Fast FSR variable selection with applications to clinical trials

Flexible variable selection in the presence of missing data

In many applications, it is of interest to identify a parsimonious set of features, or panel, from multiple candidates that achieves a desired level of performance in predicting a response. This task is often complicated in practice by missing data arising from the sampling design or other random mechanisms. Most recent work on variable selection in missing data contexts relies in some part on a finite-dimensional statistical model, e.g., a generalized or penalized linear model. In cases where this model is misspecified, the selected variables may not all be truly scientifically relevant and can result in panels with suboptimal classification performance. To address this limitation, we propose a nonparametric variable selection algorithm combined with multiple imputation to develop flexible panels in the presence of missing-at-random data. We outline strategies based on the proposed algorithm that achieve control of commonly used error rates. Through simulations, we show that our proposal has good operating characteristics and results in panels with higher classification and variable selection performance compared to several existing penalized regression approaches in cases where a generalized linear model is misspecified. Finally, we use the proposed method to develop biomarker panels for separating pancreatic cysts with differing malignancy potential in a setting where complicated missingness in the biomarkers arose due to limited specimen volumes.

A hierarchical integrative group least absolute shrinkage and selection operator for analyzing environmental mixtures

Environmental health studies are increasingly measuring multiple pollutants to characterize the joint health effects attributable to exposure mixtures. However, the underlying dose-response relationship between toxicants and health outcomes of interest may be highly nonlinear, with possible nonlinear interaction effects. Existing penalized regression methods that account for exposure interactions either cannot accommodate nonlinear interactions while maintaining strong heredity or are computationally unstable in applications with limited sample size. In this paper, we propose a general shrinkage and selection framework to identify noteworthy nonlinear main and interaction effects among a set of exposures. We design hierarchical integrative group least absolute shrinkage and selection operator (HiGLASSO) to (a) impose strong heredity constraints on two-way interaction effects (hierarchical), (b) incorporate adaptive weights without necessitating initial coefficient estimates (integrative), and (c) induce sparsity for variable selection while respecting group structure (group LASSO). We prove sparsistency of the proposed method and apply HiGLASSO to an environmental toxicants dataset from the LIFECODES birth cohort, where the investigators are interested in understanding the joint effects of 21 urinary toxicant biomarkers on urinary 8-isoprostane, a measure of oxidative stress. An implementation of HiGLASSO is available in the higlasso R package, accessible through the Comprehensive R Archive Network.

Bayesian approaches to variable selection: a comparative study from practical perspectives

In many clinical studies, researchers are interested in parsimonious models that simultaneously achieve consistent variable selection and optimal prediction. The resulting parsimonious models will facilitate meaningful biological interpretation and scientific findings. Variable selection via Bayesian inference has been receiving significant advancement in recent years. Despite its increasing popularity, there is limited practical guidance for implementing these Bayesian approaches and evaluating their comparative performance in clinical datasets. In this paper, we review several commonly used Bayesian approaches to variable selection, with emphasis on application and implementation through R software. These approaches can be roughly categorized into four classes: namely the Bayesian model selection, spike-and-slab priors, shrinkage priors, and the hybrid of both. To evaluate their variable selection performance under various scenarios, we compare these four classes of approaches using real and simulated datasets. These results provide practical guidance to researchers who are interested in applying Bayesian approaches for the purpose of variable selection.

