Dynamic Perioperative NPAR Change for Early Identification of Infectious Complications After Percutaneous Nephrolithotomy: Development and Internal Validation of a Simple Diagnostic Model

Background: Infectious complications are the most closely monitored adverse events after percutaneous nephrolithotomy (PCNL), and static single-timepoint markers stratify risk poorly. We evaluated whether the perioperative change in the neutrophil-percentage-to-albumin ratio (ΔNPAR), derived entirely from routine blood tests, identifies patients at higher risk of infectious complications within the first postoperative day.

Methods: We retrospectively studied 339 patients (development cohort, South Campus, n=244; validation cohort, North Campus, n=95). The outcome was a predefined infectious complication after PCNL, ascertained at 24 h. The dynamic NPAR change was calculated as the postoperative minus preoperative difference (ΔNPAR = postoperative − preoperative NPAR). Because the candidate variables and the outcome were ascertained at the same 24-h assessment, the model is diagnostic rather than prognostic in design. A four-variable logistic model (female sex, diabetes, 24-h white blood cell count and ΔNPAR) was developed, internally validated by bootstrap resampling and penalized (Firth) regression, tested for robustness by propensity-score matching and decision-curve analysis, and then applied unchanged to the North Campus.

Results: All four factors were independently associated with the outcome (all P<0.05): female sex (adjusted odds ratio 6.30, 95% confidence interval [CI] 2.44–16.28), diabetes (3.26, 1.29–8.28), 24-h white blood cell count (1.19, 1.04–1.36) and ΔNPAR (per 1 standard deviation 2.47, 1.56–3.93). The model discriminated well in the development cohort (area under the curve [AUC] 0.868, 95% CI 0.812–0.923), with minimal optimism after bootstrap correction (0.854), adequate calibration (Hosmer–Lemeshow P=0.26) and a positive net benefit across threshold probabilities of approximately 5–60%. At the Youden-optimal threshold, sensitivity was 96.9% and the negative predictive value (NPV) was 99.2%. Discrimination was maintained on the North Campus (AUC 0.896, 95% CI 0.811–0.981; NPV 98.2%). Penalized regression and propensity-score matching reproduced the same associations. Preoperative NPAR did not differ between groups (P=0.725), indicating that the discriminative signal arose specifically from the perioperative dynamic change rather than from any baseline difference.

Conclusions: A simple four-variable diagnostic model built on the dynamic change in NPAR identified PCNL patients with early infectious complications, showing good discrimination, a positive net benefit and a high negative predictive value that could support earlier discharge or de-escalation of care for low-risk patients. All inputs are obtained from routine perioperative blood tests at no additional cost. Because the variables and the outcome are ascertained at the same assessment, the model performs contemporaneous risk stratification and early identification; it does not forecast a later event, and external multicentre validation against a culture-confirmed endpoint is warranted.