Identification of high notification clusters of congenital heart disease in Colombia (2016–2020): a spatial epidemiology and machine learning approach

Abstract

Congenital heart diseases (CHD) are the most common congenital anomalies and a leading cause of morbidity and mortality in neonates and children. In Colombia, geographic variation in their notification reflects not only the true distribution of the disease, but also differences in access to prenatal and postnatal diagnosis and the performance of the healthcare system. Understanding these spatial patterns and their determinants is crucial for identifying potential inequities in detection and reporting. A retrospective ecological study (2016–2020) was conducted at the municipal level. CHD cases reported to SIVIGILA were used, with live births from DANE as the denominator. Geographic, sociodemographic, and healthcare access indicators were obtained from SISPRO and Terridata. Spatial analysis (SaTScan) was used to identify clusters of high CHD notification, and the XGBoost algorithm was applied to model cluster membership. The contribution of associated factors was assessed using SHAP values. Between 2016 and 2020, 2.4 notified CHD cases per 1,000 live births were reported in Colombia. Five high-notification clusters were identified, predominantly in the Andean region and in areas with high population density. The XGBoost model predicted cluster membership with an AUC-ROC of 0.89. Altitude, economic and urban dimensions, population density, access to improved water sources, and prenatal care coverage were the most predictive factors. The identified clusters reflect non-random spatial patterns in CHD notification in Colombia, associated with socio-geographic contextual characteristics that suggest a paradox: greater effectiveness of diagnostic, referral, and reporting systems, rather than a higher underlying occurrence of disease, may be contributing to the observed notification patterns and masking the true burden of CHD.