A flurry of interesting articles in the last few days.
Fojo and Bates on Interpretation
At JAMA Oncology, Fojo and Bates write a 3-page short article on problems and biases when interpreting Kaplan-Meier curves for conclusions like median survival.
While this may seem like one of the simplest measures, they point out that when curves grow relatively flat, a small difference in survival of a few patients can pull the "median" of the two populations far apart. That is, a small difference in the vertical data makes a big difference in the left to right data.
They raise some other topics as well, like contrasting problems with hazard ratios under some circumstances. See an AI summary of Fojo & Bates at the bottom of this blog.
https://jamanetwork.com/journals/jamaoncology/fullarticle/2834385
Bonus - See an article by Fojo in JAMA in 2023, "Journeys to failure that litter the path to developing new cancer therapeutics," here.
On March 25, Vinay Prasad (now a senior leader at FDA) published an article in Nature Reviews Clinical Oncology, refreshing us on his thoughts on suggogate endpoints. Find it here:
https://www.nature.com/articles/s41571-025-01007-z
Note that a comment letter has appeared by Vogel et al., and a reply from Prasad (both 6 June).
https://www.nature.com/articles/s41571-025-01031-z
https://www.nature.com/articles/s41571-025-01032-y
##
See David Gorski's take on FDA current thinking at Science Based Medicine blog here.
##
See a new JAMA article on standardized reporting of surrogate endpoints (as in CONSORT 2025 criteria):
https://jamanetwork.com/journals/jama/fullarticle/2822038
##
See also: articles by McCabe and by Dominitz et al. on the rising diversity of cancer screening methods:
McCabe et al, NRCO
https://www.nature.com/articles/s41571-025-01037-7
Dominitz et al., JAMA
###
###
AI Summary of Fojo and Bates (Chat GPT)
Here's a summary of the JAMA Oncology Viewpoint by Fojo and Bates (May 2025), titled “Skimming the Median and the Problem of Exaggerated Survival Gains”:
Main Argument:
Fojo and Bates challenge the reliability of median overall survival (OS) as a headline statistic in oncology trials, warning that “skimming the median” can exaggerate treatment benefit. When Kaplan-Meier survival curves linger just above or below the 50% survival threshold, small differences in curve shape or timing can create the illusion of large OS gains—even when hazard ratios (HRs) suggest otherwise.
Key Examples Highlighted:
-
CheckMate067 (Melanoma):
-
Median OS: 72.1 mo (nivo+ipi) vs. 36.9 mo (nivo alone)
-
HR: 0.84 (not significant)
-
Interpretation: A dramatic doubling of median OS was overstated due to slow tail decline of the experimental arm just above 50%.
-
-
ADRIATIC Trial (Limited-stage SCLC):
-
Median OS: 55.9 mo (durvalumab) vs. 33.4 mo (placebo)
-
HR: 0.73
-
Reality: Kaplan-Meier curves diverged by only ~11 months at 60% survival—not 22 months as suggested by medians.
-
-
ESOPEC Trial (Esophageal Adenocarcinoma):
-
Median OS: 66 mo (FLOT) vs. 37 mo (CROSS)
-
HR: 0.70
-
Observation: A 29-month median OS difference overstated real benefit, which was closer to 7.5 months at 60% survival.
-
Broader Takeaways:
-
HRs vs. Median OS: Neither metric alone suffices. HRs can obscure absolute benefit; medians can be gamed by curve shape.
-
KM Curves Matter: Visual inspection of entire Kaplan-Meier curves is crucial to avoid misinterpretation.
-
Effect Size Contextualized: An HR of ~0.73 yields vastly different absolute OS gains depending on baseline OS—~3.5 months in extensive-stage SCLC vs. ~11.5 months in limited-stage SCLC.
Implication for Practice:
Cancer clinicians and trialists should resist oversimplified metrics. Instead of relying on median OS or HRs alone, a more nuanced interpretation of survival curves is essential—particularly when shaping patient expectations or influencing regulatory/payer decisions.
This article adds an important cautionary note to how we interpret high-impact trial results—especially in the current era of immune checkpoint inhibitors and cancer immunotherapy.
###
###
###
###
Here's a concise, expert-level summary of the three-part exchange in Nature Reviews Clinical Oncology involving Vinay Prasad and Sebastian Vogel:
1. Prasad’s Original Article (NRCO, 2025)
Title: “A measured approach to surrogate end points in cancer screening and early detection”
Summary:
Prasad critiques the growing enthusiasm for early detection tools (e.g., liquid biopsies) based solely on surrogate endpoints—like stage shift or earlier diagnosis—without evidence of reduced cancer-specific or all-cause mortality. He argues:
-
Surrogates are unreliable: Earlier stage at diagnosis doesn’t necessarily reduce death if the treatment is ineffective or if lead-time and overdiagnosis biases exist.
-
Empirical validation needed: Historical examples (e.g., ovarian cancer screening, PSA) show that mortality benefit must be demonstrated in randomized trials before endorsing new screening tools.
-
Call for caution: Without solid mortality data, adoption of early detection tools risks overdiagnosis, overtreatment, and wasted resources.
2. Vogel’s Letter to the Editor
Tone: Critical but constructive.
Main Points:
-
Vogel defends early detection and stage shift as promising and biologically plausible surrogates, especially for cancers with poor late-stage prognosis (e.g., pancreatic cancer).
-
He notes that randomized mortality trials take years, and delaying implementation may hinder progress and innovation.
-
Vogel suggests that perfect shouldn’t be the enemy of good, advocating for a pragmatic, risk-adapted approach that balances early use of promising tests with real-world evidence generation.
3. Prasad’s Reply to Vogel
Tone: Firm, skeptical of compromise.
Main Arguments:
-
Prasad reiterates that plausibility is not proof: many biologically sound interventions have failed to improve survival or even caused harm when tested rigorously.
-
He warns against the “leap of faith” model of approving screening tools without RCT-level evidence, arguing this leads to medical reversals.
-
Prasad acknowledges the urgency but insists that clinical benefit must be shown, not assumed, especially in screening where harm from overdiagnosis can be substantial.
Bottom Line for Experts:
This exchange crystallizes the core debate in cancer screening: Should promising early detection tools be adopted based on surrogate endpoints like stage shift, or should the field adhere strictly to proven mortality benefit? Prasad urges scientific discipline and caution, while Vogel advocates for timely pragmatism in advancing potentially lifesaving tools.
