CHRONIC KIDNEY DISEASE IN VULNERABLE AFRO-DESCENDANT, INDIGENOUS, AND AGRICULTURAL COMMUNITIES IN LATIN AMERICA

Chronic kidney disease (CKD) is a major global public health concern, affecting an estimated 13.4% of the world’s population (Lv & Zhang, 2019). Among afro-descendant, indigenous, and agricultural communities in Latin America, studies have identified a high prevalence of CKD, frequently undiagnosed and untreated (Correa-Rotter et al., 2014; Garza & Abascal Miguel, 2025; Ulasi et al., 2025). To effectively reduce the burden of CKD in these populations, it is essential to assess how the social determinants of health (SDOH) influence kidney health (Burgos-Calderón et al., 2021). The FRENEL study is an ongoing multicenter initiative, which to date has screened 4,876 participants from vulnerable agricultural, indigenous, and afro-descendant communities in Latin America.

The present study employed a cross-sectional, multicenter design involving adults from vulnerable Afro-descendant, Indigenous, and agricultural communities across nine Latin American countries. Objetive To evaluate the risk factors associated with chronic kidney disease (CKD) in vulnerable afro-descendant, indigenous, and agricultural communities in Latin America. Sociodemographic, clinical, and lifestyle data were collected, including blood pressure, body mass index (BMI), proteinuria, and serum creatinine. The estimated glomerular filtration rate (eGFR) was measured using the Nova Max Pro point-of-care (POC) system. The binary outcome variable was defined according to the KDIGO classification, distinguishing participants with eGFR < 60 mL/min/1.73 m² (G3a–G5), indicative of at least moderate reduction in kidney function, from those with eGFR ≥ 60 mL/min/1.73 m². Multiple supervised machine learning algorithms, including logistic regression, categorical Naive Bayes, decision trees, random forest, and XGBoost, were compared. Model performance was assessed using standard classification metrics, including accuracy, precision, recall, F1-score, and the area under the receiver operating characteristic curve (AUC-ROC). Models were optimized through hyperparameter tuning using cross-validation to ensure robustness and prevent overfitting.

Among 4,876 participants, 1,105 (22.7%) presented an eGFR < 60 mL/min/1.73 m² (CKD G3a or higher). Model comparison identified XGBoost as the best-performing classifier. The model achieved a Recall of 0.65, Precision of 0.38, F1-score of 0.48, Accuracy of 0.68, and a weighted average F1-score of 0.70 on the test set. Feature importance analysis indicated that the strongest predictors were hypertension history, proteinuria, and older age. Socioeconomic factors, including subsidized/no health insurance, indigenous ethnicity, and discontinued education were also highly predictive

The study revealed a high burden of CKD among Afro-descendant, Indigenous, and agricultural populations, with a prevalence significantly higher than that observed in the general population, indicating an elevated risk in these groups. These findings highlight the need to implement preventive strategies that address both clinical factors and SDOH. The FRENEL registry is continuously expanding, with a target enrollment of 20,000 participants, and ongoing efforts are focused on developing improved predictive models that address class imbalance to better capture CKD risk patterns.