Estimation of life expectancy — cohort life table

Life tables are traditionally built with linear assumptions for the survival curve. Here, considering that survivors can remain at the end of the observation period, the author shows that non linear modeling is more appropriate. With data on cervix uteri cancer, e0 ≈ 12.5 years with standard error ≈ 2.8 years with infinite time horizon, but e0 ≈ 6.0 years with standard error ≈ 0.1 year in interval with finite time horizon 0, 12 years]. The average hazard function is introduced to estimate the life expectancy, and the actuarial estimate of the hazard function is showed to under‐estimate the true hazard values under the exponential distribution. Finally, a sensitivity analysis of the probabilities of death on the estimation of life expectancy completes the study.