Article type
Year
Abstract
Background: Selective cross-over (SCO), defined as the opportunity given to patients in a randomised controlled trial (RCT) to switch to the experimental arm, is an increasingly common phenomenon. Although there are various reasons justifying it, the equipoise principle is challenged and problems in the data analysis and interpretation may arise.
Objectives: (i) To assess the prevalence of SCO in scientific literature concerning the efficacy of biological and hormonal therapies for breast cancer (BC); (ii) To identify the statistical methods used to manage it.
Methods: RCTs assessing the efficacy of biological and hormonal therapies in both early and metastatic BC patients published between January 2000 and July 2012 were searched. For trials in which SCO has occurred, the following characteristics were recorded: primary end point, reason justifying the cross-over (i.e. interim analysis), total randomised patients, fraction of patients switching, statistical methods used.
Results: Figure 1 shows the flow diagram of studies. Seventy-two RCTs were identified. Cross-over occurred in 14 RCTs (19.4%) (Table 1). When SCO occurred, the methods mostly used to analyse data were: (1) Intention To Treat (ITT) analysis (2) Censored analysis (3) Inverse Probability of Censoring Weighting (IPCW) analysis All the studies in the early setting presented ITT analyses; two studies (BIG 1-98, HERA) conducted censored analysis; two studies (BIG 1-98, MA17) conducted IPCW analysis. All the studies in the metastatic setting reported the ITT analysis and two (EGF104900, Mouridsen 2003) conducted censored analysis. Censored and IPCW analyses led to results favouring the experimental drug compared to ITT (Table 2).
Conclusions: SCO entails ethical and methodological issues: since different methods lead to different results, further research on the impact of the various strategies is needed. It is important to keep the phenomenon monitored—not just in BC.
Objectives: (i) To assess the prevalence of SCO in scientific literature concerning the efficacy of biological and hormonal therapies for breast cancer (BC); (ii) To identify the statistical methods used to manage it.
Methods: RCTs assessing the efficacy of biological and hormonal therapies in both early and metastatic BC patients published between January 2000 and July 2012 were searched. For trials in which SCO has occurred, the following characteristics were recorded: primary end point, reason justifying the cross-over (i.e. interim analysis), total randomised patients, fraction of patients switching, statistical methods used.
Results: Figure 1 shows the flow diagram of studies. Seventy-two RCTs were identified. Cross-over occurred in 14 RCTs (19.4%) (Table 1). When SCO occurred, the methods mostly used to analyse data were: (1) Intention To Treat (ITT) analysis (2) Censored analysis (3) Inverse Probability of Censoring Weighting (IPCW) analysis All the studies in the early setting presented ITT analyses; two studies (BIG 1-98, HERA) conducted censored analysis; two studies (BIG 1-98, MA17) conducted IPCW analysis. All the studies in the metastatic setting reported the ITT analysis and two (EGF104900, Mouridsen 2003) conducted censored analysis. Censored and IPCW analyses led to results favouring the experimental drug compared to ITT (Table 2).
Conclusions: SCO entails ethical and methodological issues: since different methods lead to different results, further research on the impact of the various strategies is needed. It is important to keep the phenomenon monitored—not just in BC.
