Abstract: OBJECTIVE To explore the epidemiological characteristics and prevalence trend of extended-spectrum β-lactamases (ESBLs)-producing Klebsiella pneumoniae in intensive care unit (ICU) of a three-A hospital in Anhui Province from 2020 to 2024 so as to provide prospective early warning bases for prevention and control of health care-associated infections (HAIs) in ICU. METHODS The data regarding to the K. pneumoniae strains isolated from submitted specimens of ICU patients were collected from Fuyang People's Hospital from Jan. 2020 to Dec. 2024. The monthly isolation rates of ESBLs-producing strains were statistically analyzed, the time series were established, and the data were set as the training set to establish the Seasonal Autoregressive Integrated Moving Average Model (SARIMA). The parameters were optimized by means of Bayesian information criterion (BIC) and residual white noise test, and the actual data from Jan. 2025 to Dec. 2025 were set as the validation set to assess the predictive efficiency of the model. RESULTS Totally 1 930 strains of K. pneumoniae were isolated from the ICU patients between 2020 and 2024, of which 197 were ESBLs-producing strains, with the total isolation rate 10.21%; the isolation rate was highest (48.65%) in pediatric ICU. The annual isolation rate of the strains showed a trend of rising at first and declining afterwards (χ²_trend=34.677,P＜0.001). The sputum (75.13%) was the major source of specimens. The optimal prediction model was SARIMA(1,1,1)(0,1,1)12, the standardized BIC value was 3.380, and the residual sequences passed the Ljung-Box Q test(Q=9.179,P=0.868). The actual isolation rate of ESBLs-producing K. pneumoniae from Jan. 2025 to Dec. 2025 fell within the 95%CI of the predicted value of the model, with an average relative error -3.75%, indicating favorable fitting effect. CONCLUSION SARIMA model can effectively capture the seasonal and long-term trend of the prevalence of ESBLs-producing K. pneumoniae in ICU and achieve short-term prediction, providing quantitative tools for early identification and targeted intervention to the peak of infections and pushing forward the transition of infection control from passive monitoring to active early warning.