Selection of nonlinear interactions by a forward stepwise algorithm: Application to identifying environmental chemical mixtures affecting health outcomes

In this paper, we propose a stepwise forward selection algorithm for detecting the effects of a set of correlated exposures and their interactions on a health outcome of interest when the underlying relationship could potentially be nonlinear. Though the proposed method is very general, our application in this paper remains to be on analysis of multiple pollutants and their interactions. Simultaneous exposure to multiple environmental pollutants could affect human health in a multitude of complex ways. For understanding the health effects of multiple environmental exposures, it is often important to identify and estimate complex interactions among exposures. However, this issue becomes analytically challenging in the presence of potential nonlinearity in the outcome-exposure response surface and a set of correlated exposures. Through simulation studies and analyses of test datasets that were simulated as a part of a data challenge in multipollutant modeling organized by the National Institute of Environmental Health Sciences (http://www.niehs.nih.gov/about/events/pastmtg/2015/statistical/), we illustrate the advantages of our proposed method in comparison with existing alternative approaches. A particular strength of our method is that it demonstrates very low false positives across empirical studies. Our method is also used to analyze a dataset that was released from the Health Outcomes and Measurement of the Environment Study as a benchmark beta-tester dataset as a part of the same workshop.

Relation of Risk of Stroke in Patients With Atrial Fibrillation to Body Mass Index (from Patients Treated With Rivaroxaban and Warfarin in the Rivaroxaban Once Daily Oral Direct Factor Xa Inhibition Compared with Vitamin K Antagonism for Prevention of Stroke and Embolism Trial in Atrial Fibrillation Trial)

We investigated stroke outcomes in normal weight (body mass index [BMI] 18.50 to 24.99 kg/m²), overweight (BMI 25.00 to 29.99 kg/m²), and obese (BMI ≥30 kg/m²) patients with atrial fibrillation treated with rivaroxaban and warfarin. We compared the incidence of stroke and systemic embolic events as well as bleeding events in normal weight (n = 3,289), overweight (n = 5,535), and obese (n = 5,206) patients in a post hoc analysis of the Rivaroxaban Once Daily Oral Direct Factor Xa Inhibition Compared with Vitamin K Antagonism for Prevention of Stroke and Embolism Trial in Atrial Fibrillation trial. Stroke and systemic embolic event rates per 100 patient-years were 2.93 in the normal weight group (reference group), 2.28 in the overweight group (adjusted hazard ratio [HR] 0.81, 95% CI 0.66 to 0.99, p = 0.04) and 1.88 in the obese group (adjusted HR 0.69, 95% CI 0.55 to 0.86, p <0.001). The risk of stroke was statistically significantly lower for obese patients with BMI ≥35 than that for normal weight patients in both the rivaroxaban and warfarin groups (rivaroxaban: HR 0.62, 95% CI 0.40 to 0.96, p = 0.033; warfarin: HR 0.48, 95% CI 0.31 to 0.74, p <0.001). In conclusion, in patients with atrial fibrillation treated with anticoagulant therapy, increased BMI was associated with decreased stroke risk. Warfarin and the novel anticoagulant rivaroxaban are effective in stroke prevention in all subgroups of obese patients.

Predicting the risk of bleeding during dual antiplatelet therapy after acute coronary syndromes

OBJECTIVES

Dual antiplatelet therapy (DAPT) with aspirin + a P2Y12 inhibitor is recommended for at least 12 months for patients with acute coronary syndrome (ACS), with shorter durations considered for patients with increased bleeding risk. However, there are no decision support tools available to predict an individual patient's bleeding risk during DAPT treatment in the post-ACS setting.

METHODS

To develop a longitudinal bleeding risk prediction model, we analy sed 9240 patients with unstable angina/non-ST segment elevation myocardial infarction (NSTEMI) from the Targeted Platelet Inhibition to Clarify the Optimal Strategy to Medically Manage Acute Coronary Syndromes (TRILOGY ACS) trial, who were managed without revasculari sation and treated with DAPT for a median of 14.8 months.

