Median follow-up is often used to assess the quality of follow-up in longitudinal studies. Its relevance to evaluate the quality of the produced estimations seems obvious. However, there is no clear decision rule on its use. We evaluated the quality of the Kaplan-Meier estimator of the survival rate according to different values of the median follow-up in order to propose an objective decision rule.

We performed a sensitivity analysis via Monte Carlo simulations based on a reference sample, according to the literature. The log(–log) confidence interval of the Kaplan-Meier estimator of survival rate was evaluated in terms of coverage probability and length. The Monte Carlo simulation was performed for 10,000 replications considering 170 values for the median follow-up.

We could observe a high relationship between the quality of the estimation and the median follow-up. Heavy censorship leads to an overestimation of the survival rate and anti-conservative confidence intervals. From a threshold value of the median follow-up, the values of the criteria reach a plateau and good estimations of the survival rate are achieved.

More precisely, simulations show that reliable Kaplan-Meier estimations cannot be obtained when the median follow-up is less than two thirds of the estimation time. This provides factual elements to evaluate results for long longitudinal studies or to assess ongoing cohorts.

median follow-up, Kaplan-Meier estimator, censoring, sensitivity analysis, Monte-Carlo simulation.

Received: December 26, 2020; Accepted: January 11, 2021; Published: January 15, 2021

How to cite this article: Serge M. A. Somda, Yacouba Ouedraogo, Eric A. B. Dabone, Eve Leconte and Thomas Filleron, Assessing the sensitivity of the Kaplan-Meier confidence interval according to the median follow-up, JP Journal of Biostatistics 18(1) (2021), 67-85. DOI: 10.17654/BS018010067

This Open Access Article is Licensed under Creative Commons Attribution 4.0 International License

Reference:

[1] Edward L. Kaplan and Paul Meier, Nonparametric estimation from incomplete observations, Journal of the American Statistical Association 53(282) (1958), 457-481.
[2] Major Greenwood, The Errors of Sampling of the Survivorship Tables, Reports on Public Health and Medical Subjects, 1926.
[3] S. Sawyer, The greenwood and exponential greenwood confidence intervals in survival analysis, Applied Survival Analysis: Regression Modeling of Time to Event Data, 2003, pp. 1-14.
[4] John D. Kalbfleisch and Ross L. Prentice, The Statistical Analysis of Failure Time Data, John Wiley & Sons, 2011.
[5] John P. Klein, Small sample moments of some estimators of the variance of the Kaplan-Meier and Nelson-Aalen estimators, Scandinavian Journal of Statistics (1991), 333-340.
[6] Michael Schemper and Terry L. Smith, A note on quantifying follow-up in studies of failure time, Controlled Clinical Trials 17(4) (1996), 343-346.
[7] Stuart J. Pocock, Tim C. Clayton and Douglas G. Altman, Survival plots of time-to-event outcomes in clinical trials: good practice and pitfalls, The Lancet 359(9318) (2002), 1686-1689.
[8] R. A. Betensky, Measures of follow-up in time-to-event studies: Why provide them and what should they be?, Clinical Trials. 12(4) (2015), 403-408.
[9] Serge M. A. Somda, Stéphane Culine, Christine Chevreau, Karim Fizazi, Eve Leconte, Andrew Kramar and Thomas Filleron, A statistical approach to determine the optimal duration of post-treatment follow-up: Application to Metastatic Nonseminomatous Germ Cell Tumors, Clinical Genitourinary Cancer 15(2) (2017), 230-236.
[10] S. Culine, P. Kerbrat, A. Kramar, et al., Refining the optimal chemotherapy regimen for good-risk metastatic nonseminomatous germ-cell tumors: a randomized trial of the Genito-Urinary Group of the French Federation of Cancer Centers (GETUG T93BP), Annals of Oncology 18(5) (2007), 917-924.
[11] Stéphane Culine, Andrew Kramar, Christine Théodore, et al., Randomized trial comparing bleomycin/etoposide/cisplatin with alternating cisplatin/ cyclophosphamide/doxorubicin and vinblastine/bleomycin regimens of chemotherapy for patients with intermediate- and poor-risk metastatic nonseminomatous germ cell tumors: Genito-Urinary Group of the French Federation of Cancer Centers Trial T93MP, Journal of Clinical Oncology 26(3) (2008), 421-427.
[12] A. Agresti, Categorical Data Analysis, 2nd ed., 2013.
[13] R. Team, R. Core, et al., A Language and Environment for Statistical Computing, 2013.
[14] Terry M. Therneau and Thomas Lumley, Package survival, R Top Doc. 128 (2015), 112.
[15] Ralf Strobl and Maintainer Tobias Verbeke, Package km. ci. 2009.
[16] Nicholas Metropolis and Stanislaw Ulam, The Monte Carlo method, Journal of the American Statistical Association 44(247) (1949), 335-341.
[17] Richard Peto, MCetal Pike, Philip Armitage, et al. Design and analysis of randomized clinical trials requiring prolonged observation of each patient, II. analysis and examples, British Journal of Cancer 35(1) (1977), 1-39.
[18] David R. Thomas and Gary L. Grunkemeier, Confidence interval estimation of survival probabilities for censored data, Journal of the American Statistical Association 70(352) (1975), 865-871.
[19] Stuart Barber and Christopher Jennison, A review of inferential methods for the Kaplan-Meier estimator, Statistics Group Research Report 98(02) (1998).
[20] Sidney J. Cutler and Fred Ederer, Maximum utilization of the life table method in analyzing survival, Journal of Chronic Diseases 8(6) (1958), 699-712.
[21] Kenneth J. Rothman, Estimation of confidence limits for the cumulative probability of survival in life table analysis, Journal of Chronic Diseases 31(8) (1978), 557-560.
[22] Robert L. Strawderman, Michael I. Parzen and Martin T. Wells, Accurate confidence limits for quantiles under random censoring, Biometrics 53 (1997), 1399-1415.
[23] Craig B. Borkowf, A simple hybrid variance estimator for the Kaplan-Meier survival function, Statistics in Medicine 24(6) (2005), 827-851.
[24] Michael P. Fay, Two-sided exact tests and matching confidence intervals for discrete data, R Journal 2(1) (2010), 53-58.