Survival Analysis for a
Clinical Cohort Study

RESEARCH QUESTION

What the Study Set Out to Test

This survival analysis example follows a cardiology research team that has follow-up data on a cohort of patients tracked over five years. They want to know whether the time to a first cardiovascular event differs between two treatment groups after adjusting for baseline characteristics such as age, blood pressure and diabetes status.

The hypothesis is that the newer treatment is associated with a lower hazard of an event over the follow-up period. Because patients were not randomized, the question is framed as an association, and the analysis is planned to describe it with its uncertainty rather than claim cause, since treatment choice may reflect baseline risk.

STUDY DESIGN AND DATA

An Observational Cohort With Censoring

A prospective cohort with two treatment groups and around 600 participants. The dataset holds demographics, baseline risk factors, treatment group, time to event or last contact, and event status. Most participants were lost to follow-up or reached the end of the study without an event, so their outcome is censored: we know only that they were event-free up to a certain time.

METHOD AND ASSUMPTIONS

Why Cox Regression, and What Must Hold

Censored follow-up times cannot be analyzed with standard linear or logistic regression without bias or loss of information. Kaplan-Meier estimation describes event-free survival over time in each group, the log-rank test compares the curves, and Cox proportional hazards regression estimates the adjusted effect of treatment alongside the covariates. Censoring is assumed to be non-informative, which is checked with the team against the reasons patients left follow-up.

The Cox model rests on proportional hazards, meaning the hazard ratio stays roughly constant over time. This is examined with Schoenfeld residuals and log-minus-log plots read together with the clinical context, not a single p-value. One categorical covariate departs clearly from proportionality and is handled by stratification. Covariates are chosen from the study protocol and prior evidence, not by stepwise searching, and the number of events is checked against the number of model terms.

OUTPUT AND INTERPRETATION

What the Results Showed

The output has Kaplan-Meier curves by treatment group with numbers at risk beneath the axis, the log-rank test, and a Cox model table reporting hazard ratios with 95% confidence intervals for treatment and the other covariates, with the stratification variable described separately because the stratified model does not estimate a hazard ratio for it. The adjusted hazard ratio for treatment is explained in plain language next to the unadjusted estimate, with the confidence interval carrying the message about precision.

Sensitivity analyses test how far the conclusion depends on the choices made: a model with the non-proportional covariate as a time-varying effect, and a comparison of complete-case analysis with multiple imputation for missing baseline values. The write-up states what the estimates do and do not support for an observational design.

REPORTING

What the Researcher Received

A written report structured to the STROBE checklist for cohort studies, formatted to the target journal's requirements, with statistical wording for the methods, matched to the output, for the team to check and incorporate. The research question, substantive interpretation and conclusions remain with the research team.