Establishment and validation of a prediction model for liver injury in HIV/AIDS patients

Abstract

Objective To investigate the occurrence of liver injury in HIV/AIDS patients, and establish and verify its prediction model. Methods A total of 290 newly-diagnosed HIV/AIDS patients admitted to Mianyang 404 Hospital from January to December 2021 were subjected, and then assigned into a test group (n=203) and validation group (n=87) at a ratio of 7 ∶3. The factors that may affect the liver injury in HIV/AIDS patients were collected. The patients were divided into 2 groups according to the presence or absence of liver injury. Their gender, age, BMI and other factors were compared between the 2 groups. LASSO regression analysis was used to screen variables, then multivariate logistic regression analysis was employed to screen out independent influencing factors, and a nomogram model was constructed according to the logistic regression results. Results There were statistical differences between the patients with and without liver injury in history of alcohol drinking, CD4+T lymphocyte level, use of antifungal drugs, other types of antiretroviral drugs and sulfonamides, complications of tuberculosis, diabetes, cardiovascular disease, and nutritional status (P < 0.05). LASSO regression model screened out 7 potential influencing factors, including drinking history, CD4+T lymphocyte level, antifungal drug use, comorbid tuberculosis infection, diabetes, cardiovascular disease, and nutritional status. Multivariate logistic regression analysis showed that drinking history, CD4+T lymphocyte level, use of antifungal drugs, comorbid tuberculosis, diabetes, cardiovascular disease, and nutritional status were independent influencing factors of liver injury in HIV/AIDS patients (P < 0.05). Based on the above influencing factors, a nomogram model for predicting the risk of liver injury in HIV/AIDS patients was constructed. The results of ROC curve analysis showed that the area under the subject curve (AUC) of the model group was 0.969 (95%CI: 0.904~0.988), and the AUC of the validation group was 0.971 (95%CI: 0.905~0.992) for predicting liver injury in HIV/AIDS patients. After Bootstrapping was conducted for repeated sampling 1 000 times, and the verification group was used for verification. The results of the calibration curve indicated that the prediction curve of the model group and the verification group basically fitted the standard curve. The results of the H-L goodness of fit test suggested that the nomogram model predicted the probability of liver injury in HIV/AIDS patients was not significantly different from the actual probability (Chi-square=2.088, P=0.148). The analysis results of decision curve of model group showed that when the probability threshold of predicting liver injury of HIV/AIDS patients by the nomograph model was 0.15~0.95, the net benefit rate of patients was greater than 0. While, the analysis results of the decision curve of the validation group showed that when the probability threshold of predicting the liver injury of HIV/AIDS patients by the nomograph model was 0.10~0.95, the net benefit rate of the patients was greater than 0. Conclusion History of drinking, CD4+T lymphocyte level, use of antifungal drugs, comorbid tuberculosis infection, diabetes, cardiovascular disease, and nutritional status are independent influencing factors of liver injury in HIV/AIDS patients. Our nomogram model constructed based on the above factors has high accuracy and discrimination for HIV/AIDS liver injury risk prediction.

Keywords

• human immunodeficiency virus
• aids
• liver injury
• multivariate analysis
• nomogram model