Clostridioides difficile Infection in a Romanian Tertiary Care Hospital: Epidemiology, Comorbidity Burden, Treatment Patterns, and Mortality Outcomes (2023–2025)

Background: Real-world outcome data on Clostridioides difficile infection (CDI) from Eastern Europe are scarce, and partly digitised hospital records make surveillance laborious. We characterised the epidemiology, comorbidity, treatment and in-hospital mortality of CDI in a Romanian tertiary hospital and validated a rule-based pipeline for extracting surveillance variables from free-text records. Methods: Retrospective two-period cohort (1 September 2023 to 31 December 2025) of 295 CDI episodes at the Pius Brinzeu County Emergency Clinical Hospital, Timisoara. Thirty variables were extracted from narrative records with a lexical algorithm applied identically to both periods and validated against manual coding (n = 144). Mortality predictors were analysed by logistic regression with LASSO selection, bootstrap-corrected discrimination and calibration testing. Results: Mean age was 69.2 years. Hypertension (76.6%), heart failure (46.4%), neoplasm (39.3%), ascites or pleural effusion (61.7%), sepsis (56.9%) and pneumonia (44.4%) were common; vancomycin was documented in 83.7% of episodes and non-CDI antibiotics in 85.8%. In-hospital mortality during the index admission was 33.6%. Sepsis (adjusted OR 2.68), pneumonia (2.53) and ascites or pleural effusion (2.32) independently predicted death (optimism-corrected AUC 0.75), whereas individual chronic comorbidities lost significance after adjustment; mortality rose from 6.8% with none of the three acute markers to 61.1% with all three. Toxin index magnitude was unrelated to death. The second period had a higher comorbidity score (3.1 vs. 2.5, p = 0.002). Extraction reached at least 90% accuracy for 15 of 22 variables (median kappa 0.89). Conclusions: Acute organ involvement, much more than underlying comorbidity, determines in-hospital CDI mortality in this setting. Rule-based extraction supports reproducible CDI surveillance in partially digitised hospitals when low-agreement variables are verified by hand.