Prasad: most surrogate endpoints in cancer trials correlate poorly with survival
A systematic review of 36 trial-level meta-analyses found that in over half the surrogate endpoint (response rate, progression-free survival) had a low correlation with overall survival, and only about a quarter showed a strong correlation.
Prasad and colleagues searched for trial-level meta-analyses that quantified the correlation between a surrogate endpoint and overall survival across randomised oncology trials. They identified 36 articles covering 65 surrogate-survival associations in many tumour types and settings.
Using a conventional threshold, 52% of reported correlations were low (r below 0.7), 25% medium (0.7 to 0.85) and 23% high (0.85 or more). Correlations were weakest in the metastatic setting and for response rate. A companion analysis by the same group (Kim and Prasad, JAMA Internal Medicine 2015) showed that of 54 FDA cancer drug approvals in 2008-2012, 36 were based on surrogates, and after a median 4.4 years only 5 had demonstrated an overall survival benefit.
The work catalysed the debate over accelerated approval, the FDA's 2023-2025 reforms requiring confirmatory trials to be under way at approval, and the growing use of quality-of-life and patient-reported endpoints.
- 36 trial-level meta-analyses covering 65 surrogate-OS associations reviewed
- 52% of surrogate-survival correlations were low (r below 0.7), 25% medium, 23% high
- Response rate and PFS in the metastatic setting were the weakest surrogates
- Companion analysis: of 36 approvals on surrogates (2008-2012), 5 later showed an OS benefit, 18 failed to or were not tested, and 13 remained unknown
A drug that shrinks tumours or delays progression on scans has not necessarily been shown to help patients live longer or better. Patients and clinicians should ask what the endpoint was; regulators should insist on timely confirmatory trials; and trialists should validate surrogates before relying on them.
- Trial-level correlation is only one way to validate a surrogate; some settings (adjuvant DFS in colon cancer) have well-validated surrogates
- Low correlation may reflect crossover and post-progression therapy diluting the OS signal rather than a useless surrogate
- The review depended on published meta-analyses and their heterogeneous methods
- PFS can be a meaningful endpoint in itself when progression is symptomatic and treatment is tolerable