A PENALIZED LIKELIHOOD APPROACH FOR ARBITRARILY CENSORED AND TRUNCATED DATA - APPLICATION TO AGE-SPECIFIC INCIDENCE OF DEMENTIA

The Cox model is the model of choice when analyzing survival data pres enting only right censoring and left truncation. There is a need for m ethods that can accommodate more complex observation schemes involving general censoring and truncation. In addition, it is important in man y epidemiological applications to have a smooth estimate of the hazard function. We show that the penalized likelihood approach gives a solu tion to these problems. The solution of the maximum of the penalized l ikelihood is approximated on a basis of splines. The smoothing paramet er is estimated using approximate cross-validation; confidence bands c an be given. A simulation study shows that this approach gives better results than the smoothed Nelson-Aalen estimator. We apply this method to the analysis of data from a large cohort study on cerebral aging. The age-specific incidence of dementia is estimated and risk factors o f dementia studied.