A discrete latent factor model for smoking, cancer and mortality.

This paper investigates the relationships between social circumstances, individual behaviours, and ill-health later in life, with a particular focus on the development of cancer. A discrete latent factor model incorporating individuals' smoking and health outcomes (lifespan and time-to-cancer) is jointly estimated, using the 1984/5 British Health and Lifestyle Survey (HALS) dataset and its July 2009 follow-up, allowing for unobservable factors to affect decisions regarding smoking behaviours as well as health outcomes. Results from this discrete latent factor model are found to be substantially different to those derived from single-equation modelling, suggesting the presence of unobserved heterogeneity. Contrary to previous work on the relationship between circumstances and the development of cancer, a social gradient in time-to-cancer is observed, with individuals in the lowest two social classes developing cancer significantly sooner than individuals in the highest social class. The reduction in estimated median time-to-cancer between individuals in the highest social class, and those in the lowest social class, is found to be between 4 and 4.5 years; approximately twice as many individuals in the lowest social classes as in the highest social class are predicted to develop cancer by an observed age of 75. Those in lower social classes are found to be more likely to smoke, smoke earlier in life, and smoke more cigarettes before quitting.