Clinical Trial Readouts: HR, CI, P-Values, KM Curves

You’re staring at a trial readout. The press release says “Phase 3 met primary endpoint, hazard ratio 0.47, p<0.001.” Your broker is tweeting “BULLISH.” But you have no idea whether those numbers are good, great, or mediocre. Here’s how to read the data yourself.

The one-sentence answer: learn eight terms, check the primary endpoint first, and never trust a press release that doesn’t show the confidence interval. For a companion guide on spotting spin in the press release itself, see how to read a clinical trial press release without getting fooled.

Step 1: Find the primary endpoint

The primary endpoint is the one question the trial was designed to answer. It’s locked before enrollment begins. You can verify it on ClinicalTrials.gov under the trial’s “History of Changes” tab. If the readout says “met primary endpoint,” that’s a pass. If it says “met key secondary endpoints” and goes quiet on the primary, that’s a fail dressed up in good news.

Real example: Biogen’s diranersen (BIIB) CELIA trial in Alzheimer’s disease. The primary endpoint was dose response on the CDR-SB cognitive scale at Week 76. It missed: no dose response. The press release led with “first study to show tau reduction and cognitive benefit.” Both true, but the trial failed its primary endpoint. An investor who read only the headline would think the drug worked. An investor who checked ClinicalTrials.gov would see the primary endpoint was dose response, and it missed. We covered this in detail in our diranersen CELIA analysis.

Step 2: Read the hazard ratio

A hazard ratio (HR) compares the risk of an event (death, disease progression) between treatment and control. HR 1.0 means no difference. In oncology, HR 0.65 means the treatment group had 35% fewer events than control. Lower is better when the event is bad.

Real example: Enhertu (trastuzumab deruxtecan) in the DESTINY-Breast05 Phase 3 trial showed a 53% reduction in risk of invasive disease recurrence or death versus trastuzumab + pertuzumab in early HER2+ breast cancer. That’s HR 0.47, a large and clinically meaningful effect. The FDA approved the sBLA on July 7, 2026. We broke down the data in our Enhertu sBLA analysis.

Step 3: Check the confidence interval

The 95% confidence interval (CI) tells you the plausible range for the true effect. If the HR is 0.65 with a 95% CI of 0.52 to 0.81, the result is statistically significant because the entire range is below 1.0. But if the CI is 0.85 to 1.15, the range crosses 1.0 and the result is not statistically significant, even if the point estimate looks favorable.

A narrow CI means a large trial with precise estimates. A wide CI means a small trial with uncertainty. Companies sometimes highlight the point estimate and bury the CI. Always find both numbers.

Step 4: Understand the p-value

The p-value is the probability of seeing this result if the drug had no effect. The conventional threshold is p < 0.05: below that, the result is “statistically significant.” But statistical significance is not the same as clinical significance. A trial of 10,000 patients can show a statistically significant 2% improvement that helps no one.

Effect size matters more than the p-value. A hazard ratio of 0.47 at p<0.001 is both statistically and clinically significant. A hazard ratio of 0.95 at p<0.05 is statistically significant but probably meaningless in practice.

Step 5: Read the Kaplan-Meier curve

A Kaplan-Meier curve plots survival probability over time. The x-axis is time (months or years). The y-axis is the proportion of patients still event-free. Each curve represents one treatment group. The separation between the curves is the treatment effect.

Curves that separate early and stay apart are more convincing than curves that separate late. Late separation can mean the benefit only kicks in for patients who survive long enough, a smaller, sicker subgroup. Tick marks on the curve represent censored patients (those who left the trial or were lost to follow-up). Heavy early censoring is a red flag because it can skew the curve and means fewer patients contribute data at later time points.

Step 6: ITT vs mITT vs per-protocol

Intention-to-treat (ITT) analysis includes every randomized patient, regardless of whether they completed treatment. It’s the gold standard because it reflects real-world use: some patients stop treatment, switch arms, or drop out.

Modified ITT (mITT) excludes certain patients based on pre-specified criteria. Per-protocol (PP) includes only patients who followed the protocol perfectly. Both can inflate the treatment effect by removing patients who didn’t respond or had side effects.

If a company reports PP but not ITT, or mITT but not ITT, that’s a red flag. The ITT result is the one the FDA weighs most heavily.

Step 7: ORR, CR, PR, and duration of response

In oncology trials, objective response rate (ORR) measures the percentage of patients whose tumors shrank. It combines complete response (CR, tumor disappeared) and partial response (PR, tumor shrank by at least 30%). Stable disease (SD) and progressive disease (PD) are not responses.

ORR alone is insufficient. A 40% response rate means nothing if responses last three weeks. Always check the median duration of response (DOR): how long responses lasted. A 40% ORR with a 14-month median DOR is a strong result. A 40% ORR with a 2-month DOR is weak.

For more on endpoint definitions, see our clinical trial endpoints guide.

Step 8: Read the forest plot

A forest plot shows the treatment effect across subgroups: age, biomarker status, disease stage, region. Each row is a subgroup. The diamond is the point estimate (hazard ratio), and the horizontal line is the 95% CI. If the diamond is left of 1.0 and the line doesn’t cross 1.0, that subgroup benefited.

But subgroup analysis is hypothesis-generating, not confirmatory. The trial wasn’t powered to detect effects in subgroups. If the overall trial failed but one subgroup “showed benefit,” be skeptical. With enough subgroup comparisons, one will look positive by chance alone. The FDA rarely approves based on subgroup analysis of a failed trial.

Common mistakes

• Trusting the headline without reading the primary endpoint. The Biogen diranersen press release framed a failed trial as a success. The primary endpoint was dose response, and it missed. Check ClinicalTrials.gov before you check the stock price.
• Ignoring the confidence interval. A hazard ratio of 0.65 with a CI of 0.45 to 0.85 is strong. A hazard ratio of 0.65 with a CI of 0.42 to 1.02 is not statistically significant. The CI matters more than the point estimate.
• Confusing ORR with survival benefit. Tumors that shrink may not translate to longer overall survival. ORR is a surrogate endpoint; overall survival (OS) is the gold standard. Check whether the trial measured OS or just ORR.
• Buying on “clinically meaningful” without a p-value. GSK’s camlipixant CALM-2 trial in chronic cough missed its primary endpoint, but the press release highlighted “numerical improvement” in patient-reported outcomes. Numerical improvement without statistical significance is an observation, not a result. See our camlipixant CALM-2 failure analysis.
• Believing subgroup rescue stories. If the trial failed overall but the company highlights a subgroup that worked, remember: with 10 subgroup comparisons, there’s a 40% chance at least one looks positive by chance. Subgroups generate hypotheses; they don’t confirm them.

Final checklist

• Primary endpoint met? (Check ClinicalTrials.gov, not just the press release)
• Hazard ratio below 1.0 for oncology?
• 95% CI does not cross 1.0?
• p-value below 0.05 AND effect size is clinically meaningful?
• ITT analysis reported, not just mITT or per-protocol?
• Kaplan-Meier curves separate early and stay apart?
• Median duration of response reported alongside ORR?
• Subgroup results treated as exploratory, not confirmatory?

Once you can read the numbers, the next step is turning them into a trading decision. See our 8-step guide to trading Phase 3 readouts for the mechanics of sizing positions and timing entries around catalysts.