In oncology trials, we rely heavily on two time-to-event endpoints: progression-free survival (PFS) and overall survival (OS). At first glance, they seem closely related—both measure time-to-event outcomes, both use similar statistical methods, and both aim to evaluate treatment benefit [1]. Historically, improvements in PFS were often assumed to foreshadow improvements in OS, but increasingly we see trials where the statistical conclusions diverge [2]:
- PFS is statistically significant, OS is not statistically significant.
- Or, less commonly, OS is statistically significant while PFS is not statistically significant.
But why is this the case? This blog explores why PFS and OS, two of the most important endpoints in oncology trials, can sometimes lead to different conclusions and what these discrepancies mean for evaluating treatment benefit.
What are PFS and OS?
PFS measures the time from randomisation until disease progression (using radiographic imaging), or death from any cause, whichever occurs first, censoring patients who are lost to follow-up or who are alive at the time of the evaluation. OS measures the time from randomisation until death from any cause, censoring any patients who are still alive at the time of evaluation or who are lost to follow-up. PFS captures earlier events (progression), while OS only captures death [3].
Why PFS is significant, but OS is not?
Statistical Power and Event Rates
PFS is often considered to be a more suitable primary endpoint than OS due to PFS events occurring earlier and more frequently. As a result of this, more events occur sooner in PFS analyses, statistical power is reached earlier, and smaller sample sizes are often sufficient [4]. OS in contrast requires a longer follow-up period, needs more deaths to reach statistical significance, data may take years to mature and therefore may be underpowered if fewer deaths occur than expected. The timing difference alone can lead to discordant results, especially if the treatment delays progression but does not substantially extend life [5]. Endpoints like PFS allow faster evaluation of treatments and are often used in drug development or accelerated approvals [6].
If a trial is powered primarily for PFS, OS may be considered a secondary endpoint. The sample size may not be sufficient to detect modest survival improvements, which in turn leads to significant PFS and non-significant OS. Even if there is a real survival benefit, the study may lack statistical power to detect it [7].
Treatment Toxicity
The risk of PFS vs OS discordant results is higher when the gains in PFS are relatively small, but the treatment causes substantial toxicity. Even when a therapy produces larger improvements in PFS, OS can look worse if the treatment has serious side effects that lead to treatment-related deaths. Importantly, side effects don’t have to be fatal to affect survival outcomes. Toxicity can prevent patients from completing the full course of an otherwise effective treatment or make them less able to receive other beneficial therapies later in the trial [8].
Crossover Effects
In many oncology trials designs, patients in the control arm are allowed to switch to the experimental treatment arm after progression [9]. Ethically, this is often appropriate. Statistically, however, it can reduce the ability to detect OS differences. If many control patients receive the experimental drug later, the OS survival curves converge, the hazard ratio moves toward 1 and statistical significance of OS may then be lost. PFS, however, is unaffected by crossover because it measures time before progression [10].
The Post-Progression Treatment Effect
Oncology trials rarely involve a single treatment line. One of the biggest drivers of OS–PFS discrepancies is subsequent therapy. Like the cross-over effect, after disease progression, patients may receive additional treatments which can “dilute” OS differences between arms. So, while the experimental drug clearly delays tumor growth, it may not show a survival advantage [11].
Measurement Differences and Bias
PFS relies on radiographic assessments, which introduce variability in imaging schedule timing, reader variability, assessment bias (especially in open-label trials) and RECIST criteria interpretation. OS is more objective since death is a clear event. In some cases, PFS may appear improved due to assessment timing. OS, being less subjective, may not reflect the same magnitude of benefit [8].
Biological Impact of the Drug
Some treatments strongly delay tumor growth, improve symptoms, stabilize the disease, but on the other hand it may not affect tumor biology long-term or resistance development. If the drug mainly delays progression without changing ultimate disease trajectory, PFS improves but OS does not [11].
Why OS is significant, but PFS is not?
Delayed Treatment Effects
If a treatment takes time to start working, many patients may progress early, so PFS does not show an improvement of treatment over control. However, over time, the treatment can lead to long-lasting responses or durable disease control, which eventually improves OS [8]. This is common with immunotherapies, where tumors may not shrink immediately but can trigger the immune system to control cancer over the long term, unlike chemotherapy which is designed to shrink tumors quickly [12].
