Non conventional risc factors for coronary heart disease in end stage renal disease patients
M C Cafka, M R R Rroji, E S H Shkembi, J D Djamandi, D C Cafka, A S Strakosha, A D DibraAbstract
Background
Risk stratification for angiographic coronary artery disease (CAD) in hemodialysis remains difficult.
Methods
We analyzed 49 hemodialysis patients undergoing coronarography (42 CAD+, 7 normal). CAD+ was any 1/2/3-vessel involvement; "Normal coronary artery" was negative. From the record we structured ischemic history (MI, PCI) and standard labs; continuous variables were standardized. Univariate screening used Fisher’s exact and Mann–Whitney U. Multivariable analysis used ridge logistic regression with 5-fold stratified cross-validation; discrimination by ROC/PR, calibration by Brier score/Hosmer–Lemeshow, and clinical utility by decision-curve analysis.
Results
The cohort was 67.3% male and 32.7% smokers; prior MI and PCI occurred in 57.1% and 63.3%, respectively. Smoking prevalence was 38.1% in CAD+ vs 0% in normals. Key continuous contrasts (median [IQR], CAD+ vs normal):
• Dialysis vintage (months): 22.0 [4.5–105.0] vs 24.0 [2.0–24.0]; p=0.33; Uricemia (mg/dL): 7.65 [6.15–8.90] vs 5.90 [5.10–7.15]; p=0.038; Hemoglobin (g/dL): 10.70 [9.65–11.55] vs 10.90 [10.40–11.85]; p=0.579; Albumin (g/dL): 3.30 [3.10–3.70] vs 3.15 [2.92–3.38]; p=0.391; Calcium (mg/dL): 8.05 [7.55–8.85] vs 8.40 [8.10–9.12]; p=0.178
In the multivariable model, smoking, prior MI/PCI, higher uricemia, and longer dialysis vintage were associated with higher odds of CAD+, while higher calcium, and trends toward higher hemoglobin/albumin, were protective. The cross-validated model achieved ROC AUC = 0.93, with acceptable calibration and positive net benefit across clinically reasonable thresholds versus treat-all/none strategies.
Conclusions
In hemodialysis patients, ischemic history, smoking, uric acid, and dialysis vintage are key correlates of angiographic CAD, while higher calcium and better hematinic/nutritional markers trend protective. A penalized, cross-validated model provided strong discrimination (AUC 0.93), supporting these predictors for risk stratification pending validation in larger, balanced cohorts.