Sampling weighting strategies in causal mediation analysis.

BACKGROUND: Causal mediation analysis plays a crucial role in examining causal effects and causal mechanisms. Yet, limited work has taken into consideration the use of sampling weights in causal mediation analysis. In this study, we compared different strategies of incorporating sampling weights into causal mediation analysis. METHODS: We conducted a simulation study to assess 4 different sampling weighting strategies-1) not using sampling weights, 2) incorporating sampling weights into mediation "cross-world" weights, 3) using sampling weights when estimating the outcome model, and 4) using sampling weights in both stages. We generated 8 simulated population scenarios comprising an exposure (A), an outcome (Y), a mediator (M), and six covariates (C), all of which were binary. The data were generated so that the true model of A given C and the true model of A given M and C were both logit models. We crossed these 8 population scenarios with 4 different sampling methods to obtain 32 total simulation conditions. For each simulation condition, we assessed the performance of 4 sampling weighting strategies when calculating sample-based estimates of the total, direct, and indirect effects. We also applied the four sampling weighting strategies to a case study using data from the National Survey on Drug Use and Health (NSDUH). RESULTS: Using sampling weights in both stages (mediation weight estimation and outcome models) had the lowest bias under most simulation conditions examined. Using sampling weights in only one stage led to greater bias for multiple simulation conditions. DISCUSSION: Using sampling weights in both stages is an effective approach to reduce bias in causal mediation analyses under a variety of conditions regarding the structure of the population data and sampling methods.

Comparison between asymptotic and re-randomisation tests under non-proportional hazards in a randomised controlled trial using the minimisation method.

BACKGROUND: Pocock-Simon's minimisation method has been widely used to balance treatment assignments across prognostic factors in randomised controlled trials (RCTs). Previous studies focusing on the survival outcomes have demonstrated that the conservativeness of asymptotic tests without adjusting for stratification factors, as well as the inflated type I error rate of adjusted asymptotic tests conducted in a small sample of patients, can be relaxed using re-randomisation tests. Although several RCTs using minimisation have suggested the presence of non-proportional hazards (non-PH) effects, the application of re-randomisation tests has been limited to the log-rank test and Cox PH models, which may result in diminished statistical power when confronted with non-PH scenarios. To address this issue, we proposed two re-randomisation tests based on a maximum combination of weighted log-rank tests (MaxCombo test) and the difference in restricted mean survival time (dRMST) up to a fixed time point τ , both of which can be extended to adjust for randomisation stratification factors. METHODS: We compared the performance of asymptotic and re-randomisation tests using the MaxCombo test, dRMST, log-rank test, and Cox PH models, assuming various non-PH situations for RCTs using minimisation, with total sample sizes of 50, 100, and 500 at a 1:1 allocation ratio. We mainly considered null, and alternative scenarios featuring delayed, crossing, and diminishing treatment effects. RESULTS: Across all examined null scenarios, re-randomisation tests maintained the type I error rates at the nominal level. Conversely, unadjusted asymptotic tests indicated excessive conservatism, while adjusted asymptotic tests in both the Cox PH models and dRMST indicated inflated type I error rates for total sample sizes of 50. The stratified MaxCombo-based re-randomisation test consistently exhibited robust power across all examined scenarios. CONCLUSIONS: The re-randomisation test is a useful alternative in non-PH situations for RCTs with minimisation using the stratified MaxCombo test, suggesting its robust power in various scenarios.

The impact of heterogeneity on the analysis of platform trials with normally distributed outcomes.

BACKGROUND: A platform trial approach allows adding arms to on-going trials to speed up intervention discovery programs. A control arm remains open for recruitment in a platform trial while intervention arms may be added after the onset of the study and could be terminated early for efficacy and/or futility when early stopping is allowed. The topic of utilising non-concurrent control data in the analysis of platform trials has been explored and discussed extensively. A less familiar issue is the presence of heterogeneity, which may exist for example due to modification of enrolment criteria and recruitment strategy. METHOD: We conduct a simulation study to explore the impact of heterogeneity on the analysis of a two-stage platform trial design. We consider heterogeneity in treatment effects and heteroscedasticity in outcome data across stages for a normally distributed endpoint. We examine the performance of some hypothesis testing procedures and modelling strategies. The use of non-concurrent control data is also considered accordingly. Alongside standard regression analysis, we examine the performance of a novel method that was known as the pairwise trials analysis. It is similar to a network meta-analysis approach but adjusts for treatment comparisons instead of individual studies using fixed effects. RESULTS: Several testing strategies with concurrent control data seem to control the type I error rate at the required level when there is heteroscedasticity in outcome data across stages and/or a random cohort effect. The main parameter of treatment effects in some analysis models correspond to overall treatment effects weighted by stage wise sample sizes; while others correspond to the effect observed within a single stage. The characteristics of the estimates are not affected significantly by the presence of a random cohort effect and/ or heteroscedasticity. CONCLUSION: In view of heterogeneity in treatment effect across stages, the specification of null hypotheses in platform trials may need to be more subtle. We suggest employing testing procedure of adaptive design as opposed to testing the statistics from regression models; comparing the estimates from the pairwise trials analysis method and the regression model with interaction terms may indicate if heterogeneity is negligible.

