A Novel Method to Adjust Efficacy Estimates for Uptake of Other Active Treatments in Long-Term Clinical Trials

Abstract

BackgroundWhen rates of uptake of other drugs differ between treatment arms in long-term trials, the true benefit or harm of the treatment may be underestimated. Methods to allow for such contamination have often been limited by failing to preserve the randomization comparisons. In the Fenofibrate Intervention and Event Lowering in Diabetes (FIELD) study, patients were randomized to fenofibrate or placebo, but during the trial many started additional drugs, particularly statins, more so in the placebo group. The effects of fenofibrate estimated by intention-to-treat were likely to have been attenuated. We aimed to quantify this effect and to develop a method for use in other long-term trials.Methodology/Principal FindingsWe applied efficacies of statins and other cardiovascular drugs from meta-analyses of randomized trials to adjust the effect of fenofibrate in a penalized Cox model. We assumed that future cardiovascular disease events were reduced by an average of 24% by statins, and 20% by a first other major cardiovascular drug. We applied these estimates to each patient who took these drugs for the period they were on them. We also adjusted the analysis by the rate of discontinuing fenofibrate. Among 4,900 placebo patients, average statin use was 16% over five years. Among 4,895 assigned fenofibrate, statin use was 8% and nonuse of fenofibrate was 10%. In placebo patients, use of cardiovascular drugs was 1% to 3% higher. Before adjustment, fenofibrate was associated with an 11% reduction in coronary events (coronary heart disease death or myocardial infarction) (P = 0.16) and an 11% reduction in cardiovascular disease events (P = 0.04). After adjustment, the effects of fenofibrate on coronary events and cardiovascular disease events were 16% (P = 0.06) and 15% (P = 0.008), respectively.Conclusions/SignificanceThis novel application of a penalized Cox model for adjustment of a trial estimate of treatment efficacy incorporates evidence-based estimates for other therapies, preserves comparisons between the randomized groups, and is applicable to other long-term trials. In the FIELD study example, the effects of fenofibrate on the risks of coronary heart disease and cardiovascular disease events were underestimated by up to one-third in the original analysis.Trial RegistrationControlled-Trials.com ISRCTN64783481

Favourable effects of fenofibrate on lipids and cardiovascular disease in women with type 2 diabetes: results from the Fenofibrate Intervention and Event Lowering in Diabetes (FIELD) study

“…Nor did insulin use or differences in statin uptake between the sexes explain the difference in LDL-cholesterol and fenofibrate response (see ESM Fig. 1 and Simes et al [11]). Uptake of statin therapy during the trial may have confounded the treatment effects.…”

“…HRs and 95% CIs were generated from Cox proportional-hazards models [13]. Analyses were adjusted for covariates (including ethnicity, age, diabetes duration, BMI, waist-to-hip ratio, systolic and diastolic blood pressure, smoking, prior CVD, prior coronary revascularisation, hypertension, microvascular disease, baseline total cholesterol, LDL-cholesterol, HDL-cholesterol, creatinine, homocysteine, dyslipidaemia, microalbuminuria, macroalbuminuria, use of metformin, and use of sulfonylurea) and on-trial statin uptake [11]. ANOVA methods were used to assess the treatment-effect difference in the change in lipid and apolipoprotein levels.…”

“…Adjusted effects of fenofibrate treatment With adjustment for statin use and other baseline covariates, [11] allocation to fenofibrate reduced total CVD events by 17% overall (95% CI 6%, 27%, and in women by 30% (95% CI 8%, 46%) although not significantly in men (13%; 95% CI -1%, 24%), with no statistical evidence of heterogeneity of effect by sex ( p=0.17). There was no significant reduction in non-fatal MI among women (35%; 95% CI -5%, 61%) but there was for men (28%; 95% CI 5%, 47%).…”

“…In FIELD, fenofibrate did not significantly reduce the primary endpoint of CHD events (non-fatal MI and CHD death; HR 0.89; 95% CI 0.75, 1.05, p=0.16) [4]. After adjustment for fenofibrate adherence and imbalanced uptake of statins and other cardiovascular drugs, this effect increased but remained non-significant (HR 0.84; 95% CI 0.69, 1.01, p=0.06) [11]. There was no significant interaction between fenofibrate treatment and sex for total CVD events (non-fatal MI, stroke, all CVD death, and coronary and carotid revascularisation), the pre-specified endpoint for all subgroup analyses [4].…”

Low-Dose Aspirin for Preventing Recurrent Venous Thromboembolism

“…In addition, we estimated the efficacy of aspirin in a fully adherent group by adjusting the treatment effect in the intention-to-treat analysis for the nonadherence rates averaged over the study period; the nonadherence rates were defined as the proportion of patients assigned to aspirin who discontinued it and the proportion of patients assigned to placebo who initiated antiplatelet or anticoagulation treatment. 17 We combined the results from the ASPIRE and WARFASA studies by performing a meta-analysis of the log hazard ratios for the treatment effect on vascular events from each study weighted by their inverse variances. 18 All analyses were performed with the use of SAS software, version 9.3 (SAS Institute).…”