Construction and verification of a prediction model for arteriovenous fistula failure in elderly dialysis patients based on clinical characteristics and laboratory indexes

In elderly patients (≥ 60 years) undergoing autogenous arteriovenous fistula (AVF) creation, early prediction of 12-month failure can guide optimisation and surveillance. We conducted a single-centre retrospective cohort study with a temporal split: patients from June 2022–June 2023 comprised the development cohort (n = 153), and those from July 2023–June 2024 comprised the validation cohort (n = 77). All predictors were measured within 7 days before surgery. Candidate variables were prespecified from prior evidence. Logistic regression with stepwise selection retained age ≥ 70 years, diabetes mellitus, serum IL-6, and CRP. We assessed discrimination (AUROC with 95% CI), calibration (slope, intercept, Hosmer–Lemeshow), overall accuracy (Brier score), and clinical utility (decision-curve analysis). Bootstrap resampling (B = 200) was used for internal optimism correction. Among 230 patients, the 12-month AVF failure rate was 40.0% (92/230). Independent predictors were age ≥ 70 years (OR = 1.058, 95% CI 1.009–1.109), diabetes (OR = 1.654, 95% CI 1.104–2.476), IL-6 (OR = 1.034, 95% CI 1.004–1.064), and CRP (OR = 1.133, 95% CI 1.019–1.260). The final model equation was: logit(p) = − 0.842 + 0.056(Age) + 0.503(Diabetes) + 0.033(IL-6) + 0.125(CRP). Discrimination was strong: AUROC = 0.889 (95% CI 0.816–0.962) in the training cohort and 0.835 (95% CI 0.744–0.926) in validation. Optimism-corrected C-statistic = 0.872. Calibration slope was 0.93, intercept − 0.02; Hosmer–Lemeshow χ²=6.78, P = 0.276. Brier score = 0.13. DCA confirmed positive net benefit between 10 and 30% risk thresholds. In elderly AVF candidates, a four-predictor model integrating age, diabetes, IL-6, and CRP showed strong discrimination and acceptable calibration. While clinically promising, transportability requires multicentre validation, and the added value of IL-6 should be weighed against local testing availability.