Evolving Improved Sampling Protocols for Dose-Response Modelling Using Genetic Algorithms with a Profile-Likelihood Metric.

Practical limitations of quality and quantity of data can limit the precision of parameter identification in mathematical models. Model-based experimental design approaches have been developed to minimise parameter uncertainty, but the majority of these approaches have relied on first-order approximations of model sensitivity at a local point in parameter space. Practical identifiability approaches such as profile-likelihood have shown potential for quantifying parameter uncertainty beyond linear approximations. This research presents a genetic algorithm approach to optimise sample timing across various parameterisations of a demonstrative PK-PD model with the goal of aiding experimental design. The optimisation relies on a chosen metric of parameter uncertainty that is based on the profile-likelihood method. Additionally, the approach considers cases where multiple parameter scenarios may require simultaneous optimisation. The genetic algorithm approach was able to locate near-optimal sampling protocols for a wide range of sample number (n = 3-20), and it reduced the parameter variance metric by 33-37% on average. The profile-likelihood metric also correlated well with an existing Monte Carlo-based metric (with a worst-case r > 0.89), while reducing computational cost by an order of magnitude. The combination of the new profile-likelihood metric and the genetic algorithm demonstrate the feasibility of considering the nonlinear nature of models in optimal experimental design at a reasonable computational cost. The outputs of such a process could allow for experimenters to either improve parameter certainty given a fixed number of samples, or reduce sample quantity while retaining the same level of parameter certainty.

Bayesian Analysis of Multi-Factorial Experimental Designs Using SEM.

Latent repeated measures ANOVA (L-RM-ANOVA) has recently been proposed as an alternative to traditional repeated measures ANOVA. L-RM-ANOVA builds upon structural equation modeling and enables researchers to investigate interindividual differences in main/interaction effects, examine custom contrasts, incorporate a measurement model, and account for missing data. However, L-RM-ANOVA uses maximum likelihood and thus cannot incorporate prior information and can have poor statistical properties in small samples. We show how L-RM-ANOVA can be used with Bayesian estimation to resolve the aforementioned issues. We demonstrate how to place informative priors on model parameters that constitute main and interaction effects. We further show how to place weakly informative priors on standardized parameters which can be used when no prior information is available. We conclude that Bayesian estimation can lower Type 1 error and bias, and increase power and efficiency when priors are chosen adequately. We demonstrate the approach using a real empirical example and guide the readers through specification of the model. We argue that ANOVA tables and incomplete descriptive statistics are not sufficient information to specify informative priors, and we identify which parameter estimates should be reported in future research; thereby promoting cumulative research.

Sample size calculation for comparing two ROC curves.

Biomarkers are key components of personalized medicine. In this paper, we consider biomarkers taking continuous values that are associated with disease status, called case and control. The performance of such a biomarker is evaluated by the area under the curve (AUC) of its receiver operating characteristic curve. Oftentimes, two biomarkers are collected from each subject to test if one has a larger AUC than the other. We propose a simple non-parametric statistical test for comparing the performance of two biomarkers. We also present a simple sample size calculation method for this test statistic. Our sample size formula requires specification of AUC values (or the standardized effect size of each biomarker between cases and controls together with the correlation coefficient between two biomarkers), prevalence of cases in the study population, type I error rate, and power. Through simulations, we show that the testing on two biomarkers controls type I error rate accurately and the proposed sample size closely maintains specified statistical power.

Sample size calculation for mixture model based on geometric average hazard ratio and its applications to nonproportional hazard.