RESULTS

We identified 10 significant baseline predictors of non-coronary artery bypass grafting (CABG)-related Global Use of Strategies to Open Occluded Arteries (GUSTO) severe/life-threatening/moderate bleeding: age, sex, weight, NSTEMI (vs unstable angina), angiography performed before randomi sation, prior peptic ulcer disease, creatinine, systolic blood pressure, haemoglobin and treatment with beta-blocker. The five significant baseline predictors of Thrombolysis In Myocardial Infarction (TIMI) major or minor bleeding included age, sex, angiography performed before randomi sation, creatinine and haemoglobin. The models showed good predictive accuracy with Therneau's C-indices: 0.78 (SE = 0.024) for the GUSTO model and 0.67 (SE = 0.023) for the TIMI model. Internal validation with bootstrapping gave similar C-indices of 0.77 and 0.65, respectively. External validation demonstrated an attenuated C-index for the GUSTO model (0.69) but not the TIMI model (0.68).

CONCLUSIONS

Longitudinal bleeding risks during treatment with DAPT in patients with ACS can be reliably predicted using selected baseline characteristics. The TRILOGY ACS bleeding models can inform risk -benefit considerations regarding the duration of DAPT following ACS.

TRIAL REGISTRATION

ClinicalTrials.gov identifier: https://clinicaltrials.gov/ct2/show/NCT00699998.

Spontaneous MI After Non-ST-Segment Elevation Acute Coronary Syndrome Managed Without Revascularization: The TRILOGY ACS Trial

BACKGROUND

Patients with acute coronary syndrome (ACS), especially those receiving medical management without revascularization, are at high risk for spontaneous myocardial infarction (MI), but its frequency and predictors are unknown.

OBJECTIVES

This study sought to characterize spontaneous MI events in a randomized population during 30 months of follow-up and develop a prediction model for spontaneous MI to assign risk of spontaneous MI events in ACS populations.

METHODS

We analyzed data from the randomized TRILOGY ACS (TaRgeted platelet Inhibition to cLarify the Optimal strateGy to medically manage Acute Coronary Syndromes) trial of aspirin plus prasugrel or clopidogrel following ACS. The trial included 9,326 patients with non-ST-segment elevation myocardial infarction (NSTEMI)/unstable angina (UA) who were managed medically without planned revascularization. Our study population included 9,294 patients. A multivariable Cox proportional hazards model was developed to determine predictors of time to first spontaneous MI event through 30 months. After model validation, we developed a calculator for model implementation.

RESULTS

Among 9,294 patients, 695 spontaneous MI events occurred over a median of 17 months, representing 94% of adjudicated MI events (n = 737). The Kaplan-Meier event rate of spontaneous MI through 30 months was 10.7%. The strongest predictors of spontaneous MI were older age, NSTEMI versus UA as index event, diabetes mellitus, no pre-randomization angiography, and higher baseline creatinine values. The model exhibited good predictive capabilities (c-index = 0.732) and had good calibration, especially for patients with low-to-moderate risk of spontaneous MI.

CONCLUSIONS

Spontaneous MI following a medically managed UA/NSTEMI event is common. Baseline characteristics can be used to predict subsequent risk of spontaneous MI in this population. These findings provide insight into the long-term natural history of medically managed UA/NSTEMI patients and could be used to optimize risk stratification and treatment of these patients. (A Comparison of Prasugrel and Clopidogrel in Acute Coronary Syndrome Subjects [TRILOGY ACS]; NCT00699998).

Frequency and Predictors of Internal Mammary Artery Graft Failure and Subsequent Clinical Outcomes: Insights From the Project of Ex-vivo Vein Graft Engineering via Transfection (PREVENT) IV Trial

BACKGROUND

The internal mammary artery (IMA) is the preferred conduit for bypassing the left anterior descending (LAD) artery in patients undergoing coronary artery bypass grafting. Systematic evaluation of the frequency and predictors of IMA failure and long-term outcomes is lacking.

