In August 2025, FDA issued the draft guidance, Approaches to Assessment of Overall Survival in Oncology Clinical Trials. This guidance is based in part on discussions held at a joint FDA, American Association for Cancer Research (AACR), and American Statistical Association (ASA) public workshop held in July 2023. It also puts into writing feedback that I have received from the Agency on oncology clinical trials and guidance that I have related to our oncology clients.
The focus of the new guidance is overall survival in randomized trials where overall survival is not the primary endpoint. Overall survival is still the ‘gold standard’ for efficacy and safety but may be impractical as a primary endpoint in diseases with a natural, untreated survival of years. Single-arm trials are not discussed as it is difficult to interpret time-to-event endpoints in such trials.
I am reminded of what I learned when I first started working on oncology products early in my career in regulatory affairs. This was back in the dark ages where the mainstay of treatment was cytotoxic agents. The main mantra at the time was that in order for a product to be approved, it either needed to make the patient feel better or live longer.
We have come a long way since then, have turned some cancers from uniformly fatal to potentially curable, and, for many cancers, have developed an arsenal of potential treatment options. As patients cycle through available treatment options, we have relied on and successfully used measures of efficacy other than overall survival, such as progression free survival (PFS), as the primary endpoint in many registration studies. These other endpoints have been very helpful for determining a product’s efficacy, especially when patients receive additional therapy after progression, which make the assessment of overall survival more challenging.
With this in mind, one may question why the emphasis on survival now. The answer is simple but complex at the same time. In oncology, survival is an efficacy measure but also a key safety parameter and should be included as a pre-specified endpoint in all randomized protocols.
There are a number of examples in oncology drug development where one might see an improvement in a measure such as PFS but the survival did not follow the same trend. When this happens and the survival is longer in the control versus investigational arm, the data may be insufficient for regulatory approval. Perhaps the investigational agent is doing more harm than good.
Overall Survival Analysis Plans in Protocols and Statistical Analysis Plans
New to this guidance, and very important, is the recommendation for a pre-specified plan to assess overall survival as a safety endpoint, with an aim to evaluate for potential harm due to the investigational product. FDA recommends that overall survival analyses be included in the clinical protocol and Statistical Analysis Plan (SAP) and discussed with FDA in advance.
In development of the protocol and the SAP, it is important to consider the amount of survival data to be collected, as well as the timing of the analyses, as the use of early or immature data can cause high uncertainty in treatment effect estimates. Analyses should be conducted only after a pre-specified number of deaths have occurred to allow sufficient precision around survival estimates.
It is recommended that interim analyses of overall survival be included, in most trials, and that independent data monitoring committees oversee these analyses. However, in trials where the mortality rate may be low, interim analyses may not be feasible.
Not unexpectedly, the guidance points out that crossover to another treatment arm can impact the interpretation of overall survival results; therefore, use of crossover designs should be limited.
Study Design and Statistical Analysis Considerations
The guidance includes recommendations for study design features and statistical analyses important in assessing overall survival. The necessity of including details of the survival analyses in the protocol and SAP is emphasized, as any analyses not pre-specified in these documents will be considered exploratory. General recommendations include defining the primary analysis set as all randomized patients, providing hazard ratios and 95% confidence intervals (CI) as summary measures, and presenting Kaplan-Meier plots of overall survival. Justification for the planned length of follow-up and number of events required for survival estimates, and methods for handling missing data, intercurrent events, and multiple testing should all be addressed. Details of planned interim analyses, where applicable, should also be included. If hypothesis testing of overall survival is planned, then the methods for alpha spending and control of the study-wise Type I error rate when testing multiple times and/or multiple endpoints should be described in the SAP.
Specific to this guidance, and a key focus, is a discussion of the study design and statistical considerations necessary for effectively evaluating the assessment of harm. In particular, the objective is to collect sufficient survival data to allow estimation of an overall survival measure with a pre-specified degree of precision. For example, assessment of harm can be based on a point estimate of the overall survival hazard ratio and its CI, where a hazard ratio greater than 1 indicates an increased risk of death occurring in the treatment group compared to the control group. If a clinically relevant threshold for the hazard ratio (i.e., a number greater than 1) is not contained within the CI, then it is concluded that there does not appear to be additional harm to patients from the treatment versus the control. The choice of the clinically relevant threshold must be pre-defined and justified, and the study must collect sufficient survival data to allow adequate precision of the CI around the hazard ratio estimate.
If there are subgroups of patients that are known or suspected to be more vulnerable and their overall survival may differ from the overall population, then they should be identified and subgroup analyses should planned in the protocol and SAP. The study design should allow sufficient data collection in the subgroups of interest for estimation of survival measures, and where appropriate, hypothesis testing. In some cases, stratified randomization may be helpful.
Regulatory Considerations
The last section of the guidance, entitled Regulatory Considerations, addresses the Benefit-Risk Determination. For trials that are designed to support approval, the SAP should include the timing of the primary analysis of the primary endpoint and the timing of the analysis of overall survival to assess potential harm. When discussing trial design with FDA, one can expect that FDA will want to know how mature the survival data will be at the time of the primary analysis.
In this section of the guidance, FDA points out that when efficacy is supported by a clinical endpoint but there is significant uncertainty in the overall survival results, FDA may choose to approve the product under accelerated approval. The approval could be converted to full approval when there are robust and interpretable overall survival results. If overall survival data are immature at the time of approval, FDA may make the collection of additional post marketing overall survival data a post marketing requirement (PMR) or post marketing commitment (PMC).
We at PROMETRIKA have designed, implemented, analyzed, and reported on many oncology studies and have extensive experience with the ways to implement the recommendations in the new guideline. We also offer independent data monitoring committee establishment and management services to oversee study conduct and required interim analysis.