With the advent of cancer immunotherapy, some special features including delayed treatment effect, cure rate, diminishing treatment effect and crossing survival are often observed in survival analysis. They violate the proportional hazard model assumption and pose a unique challenge for the conventional trial design and analysis strategies. Many methods like cure rate model have been developed based on mixture model to incorporate some of these features. In this work, we extend the mixture model to deal with multiple non-proportional patterns and develop its geometric average hazard ratio (gAHR) to quantify the treatment effect. We further derive a sample size and power formula based on the non-centrality parameter of the log-rank test and conduct a thorough analysis of the impact of each parameter on performance. Simulation studies showed a clear advantage of our new method over the proportional hazard based calculation across different non-proportional hazard scenarios. Moreover, the mixture modeling of two real trials demonstrates how to use the prior information on the survival distribution among patients with different biomarker and early efficacy results in practice. By comparison with a simulation-based design, the new method provided a more efficient way to compute the power and sample size with high accuracy of estimation. Overall, both theoretical derivation and empirical studies demonstrate the promise of the proposed method in powering future innovative trial designs.

Cautionary note on regional consistency evaluation in multiregional clinical trials with binary outcomes.

Multiregional clinical trials (MRCTs) have become increasingly common during the development of new drugs to obtain simultaneous drug approvals worldwide. When planning MRCTs, a major statistical challenge is determination of the regional sample size. In general, the regional sample size must be determined as the sample size such that the regional consistency probability, defined as the probability of meeting the regional consistency criterion, is greater than a prespecified value. The Japanese Ministry of Health, Labour and Welfare proposed two criteria for regional consistency. Moreover, many researchers have proposed corresponding closed-form formulas for calculating regional consistency probabilities when the primary outcome is continuous. Although some researchers have argued that those formulas are also applicable to cases with binary outcomes, it remains questionable whether such an argument can be true. Based on simulation results, we demonstrate that the existing formulas are inappropriate for binary cases, even when the regional sample size is sufficiently large. To address this issue, we develop alternative formulas and use simulation to show that they provide accurate regional consistency probabilities. Furthermore, we present an application of our proposed formulas for an MRCT of advanced or metastatic clear-cell renal cell carcinoma.

What is replication?

Credibility of scientific claims is established with evidence for their replicability using new data. According to common understanding, replication is repeating a study's procedure and observing whether the prior finding recurs. This definition is intuitive, easy to apply, and incorrect. We propose that replication is a study for which any outcome would be considered diagnostic evidence about a claim from prior research. This definition reduces emphasis on operational characteristics of the study and increases emphasis on the interpretation of possible outcomes. The purpose of replication is to advance theory by confronting existing understanding with new evidence. Ironically, the value of replication may be strongest when existing understanding is weakest. Successful replication provides evidence of generalizability across the conditions that inevitably differ from the original study; Unsuccessful replication indicates that the reliability of the finding may be more constrained than recognized previously. Defining replication as a confrontation of current theoretical expectations clarifies its important, exciting, and generative role in scientific progress.

Ensuring the quality and specificity of preregistrations.

Researchers face many, often seemingly arbitrary, choices in formulating hypotheses, designing protocols, collecting data, analyzing data, and reporting results. Opportunistic use of "researcher degrees of freedom" aimed at obtaining statistical significance increases the likelihood of obtaining and publishing false-positive results and overestimated effect sizes. Preregistration is a mechanism for reducing such degrees of freedom by specifying designs and analysis plans before observing the research outcomes. The effectiveness of preregistration may depend, in part, on whether the process facilitates sufficiently specific articulation of such plans. In this preregistered study, we compared 2 formats of preregistration available on the OSF: Standard Pre-Data Collection Registration and Prereg Challenge Registration (now called "OSF Preregistration," http://osf.io/prereg/). The Prereg Challenge format was a "structured" workflow with detailed instructions and an independent review to confirm completeness; the "Standard" format was "unstructured" with minimal direct guidance to give researchers flexibility for what to prespecify. Results of comparing random samples of 53 preregistrations from each format indicate that the "structured" format restricted the opportunistic use of researcher degrees of freedom better (Cliff's Delta = 0.49) than the "unstructured" format, but neither eliminated all researcher degrees of freedom. We also observed very low concordance among coders about the number of hypotheses (14%), indicating that they are often not clearly stated. We conclude that effective preregistration is challenging, and registration formats that provide effective guidance may improve the quality of research.

Reporting bias in clinical trials: Progress toward transparency and next steps.

Mayookha Mitra-Majumdar and Aaron Kesselheim reflect on steps taken to combat reporting bias in clinical trials over the last two decades.

A simple clinical score to promote and enhance ferroportin disease screening.

