Objective To investigate the influence of meteorological factors on the number of non-accidental deaths in Mengzi city, China, and to establish a health risk early warning model for dry-wet effects.

Methods The distributed lag non-linear model (DLNM) was used to analyze the correlation and hysteresis between meteorological factors and the number of non-accidental deaths and evaluate the interaction of cold-hot and dry-wet effects. On this basis, the health risk early warning model was established by the random forest algorithm, the extended Shepard interpolation method and the constructed nonlinear function regression method to examine its early warning effect for the health risk level of dry-wet effects.

Results Through Spearman rank correlation analysis among meteorological elements, air pollutants, and the number of non-accidental deaths, seven factors, i.e., mean air temperature, diurnal temperature range, relative humidity, sunshine duration, surface pressure, mean wind speed, and O₃ were included in the DLNM. The exposure-response curve between the mean air temperature and relative humidity and the number of non-accidental deaths was U-shaped, with the minimum risk of death at 18.0 ℃and 76%, respectively.The comfort ranges of mean air temperature and relative humidity were (11.0-22.5)℃ and 58%-79%, respectively.Through random forest quantitative analysis, relative humidity, diurnal temperature difference, mean air temperature and surface pressure were included in the early warning model as optimal factors.In the interaction between cold-hot and dry-wet effects, dry-cold effect did not appear, but wet-cold and dry-hot effects were significant.For the early warning result for the health risk level of dry-wet effect, the extended Shepard interpolation method and the constructed nonlinear function method both had an independent sample early warning accuracy of >50%, with the highest accuracy of 77.8%.

Conclusion The number of non-accidental deaths in Mengzi city aremainly affected by high temperature, low temperature, large diurnal temperature difference, dryness, and high humidity, and the wet-cold and dry-hot effects also have a significant effect on the number of non-accidental deaths. The health risk level of dry-wet effect is reasonable. The constructed nonlinear function method is more accurate for the early warning of low risk, and the extended Shepard interpolation method is more effective for the early warning of medium-high risk. The two method have their own advantages and can be used in combination.