Statistical methods to improve OS analyses
Statistical approaches such as time-dependent Cox regression, multistate modelling, sensitivity analyses, and measurement-error models can be used to account for post-progression treatment effects and potential differences or biases in outcome measurement. Inverse probability of censoring weighting (IPCW) can be particularly useful where post-progression treatment or treatment switching is an issue.
Clinical examples of PFS-OS discordance
Evidence from clinical trials suggests that improvements in PFS do not always translate into corresponding improvements in OS. Hess et al. found that approximately 31% of treatment comparisons in randomized oncology trials involving targeted and biologic agents showed discordance between PFS and OS, with one endpoint improving while the other did not [13]. Similarly, Milrod et al. reported only a modest correlation between PFS and OS treatment effects in follicular lymphoma trials [14]. A systematic review by Ciani et al. also found that the relationship between PFS and OS varies across tumor types and treatment settings, indicating that PFS may not consistently serve as a reliable surrogate for OS [15]. This discordance can arise for several reasons, as described above. The following examples highlight how PFS and OS outcomes may differ:
Immunotherapy Trials [16] [17]:
- Drugs: Pembrolizumab (PD-1 inhibitor used in head & neck cancers), Tebentafusp (used in uveal melanoma).
- Observation: Many patients did not show early tumor shrinkage and had little difference in PFS compared to chemotherapy.
- Result: Over time, a subset of patients experienced durable responses that continued for a long time. This led to a statistically significant improvement in OS.
- Key Lesson: PFS may underestimate benefit when a treatment takes time to activate the immune system. Early progression does not necessarily mean the therapy is ineffective.
PI3K Inhibitors in Blood Cancers [18]:
- Drugs: Duvelisib (PI3K inhibitor used in chronic lymphocytic leukemia (CLL)).
- Observation: Trials showed improved PFS, meaning patients’ disease stayed controlled longer initially.
- Result: No OS improvement was seen. High toxicity caused dose reductions, treatment interruptions, and some treatment-related deaths. Additionally, the OS benefit was masked by a high rate of crossover (57% crossing from control arm to duvelisib).
- Key Lesson: Early tumor control (PFS) can be negated by treatment toxicity, which affects long-term survival.
While progression-free survival and overall survival are both important measures of treatment benefit, they answer different clinical questions and can be influenced by different factors. Differences in statistical power, treatment toxicity, crossover, subsequent therapies and disease biology can all contribute to discordant results. Understanding these factors is essential when designing oncology trials and interpreting their outcomes. Rather than viewing PFS and OS in isolation, sponsors should consider the broader clinical context and treatment landscape to determine whether a therapy delivers meaningful benefit to patients.
References
[1] Dudley NW, Wickham R, Coombs N. (2016) “An Introduction to Survival Statistics: Kaplan-Meier Analysis”. Journal of the Advanced Practitioner in Oncology (JADPRO) [Internet]. Available at: https://pmc.ncbi.nlm.nih.gov/articles/PMC5045282/ (Accessed: 01JUL2026).
[2] Broglio RW, Berry AD. (2009) “Detecting an Overall Survival Benefit that Is Derived From Progression-Free Survival”. JOURNAL of the NATIONAL CANCER INSTITUTE [Internet]. Available at: https://pmc.ncbi.nlm.nih.gov/articles/PMC4137232/ (Accessed: 01JUL2026).
[3] Delgado A, Guddati KA. (2021) “Clinical endpoints in oncology – a primer”. American Journal of Cancer Research [Internet]. Available at: https://pmc.ncbi.nlm.nih.gov/articles/PMC8085844/ (Accessed: 01JUL2026).
[4] Belin L, Tan A, Rycke DY, Dechartres A. (2020) “Progression-free survival as a surrogate for overall survival in oncology trials: a methodological systematic review”. British Journal of Cancer [Internet]. Available at: https://www.nature.com/articles/s41416-020-0805-y (Accessed: 01JUL2026).
[5] U.S. Department of Health and Human Services Food and Drug Administration. Oncology Center of Excellence. Center for Drug Evaluation and Research (CDER). Center for Biologics Evaluation and Research (CBER). (2018) “Promoting the transparency of Accelerated Approval for oncology indications”. [Internet]. Available at: https://www.fda.gov/media/71195/download (Accessed: 01JUL2026).
[6] U.S FOOD & DRUG ADMINISTRATION. (2024) “Clinical Trial Endpoints for the Approval of Cancer Drugs and Biologics Guidance for Industry”. [Internet]. Available at: https://www.fda.gov/about-fda/oncology-center-excellence/project-confirm (Accessed: 01JUL2026).