METHODS AND RESULTS

The Project of Ex-vivo Vein Graft Engineering via Transfection (PREVENT) IV trial participants who underwent IMA-LAD revascularization and had 12- to 18-month angiographic follow-up (n=1539) were included. Logistic regression with fast false selection rate methods was used to identify characteristics associated with IMA failure (≥75% stenosis). The relationship between IMA failure and long-term outcomes, including death, myocardial infarction, and repeat revascularization, was assessed with Cox regression. IMA failure occurred in 132 participants (8.6%). Predictors of IMA graft failure were LAD stenosis <75% (odds ratio, 1.76; 95% confidence interval, 1.19-2.59), additional bypass graft to diagonal branch (odds ratio, 1.92; 95% confidence interval, 1.33-2.76), and not having diabetes mellitus (odds ratio, 1.82; 95% confidence interval, 1.20-2.78). LAD stenosis and additional diagonal graft remained predictive of IMA failure in an alternative model that included angiographic failure or death before angiography as the outcome. IMA failure was associated with a significantly higher incidence of subsequent acute (<14 days of angiography) clinical events, mostly as a result of a higher rate of repeat revascularization.

CONCLUSIONS

IMA failure was common and associated with higher rates of repeat revascularization, and patients with intermediate LAD stenosis or with an additional bypass graft to the diagonal branch had increased risk for IMA failure. These findings raise concerns about competitive flow and the benefit of coronary artery bypass grafting in intermediate LAD stenosis without functional evidence of ischemia.

CLINICAL TRIAL REGISTRATION

URL: http:/www.clinicaltrials.gov. Unique identifier: NCT00042081.

Ascertainment, classification, and impact of neoplasm detection during prolonged treatment with dual antiplatelet therapy with prasugrel vs. clopidogrel following acute coronary syndrome

AIMS

Studies have suggested increased cancer incidence associated with long-term dual antiplatelet therapy (DAPT) for acute coronary syndrome (ACS). We evaluated cancer incidence and treatment-related differences in an analysis of DAPT for ACS.

METHODS AND RESULTS

The Targeted Platelet Inhibition to Clarify the Optimal Strategy to Medically Manage Acute Coronary Syndromes trial enrolled 9326 participants with ACS, who received aspirin plus clopidogrel or prasugrel. Median treatment exposure was 15 months. Cancer history and screening procedures were collected. Suspected non-benign neoplasm events were reported and adjudicated. The primary outcome was detection of new, non-benign neoplasm. Factors associated with neoplasm events, the relationship of these events to cardiovascular and bleeding endpoints, and treatment-related differences in neoplasm detection were studied. Among 9240 participants who received ≥1 dose of study drug, 1.8% had a confirmed neoplasm event. The efficacy composite of cardiovascular death, myocardial infarction, or stroke occurred more frequently among those with a neoplasm event vs. those without (18.2 vs. 13.5%) as did Global Use of Strategies to Open Occluded Coronary Arteries severe/moderate bleeding (11.2 vs. 1.5%). Screening rates were substantially higher in North America and Western Europe/Scandinavia vs. other regions. Factors most strongly associated with detection of neoplasm events were older age, region, male sex, and current/recent smoking. Among the pre-specified population without a history of neoplasm or previous curative treatment for neoplasm (n = 9105), the incidence of neoplasm events was similar with prasugrel vs. clopidogrel (1.8 vs. 1.7%; HR = 1.04; 95% CI 0.77-1.42; P = 0.79).

CONCLUSIONS

Neoplasm events were infrequent during long-term DAPT after ACS, were associated with differential cancer-screening practices across regions, and the frequency of neoplasm detection was similar with prasugrel vs. clopidogrel.

TRIAL REGISTRATION

ClinicalTrials.gov identifier: NCT00699998.

Saphenous vein graft failure after coronary artery bypass surgery: insights from PREVENT IV

BACKGROUND

Coronary artery bypass grafting success is limited by vein graft failure (VGF). Understanding the factors associated with VGF may improve patient outcomes.

