Spatial resolution and temperature-attributable mortality estimates: a nationwide study in Japan

Temperature-attributable mortality estimates are widely used to assess climate-related health risks, yet their comparability across spatial and analytical settings remains uncertain. We compared municipality- and prefecture-level estimates across Japan using daily mortality and temperature data from 1,845 municipalities during 2010-2019. Municipality-level analyses used approximately 1-km gridded temperature and a case-time-series framework, whereas prefecture-level analyses used representative weather-station temperature and conventional time-series regression. Additional implementations examined differences associated with temperature representation, population weighting, analytical framework, and the geographic level at which temperature and mortality were linked. The prefecture-level analysis estimated 1.4% more cold-attributable deaths but 39.2% more heat-attributable deaths than the municipality-level analysis. Cold-related differences were generally smaller than heat-related differences across alternative implementations and sensitivity analyses, although their magnitudes varied with model specification. At the prefecture level, heat-attributable mortality varied by less than 1.5% across representative weather-station, prefecture-mean gridded, and population-weighted gridded temperatures. In contrast, replacing unweighted with population-weighted municipality temperature increased the municipality-level heat estimate by 16.0%. When both spatial levels used the same gridded temperature dataset and conventional time-series framework, the prefecture-level heat estimate remained 17.6% higher. These findings suggest that finer-resolution temperature data alone may have limited influence when summarized and linked to mortality at a broader geographic level. Retaining the local correspondence between temperature and mortality may be particularly important for heat-related burden estimation.