Reading a Positive Progression-Free Survival Trial at Tumor Board
A colleague presents a phase III trial at tumor board. Patients with newly diagnosed advanced ovarian cancer were randomized to standard chemotherapy plus a new agent or chemotherapy plus placebo, with progression-free survival as the primary endpoint. Results: median progression-free survival 22 versus 18 months, hazard ratio 0.78 (95% CI 0.65-0.94), p = 0.008. Median overall survival 51 versus 47 months, hazard ratio 0.89 (95% CI 0.72-1.10); 55% of events had occurred, and crossover was permitted, with 40% of control patients ultimately receiving the agent. Grade 3 or higher neutropenia occurred in 42% versus 18%, 12% discontinued for toxicity, and the drug costs about $180,000 per year. Your patient is 74 years old, ECOG 1, with stage IIIC disease and borderline renal function, and she asks whether the drug will let her live longer.
1. Interpret the progression-free and overall survival results. Is the overall survival result negative or immature?
Read the worked answer
The progression-free survival result is statistically significant and internally coherent: a 4-month median gain with a hazard ratio of 0.78 and a confidence interval excluding 1.
The overall survival result should not be called negative. With only 55% of events, the confidence interval (0.72-1.10) is wide and includes a clinically meaningful reduction in death risk as well as a small harm. The trial is underpowered for overall survival at this analysis. Crossover in 40% of controls further attenuates any true survival effect in an intention-to-treat comparison, because control patients receive the active drug anyway.
What I would want to know before drawing conclusions:
- Was overall survival in the prespecified testing hierarchy, with alpha controlled across interim analyses? If it was tested at a nominal level at an immature interim, a non-significant result is uninformative, not evidence of no benefit.
- Proportional hazards assumption: if the curves separate late, the log-rank test and hazard ratio understate the benefit; restricted mean survival time or milestone survival may be more honest.
- Subsequent therapy: how much PARP inhibitor or bevacizumab was used after progression in each arm, since that affects overall survival independent of the randomized drug.
- Maturity target: typically 70% or more events are needed before a definitive overall survival claim.
Progression-free survival is an accepted primary endpoint in ovarian cancer because of crossover and subsequent-line confounding, but it does not by itself prove a survival benefit.
2. How would you translate these numbers for this 74-year-old patient, including absolute benefit and harm?
Read the worked answer
Move from relative to absolute terms and make the uncertainty explicit.
- What is known: the drug delays progression by a median of about 4 months. Median gains understate or overstate individual experience, so use proportions: if 3-year progression-free survival were 30% versus 20%, the number needed to treat to prevent one progression at 3 years would be 10.
- What is not known: whether it extends life. Say plainly that the survival data are immature and that no overall survival benefit has been demonstrated.
- Harms: grade 3-4 neutropenia 42% versus 18%, 12% discontinuing for toxicity, growth factor injections, monitoring visits, infusion time, infection risk, and cost. Her borderline renal function may complicate some agents and requires checking before starting.
- Her values: at 74 with ECOG 1, time at home and freedom from hospitalization may matter more than a 4-month median PFS gain.
A usable script: 'This drug lengthens the time before the cancer grows by about four months on average. We do not yet know whether it helps people live longer. It roughly doubles the risk of seriously low white counts and about one in eight people stop it because of side effects.' Offer a clinical trial or observation as legitimate alternatives, and document the discussion.
3. What design and reporting criticisms would you raise about this trial?
Read the worked answer
Several, in order of importance.
- Endpoint hierarchy and alpha control: progression-free survival as primary is acceptable, but if overall survival was analyzed at an interim with no prespecified alpha spending, the reported p-values are descriptive only. The testing strategy must be stated.
- Non-proportional hazards: if the survival curves separate after 12 months, the log-rank test loses power and the hazard ratio is a misleading summary. Restricted mean survival time or a milestone analysis should be prespecified.
- Subgroup claims: any claim of benefit in a biomarker-defined subgroup must be prespecified with an interaction test, not a set of subset p-values.
- Control arm relevance: if the control arm did not include maintenance PARP inhibition or bevacizumab for eligible patients, the comparator is outdated and the result may not generalize.
- Crossover analysis: with 40% crossover, a naive intention-to-treat overall survival analysis is biased toward the null; a prespecified sensitivity analysis (for example rank-preserving structural failure time) should be reported.
- Patient-reported outcomes and cost-effectiveness: missing quality-of-life data and a $180,000 annual price are part of the evidence, not an afterthought.
- Generalisability: the trial's age and comorbidity profile should be compared with the patient in front of you; older adults with renal impairment are often underrepresented.