OBJECTIVES: Trials are too short to fully quantify survival, yet detailed life expectancy estimates are required to accurately assess the cost-effectiveness of interventions for cardiovascular disease. Whether the overall life expectancy of a cohort provides an adequate proxy for patients suffering events during follow up is examined in this study. METHODS: Health records of Saskatchewan residents with cardiovascular disease between 1990 and 1995 were obtained. Data were available from January 1980 through December 2002 or date of death. Life expectancy was estimated for the entire cohort of patients and separately for patients with subsequent non-fatal myocardial infarction (MI) or stroke, and with no further events. Piece-wise parametric regressions were used to derive mortality hazard functions; time-dependent Cox proportional hazards were used to adjust for covariates. The resulting hazard functions were used to derive the survival curves. RESULTS: Of 53,983 patients (56% male; mean age 70.4 years) 3,898 suffered subsequent non-fatal MI and 3,714 non-fatal stroke. Accounting for the occurrence of these events lead to considerably shorter life expectancy compared to projected survival derived from the full cohort. For instance, the life expectancy of a 60 year old male who survived an MI is estimated to be 16.6 years when derived from the full cohort versus 13.5 years when this event is considered. The impact of stroke is even greater, as the expected survival after this event is 9.9 years. Differences between full-cohort and event-based estimates were larger for women. For a 60 year old female, the full-cohort estimate is 19.7 years and 15.1 and 11.3 years after MI and stroke, respectively. CONCLUSIONS: Events that occur during follow-up strongly impact the life expectancy of those who survive beyond the study window. Ignoring such differences by oversimplifying model inputs can lead to inaccurate results and dilute treatment effects.