Zhanjie Li,1,* Jian Li,2,* Chuanlong Zhu,3 Shengyuan Jiao1,2 

1Department of Infection Control, The First Affiliated Hospital of Nanjing Medical University, Nanjing, Jiangsu, 210029, People’s Republic of China; 2Department of Disease Prevention and Control, Air Force Hospital of Eastern Theater, Nanjing, People’s Republic of China; 3Department of Infections Disease, The First Affiliated Hospital of Nanjing Medical University, Nanjing, People’s Republic of China

*These authors contributed equally to this work

Correspondence: Shengyuan Jiao, Department of Infection Control, The First Affiliated Hospital of Nanjing Medical University, Nanjing, Jiangsu, 210029, People’s Republic of China, Tel +8618021136781, Email 17319884906@163.com Chuanlong Zhu, Department of Infections Disease, The First Affiliated Hospital of Nanjing Medical University, Nanjing, Jiangsu, 210009, People’s Republic of China, Tel +8617714316539, Email zhuchuanlong@jsph.org.cn

Background: To identify risk factors for acquiring HAIs in COVID-19 patients and establish visual prediction model.
Methods: Data was extracted from Xinglin Hospital Infection Monitoring System to analyze COVID-19 patients diagnosed between December 1, 2022, and March 1, 2023. Univariate and multivariate analyses were conducted to identify risk factors. Predictive signature was developed by selected variables from lasso, logistic regression, and their intersection and union. Models were compared using DeLong’s t-tests. Likelihood ratio (LR) and Youden’s index was used to evaluate the predictive performance. Nomogram was constructed using optimal variables ensemble, prediction accuracy was evaluated using AUC, DCA and calibration curve.
Results: Total of 739 patients met the criteria, of which 53 (7.2%) were HAIs. NSAIDs, surgery, fungi and MDRO detected, hormone drugs and LYMR were independent risk factors. Lasso model screened seven variables, and logistic model identified six risk factors. Union model performed the best with the maximum of the Youden’s index is 0.703, the sensitivity is 95.6%, the specificity is 74.7%, the LR is 3.778. The best AUC of union model is 0.953 (0.928– 0.978), and the accuracy is 87.5%. DCA indicated that the union model provided the best net benefits and calibration curve demonstrated good predictive agreement.
Conclusions: HAIs prediction in COVID-19 patients is feasible and beneficial to improve prognosis. Physicians can use this nomogram to identify high-risk COVID-19 populations for HAIs and tailor follow-up strategies.

Keywords: risk factors, healthcare-associated infection, COVID-19, nomogram, prediction model