BACKGROUND & AIMS: Ferroportin disease is a rare genetic iron overload disorder which may be underdiagnosed, with recent data suggesting it occurs at a higher prevalence than suspected. Costs and the lack of defined criteria to prompt genetic testing preclude large-scale molecular screening. Hence, we aimed to develop a readily available scoring system to promote and enhance ferroportin disease screening. METHODS: Our derivation cohort included probands tested for ferroportin disease from 2008 to 2016 in our rare disease network. Data were prospectively recorded. Univariate and multivariate logistic regression were used to determine significant criteria, and odds ratios were used to build a weighted score. A cut-off value was defined using a ROC curve with a predefined aim of 90% sensitivity. An independent cohort was used for cross validation. RESULTS: Our derivation cohort included 1,306 patients. Mean age was 55±14 years, ferritin 1,351±1,357 µg/L, and liver iron concentration (LIC) 166±77 µmol/g. Pathogenic variants (n = 32) were identified in 71 patients. In multivariate analysis: female sex, younger age, higher ferritin, higher LIC and the absence of hypertension or diabetes were significantly associated with the diagnosis of ferroportin disease (AUROC in whole derivation cohort 0.83 [0.78-0.88]). The weighted score was based on sex, age, the presence of hypertension or diabetes, ferritin level and LIC. An AUROC of 0.83 (0.77-0.88) was obtained in the derivation cohort without missing values. Using 9.5 as a cut-off, sensitivity was 93.6 (91.7-98.3) %, specificity 49.5 (45.5-53.6) %, positive likelihood ratio 1.8 (1.6-2.0) and negative likelihood ratio 0.17 (0.04-0.37). CONCLUSION: We describe a readily available score with simple criteria and good diagnostic performance that could be used to screen patients for ferroportin disease in routine clinical practice. LAY SUMMARY: Increased iron burden associated with metabolic syndrome is a very common condition. Ferroportin disease is a dominant genetic iron overload disorder whose prevalence is higher than initially thought. They can be difficult to distinguish from each other, but the limited availability of genetic testing and the lack of definitive guidelines prevent adequate screening. We herein describe a simple and definitive clinical score to help clinicians decide whether to perform genetic testing.

Analysis of spin in vascular surgery randomized controlled trials with nonsignificant outcomes.

OBJECTIVE: Spin is the manipulation of language that distorts the interpretation of objective findings. The purpose of this study is to describe the characteristics of spin found in statistically nonsignificant randomized controlled trials (RCT) comparing carotid endarterectomy with carotid artery stenting for carotid artery stenosis (CS), and endovascular repair with open repair (OR) for abdominal aortic aneurysms (AAA). METHODS: A search of MEDLINE, EMBASE, and the Cochrane Controlled Register of Trials was performed in June 2020 for studies published describing AAA or CS. All phase III RCTs with nonsignificant primary outcomes comparing open repair with endovascular repair or carotid endarterectomy to carotid artery stenting were included. Studies were appraised for the characteristics and severity of spin using a validated tool. Binary logistic regression was performed to assess the association of spin grade to (1) funding source (commercial vs noncommercial) and (2) the publishing journal's impact factor. RESULTS: Thirty-one of 355 articles captured were included for analysis. Spin was identified in 9 abstracts (9/18) and 13 main texts (13/18) of AAA articles and 7 abstracts (7/13) and 10 main texts (10/13) of CS articles. For both AAA and CS articles, spin was most commonly found in the discussion section, with the most commonly used strategy being the interpretation of statistically nonsignificant primary results to show treatment equivalence or rule out adverse treatment effects. Increasing journal impact factor was associated with a statistically significant lower likelihood of spin in the study title or abstract conclusion (ß odds ratio, 0.96; 95% confidence interval, 0.94-0.98; P < .01); no significant association could be found with funding source (ß odds ratio, 1.33; 95% confidence interval, 0.30-5.92; P = .71). CONCLUSIONS: A large proportion of statistically nonsignificant RCTs contain interpretations that are inconsistent with their results. These findings should prompt authors and readers to appraise study findings independently and to limit the use of spin in study interpretations.

Guidelines for Designing and Evaluating Feasibility Pilot Studies.

