To analyze risk factors of bloodstream infection in patients treated with extracorporeal membrane oxygenation (ECMO), and to establish a risk prediction model.

The clinical data of 108 patients receiving ECMO adjuvant therapy in the ICU of Affiliated Hangzhou First People's Hospital, Zhejiang University School of Medicine from January 2016 to December 2021 were retrospectively analyzed. The incidence of ECMO-related bloodstream infection and distribution of pathogens were explored, univariate analysis and multivariate Logistic regression analysis were performed for risk factors of ECMO-related bloodstream infection, and then a risk prediction model was established. The fitting degree of the model was evaluated by the Hosmer-Lemeshow test, and its predictive value was analyzed by means of the receiver operating characteristic (ROC) curve.

Of 108 patients receiving ECMO, 31 patients experienced ECMO-related bloodstream infection, with an infection rate of 28.70%. The ECMO treatment was applied for 1 076 d, and the ECMO-related bloodstream infection rate was 28.81 per thousand ECMO days. In these 31 patients, a total of 43 pathogens were isolated, including 15 strains of gram-positive bacteria (34.88%), 24 strains of gram-negative bacteria (55.81%) and four strains of fungus (9.30%). The multivariate Logistic regression analysis showed that the ECMO running time [odds ratio (OR) = 1.154, 95% confidence interval (CI) (1.013, 1.314), P = 0.031], number of intravascular catheters ≥ 4 [OR = 8.607, 95%CI (2.176, 34.046), P = 0.002], intra-aortic balloon pump application [OR = 4.467, 95%CI (1.111, 17.957), P = 0.035], continuous renal replacement therapy treatment [OR = 4.963, 95%CI (1.241, 19.843), P = 0.023] and procalcitonin [OR = 1.052, 95%CI (1.004, 1.103), P = 0.035] were independent risk factors for ECMO-related bloodstream infection. A risk prediction model was established based on these Logistic regression results, with a high goodness-of-fit according to the Hosmer-Lemeshow test (χ² = 3.672, P = 0.885). ROC curve analysis showed that the area under the curve for the risk prediction model of ECMO-related bloodstream infection was 0.910 [95%CI (0.856, 0.963), P < 0.001].

Bloodstream infection is one of common critical complications in ECMO treatment. The gram-negative bacteria are dominant among the pathogens causing infection. The multivariate Logistic regression model can achieve favourable effect on the prediction of ECMO-related bloodstream infection. In clinical practice, the monitoring of risk factors of ECMO-related bloodstream infection should be strengthened and corresponding preventive control measures should be formulated to reduce its incidence and improve patients' prognosis.