Skip to main content

Conference Program

Subpage Hero

Loading

Utilization of predictive modeling to maximize benefits of genetic testing for patients with chronic kidney disease

Clinical Genetics and Therapeutics
  • Primary Categories:
    • Clinical Genetics
  • Secondary Categories:
    • Clinical Genetics
Introduction:
Genetic testing is increasingly important in the diagnosis and management of patients with chronic kidney disease (CKD) but remains under-utilized in nephrology practice. Although more than 600 genes have been associated with CKD, identifying patients most likely to benefit from genetic testing and tailoring genetic testing strategies to individuals remain significant challenges. We aim to a model to assist clinicians in selecting candidates for genetic testing based on individual patient factors.

Methods:
Comprehensive demographic, clinical, and laboratory data were collected from 645 patients evaluated by the Cleveland Clinic Foundation (CCF) Renal Genetics Clinic (RGC) from December 2018-July 2024. For patients wth multiple genetic testing results, a hierarchical classification system was used with clinically relevant findings taking precedence. Statistical analysis included Fisher’s exact testing for categorical variables and two-sample t-tests for continuous variables. False discovery rate (FDR) was used for multiple testing correction. Predictive modeling was conducted using XGBoost, a scalable tree-boosting algorithm, with an 80:20 train-test split, a tree depth of 6, and 400 boosting rounds (training model iterations). Receiver operating characteristic (ROC) area under the curve (AUC) was used to assess model performance. All statistical analysis was performed using R version 4.1.3.

Results:
Of the 645 patients, genetic testing revealed positive results in 226 (35%) patients, negative results in 177 (27.4%) patients, variants of uncertain significance (VUS) in 133 (20.6%) patients, carriers of autosomal recessive conditions (carriers) in 98 (15.2%) patients, and APOL1 risk genotypes in 11 (1.7%) patients. Only patients with positive or negative results (n=403) were included in model building. Patients with positive family history (q=0.017), diagnosis of a glomerular disease (q<0.001), or biopsy findings of focal segmental glomerulosclerosis (FSGS) (q=0.017) or thin glomerular basement membrane (GBM) disease (q=0.025) were significantly more likely to receive positive genetic testing results after multiple testing correction. Conversely, patients with diagnosis of atypical hemolytic uremic syndrome (aHUS) or thrombotic microangiopathy (TMA) (q=0.047), congenital anomalies of the kidneys and urinary tracts (CAKUT) (q=0.049), or electrolytes disorders (q=0.022) were significantly more likely to receive negative genetic testing results after multiple testing correction. XGBoost predictive modeling performed with a ROC-AUC of 0.76. The variables eGFR, glomerular disease, age at index visit, family history, and urine protein to creatinine ratio (UPCR) were 5 most important variables in model performance as determined by mean Shapley Additive Explanations (SHAP) values.

Conclusion:
This is the first study to report a model capable of reliably predicting genetic testing outcomes for CKD patients. XGBoost is a unique in its ability to incorporate missing values, which was important for integrating diverse clinical data. An ROC-AUC of 0.76 represents robust predictive accuracy, setting the stage for a future web-based tool to guide nephrologists in genetic testing decisions. Importantly, as this model was trained on CKD patients referred specifically for genetics evaluation, validating model performance in a broader CKD cohort will be an essential next step.

Agenda

Sponsors