdown), stay-at-home order, mandatory facial mask in public in response to the rapid spread of COVID-19. To evaluate the effectiveness of these NPIs, we propose a nested case-control design with propensity score weighting under the quasi-experiment framework to estimate the average intervention effect on disease transmission across states. We further develop a method to test for factors that moderate intervention effect to assist precision public health intervention. Our method takes account of the underlying dynamics of disease transmission and balance state-level pre-intervention characteristics. We prove that our estimator provides causal intervention effect under assumptions. We apply this method to analyze US COVID-19 incidence cases to estimate the effects of six interventions. We show that lockdown has the largest effect on reducing transmission and reopening bars significantly increase transmission. States with a higher percentage of non-White population are at greater risk of increased R t $$ {R}_t $$ associated with reopening bars.

Longitudinal phenotypes in patients with acute respiratory distress syndrome: a multi-database study.

BACKGROUND: Previously identified phenotypes of acute respiratory distress syndrome (ARDS) have been limited by a disregard for temporal dynamics. We aimed to identify longitudinal phenotypes in ARDS to test the prognostic and predictive enrichment of longitudinal phenotypes, and to develop simplified models for phenotype identification. METHODS: We conducted a multi-database study based on the Chinese Database in Intensive Care (CDIC) and four ARDS randomized clinical trials (RCTs). We employed latent class analysis (LCA) to identify longitudinal phenotypes using 24-hourly data from the first four days of invasive ventilation. We used the Cox regression model to explore the association between time-varying respiratory parameters and 28-day mortality across phenotypes. Phenotypes were validated in four RCTs, and the heterogeneity of treatment effect (HTE) was investigated. We also constructed two multinomial logistical regression analyses to develop the probabilistic models. FINDINGS: A total of 605 ARDS patients in CDIC were enrolled. The three-class LCA model was identified and had the optimal fit, as follows: Class 1 (n = 400, 66.1% of the cohort) was the largest phenotype over all study days, and had fewer abnormal values, less organ dysfunction and the lowest 28-day mortality rate (30.5%). Class 2 (n = 102, 16.9% of the cohort) was characterized by pulmonary mechanical dysfunction and had the highest proportion of poorly aerated lung volume, the 28-day mortality rate was 47.1%. Class 3 (n = 103, 17% of the cohort) was correlated with extra-pulmonary dysfunction and had the highest 28-day mortality rate (56.3%). Time-varying mechanical power was more significantly associated with 28-day mortality in Class 2 patients compared to other phenotypes. Similar phenotypes were identified in four RCTs. A significant HTE between phenotypes and treatment strategies was observed in the ALVEOLI (high PEEP vs. low PEEP) and the FACTT trials (conservative vs. liberal fluid management). Two parsimonious probabilistic models were constructed to identify longitudinal phenotypes. INTERPRETATION: We identified and validated three novel longitudinal phenotypes for ARDS patients, with both prognostic and predictive enrichment. The phenotypes of ARDS can be accurately identified with simple classifier models, except for Class 3.

Personalization of renal replacement therapy initiation: a secondary analysis of the AKIKI and IDEAL-ICU trials.

BACKGROUND: Trials comparing early and delayed strategies of renal replacement therapy in patients with severe acute kidney injury may have missed differences in survival as a result of mixing together patients at heterogeneous levels of risks. Our aim was to evaluate the heterogeneity of treatment effect on 60-day mortality from an early vs a delayed strategy across levels of risk for renal replacement therapy initiation under a delayed strategy. METHODS: We used data from the AKIKI, and IDEAL-ICU randomized controlled trials to develop a multivariable logistic regression model for renal replacement therapy initiation within 48 h after allocation to a delayed strategy. We then used an interaction with spline terms in a Cox model to estimate treatment effects across the predicted risks of RRT initiation. RESULTS: We analyzed data from 1107 patients (619 and 488 in the AKIKI and IDEAL-ICU trial respectively). In the pooled sample, we found evidence for heterogeneous treatment effects (P = 0.023). Patients at an intermediate-high risk of renal replacement therapy initiation within 48 h may have benefited from an early strategy (absolute risk difference, - 14%; 95% confidence interval, - 27% to - 1%). For other patients, we found no evidence of benefit from an early strategy of renal replacement therapy initiation but a trend for harm (absolute risk difference, 8%; 95% confidence interval, - 5% to 21% in patients at intermediate-low risk). CONCLUSIONS: We have identified a clinically sound heterogeneity of treatment effect of an early vs a delayed strategy of renal replacement therapy initiation that may reflect varying degrees of kidney demand-capacity mismatch.

Blood Purification Studies in the ICU: What Endpoints Should We Use?