[7] Tang Y, Bycott P, Akerborg O, Jonsson L, Negrier S, Chen C. (2014) “Interpreting overall survival results when progression-free survival benefits exist in today’s oncology landscape: a metastatic renal cell carcinoma case study”. Dovepress Taylor & Francis Group. [Internet]. Available at: https://pmc.ncbi.nlm.nih.gov/articles/PMC4179830/ (Accessed: 01JUL2026).
[8] Merino M, Kasamon Y, Theoret M, Pazdur R, Kluetz P, Gormley N. (2023) “Irreconcilable Differences: The Divorce Between Response Rates, Progression-Free Survival, and Overall Survival”. Journal of Clinical Oncology. [Internet]. Available at: https://ascopubs.org/doi/10.1200/JCO.23.00225 (Accessed: 01JUL2026).
[9] Haslam A, Prasad V. (2018) “When is crossover desirable in cancer drug trials and when is it problematic?”. Annals of Oncology. [Internet]. Available at: https://pmc.ncbi.nlm.nih.gov/articles/PMC5961160/ (Accessed: 01JUL2026).
[10] Rong C. (2025) “RMST-Based Method to Adjust Crossover in Oncology Clinical Trials”. The National High School of Journal of Science. [Internet]. Available at: https://nhsjs.com/2025/rmst-based-method-to-adjust-crossover-in-oncology-clinical-trials/ (Accessed: 01JUL2026).
[11] Gutman SI, Piper M, Grant MD, et al. “Progression-Free Survival: What Does It Mean for Psychological Well-Being or Quality of Life?” [Internet] Rockville (MD): Agency for Healthcare Research and Quality (US); 2013 Apr. Available from: https://www.ncbi.nlm.nih.gov/books/NBK137759/ (Accessed 01JUL2026).
[12] CRI staff. (2025). “Immunotherapy vs. Chemotherapy: What’s the Difference?”. Cancer Research Institute. [Internet]. Available from: https://www.cancerresearch.org/blog/difference-cancer-immunotherapy-and-chemotherapy (Accessed 01JUL2026).
[13] Hess ML, Brnabic A, Mason O, Lee P, Barker S (2019). “Relationship between Progression-free Survival and Overall Survival in Randomized Clinical Trials of Targeted and Biologic Agents in Oncology“. Journal of Cancer. [Internet]. Available from: https://pmc.ncbi.nlm.nih.gov/articles/PMC6636299/ (Accessed 24AUG2026).
[14] Milrod C, Kim KW, Ollila T, Olszewski AJ, Pelcovits A (2023). “Correlation between PFS and OS in FL: Implications for Clinical Trial Endpoint Selection”. Blood. [Internet]. Available from:
[15] Ciani O, Davis S, Tappenden P, Garside R, Stein K, Cantrell A, et al. (2014). “Validation of surrogate endpoints in advanced solid tumors: systematic review of statistical methods, results, and implications for policy makers”. International Journal of Technology Assessment in Health Care. [Internet]. Available from: https://pubmed.ncbi.nlm.nih.gov/25308694/ (Accessed 24AUG2026).
[16] Nathan P, Hassel JC, Rutkowski P, Baurain JF, Butler MO, Schlaak M, et al. (2021). “Overall Survival Benefit with Tebentafusp in Metastatic Uveal Melanoma”. The New England Journal of Medicine. [Internet]. Available from: https://pubmed.ncbi.nlm.nih.gov/34551229/ (Accessed 01JUL2026).
[17] Burtness B, Harrington KJ, Greil R, Soulieres D, Tahara M, Castro Jr G, et al. (2019). “Pembrolizumab alone or with chemotherapy versus cetuximab with chemotherapy for recurrent or metastatic squamous cell carcinoma of the head and neck (KEYNOTE-048): a randomised, open-label, phase 3 study” Lancet. [Internet]. Available from: https://pubmed.ncbi.nlm.nih.gov/31679945/ (Accessed 01JUL2026).
[18] Danilov A, Flinn IW, Davids MS, Gregory B, Bentur O, Sidransky D, et al. (2024). “The phase III DUO trial of PI3K inhibitor duvelisib versus ofatumumab in relapsed/refractory chronic lymphocytic leukemia/small lymphocytic lymphoma: final analysis including overall survival” Haematologica. [Internet]. Available from: https://pmc.ncbi.nlm.nih.gov/articles/PMC11443371/ (Accessed 01JUL2026).
