Application of Illness-Death Model for Semi-Competing Risks in Analysis Gastric Cancer Data

In cancer studies, sometimes events are semi-competing. In this case, each patient experiences more than one event, and only one of those is a terminal event (for example, death), and others are non-terminal events (for example, relapse of the disease after treatment) where the risk of death competes with the risk of a non-terminal event of relapse but not vice versa. The aim of this study was to determine the effect of risk factors on relapse time, death time without relapse, and death time after relapse in gastric cancer patients using an illness-death model.
Materials & In this historical cohort study, 200 patients with gastric cancer were studied. These patients had referred to Imam Khomeini hospital 2003 - 2007 and were followed up to October 2009. The data was analyzed by illness-death model, with and without frailty by considering semi-Markov assumption in R 3.4.3 software. Of the patients, 61 (30.5%) had relapses, 71 (35.5%) deaths, 27 (13%) relapsed and death, 34 (17%) relapsed without death and 44 (22%) was died without relapsed. In multivariate analysis, based on confidence intervals, the lymph nodes had a significant effect on the relapse time; metastasis had a significant effect on the time of death without relapse and stage of disease had a significant effect on the death time after relapse. Estimates of the effect of variables were different at the stages in the model. In other words, the effect of risk factors on the time of death was different before and after of relapse occurrence. Therefore, the relapse event as an intermediate event is very important and should be considered by therapists.