Hazard ratios and Kaplan–Meier curves in peptide trials: a reader's guide

01

Hazard, risk, and time are different quantities

Risk is the probability that an event occurs over a stated period. Hazard describes the event rate at a particular moment among people who have not yet had the event and remain under observation. A hazard ratio compares those rates between groups over follow-up, commonly using a Cox proportional-hazards model. It uses timing information rather than only the final event count.

Because hazard is not a probability, a hazard ratio cannot be read as a percentage-point difference. A reported ratio of 0.70 for an adverse outcome is often summarized as a 30% lower hazard under the model, not as 30% fewer events or a 30% reduction in each participant's risk. Absolute probabilities at useful time points answer a different question.

02

First identify the event and favorable direction

A ratio below 1 can favor treatment when the event is undesirable, such as hospitalization or disease progression. If the event is desirable, such as symptom resolution, a ratio above 1 may indicate faster recovery. Never interpret the number without copying the exact endpoint, comparison order, and model definition. Marketing language often omits these details and creates a directional error.

Also check whether the endpoint is a single event or a composite. Time to first hospitalization, time to first composite event, progression-free survival, and time to treatment discontinuation answer different questions. The same numerical hazard ratio can carry very different clinical meaning depending on the event, population, background therapy, and follow-up period.

03

How to read a Kaplan–Meier curve

Kaplan–Meier curves estimate the proportion remaining event-free or surviving over time while accounting for varying follow-up. Read the horizontal time axis, vertical probability scale, group labels, and event definition. Look at when curves separate, whether they remain apart, and the uncertainty shown. A visually dramatic gap can be unstable when few participants remain under observation.

The numbers-at-risk table shows how many participants are still contributing information at successive times. Late portions of a curve may represent a small fraction of the original sample. Tick marks can indicate censoring, meaning follow-up ended without the event being observed by that time. Censoring is not proof that a participant never experienced the event.

04

Confidence intervals and model assumptions matter

The confidence interval describes statistical uncertainty around the hazard ratio. For the usual ratio scale, an interval crossing 1 is compatible with no difference under the model, although clinical importance and study design require more than a binary significance label. A narrow interval can still surround an effect too small to matter, while a wide interval can include both meaningful benefit and harm.

The common Cox model assumes proportional hazards: roughly, that the hazard ratio is sufficiently stable over time for one summary to be meaningful. Strongly crossing or converging curves can challenge that interpretation. Reports may use time-varying effects, milestone probabilities, median times, or restricted mean survival time as complementary summaries. Readers should not force a single ratio to answer every timing question.

05

Censoring and competing events can change the story

Censoring allows inclusion of a participant's observed follow-up until loss to follow-up, study end, or another defined point. Standard methods rely on assumptions about whether censoring is informative. If people leave because of adverse effects, lack of benefit, access barriers, or worsening disease, the missing future experience may differ systematically from that of participants who remain.

A competing event can prevent the outcome of interest from occurring or being observed in the same way. Death, for example, can compete with a nonfatal outcome. Trial reports should explain the analysis. A clinic's claim that a product delayed one event may be incomplete if it ignores withdrawals, deaths, rescue treatment, or other events that altered who remained at risk.

06

A complete time-to-event claim check

Write down the endpoint, whether it is favorable or adverse, the treatment-to-control comparison order, hazard ratio, confidence interval, time horizon, median follow-up, event counts, milestone probabilities, numbers at risk, and censoring rules. Inspect the curves and note whether their shape supports a single proportional-hazards summary. Compare with absolute outcomes at clinically relevant times.

Then verify product identity and trial applicability. A hazard ratio from an approved peptide drug cannot validate an unstudied compounded formulation, another route, or a wellness program. It also cannot forecast an individual's event time. Use the statistic to understand the group comparison, and discuss treatment implications with a licensed clinician who can weigh the full label, evidence, alternatives, and personal risks.

If a report provides only the ratio, look for the full paper, supplement, registry results, or regulatory review. Without arm-level event counts and time-specific probabilities, readers cannot tell whether a large relative contrast corresponds to a small absolute difference. Also note the analysis cutoff and data maturity: a later update can change the curve, median follow-up, number of events, and precision without implying that the earlier calculation was dishonest.

• →Exact event definition
• →Which direction favors treatment
• →Hazard ratio and confidence interval
• →Kaplan–Meier curve shape
• →Numbers at risk
• →Absolute probabilities at stated times
• →Censoring and competing-event rules