Abstract
STUDY OBJECTIVES: The study aimed to develop and evaluate a method for small area analysis of different non-random components of the variation in death rates. The method was applied to incidence and mortality data for selected malignant neoplasms in 26 administrative health areas in Sweden. DESIGN: Variation in mortality and incidence rates for malignant neoplasms of the trachea, bronchus, and lung; colon; rectum; and cervix uteri in the 26 health areas were analysed after standardisation for age. In addition, the systematic and random components of variance were estimated. The systematic component of variance in mortality was divided into two additive components-one component was dependent on the variation in the corresponding incidence rates and the other remained after adjustment for incidence. SETTING AND PARTICIPANT: All cases diagnosed between 1972 and 1983 and all deaths between 1974 and 1985 for selected malignant neoplasms in Swedish citizens and other residents in Sweden, aged between 0 and 64 years, were analysed. MAIN RESULTS: Much of the observed variation in mortality was explained by the estimated random variation. For malignant neoplasms of the trachea, bronchus, and lung the systematic variation in mortality was mainly explained by the variation in incidence. For cancer of the cervix uteri, alone, there was significant systematic variation of moderate magnitude that was not explained by the incidence rates. CONCLUSIONS: These methods made it possible to divide the observed variance in mortality into different components. Random effects and variance in incidence rates were found to be of great importance when analysing the variance in death rates between health areas. By studying different systematic components of variation it is possible to identify fields for in depth studies on the quality of prevention and treatment.