A modified landmark method for dealing with an event observed during a follow-up period.

Abstract

Events observed during a follow-up period are generally more informative when predicting the prognosis of patients than the clinical data available at the start of follow-up. The present paper has aimed to introduce a modified landmark method for estimating a survival curve for patients in whom an event related to their survival had occurred at a given landmark time during the follow-up period. Survival curves were compared graphically by the conventional method, the Mantel-Byar method, the landmark method and the modified landmark method. The subjects were ninety-five patients with myelodysplastic syndrome who had been diagnosed between 1979 and 1992 at Tokyo Medical College Hospital, and followed-up until September, 1993. Disease evolution was used as the event. The survival curves for patients with and without the event, calculated by the conventional method, were biased and not applicable to the actual patients at the start of follow-up before the occurrence of the event. The curves obtained using the Mantel-Byar method were not biased but similarly not applicable to the actual patients. The landmark method gave unbiased curves applicable to patients with and without the event at a given landmark time, but the curve for those with the event included a lead time from the onset of the event to the landmark time. The modified method gave an unbiased estimation of survival for patients in whom the event had just occurred around a given landmark time.