| Abstract |
We propose a Weibull mixture model considering both covariates and unobserved heterogeneity to examine how demographic variables affect individual survival times and to derive the annual number of deaths. We analyze the records of patients diagnosed with lung and bronchus cancer, the most common cancer in the United States. The result shows that the diagnosis year as well as age, gender, race, and registry significantly affect individual survival times. Including unobserved heterogeneity, we remove a bias in hazard rates and provide better performance in forecasting the annual number of cancer deaths than other benchmarks. Furthermore, from segmenting patients into several groups, we specify the difference between groups and assess their group-specific survival probabilities within a given period. Our study is distinctive in that a bottom-up strategy is adopted to predict aggregate-level units. This makes health forecasting available in two sides: public and private sector. For the public sector, our study enables a more precise allocation of the government’s health and welfare budget. Also for the private sector, our segmentation results provide guidance to the insurance industry for targeting customers more specifically. |