P N Lee1, B A Forey. 1. P N Lee Statistics and Computing Ltd., Sutton, Surrey, UK.
Abstract
STUDY OBJECTIVES: To determine whether British lung cancer (LC) trends are adequately explained by cigarette smoking trends, and whether modelling using aggregated smoking prevalence estimates can validly replace modelling using individual smoking histories. METHODS: Observed LC trends for 1955-1985 for both sexes and three age groups were compared with multistage model predictions using smoking history data from two surveys (HALS, AHIP). The modelling used the individual smoking data directly or aggregated prevalence estimates. It allowed for variation in age of starting and stopping smoking, amount smoked, tar levels, and environmental tobacco smoke (ETS) exposure. RESULTS: Observed male LC rates fell faster than predicted by a model (with the first and penultimate stages assumed affected by smoking) that allowed for variation in amount smoked and in tar level (with some provision for "compensation"), and was based on aggregated smoking data from HALS. The discrepancy equated to an annual change unexplained by smoking of -2.4%, -2.8%, and -1.9% for ages 35-44, 45-54, and 55-64. The annual unexplained changes were less in women, and reversed at age 55-64; -1.7%, -0.8%, and +0.8% for the three ages. They were similar using individual smoking histories (-2.6%, -1.8%, and -1.6%; women, -0.9%, -0.5%, and +0.2%). The discrepancies were unexplained by plausible alternative multistage parameters, full allowance for tar reduction, alternative estimates of amount smoked, or ETS. CONCLUSIONS: British LC trends cannot be fully explained by cigarette consumption trends, implying factors other than cigarette smoking contribute importantly to overall risk. Predictions using aggregated prevalence estimates provide useful information.
STUDY OBJECTIVES: To determine whether British lung cancer (LC) trends are adequately explained by cigarette smoking trends, and whether modelling using aggregated smoking prevalence estimates can validly replace modelling using individual smoking histories. METHODS: Observed LC trends for 1955-1985 for both sexes and three age groups were compared with multistage model predictions using smoking history data from two surveys (HALS, AHIP). The modelling used the individual smoking data directly or aggregated prevalence estimates. It allowed for variation in age of starting and stopping smoking, amount smoked, tar levels, and environmental tobacco smoke (ETS) exposure. RESULTS: Observed male LC rates fell faster than predicted by a model (with the first and penultimate stages assumed affected by smoking) that allowed for variation in amount smoked and in tar level (with some provision for "compensation"), and was based on aggregated smoking data from HALS. The discrepancy equated to an annual change unexplained by smoking of -2.4%, -2.8%, and -1.9% for ages 35-44, 45-54, and 55-64. The annual unexplained changes were less in women, and reversed at age 55-64; -1.7%, -0.8%, and +0.8% for the three ages. They were similar using individual smoking histories (-2.6%, -1.8%, and -1.6%; women, -0.9%, -0.5%, and +0.2%). The discrepancies were unexplained by plausible alternative multistage parameters, full allowance for tar reduction, alternative estimates of amount smoked, or ETS. CONCLUSIONS: British LC trends cannot be fully explained by cigarette consumption trends, implying factors other than cigarette smoking contribute importantly to overall risk. Predictions using aggregated prevalence estimates provide useful information.