BACKGROUND: Pilot studies test the feasibility of methods and procedures to be used in larger-scale studies. Although numerous articles describe guidelines for the conduct of pilot studies, few have included specific feasibility indicators or strategies for evaluating multiple aspects of feasibility. In addition, using pilot studies to estimate effect sizes to plan sample sizes for subsequent randomized controlled trials has been challenged; however, there has been little consensus on alternative strategies. METHODS: In Section 1, specific indicators (recruitment, retention, intervention fidelity, acceptability, adherence, and engagement) are presented for feasibility assessment of data collection methods and intervention implementation. Section 1 also highlights the importance of examining feasibility when adapting an intervention tested in mainstream populations to a new more diverse group. In Section 2, statistical and design issues are presented, including sample sizes for pilot studies, estimates of minimally important differences, design effects, confidence intervals (CI) and nonparametric statistics. An in-depth treatment of the limits of effect size estimation as well as process variables is presented. Tables showing CI around parameters are provided. With small samples, effect size, completion and adherence rate estimates will have large CI. CONCLUSION: This commentary offers examples of indicators for evaluating feasibility, and of the limits of effect size estimation in pilot studies. As demonstrated, most pilot studies should not be used to estimate effect sizes, provide power calculations for statistical tests or perform exploratory analyses of efficacy. It is hoped that these guidelines will be useful to those planning pilot/feasibility studies before a larger-scale study.

Thought-based interaction: Same data, same methods, different results?

Restoration of communication in people with complete motor paralysis-a condition called complete locked-in state (CLIS)-is one of the greatest challenges of brain-computer interface (BCI) research. New findings have recently been presented that bring us one step closer to this goal. However, the validity of the evidence has been questioned: independent reanalysis of the same data yielded significantly different results. Reasons for the failure to replicate the findings must be of a methodological nature. What is the best practice to ensure that results are stringent and conclusive and analyses replicable? Confirmation bias and the counterintuitive nature of probability may lead to an overly optimistic interpretation of new evidence. Lack of detail complicates replicability.

Exact replication: Foundation of science or game of chance?

The need for replication of initial results has been rediscovered only recently in many fields of research. In preclinical biomedical research, it is common practice to conduct exact replications with the same sample sizes as those used in the initial experiments. Such replication attempts, however, have lower probability of replication than is generally appreciated. Indeed, in the common scenario of an effect just reaching statistical significance, the statistical power of the replication experiment assuming the same effect size is approximately 50%-in essence, a coin toss. Accordingly, we use the provocative analogy of "replicating" a neuroprotective drug animal study with a coin flip to highlight the need for larger sample sizes in replication experiments. Additionally, we provide detailed background for the probability of obtaining a significant p value in a replication experiment and discuss the variability of p values as well as pitfalls of simple binary significance testing in both initial preclinical experiments and replication studies with small sample sizes. We conclude that power analysis for determining the sample size for a replication study is obligatory within the currently dominant hypothesis testing framework. Moreover, publications should include effect size point estimates and corresponding measures of precision, e.g., confidence intervals, to allow readers to assess the magnitude and direction of reported effects and to potentially combine the results of initial and replication study later through Bayesian or meta-analytic approaches.

Statistical methods for comparing test positivity rates between countries: Which method should be used and why?

The test positivity (TP) rate has emerged as an important metric for gauging the illness burden due to COVID-19. Given the importance of COVID-19 TP rates for understanding COVID-related morbidity, researchers and clinicians have become increasingly interested in comparing TP rates across countries. The statistical methods for performing such comparisons fall into two general categories: frequentist tests and Bayesian methods. Using data from Our World in Data (ourworldindata.org), we performed comparisons for two prototypical yet disparate pairs of countries: Bolivia versus the United States (large vs. small-to-moderate TP rates), and South Korea vs. Uruguay (two very small TP rates of similar magnitude). Three different statistical procedures were used: two frequentist tests (an asymptotic z-test and the 'N-1' chi-square test), and a Bayesian method for comparing two proportions (TP rates are proportions). Results indicated that for the case of large vs. small-to-moderate TP rates (Bolivia versus the United States), the frequentist and Bayesian approaches both indicated that the two rates were substantially different. When the TP rates were very small and of similar magnitude (values of 0.009 and 0.007 for South Korea and Uruguay, respectively), the frequentist tests indicated a highly significant contrast, despite the apparent trivial amount by which the two rates differ. The Bayesian method, in comparison, suggested that the TP rates were practically equivalent-a finding that seems more consistent with the observed data. When TP rates are highly similar in magnitude, frequentist tests can lead to erroneous interpretations. A Bayesian approach, on the other hand, can help ensure more accurate inferences and thereby avoid potential decision errors that could lead to costly public health and policy-related consequences.

Research Priorities for Patients with Heart Failure and Central Sleep Apnea. An Official American Thoracic Society Research Statement.