METHODS AND RESULTS

We examined 1828 participants in the Project of Ex Vivo Vein Graft Engineering via Transfection IV (PREVENT IV) trial undergoing protocol-mandated follow-up angiography 12 to 18 months post-coronary artery bypass grafting or earlier clinically driven angiography. Outcomes included patient- and graft-level angiographic VGF (≥75% stenosis or occlusion). Variables were selected by using Fast False Selection Rate methodology. We examined relationships between variables and VGF in patient- and graft-level models by using logistic regression without and with generalized estimating equations. At 12 to 18 months post-coronary artery bypass grafting, 782 of 1828 (42.8%) patients had VGF, and 1096 of 4343 (25.2%) vein grafts had failed. Demographic and clinical characteristics were similar between patients with and without VGF, although VGF patients had longer surgical times, worse target artery quality, longer graft length, and they more frequently underwent endoscopic vein harvesting. After multivariable adjustment, longer surgical duration (odds ratio per 10-minute increase, 1.05; 95% confidence interval, 1.03-1.07), endoscopic vein harvesting (odds ratio, 1.41; 95% confidence interval, 1.16-1.71), poor target artery quality (odds ratio, 1.43; 95% confidence interval, 1.11-1.84), and postoperative use of clopidogrel or ticlopidine (odds ratio, 1.35; 95% confidence interval, 1.07-1.69) were associated with patient-level VGF. The predicted likelihood of VGF in the graft-level model ranged from 12.1% to 63.6%.

CONCLUSIONS

VGF is common and associated with patient and surgical factors. These findings may help identify patients with risk factors for VGF and inform the development of interventions to reduce VGF.

CLINICAL TRIAL REGISTRATION URL

http://www.clinicaltrials.gov. Unique identifier: NCT00042081.

Edifoligide and long-term outcomes after coronary artery bypass grafting: PRoject of Ex-vivo Vein graft ENgineering via Transfection IV (PREVENT IV) 5-year results

BACKGROUND

Edifoligide, an E2F transcription factor decoy, does not prevent vein graft failure or adverse clinical outcomes at 1 year in patients undergoing coronary artery bypass grafting (CABG). We compared the 5-year clinical outcomes of patients in PREVENT IV treated with edifoligide and placebo to identify predictors of long-term clinical outcomes.

METHODS

A total of 3,014 patients undergoing CABG with at least 2 planned vein grafts were enrolled. Kaplan-Meier curves were generated to compare the long-term effects of edifoligide and placebo. A Cox proportional hazards model was constructed to identify factors associated with 5-year post-CABG outcomes. The main outcome measures were death, myocardial infarction (MI), repeat revascularization, and rehospitalization through 5 years.

RESULTS

Five-year follow-up was complete in 2,865 patients (95.1%). At 5 years, patients randomized to edifoligide and placebo had similar rates of death (11.7% and 10.7%, respectively), MI (2.3% and 3.2%), revascularization (14.1% and 13.9%), and rehospitalization (61.6% and 62.5%). The composite outcome of death, MI, or revascularization occurred at similar frequency in patients assigned to edifoligide and placebo (26.3% and 25.5%, respectively; hazard ratio 1.03 [95% CI 0.89-1.18], P = .721). Factors associated with death, MI, or revascularization at 5 years included peripheral and/or cerebrovascular disease, time on cardiopulmonary bypass, lung disease, diabetes mellitus, and congestive heart failure.

CONCLUSIONS

Up to a quarter of patients undergoing CABG will have a major cardiac event or repeat revascularization procedure within 5 years of surgery. Edifoligide does not affect outcomes after CABG; however, common identifiable baseline and procedural risk factors are associated with long-term outcomes after CABG.

Cost effectiveness of drug-eluting stents in acute myocardial infarction patients in Germany: results from administrative data using a propensity score-matching approach

BACKGROUND

The high number of patients with acute myocardial infarction (AMI) has facilitated greater research, resulting in the development of innovative medical devices. So far, results from economic evaluations that compared drug-eluting stents (DES) and bare-metal stents (BMS) have not shown clear evidence that one intervention is more cost effective than the other.

