Abstract: Objective Based on big data, the risk prediction model of multidrug resistant organisms infection was constructed and its application value was evaluated. Method From January 2018 to December 2022, 405 patients in the Fifth Affiliated Hospital of Xinjiang Medical University were divided into non-MDRO group（n=324） and MDRO group（n=324） according to whether multidrug-resistant organisms (MDRO) were infected, and the correlation between each index and the risk of MDRO was compared and analyzed. Build a big data risk prediction model, analyzing the importance of each index and verify its accuracy. Result The proportion of patients with diabete and primary lung infection in MDRO group, the time of mechanical ventilation, the use of broad-spectrum antibiotics and the level of procalcitonin were significantly higher than those in non-MDRO group, while the levels of hemoglobin and albumin were significantly lower than those in non-MDRO group (all P<0.05). Correlation analysis showed that the risk of MDRO was highly correlated with factors such as diabete mellitus and primary lung infection, and there was a high correlation between diabete mellitus and primary lung infection and combined use of antibiotics. The big data model shows that factors such as antibiotic using time and swallowing dysfunction are more important, while hemoglobin and albumin are less important. The AUC of the big data model in predicting the risk of MDRO is significantly higher than that of the Logistic regression model (Z=2.415, P=0.016), and there is no statistical difference in the prediction accuracy of the training set between the two prediction models (P>0.05). However, the prediction accuracy, sensitivity and specificity of test set are significantly higher than those of Logistic regression model (χ²=9.062, 5.385, 4.267;All P<0.05). Conclusion Supplementary factors such as diabetes mellitus, primary lung infection and combined use of antibiotics have certain correlation with the risk of MDRO, and the big data model based on MDRO risk factor indicators has high predictive value for the risk of MDRO.

Key words: Multi-drug resistant organisms, Risk factors, Machine learning, Screening, Prediction model