Background: Central sleep apnea (CSA) is common among patients with heart failure and has been strongly linked to adverse outcomes. However, progress toward improving outcomes for such patients has been limited. The purpose of this official statement from the American Thoracic Society is to identify key areas to prioritize for future research regarding CSA in heart failure.Methods: An international multidisciplinary group with expertise in sleep medicine, pulmonary medicine, heart failure, clinical research, and health outcomes was convened. The group met at the American Thoracic Society 2019 International Conference to determine research priority areas. A statement summarizing the findings of the group was subsequently authored using input from all members.Results: The workgroup identified 11 specific research priorities in several key areas: 1) control of breathing and pathophysiology leading to CSA, 2) variability across individuals and over time, 3) techniques to examine CSA pathogenesis and outcomes, 4) impact of device and pharmacological treatment, and 5) implementing CSA treatment for all individualsConclusions: Advancing care for patients with CSA in the context of heart failure will require progress in the arenas of translational (basic through clinical), epidemiological, and patient-centered outcome research. Given the increasing prevalence of heart failure and its associated substantial burden to individuals, society, and the healthcare system, targeted research to improve knowledge of CSA pathogenesis and treatment is a priority.

Reporting of Results by Sex in Randomized Controlled Trials of Acute Stroke Therapies (2010-2020).

Background and Purpose: When reporting primary results from randomized controlled trials, recommendations include reporting results by sex. We reviewed the reporting of results by sex in contemporary acute stroke randomized controlled trials. Methods: We searched MEDLINE for articles reporting the primary results of phase 2 or 3 stroke randomized controlled trials published between 2010 and June 2020 in one of nine major clinical journals. Eligible trials were restricted to those with a therapeutic intervention initiated within one month of stroke onset. Of primary interest was the reporting of results by sex for the primary outcome. We performed bivariate analyses using Fisher exact tests to identify study-level factors associated with reporting by sex and investigated temporal trends using an exact test for trend. Results: Of the 115 studies identified, primary results were reported by sex in 37% (n=42). Reporting varied significantly by journal, with the New England Journal of Medicine (61%) and Lancet journals (40%) having the highest rates (P=0.03). Reporting also differed significantly by geographic region (21% Europe versus 48% Americas, P=0.03), trial phase (13% phase 2 versus 40% phase 3, P=0.05), and sample size (24% <250 participants versus 61% >750 participants, P<0.01). Although not statistically significant (P=0.11), there was a temporal trend in favor of greater reporting among later publications (25% 2010­2012 versus 48% 2019­2020). Conclusions: Although reporting of primary trial results by sex improved from 2010 to 2020, the prevalence of reporting in major journals is still low. Further efforts are required to encourage journals and authors to comply with current reporting recommendations.

Unfinished business: Terminated cancer trials and the relevance of treatment intent, sponsors and intervention types.

The aim of the study was to report on the association of trial sponsors with intervention type, treatment intent, recruitment success and reasons to terminate cancer trials. The ClinicalTrials database was searched for interventional Phase 3 cancer trials (01/2006-05/2017). Noncancer studies and ongoing studies were excluded, permanently suspended studies were counted as terminated. Trials were stratified according to sponsors (industry/nonindustry), intervention type, setting (curative/palliative) and intent of intervention (curative/symptom-control/life-extending). We identified 345 terminated trials and 1137 completed studies as a control group. The frequency of premature termination did not differ significantly between sponsors. Time to termination was shorter but recruitment per month prior to termination was higher in industry-sponsored studies (7.0 vs 2.2 patients/month; P < .001). Drug interventions were more common in industry-sponsored, all other interventions in nonindustry-sponsored settings (P < .001). Life-extending palliative interventions occurred more frequently, symptom-control interventions in a curative setting less frequently in industry-sponsored trials (both P < .001). Intervention, setting and intent were not associated with termination in industry-sponsored trials. In nonindustry-sponsored trials, the frequency of drug interventions and life-extending (noncurative) interventions were increased in terminated trials (both P < .05); symptom-control interventions in curative settings occurred more frequently in completed studies. Industry-sponsored trials were more often terminated due to toxicity/inefficacy while lack of accrual occurred more frequently in nonindustry-sponsored trials (P < .01). Interventions, treatment setting/intent and reasons for termination differed between sponsor types. In nonindustry-sponsored trials, drug interventions and life-extending (noncurative) interventions were associated with premature termination and symptom-control interventions (curative setting) were associated with trial completion.