The potential for treatment of the critically ill using blood purification techniques has been discussed for several decades. However, since the first attempts at applying extracorporeal techniques to patients with sepsis were described, there has been considerable hesitancy towards the widespread adoption of such methods, given the lack of mortality benefit observed and indeed the paucity of randomized controlled studies. However, this is not unique so far as studies on the critically ill are concerned where there is a dearth of studies providing a positive finding to influence clinical practice. Consequently, as well as targeted patient selection, it is perhaps time to consider endpoints other than mortality in studies on the critically ill, particularly in blood purification studies where, to-date, such heterogeneous groups of patients have been studied.

Endotypes and the Path to Precision in Moderate and Severe Traumatic Brain Injury.

Heterogeneity is recognized as a major barrier in efforts to improve the care and outcomes of patients with traumatic brain injury (TBI). Even within the narrower stratum of moderate and severe TBI, current management approaches do not capture the complexity of this condition characterized by manifold clinical, anatomical, and pathophysiologic features. One approach to heterogeneity may be to resolve undifferentiated TBI populations into endotypes, subclasses that are distinguished by shared biological characteristics. The endotype paradigm has been explored in a range of medical domains, including psychiatry, oncology, immunology, and pulmonology. In intensive care, endotypes are being investigated for syndromes such as sepsis and acute respiratory distress syndrome. This review provides an overview of the endotype paradigm as well as some of its methods and use cases. A conceptual framework is proposed for endotype research in moderate and severe TBI, together with a scientific road map for endotype discovery and validation in this population.

Individualized treatment effects with censored data via fully nonparametric Bayesian accelerated failure time models.

Individuals often respond differently to identical treatments, and characterizing such variability in treatment response is an important aim in the practice of personalized medicine. In this article, we describe a nonparametric accelerated failure time model that can be used to analyze heterogeneous treatment effects (HTE) when patient outcomes are time-to-event. By utilizing Bayesian additive regression trees and a mean-constrained Dirichlet process mixture model, our approach offers a flexible model for the regression function while placing few restrictions on the baseline hazard. Our nonparametric method leads to natural estimates of individual treatment effect and has the flexibility to address many major goals of HTE assessment. Moreover, our method requires little user input in terms of model specification for treatment covariate interactions or for tuning parameter selection. Our procedure shows strong predictive performance while also exhibiting good frequentist properties in terms of parameter coverage and mitigation of spurious findings of HTE. We illustrate the merits of our proposed approach with a detailed analysis of two large clinical trials (N = 6769) for the prevention and treatment of congestive heart failure using an angiotensin-converting enzyme inhibitor. The analysis revealed considerable evidence for the presence of HTE in both trials as demonstrated by substantial estimated variation in treatment effect and by high proportions of patients exhibiting strong evidence of having treatment effects which differ from the overall treatment effect.

Lasso estimation of hierarchical interactions for analyzing heterogeneity of treatment effect.

Individuals differ in how they respond to a given treatment. In an effort to predict the treatment response and analyze the heterogeneity of treatment effect, we propose a general modeling framework by identifying treatment-covariate interactions honoring a hierarchical condition. We construct a single-step l1 norm penalty procedure that maintains the hierarchical structure of interactions in the sense that a treatment-covariate interaction term is included in the model only when either the covariate or both the covariate and treatment have nonzero main effects. We developed a constrained Lasso approach with two parameterization schemes that enforce the hierarchical interaction restriction differently. We solved the resulting constrained optimization problem using a spectral projected gradient method. We compared our methods to the unstructured Lasso using simulation studies including a scenario that violates the hierarchical condition (misspecified model). The simulations showed that our methods yielded more parsimonious models and outperformed the unstructured Lasso for correctly identifying nonzero treatment-covariate interactions. The superior performance of our methods are also corroborated by an application to a large randomized clinical trial data investigating a drug for treating congestive heart failure (N = 2569). Our methods provide a well-suited approach for doing secondary analysis in clinical trials to analyze heterogeneous treatment effects and to identify predictive biomarkers.

Generalized boosted modeling to identify subgroups where effect of dabigatran versus warfarin may differ: An observational cohort study of patients with atrial fibrillation.

PURPOSE: To explore generalized boosted modeling (GBM) as a method for identifying subgroups with greater benefit or harm with dabigatran versus warfarin for treatment of atrial fibrillation. METHODS: We identified new initiators of warfarin or dabigatran with nonvalvular atrial fibrillation in 2 healthcare claims databases (2009-2013) and used GBM within 1 data source (development cohort) to explore subgroups where their effect on thromboembolism and major bleeding may differ. Identified subgroups were evaluated in the second data source (validation cohort) with stabilized-inverse-probability-of-treatment weights to adjust for confounding. RESULTS: Development and validation cohorts included 13 624 (28% dabigatran) and 62 596 (29% dabigatran) initiators, respectively. In development data, the strongest exposure interactions were prior thromboembolism and renal disease. In validation data, reduction in thromboembolism with dabigatran was greater for patients with versus without a history of thromboembolism by 2.8 (95% CI, -0.5 to 5.4) events per 100 patient-years. Major bleeding was reduced by 1.6/100 patient-years for dabigatran compared to warfarin initiators, without evidence of variation by renal disease. CONCLUSIONS: We explored use of GBM to identify potential subgroups with different treatment effect. Dabigatran's superiority to warfarin at prevention of thromboembolism may be greater in secondary than primary prevention. In practice, secondary prevention patients are more often treated with warfarin.

