One of the key factors that influence treatment decision for patients with cancer is survival analyses. Network meta-analyses (NMA) of survival data are usually performed using the individual summary statistic—the hazard ratio (HR) from each study as a measure of relative treatment effect. This approach assumes that the proportional hazard (PH) assumption holds—that is, the relative hazards of the event are constant over time. In a recently published paper in PLOS ONE[1], my colleagues and I review the methods and reporting of survival analyses in randomised controlled trials (RCTs) in the oncology setting and assess the suitability and relevance of survival data reported in RCTs for inclusion in NMA.

We systematically searched five oncology journals and identified 32 of the most recent Phase III RCTs which analysed a survival outcome (also known as time-to-event outcomes). Details of the reporting of the survival analyses were collected and reviewed across the studies to include the methods of analysis and presentation of the results.

We found that none of the RCT publications reported details relating to a strategy for statistical model building, the goodness of fit of the final model, or final model validation for the analysis of survival outcomes. The majority of studies reported the use of Cox PH regression to analyse survival endpoints (reporting HRs as measures of relative treatment effect). However, most publications failed to report the validation of the statistical models in terms of the PH assumption. Furthermore, our crude assessment of the reported survival Kaplan Meier curves suggested that the PH assumption was unlikely to be appropriate for many of the survival outcomes analysed across the studies.

This research highlights deficiencies in the reporting of the methodology and validity of survival analyses within oncology RCTs. In many cases, the PH assumption which underpins the most common strategy for the evidence synthesis of survival outcomes may be inappropriate thus impacting payer decisions based upon cost-effectiveness analyses. We conclude that alternative approaches for the NMA of survival outcomes when the PH assumption is violated are required if valid clinical decisions are to be made. For further information on potential alternative approaches to single parameter NMA of survival data contact us.

[1] Batson S, Greenall G, Hudson P (2016). Review of the Reporting of Survival Analyses within Randomised Controlled Trials and the Implications for Meta-Analysis. PLoS One 10(15):e0154870.

 

How Glympse Bio oversubscribed their Series B funding amidst the pandemic

View Now