Abstract
BACKGROUND: Stroke-associated pneumonia (SAP) complicates 10-30% of acute ischemic stroke (AIS) cases and worsens outcomes. The C-reactive protein to albumin ratio (CAR) integrates inflammation and nutritional status and may serve as a prognostic biomarker. However, non-linear relationships and threshold effects for CAR in SAP prediction remain unclear. METHODS: We retrospectively analyzed 1,595 consecutive AIS patients admitted between September 2016 and September 2022. CAR was calculated from admission CRP and albumin. Associations between CAR and SAP were assessed using multivariable logistic regression, generalized additive models (GAM), and two-piecewise regression to identify thresholds. Predictive performance was evaluated by ROC analysis and DeLong's test. RESULTS: A total of 1,595 patients were included (58.6% male; mean age 70.1 ± 12.2 years). The median admission NIHSS score was 3.0 (IQR 1.0-6.0). SAP occurred in 376 patients (23.6%). Log₂-CAR was strongly associated with SAP risk (P-trend < 0.0001), with the highest quartile showing a fully adjusted OR of 6.11 (95% CI: 3.63-10.27, p < 0.0001). This association was consistent across all subgroups (all P-interaction > 0.05). A non-linear threshold was identified at CAR ≈ 0.14; below this, the association was modest (OR = 1.22, 95% CI: 1.02-1.45), while above it, risk increased substantially (OR = 2.03, 95% CI: 1.70-2.42). The threshold model outperformed the linear model (p < 0.001). At the optimal cutoff of CAR = 0.18 (Youden's index), CAR had superior predictive performance for SAP (AUC = 0.832, 95% CI: 0.807-0.858) compared to CRP (AUC = 0.827, p = 0.0002), A2DS2 (AUC = 0.764, p < 0.001), and WBC (AUC = 0.780, p = 0.0029). Adding CAR to baseline models improved prediction accuracy (all p < 0.05). CONCLUSION: CAR is a strong and independent predictor of stroke-associated pneumonia, outperforming traditional markers and improving risk prediction in acute stroke patients.