The press release hits the wire at 4:05pm ET. You have maybe twenty minutes before the sell-side notes start landing and the AH print tells you what the fast money thinks. The question is not whether the trial "worked." The question is whether the data clears the bar that matters — the pre-specified primary endpoint, the statistical hierarchy, the safety profile, and the regulatory lens the FDA will actually apply.
This is the order I read a topline in.
Start with the primary endpoint: hit or miss, and by how much
Every Phase 3 has a pre-specified primary endpoint. It was locked in the SAP (Statistical Analysis Plan) before unblinding. You find it in the protocol summary, in ClinicalTrials.gov, and almost always in the first two sentences of the press release.
The words to look for: "met its primary endpoint," "achieved statistical significance," and the effect size. A headline of "statistically significant improvement" with no effect size is a yellow flag. You want the number.
For a time-to-event endpoint (overall survival, progression-free survival), the primary read is the hazard ratio and its 95% confidence interval. An HR of 0.72 (95% CI 0.58–0.89, p=0.002) means a 28% reduction in the hazard of the event in the treatment arm, and the CI tells you how tight the estimate is. If the upper bound of the CI is comfortably below 1.0, the signal is real. If the upper bound is 0.98, the p-value squeaked in but the precision is weak.
For a continuous endpoint (change from baseline in a score), you want the least-squares mean difference, the CI, and the p-value. And you want to know whether that difference is clinically meaningful, not just statistically significant. A 2.1-point improvement on a 70-point scale may clear p<0.05 in a 1,200-patient trial and still not move prescribing behavior. The FDA reviews for both statistical and clinical significance, and the advisory committee will absolutely ask.
For a responder endpoint (proportion of patients achieving a threshold), the read is the difference in response rates and the CI on that difference. A 42% vs 28% response (14-point delta, 95% CI 8–20, p<0.001) is a different story than a 31% vs 27% response (4-point delta, p=0.04).
The stat gotchas that separate a clean win from a messy one
The primary endpoint hit. Good. Now you check the stat machinery around it, because this is where buy-side misses happen.
Pre-specified hierarchies and alpha spending
Most Phase 3 trials with multiple endpoints use a fixed-sequence hierarchy: you test the primary first, and only if it hits do you get to test the first key secondary, and so on. If the primary misses, every downstream endpoint is officially "nominal" or "descriptive" — meaning the p-values are not adjusted for multiplicity and cannot support a label claim. A press release that leads with a secondary endpoint is almost always covering a primary endpoint that missed or was weak.
For group-sequential designs with interim analyses, alpha spending (O'Brien-Fleming, Pocock, or Lan-DeMets) divides the overall Type I error across looks. If a trial hit at a pre-planned interim, the effect usually has to be larger than the final-analysis bar would have required. Confirm the trial was stopped for efficacy at a DMC-approved interim, not because the sponsor ran out of cash.
Missing data handling
This is the most underread section of a topline. Look for the estimand and the missing-data strategy. Is the primary analysis ITT (intention-to-treat, everyone randomized), modified ITT, or per-protocol? ITT is the FDA's default. Per-protocol analyses tend to flatter the drug because they exclude dropouts, who are often sicker or non-responders. If a sponsor leads with per-protocol and relegates ITT to a secondary, that is a signal.
Multiple imputation and jump-to-reference are common sensitivity analyses. If the primary holds across sensitivity analyses, the signal is robust. If the press release does not mention sensitivity analyses, assume they are mixed until the full data drop.
Subgroup-only wins
"The trial met its primary endpoint in the pre-specified biomarker-positive subgroup." Read this twice. Was the biomarker-positive subgroup the primary analysis population (a pre-specified enrichment design), or was it a subgroup of a broader ITT population that failed? The former is a clean win. The latter is a salvage narrative.
The secondary endpoints that sometimes matter more than the primary
Secondary endpoints can move a stock more than the primary if they hit the FDA's review lens or the commercial positioning.
- Overall survival as a key secondary in an oncology trial with a PFS primary. OS is the gold standard. A PFS win with an OS trend (HR<1.0, CI crossing 1.0) is common at interim; a PFS win with a flat or adverse OS is a serious concern for the advisory committee.
- Safety and tolerability. Look for the rate of treatment-related Grade 3+ adverse events, the discontinuation rate due to AEs, and any deaths on study attributed to the drug. A 6-point efficacy win paired with a 22% discontinuation rate is a harder label than the headline suggests.
- Durability. In response-based endpoints, the median duration of response and Kaplan-Meier tails matter. A 60% response rate that melts away by month 8 is a different drug than a 50% response rate that plateaus.
- Pre-specified subgroups. Consistency across age, sex, race, geography, and disease severity is what the FDA's statistical reviewer will graph first. A forest plot with one subgroup dramatically out of line invites questions.
The FDA review lens
The market reads the p-value. The FDA reads the totality of evidence, and the gap between the two is where the post-readout trade lives.
Three framings to apply. First, regulatory precedent in the indication: what did the last two approvals in this space show, and does this data look better, worse, or roughly the same? Second, the label the sponsor can credibly ask for: broad vs. restricted to the biomarker subgroup, first-line vs. later-line, monotherapy vs. combination. A clean primary in a narrow population is often a narrow label. Third, advisory committee risk: novel mechanisms, surrogate endpoints not previously validated by the FDA, and modest effect sizes with meaningful safety signals are the combinations that get referred to an AdCom. AdCom risk is often more dilutive to the stock than the data itself.
A worked example
Say fictional sponsor Meridian Bio (ticker: MRDN) reports topline from PIVOT-3, a 780-patient Phase 3 of its oral SGLT-adjacent agent meridiglide in chronic kidney disease with type 2 diabetes. Primary endpoint: time to a composite of ≥50% eGFR decline, ESRD, or renal death. Result: HR 0.69 (95% CI 0.55–0.86, p=0.0008), ITT population. Key secondary (in hierarchy): cardiovascular death, MI, or stroke — HR 0.84 (95% CI 0.69–1.02, p=0.08), missed at the 0.05 threshold. Safety: Grade 3+ AEs 18% vs 14%; discontinuation due to AE 9% vs 7%; no imbalance in deaths.
The read: primary is a clean win on a regulator-favored composite, with a CI that does not kiss 1.0 and a p-value an order of magnitude below 0.05. The MACE secondary missed, which kills the cardiovascular label expansion narrative some sell-side models were carrying, but does not threaten the renal approval. Safety is manageable and in line with class. Likely path: full approval in the renal indication, label roughly mirrors the enrolled population, low AdCom probability given a previously validated endpoint, and the fair-value move is positive but capped by the loss of the CV optionality. The AH trade is long the renal thesis, short the CV bull case.
What to do with this in twenty minutes
Read the primary effect size and CI. Check the hierarchy and whether the press release leads with the right endpoint. Scan the ITT vs per-protocol framing. Pull the safety top line. Compare to the nearest precedent in the indication. Then — and only then — look at the AH tape.
For a daily, analyst-grade read on Phase 3 readouts, FDA decisions, and catalyst setups, subscribe to Biotech Catalyst Daily. Written for investors and PMs who need the data before the narrative.