A CASE-CROSSOVER STUDY ON THE RELATION BETWEEN TEMPERATURE AND MORTALITY IN JAPAN

Abstract

ISEE-223 Introduction: One of the main drawbacks of the usual time-series analysis on the relation between ambient temperature and mortality rate is that it cannot control for factors that modify the effect-measure. In order to overcome this problem, we plan to conduct case-crossover study nested in a cohort study. As the first step of this study, we evaluated the relation using all residents of the municipalities from which the cohort study members were recruited. Methods: We used computerized daily mortality data with permission by the government, and computerized ambient temperature data (obtained from Automated Meteorological Data Acquisition System) from 1985 to 1995 for the areas of the cohort study. Using the data, we conducted a case-crossover study with symmetrical bidirectional sampling as a control selection method. The interval between the day a case deceased and the control days was one week, such that it avoids the weekday-weekend difference. Mantel-Haenszel odds ratio (MHOR) with each person as a stratum was the measure of association. Mortality comparison was between the days with more than 33 degrees C and the days with less than 33 degrees C. Results: As shown in the Table, the results varied with the area; in Iwate, Nagano and Okinawa area, we observed effect-measure modification by age. Akita’s effect-measure modification was moderate. On the other hand, Tokyo area showed null OR in both of the age groups. We did not use semi-symmetrical bidirectional sampling because we prefer statistical stability, but since the mortality rate has been decreasing in Japan, it may be biased by time trend. Another problem with the present evaluation is that it lacks precision. It may not be feasible to conduct a nested case-crossover study, but the present evaluation shows that area- and age-difference should be taken into account.Table: Area difference in Mortality: 33+ degrees C compared with <33 degrees C