OBJECTIVE

The aim of this study was to measure the cost effectiveness of DES compared with BMS in routine care.

METHODS

We used administrative data from a large German sickness fund to compare the costs and effectiveness of DES and BMS in patients with AMI. Patients with hospital admission after AMI in 2004 and 2005 were followed up for 1 year after hospital discharge. The cost of treatment and survival after 365 days were compared for patients treated with DES and BMS. We adjusted for covariates defined according to the Ontario Acute Myocardial Infarction Mortality Prediction Rules using propensity score matching. After matching, we calculated incremental cost-effectiveness ratios (ICERs) by (i) using sample means based on bootstrapping procedures and (ii) estimating generalized linear mixed models for costs and survival.

RESULTS

After propensity score matching, the sample included 719 patients treated with DES and 719 patients treated with BMS. A comparison of sample means resulted in average costs of € 12 714 and € 11 714 for DES and BMS, respectively, in 2005 German euros. Difference in 365-day survival was not statistically significant (700 patients with DES and 701 with BMS). The ICER of DES versus BMS was -€ 718 709 per life saved. Bootstrapping resulted in DES being dominated by BMS in 54.5% of replications and DES being a dominant strategy in 2.7% of replications. Results from regression models and sensitivity analyses confirm these results.

CONCLUSION

Treatment with DES after admission with AMI is less cost effective than treatment with BMS. Our results are in line with other cost-effectiveness analyses that used administrative data, i.e. under routine care conditions. However, our results do not preclude that DES may be cost effective in specific patient subgroups.

Variable selection for covariate-adjusted semiparametric inference in randomized clinical trials

Extensive baseline covariate information is routinely collected on participants in randomized clinical trials, and it is well recognized that a proper covariate-adjusted analysis can improve the efficiency of inference on the treatment effect. However, such covariate adjustment has engendered considerable controversy, as post hoc selection of covariates may involve subjectivity and may lead to biased inference, whereas prior specification of the adjustment may exclude important variables from consideration. Accordingly, how to select covariates objectively to gain maximal efficiency is of broad interest. We propose and study the use of modern variable selection methods for this purpose in the context of a semiparametric framework, under which variable selection in modeling the relationship between outcome and covariates is separated from estimation of the treatment effect, circumventing the potential for selection bias associated with standard analysis of covariance methods. We demonstrate that such objective variable selection techniques combined with this framework can identify key variables and lead to unbiased and efficient inference on the treatment effect. A critical issue in finite samples is validity of estimators of uncertainty, such as standard errors and confidence intervals for the treatment effect. We propose an approach to estimation of sampling variation of estimated treatment effect and show its superior performance relative to that of existing methods.

FSR Methods for Second-Order Regression Models

Most variable selection techniques focus on first-order linear regression models. Often, interaction and quadratic terms are also of interest, but the number of candidate predictors grows very fast with the number of original predictors, making variable selection more difficult. Forward selection algorithms are thus developed that enforce natural hierarchies in second-order models to control the entry rate of uninformative effects and to equalize the false selection rates from first-order and second-order terms. Method performance is compared through Monte Carlo simulation and illustrated with data from a Cox regression and from a response surface experiment.

Costs and quality of hospitals in different health care systems: a multi-level approach with propensity score matching

Cross-country comparisons of costs and quality between hospitals are often made at the macro level. The goal of this study was to explore methods to compare micro-level data from hospitals in different health care systems. To do so, we developed a multi-level framework in combination with a propensity score matching technique using similarly structured data for patients receiving treatment for acute myocardial infarction in German and US Veterans Health Administration hospitals. Our case study shows important differences in results between multi-level regressions based on matched and unmatched samples. We conclude that propensity score matching techniques are an appropriate way to deal with the usual baseline imbalances across the samples from different countries. Multi-level models are recommendable to consider the clustered structure of the data when patient-level data from different hospitals and health care systems are compared. The results provide an important justification for exploring new ways in performing health system comparisons.