Elsevier

Annals of Epidemiology

Volume 16, Issue 1, January 2006, Pages 41-48
Annals of Epidemiology

Randomized Controlled Trials With Time-to-Event Outcomes: How Much Does Prespecified Covariate Adjustment Increase Power?

https://doi.org/10.1016/j.annepidem.2005.09.007Get rights and content

Purpose

We evaluated the effects of various strategies of covariate adjustment on type I error, power, and potential reduction in sample size in randomized controlled trials (RCTs) with time-to-event outcomes.

Methods

We used Cox models in simulated data sets with different treatment effects (hazard ratios [HRs] = 1, 1.4, and 1.7), covariate effects (HRs = 1, 2, and 5), covariate prevalences (10% and 50%), and censoring levels (no, low, and high). Treatment and a single covariate were dichotomous. We examined the sample size that gives the same power as an unadjusted analysis for three strategies: prespecified, significant predictive, and significant imbalance.

Results

Type I error generally was at the nominal level. The power to detect a true treatment effect was greater with adjusted than unadjusted analyses, especially with prespecified and significant-predictive strategies. Potential reductions in sample size with a covariate HR between 2 and 5 were between 15% and 44% (covariate prevalence 50%) and between 4% and 12% (covariate prevalence 10%). The significant-imbalance strategy yielded small reductions. The reduction was greater with stronger covariate effects, but was independent of treatment effect, sample size, and censoring level.

Conclusions

Adjustment for one predictive baseline characteristic yields greater power to detect a true treatment effect than unadjusted analysis, without inflation of type I error and with potentially moderate reductions in sample size. Analysis of RCTs with time-to-event outcomes should adjust for predictive covariates.

Introduction

Randomized controlled trials (RCTs) are important research tools to evaluate the usefulness of treatments and interventions (1). Heterogeneity is common among patients participating in RCTs with time-to-event outcomes (2). Prognosis commonly varies according to patient baseline characteristics, which are recorded routinely in RCTs. After proper randomization, imbalance in patient characteristics may arise by chance (3).

Covariate adjustment for prognostic baseline characteristics usually is performed with Cox proportional hazards model in RCTs with time-to-event outcomes 3, 4, 5, 6, 7, 8, 9, 10. Inclusion of a strongly predictive covariate in addition to the treatment variable in a Cox model provides three important benefits: correction for imbalance 3, 4, 6, 9, acquisition of more individualized treatment effects 3, 7, 9, and increase in statistical power, i.e., the ability to detect a treatment effect when it really exists 2, 5, 6, 7, 8, 9. Moreover, omission or misspecification of prognostic covariates in the analysis produces deviations from the proportional hazards assumptions 5, 10, 11, 12, 13, 14, 15.

The power of covariate-adjustment strategies in RCTs with time-to-event outcomes depends on various characteristics: strength of treatment effect, strength of covariate effect, covariate prevalence, and censoring level 2, 5, 6, 7, 9, 11, 12, 13, 16. Effects of covariate-adjustment strategies on statistical power and type I error, using plausible clinical scenarios, have been insufficiently studied 17, 18. Some examples of covariate adjustment in RCTs with survival outcomes are available in the medical literature, especially in oncology and cardiology 19, 20, 21, 22.

We used various strategies for choice of covariates (prespecified, predictive, and imbalance strategies) in simulated Cox proportional hazards models with one dichotomous covariate, using different treatment effects, covariate effects, covariate prevalences, and censoring levels. We aim to identify the pros and cons of each covariate adjustment strategy, with a focus on the quantification of changes in statistical power. We expressed the gain in statistical power in the decrease in sample size that gives the same power as an unadjusted analysis.

Section snippets

Models and Adjustment Strategies

Cox proportional hazards models were used to analyze the effects of treatment on a time-to-event outcome (e.g., time to death). For simplicity, a single dichotomous baseline characteristic was considered as covariate to adjust the treatment effect. We expected that our results were generalizable to more complex cases (i.e., more covariates included), as shown by others 2, 5, 6, 7, 9, 11, 13, 16. Cox model coefficients and SEs were estimated by using standard maximum-likelihood procedures.

We

Covariate Prevalence 50% and No Censoring

When there was no treatment effect, type I error was similar in all adjustment strategies and with all covariate effects (Table 1). Type I error mainly was slightly less than 5%, especially when the covariate effect was very strong (3.8% for imbalance-adjusted strategy at covariate HR = 10). This conservative estimate implies that fewer false-positive effects were identified. When there was no covariate effect, covariate adjustment strategies on average did not change treatment-effect

Discussion

Covariate adjustment yields greater statistical power than an unadjusted analysis of an RCT with time-to-event outcomes. In our study, the gain in power was translated into potential reductions in sample size, without inflation of type I error. We used several approaches of covariate adjustment in simple simulated Cox models with different treatment effects, covariate effects, covariate prevalences, and censoring levels. We found that prespecified and significant predictor-adjustment

References (26)

  • C. Schmoor et al.

    Effects of covariate omission and categorization when analysing randomized trials with the Cox model

    Stat Med

    (1997)
  • I. Ford et al.

    Model inconsistency, illustrated by the Cox proportional hazards model

    Stat Med

    (1995)
  • Z. Li

    Covariate adjustment for non-parametric tests for censored survival data

    Stat Med

    (2001)
  • Cited by (55)

    • Covariate adjustments in randomized controlled trials increased study power and reduced biasedness of effect size estimation

      2016, Journal of Clinical Epidemiology
      Citation Excerpt :

      Furthermore, adjusted analyses are uncommon in the literature, in which only 20–30% of articles published from 2001 to 2010 did so [1–5], despite the ample evidence from both simulation studies and sensitivity analyses with real data showing that adjusting covariates would increase the power of the statistical analysis. Various simulation studies have shown that adjustment of covariates could reduce the required sample size to achieve a prespecified power [12–16]. Furthermore, the strength of this reduction increased with the association between the covariates and the outcome and was independent of sample size and the balance across treatment and control groups [12,14].

    View all citing articles on Scopus

    A.V.H. was supported by the Netherlands Organization for Scientific Research (ZON/MW 908-02-117) and E.W.S. was supported by the Royal Netherlands Academy of Arts and Sciences.

    View full text