Skip to main content

Table 6 Recommendation for reporting survival analyses methods

From: A systematic review of the quality of conduct and reporting of survival analyses of tuberculosis outcomes in Africa

Section Recommendation
Study design • Define the study time-to-event outcome.
• Report the sample size and sample estimation methods providing all the assumptions made in calculating sample size.
• Report the planned fixed length of follow-up (days, months, years).
Statistical methods and results • Report beginning and end dates of each event under observation.
• Report the total time under observation using standard epidemiological units like person-years and median time. Where the aim is to compare groups of participants, in additional to total time observed, report total time and median time stratified by the groups.
• Report number of participants lost-to-follow-up, how censoring was done and if non-informative censoring assumption was evaluated.
• Report total number of time-to-event outcome events observed, and events per groups.
• Provide the survival probabilities at specific follow-up time points (outcome free probabilities where the outcome is not death), median survival time and 95% confidence interval is preferred for comparison with other studies. This should be provided for each group as well when the objective is to compare groups.
• Report the method used to estimate the survival probabilities and plot the survival curves using appropriate graphs like Kaplan-Meier or Nelson-aalen cumulative curve stratified by groups when necessary. Include the following information in the survival curves: number of participants at risk at each specified timepoint, indicate when participants were censored, use different colors/type of lines to distinguish group curves and clearly label the x-axis as time under observation and y-axis appropriately.
• When testing hypothesis of differences in survival probabilities between/among groups, report the method used, the test results and a P-value.
• When survival regression is performed, report the methods used to test underlying assumptions (test for Proportion Hazard assumption for Cox regression and test of used probability distribution for parametric methods) and the test result.
• Report the measure of effect (e.g Hazard ratios, sub-distribution hazard ratios, time ratios), their measure of uncertainty (e.g 95% confidence intervals, standard errors) and P-values from the regression model.
• Like other statistical regression modeling, report all the covariates assessed, method of selecting features to be included in the multivariable survival regression model, methods used to assess the multivariable regression goodness of fit, proportion of missing data in the outcome and covariates assessed plus how missing data were handled, methods used to test for interaction and methods used to control for clustering in multilevel studies.