Abstract
Abstract
Background.
This study examines the association between ambient temperature and road-traffic mortality across Thailand and characterises its pattern: the shape of the association over the local temperature distribution, and the share of the associated deaths that arises on moderately warm rather than extreme days. Temperature–mortality research has concentrated on non-accidental causes. Recent work in Latin America, Japan and China has quantified heat effects on road-traffic and other external-cause deaths, but nationwide evidence from Southeast Asia remains scarce, and few studies report where across the temperature distribution the associated burden accrues.
Methods.
We analysed 283,847 road-traffic deaths in Thailand's 77 provinces between 1 January 2011 and 24 August 2025, from the integrated RTDDI death file. The exposure dataset was constructed for this study: 4,514,684 daily station observations from 1,347 stations, harmonised from two national meteorological networks with per-station quality control, and a 0.1° gridded reconstruction of daily maximum temperature by inverse-distance weighting from 1,145 stations. Exposure entered as province-specific percentile bands. We fitted a time-stratified case-crossover design as a conditional Poisson model, stratifying on province × year × month × day-of-week, with cluster-robust variance on province, adjusting for precipitation and the Songkran and New Year holiday periods; the estimator was implemented for this study and validated against a Poisson GLM with dense stratum fixed effects (agreement to 5.8 × 10
− 15
). Five pre-specified sensitivity analyses addressed spatial exposure resolution, exposure assigned by crash rather than death date, exclusion of the Songkran window, consecutive-day heatwave definitions, and composite humid-heat metrics.
Results.
Mortality risk per day rose from the cold end of the distribution to approximately the 75th local percentile and showed no further increase above it (P75–90 RR 1.045, 95% CI 1.023–1.068; P95–99 RR 1.050, 1.018–1.083; ≥P99 RR 1.041, 0.979–1.108; reference P40–60). A binary extreme-heat-day contrast gave RR 1.015 (0.992–1.039), because its comparison group already contains the elevated warm days. The shape persisted when exposure was reassigned at 0.1° spatial resolution using crash coordinates on the identical 107,313 geocoded deaths, despite a mean absolute difference of 2.95°C between the two exposure measures, and when exposure was assigned by crash date rather than death date in the 94,437 deaths where both were available. It was also unchanged by excluding the Songkran holiday period, by a consecutive-day heatwave framing across 15 definitions, and by substituting composite humid-heat indices for dry-bulb temperature. Excess mortality associated with temperatures above the P40–60 reference band was 4,509 deaths (95% CI 2,581–6,442), 1.59% of all road-traffic deaths.
Approximately 72% of that excess occurred between the 60th and 90th percentiles, whereas only 14.8% occurred above the 95th percentile.
Conclusions.
Road-traffic mortality was higher on warm days than on days of median temperature, with risk rising from moderately warm temperatures onward and levelling off above approximately the 75th local percentile. Because moderately warm days are common, they carried most of the temperature-associated excess: about 72% arose between the 60th and 90th percentiles and 14.8% above the 95th. Warm weather therefore warrants sustained road-safety attention across the whole warm season, and not only on the extreme days that heat–health alert systems are designed to detect, which account for a small share of this burden.