PURPOSE: To construct lifetables of background mortality rates as a function of smoking and obesity for 1970-2003. These tables are designed for simulation models that aim to explain past trends or forecast future trends of diseases constituting substantial shares of all deaths, e.g., obesity & smoking not only increase colorectal cancer (CRC) but also predispose individuals to other causes of death including heart diseases and other cancers.

METHODS: Separating other-cause mortality rates by obesity/smoking requires three types of inputs: first, we obtained population-wide non-CRC mortality rates by removing CRC from annual US all-cause lifetables (1970-2003), by sex and age. Second, we defined 15 risk-factor strata from 3 smoking (never, past, and current smoker) and 5 body mass index (BMI) classes (<18.5, 18.5-24.9, 25-29.9, 30-34.9, 35+ kg/m2). We fit generalized logit models to 4 cross-sectional national surveys of adults age 20+ (National Health and Nutrition Examination Surveys (NHANES) 1971-2004, N=32,318) and estimated probability distributions of US adults falling into each BMI/smoking category, by age and year 1970-2003. Third, we fit Poisson regression models to NHANES linked-mortality data to obtain smoothed hazard functions for non-CRC deaths, conditional on BMI/smoking at baseline, sex, age, and year. Finally, we estimated non-CRC mortality rates for each BMI/smoking category to demonstrate differential mortality trends. Similar approach is applied to lung (LC) and breast cancers (BC).

RESULTS: The impact of removing selected cancers from all-cause mortality varies in magnitude & calendar period: removing CRC is most significant at ages 65-70 (3-5% of all-cause mortality), a proportion that increases 1970→2000 in men but decreases in women; for LC, the removal peaks ~age 60 (8% (1970)→14% (2000) in men; 4%→13% in women); for BC, 15% (1970) ~age 40→11% (2000) ~age 50. Recent smoking rates in US are 40% lower in men and 33% lower in women as compared to 1970, while obesity approximately doubled. Both risk factor trends and hazard ratios show differences by race, sex and age that may affect relative survival among BMI/smoking classes for non-CRC, non-LC, and non-BC deaths.

CONCLUSION: Common risk factors for both cancer and non-cancer causes may introduce bias if not accounted for in evaluating preventive initiatives. For modelers, our results can raise methodological considerations relating to lag time and interactions in using these lifetables to predict mortality.