Heterogeneity in the Effects of Reward- and Deposit-based Financial Incentives on Smoking Cessation.

RATIONALE: Targeting different smoking cessation programs to smokers most likely to quit when using them could reduce the burden of lung disease. OBJECTIVES: To identify smokers most likely to quit using pure reward-based financial incentives or incentive programs requiring refundable deposits to become eligible for rewards. METHODS: We conducted prespecified secondary analyses of a randomized trial in which 2,538 smokers were assigned to an $800 reward contingent on sustained abstinence from smoking, a refundable $150 deposit plus a $650 reward, or usual care. MEASUREMENTS AND MAIN RESULTS: Using logistic regression, we identified characteristics of smokers that were most strongly associated with accepting their assigned intervention and ceasing smoking for 6 months. We assessed modification of the acceptance, efficacy, and effectiveness of reward and deposit programs by 11 prospectively selected demographic, smoking-related, and psychological factors. Predictors of sustained smoking abstinence differed among participants assigned to reward- versus deposit-based incentives. However, greater readiness to quit and less steep discounting of future rewards were consistently among the most important predictors. Deposit-based programs were uniquely effective relative to usual care among men, higher-income participants, and participants who more commonly failed to pay their bills (all interaction P values < 0.10). Relative to rewards, deposits were more effective among black persons (P = 0.022) and those who more commonly failed to pay their bills (P = 0.082). Relative to rewards, deposits were more commonly accepted by higher-income participants, men, white persons, and those who less commonly failed to pay their bills (all P < 0.05). CONCLUSIONS: Heterogeneity among smokers in their acceptance and response to different forms of incentives suggests potential benefits of targeting behavior-change interventions based on patient characteristics. Clinical trial registered with www.clinicaltrials.gov (NCT 01526265).

Implications of Heterogeneity of Treatment Effect for Reporting and Analysis of Randomized Trials in Critical Care.

Randomized clinical trials (RCTs) are conducted to guide clinicians' selection of therapies for individual patients. Currently, RCTs in critical care often report an overall mean effect and selected individual subgroups. Yet work in other fields suggests that such reporting practices can be improved. Specifically, this Critical Care Perspective reviews recent work on so-called "heterogeneity of treatment effect" (HTE) by baseline risk and extends that work to examine its applicability to trials of acute respiratory failure and severe sepsis. Because patients in RCTs in critical care medicine-and patients in intensive care units-have wide variability in their risk of death, these patients will have wide variability in the absolute benefit that they can derive from a given therapy. If the side effects of the therapy are not perfectly collinear with the treatment benefits, this will result in HTE, where different patients experience quite different expected benefits of a therapy. We use simulations of RCTs to demonstrate that such HTE could result in apparent paradoxes, including: (1) positive trials of therapies that are beneficial overall but consistently harm or have little benefit to low-risk patients who met enrollment criteria, and (2) overall negative trials of therapies that still consistently benefit high-risk patients. We further show that these results persist even in the presence of causes of death unmodified by the treatment under study. These results have implications for reporting and analyzing RCT data, both to better understand how our therapies work and to improve the bedside applicability of RCTs. We suggest a plan for measurement in future RCTs in the critically ill.

Identification of subpopulations with distinct treatment benefit rate using the Bayesian tree.

Characterization of a subpopulation by the difference in marginal means of the outcome under the intervention and control may not be sufficient to provide informative guidance for individual decision and public policy making. Specifically, often we are interested in the treatment benefit rate (TBR), that is, the probability of benefitting an intervention in a meaningful way. For binary outcomes, TBR is the proportion that has "unfavorable" outcome under the control and "favorable" outcome under the intervention. Identification of subpopulations with distinct TBR by baseline characteristics will have significant implications in clinical setting where a medical intervention with potential negative health impact is under consideration for a given patient. In addition, these subpopulations with unique TBR set the basis for guidance in implementing the intervention toward a more personalized scheme of treatment. In this article, we propose a Bayesian tree based latent variable model to seek subpopulations with distinct TBR. Our method offers a nonparametric Bayesian framework that accounts for the uncertainty in estimating potential outcomes and allows more exhaustive search of the partitions of the baseline covariates space. The method is evaluated through a simulation study and applied to a randomized clinical trial of implantable cardioverter defibrillators to reduce mortality.