Observed and Forecasted Antibiotic Consumption by ATC/DDD Metrics: A Real-World Modeling Study with Implications for...

Background/Objectives: Monitoring antibiotic consumption is a central component of antimicrobial stewardship because it provides a standardized way to identify changing prescribing patterns, detect potentially excessive use, and support targeted interventions. The WHO ATC/DDD framework is widely used for hospital-based utilization studies and enables comparison across wards and time periods. Methods: This single-center observational study used antibiotic data obtained from the hospital pharmacy to compare predicted and observed consumption for 2024 and 2025. Antibiotic use was expressed as DDD per 100 bed-days and analyzed for penicillins, cephalosporins, fluoroquinolones, tetracyclines, nitroimidazole derivatives, lincosamides, and aminoglycosides. The aim of this study was to quantify antibiotic consumption using ATC/DDD methodology, evaluate how closely modeled predictions matched observed values in 2024 and 2025, and identify the antibiotic classes and wards most relevant for stewardship action. Results: Cephalosporins showed the closest agreement between forecasted and observed consumption, with absolute percentage errors of 2.16% in 2024 and 4.61% in 2025 and a two-year mean absolute percentage error of 3.39%. Consumption of fluoroquinolones exceeded predictions, particularly in gynecology, while use of nitroimidazole derivatives was lower than expected, especially in 2025. Use of tetracyclines was absent in both years, and aminoglycosides and lincosamides contributed only minimally to total consumption. Ward-level analysis showed that intensive care accounted for the highest cephalosporin exposure, whereas the largest deviations from forecast were observed in gynecology and high-risk pregnancy wards. Conclusions: Predictive modeling captured antibiotic use reasonably well for some antibiotic classes but showed lower accuracy for others. Forecast accuracy varied across antibiotic groups, indicating that the model performed differently depending on the observed consumption pattern. However, the underlying reasons for these differences cannot be determined from the present aggregated consumption analysis.