| Abstract |
This study suggests a model to forecast individual survival times with diagnosis record as covariates. The dataset is retrieved from the Surveillance, Epidemiology, and End Results (SEER) database of the National Cancer Institute. It consists of demographic variables such as individual age, gender, race, and registries and incidence-related variables (i.e. diagnosed year, survival time, and tumor progression). We used a lung and bronchus cancer which is a primary cancer in the United States in our study. Since some subjects have censored survival times, they are decomposed into two groups as of the follow-up cutoff date, December 31, 2009. The first group is comprised of subjects who have survival time records before the date and the second consists of those who are survived within the observation period. Our analysis shows that patients who are diagnosed relatively late ｏｒ young are expected to survive longer than those are not, whereas those who are female tend to survive for a shorter period than male. Using these results, we can calculate the individuals’ survival probability at each year so that we can derive annual deaths caused of the cancer. Our model is expected to contribute to construct the government’s budget into health and welfare as well as to predict the demands for insurance industry related to the cancer. |