Meeting Brochure and registration form      SMDM Homepage

Sunday, 15 October 2006


M. P. Janssen, MSc, University Medical Center Utrecht, Utrecht, Netherlands, Cees P. van der Poel, MD, PhD, Sanquin Blood Supply Foundation, Amsterdam, Netherlands, Wim P. Schaasberg, MSc, Statistics Netherlands, Voorburg, Netherlands, Gouke J. Bonsel, MD, PhD, Academic Medical Center, Amsterdam, Netherlands, and Ben A. van Hout, PhD, University Medical Center Utrecht, Utrecht, Netherlands.

Purpose: To illustrate and explain why applying Kaplan-Meier survival estimation to censored post-treatment survival data of patients who underwent multiple treatments will lead to erroneous estimates for post-treatment survival, and propose a correct estimator. For the evaluation of the cost effectiveness of safety measures that reduce low probability risks associated with repetitive treatments, estimation of loss of life due to potential treatment associated adverse events is required. In case patients receive multiple treatments over time, potential loss of life will depend on the timing of these treatments as patients' life expectancy varies over time. The potential loss of life due to treatment can be expressed in patient post-treatment survival (PTS). If timing of treatment events and deaths is available for all patients, applying the Kaplan-Meier (KM) estimator to the time intervals between treatment and death will lead to a correct estimation of PTS. The question is whether this is also the case if the survival data is censored.

Method(s): Simulation of a patient population with a fixed life expectancy and treatment regime that is subject to censoring is used to analyze the effect of censoring on the KM-estimator of post-treatment survival times. An alternative approach to the estimation of PTS, where treatment and survival are treated as separate entities, is applied to same dataset. The results are compared to the theoretical PTS.

Results: The simulations demonstrate that applying the Kaplan-Meier directly to post-treatment survival times will produce incorrect PTS-estimates whereas the alternative approach performs well. In additional simulations the effect of heterogenic patient populations, treatment regimes and incomplete data on the estimated PTS is illustrated.

Conclusions: Applying the KM-estimator to censored PTS data leads to incorrect estimation of post-treatment survival. PTS will be estimated correctly only when treatment regime and survival are estimated separately and then combined. The methodology described allows estimation of PTS, and if so desired, discounted PTS which is generally used in cost effectiveness evaluations. The methodology can therefore be used and is applicable to the evaluation of loss of life reduction and/or cost effectiveness of measures reducing treatment risks of any long-term repetitive treatment or medication.

See more of Poster Session I
See more of The 28th Annual Meeting of the Society for Medical Decision Making (October 15-18, 2006)