Machine learning reveals distinct temperature thresholds and environmental modulators for atopic dermatitis and...

Atopic dermatitis (AD) and allergic contact dermatitis (ACD) are common inflammatory skin diseases influenced by environmental factors, but disease-specific environmental pathways remain poorly defined. This study developed a machine learning model to predict monthly disease prevalence and characterize distinct environmental conditions associated with each disease. We analyzed nationwide health insurance claims data for AD, ACD, and corns (control) from six major South Korean cities from 2012 to 2017, constituting 432 city-month records per disease. The M5P model tree algorithm predicted relative monthly prevalence based on meteorological data (temperature, humidity, precipitation, diurnal temperature range) and air pollutants (SO₂, NO₂, CO, PM10), with performance evaluated using Pearson Correlation Coefficient (CC) and Mean Absolute Error (MAE). Analysis of 3,990,692 AD and 16,890,182 ACD cases showed that the combined weather-pollution model achieved high accuracy for AD (CC = 0.839, MAE = 0.038) and ACD (CC = 0.932, MAE = 0.049). Mean temperature was the primary splitting variable for both diseases, but with different thresholds and secondary modulators. For AD, the initial split occurred at 17.4°C; above this, high PM10 (>44μg/m³) was associated with higher prevalence. For ACD, a notable split was identified at 11.65°C; below this, low humidity (<62%) appeared to be a key contributing factor. PM10 was a consistent predictor for both diseases. While temperature is a universal primary driver for both AD and ACD, the diseases follow distinct environmental pathways. AD is modulated by air pollution in warmer conditions, whereas ACD is sensitive to humidity in cooler conditions. This data-driven approach provides insights into disease-specific environmental triggers for public health interventions.