Retrospectively reported indoor exposures, allergic multimorbidity, and questionnaire-based stratification among children aged 3–8 years in Northwest China: a cross-sectional study

Allergic diseases often coexist in childhood. We described allergic multimorbidity among children aged 3–8 years in Urumqi and Hami, examined associations with retrospectively reported indoor exposures, and evaluated questionnaire models for identifying parent-reported physician-diagnosed allergic disease. We conducted a stratified, cluster-sampled cross-sectional survey in Xinjiang, China, in 2025–2026. Parents reported prenatal, infancy and past-12-month indoor exposures and allergic diagnoses. Allergic multimorbidity was defined as two or more physician-diagnosed allergic diseases. We estimated exposure associations using multivariable models and compared ten prediction algorithms. Urumqi provided training and internal-validation data; Hami was excluded from fitting and preprocessing but informed the model comparison. Additional analyses used a separately fitted LightGBM model, with its parameters held fixed for interpretation and score stratification. Among 9033 children, 2356 (26.1%) had at least one physician-diagnosed allergic disease and 704 (7.8%) had allergic multimorbidity. In the original Urumqi analysis, infancy second-hand smoke (OR 1.57, 95% CI 1.17–2.11) and current second-hand smoke (1.55, 1.26–1.91) were associated with multimorbidity. Prenatal-renovation and infancy-furniture estimates were borderline; the prenatal-renovation interval included the null after weighting for exposure completeness. In the original model comparison, LightGBM with random oversampling had the highest Hami AUC point estimate (0.728, 95% CI 0.669–0.786). The supplementary model had AUCs of 0.726 internally and 0.735 in Hami, but overpredicted diagnosis probabilities (Hami observed-to-expected ratio 0.563). The cross-sectional exposure associations are hypothesis-generating and cannot establish causal effects. Questionnaire models showed moderate discrimination but overpredicted diagnosis probabilities. Because Hami informed the algorithm choice, evaluation in a new cohort and recalibration are needed before considering individual screening. Longitudinal studies with objective exposure measurements are also required.