Background and Objectives
We investigate three statistical and methodological issues within the context of non-inferiority randomized controlled trials (RCTs), specifically those of radiotherapy regimens for the prevention of local recurrence in patients with early stage breast cancer who have undergone breast conserving surgery. These issues are: (1) the analysis of multiple time-to-event outcomes in non-inferiority RCTs; (2) the interim analysis of a binary outcome that is repeatedly assessed at pre-specified times; and (3) determining the optimal analysis population for dealing with crossovers in non-inferiority RCTs.
Methods
Issue 1: We investigated and compared the properties of four statistical models (proportional hazards model, competing risk model, marginal model and frailty model) for analyzing radiotherapy non-inferiority RCTs of patients with early stage breast cancer who are at risk for and may experience multiple failure types. We applied the four methods to data from an existing trial in which subjects with breast cancer could experience local recurrence (the primary outcome), distant recurrence, death, or a combination of these events. In addition, we compared these models using simulated examples of similar non-inferiority trials with varying hazards of each failure type.
Issue 2: We investigated and compared the properties of three methods for estimating the event proportions for an interim analysis in RCTs with a binary outcome that is repeatedly assessed at pre-specified times. Generally, interim analyses are performed after half or more of the subjects have completed full follow-up. However, depending on the duration of accrual relative to the length of follow-up, this may be inefficient, since there is a possibility that the trial will have completed accrual prior to the interim analysis. We focussed our simulations on situations where delaying the interim analysis until half or more of subjects have completed full follow-up is an inefficient approach. The methods include: 1) estimation of the event proportion based on subjects who have been followed for a pre-specified time (less than the full follow-up duration) or who experienced the outcome; 2) estimation of the event proportion based on all available data from subjects randomized by the time of the interim analysis; and 3) the Kaplan-Meier approach to estimate the event proportion. We varied the risk of the outcome, the treatment effect and the probability of an event occurring at each pre-specified time. We compared the three methods in terms of overall type I and II errors, as well as the probability of stopping early for benefit.
Issue 3: We explored the effect of subject crossover from the experimental to the standard radiotherapy arm prior to treatment initiation on the intention-to-treat, per-protocol, as-treated and combined intention-to-treat and per-protocol analysis in non-inferiority RCTs of radiotherapy for the prevention of local recurrence in patients with early stage breast cancer. We varied the non-inferiority margin, the percent of subjects who cross over and evaluated random and non-random crossover. The main comparison of the methods was done using overall type I error. In addition, we compared the methods based on estimate bias and standard error of the estimate.
Results and Conclusions
Issue 1: All four models produced similar results for the existing trial (i.e. non-inferiority was observed regardless of the method used). Simulations showed that the event-specific methods yielded contrasting results when the distribution of distant recurrence or death differed between treatment groups. We conclude that multiple models should be used as part of a comprehensive analysis.
Issue 2: We showed that conducting an interim analysis when a considerable number of subjects have completed a portion of their full follow-up duration is an efficient approach under certain scenarios where event distribution probabilities are similar between treatment groups. Under these specific scenarios, all three methods preserved the type I and II errors. In these cases, we recommend using the Kaplan-Meier method because it incorporates all the available data and has greater probability of early stopping.
Issue 3: The as-treated analysis had the best performance in terms of type I error rate. However, it can be recommended only in scenarios where crossover is random. It performed poorly in scenarios with greater than 2% non-random crossover. The intention-to-treat and per-protocol analysis performed poorly under both random and non-random crossover scenarios. / Thesis / Doctor of Philosophy (PhD)
| Identifer | oai:union.ndltd.org:mcmaster.ca/oai:macsphere.mcmaster.ca:11375/16290 |
| Date | January 2014 |
| Creators | Parpia, Sameer |
| Contributors | Thabane, Lehana, Clinical Epidemiology/Clinical Epidemiology & Biostatistics |
| Source Sets | McMaster University |
| Language | English |
| Detected Language | English |
| Type | Thesis |
Page generated in 0.0018 seconds