Predictive models for acute respiratory infection in Indonesia using official surveillance data and Google Trends

Abstract

Acute respiratory infections (ARI) remain a major public health concern, with substantial regional variation in prevalence. Google Trends may provide digital search signals that complement official health survey data, but its utility for describing provincial variation in ARI prevalence in Indonesia remains insufficiently evaluated. This study evaluated whether Google Trends search indicators could serve as complementary ecological signals for describing variation in ARI prevalence across Indonesian provinces. This cross-sectional ecological study integrated symptomatic ARI prevalence reported in the 2023 Indonesian Health Survey (SKI) with Google Trends Interest by Region data for 34 provinces. The analyses included descriptive statistics, Pearson and Spearman correlations, Benjamini-Hochberg false discovery rate (FDR) correction, and multicollinearity assessment. Candidate statistical-learning and machine-learning models, including OLS, LASSO, Ridge, Elastic Net, Support Vector Regression, Random Forest, and XGBoost, were evaluated using nested cross-validation. In the main analysis of 34 provinces, correlations between search indicators and ARI prevalence were generally weak and were not statistically significant after FDR correction. All evaluated models showed limited out-of-fold generalizability. In the sensitivity analysis excluding East Nusa Tenggara and Papua, LASSO using flu, cough and cold, and shortness of breath achieved the best relative performance (RMSE 4.233, MAE 3.292, and global out-of-fold R² 0.290). Google Trends provided only limited evidence of utility as a complementary and exploratory ecological signal for describing variation in ARI prevalence across provinces. These findings do not support its use as a substitute for official health surveys, forecasting models, or early warning systems.