Reporting a survival analysis

What the methods, results, tables and figures of a prognostic study should contain, with a template built from the worked examples.

A survival analysis can only be judged if the reader knows how the times were defined and how much information lies behind each number. Reporting guidelines exist for observational studies in general (STROBE, Elm et al. (2007)), for prognostic marker studies (REMARK, McShane et al. (2005)) and for prediction models (TRIPOD, Collins et al. (2015)). The checklist below covers what they ask for in the analyses of this tutorial.

Checklist

Methods

• The time origin and the event, defined precisely (for example “overall survival: time from diagnosis to death from any cause”).
• How and when patients were censored (alive at the last contact, end of the study), and the date of closing the follow-up.
• The candidate variables, how continuous variables were entered (per unit, per 10 years …), the reference level of each factor, and how variables were chosen for the multivariable model.
• How missing data were handled (the module uses complete cases).
• The methods: Kaplan-Meier with log-transformed confidence intervals, log-rank test (and trend test for ordered groups), Cox regression, how proportional hazards were checked.
• The software and versions.

Results

• Number of patients, number of events, and the median follow-up (reverse Kaplan-Meier).
• Median survival and survival at clinically relevant times, each with a 95% CI.
• The test comparing groups, with its statistic, df and p-value.
• Hazard ratios with 95% CI, univariable and multivariable, with the number of patients and events in the model.
• The result of the proportional-hazards check.
• For interactions: the interaction test, and HRs within subgroups only if it supports them.

Figures

• Kaplan-Meier curves with the number at risk under the time axis, the y-axis from 0 to 1 (or 0–100%), and follow-up cut where few patients remain (Pocock et al. 2002; Morris et al. 2019).
• Confidence bands, or CIs at key times, so that the reader sees the uncertainty.

A template

The following paragraphs report the worked examples of this tutorial. Numbers come from the module; replace the software versions with yours (jamovi shows its own version under About jamovi, and the module’s version in the jamovi library of installed modules).

Statistical analysis. Overall survival was defined as the time from enrolment to death from any cause; patients alive at the last contact were censored. Survival was estimated with the Kaplan-Meier method, with 95% confidence intervals computed on the log scale, and follow-up was summarised with the reverse Kaplan-Meier method. Survival was compared between ECOG performance status groups with the log-rank test and the log-rank test for trend. Hazard ratios (HR) were estimated with Cox proportional hazards regression, univariable and adjusted for sex, age (per 10 years) and ECOG performance status (reference: ECOG 0); the proportional hazards assumption was assessed with scaled Schoenfeld residuals (Grambsch and Therneau 1994). Analyses used jamovi (version …) with the prognosis module (version 0.2.0), which uses the R package survival (Therneau 2026).

Results. Of 226 patients, 163 died during a median follow-up of 19.3 months. Median survival was 12.9, 10.1 and 6.5 months for ECOG 0, 1 and 2, and 1-year survival 54%, 43% and 21% (log-rank p < 0.001; trend p < 0.001). In the multivariable Cox model (Table 2), ECOG 1 and 2 were associated with higher mortality than ECOG 0 (HR 1.51, 95% CI 1.02–2.23, and 2.47, 95% CI 1.58–3.86), and women had lower mortality than men (HR 0.58, 95% CI 0.42–0.81). The crude association of age (HR per 10 years 1.20, 95% CI 1.00–1.44) weakened after adjustment (HR 1.11, 95% CI 0.93–1.34). There was no evidence against proportional hazards (global test p = 0.288).

Table 2 is the module’s hazard ratio table with Univariable and multivariable on, as you would copy it from jamovi (right-click the table → Copy, then paste into your word processor):

Hazard ratios

		Univariable					Multivariable
		HR	Lower	Upper	p	p trend	HR	Lower	Upper	p	p trend
sex	Male	1.000			.002	.002	1.000				.001
	Female	0.596	0.429	0.828	.002		0.579	0.417	0.806	.001
ecog	ECOG 0	1.000			< .001	< .001	1.000				< .001
	ECOG 1	1.446	0.979	2.134	.064		1.507	1.019	2.228	.040
	ECOG 2	2.498	1.608	3.878	< .001		2.465	1.577	3.855	< .001
age	per 10 units	1.203	1.004	1.442	.046		1.114	0.928	1.337	.247

Add a footnote with the number of patients and events, what the model was adjusted for, and what the p on the reference rows means (the test for the whole variable).

Common mistakes

• No number at risk under a Kaplan-Meier plot, or curves drawn to the last patient, where the steps are huge and the estimates meaningless.
• “Median survival not reached” reported as a number, or survival at a time beyond the follow-up.
• Only p-values, no hazard ratios or survival probabilities with CIs.
• Univariable screening (“variables with p < 0.05 entered the multivariable model”) without saying so; it biases the results and should be justified.
• Too many variables for the number of events.
• Subgroup claims without an interaction test (see the subgroup fallacy).
• Hazard ratios described as risks: “a 42% lower risk of death” for HR = 0.58 is common but loose; “a 42% lower hazard (rate) of death” is correct.
• Censoring the wrong patients: for overall survival, deaths from any cause are events, not censored observations.

References

Collins, Gary S., Johannes B. Reitsma, Douglas G. Altman, and Karel G. M. Moons. 2015. “Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD): The TRIPOD Statement.” BMJ 350: g7594. https://doi.org/10.1136/bmj.g7594.

Elm, Erik von, Douglas G. Altman, Matthias Egger, Stuart J. Pocock, Peter C. Gøtzsche, and Jan P. Vandenbroucke. 2007. “The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement: Guidelines for Reporting Observational Studies.” The Lancet 370 (9596): 1453–57. https://doi.org/10.1016/S0140-6736(07)61602-X.

Grambsch, Patricia M., and Terry M. Therneau. 1994. “Proportional Hazards Tests and Diagnostics Based on Weighted Residuals.” Biometrika 81 (3): 515–26. https://doi.org/10.1093/biomet/81.3.515.

McShane, Lisa M., Douglas G. Altman, Willi Sauerbrei, Sheila E. Taube, Massimo Gion, and Gary M. Clark. 2005. “Reporting Recommendations for Tumor Marker Prognostic Studies (REMARK).” Journal of the National Cancer Institute 97 (16): 1180–84. https://doi.org/10.1093/jnci/dji237.

Morris, Tim P., Christopher I. Jarvis, William Cragg, Patrick P. J. Phillips, Babak Choodari-Oskooei, and Matthew R. Sydes. 2019. “Proposals on Kaplan-Meier Plots in Medical Research and a Survey of Stakeholder Views: KMunicate.” BMJ Open 9 (9): e030215. https://doi.org/10.1136/bmjopen-2019-030215.

Pocock, Stuart J., Tim C. Clayton, and Douglas G. Altman. 2002. “Survival Plots of Time-to-Event Outcomes in Clinical Trials: Good Practice and Pitfalls.” The Lancet 359 (9318): 1686–89. https://doi.org/10.1016/S0140-6736(02)08594-X.

Therneau, Terry M. 2026. A Package for Survival Analysis in R. https://CRAN.R-project.org/package=survival.