Wang Xi, Liu Letian, Wang Wen, Zhang Yang, Chen Hui, Wang Zhangli, Li Jianwei, Gao Yue, Huang Yanqun, Sun Lijun, Zhang Tong, Li Aixin
Center for Infectious Diseases, Beijing Youan Hospital, Capital Medical University, Beijing, China.
School of Biomedical Engineering, Capital Medical University, Beijing, China.
Front Cell Infect Microbiol. 2025 Feb 7;14:1485231. doi: 10.3389/fcimb.2024.1485231. eCollection 2024.
Pneumocystis pneumonia (PCP) is a common and serious complication of HIV/AIDS, with a higher prevalence in patients not receiving antiretroviral therapy. Due to the high mortality rate of PCP, accurate prediction of its case fatality rate is very important for clinical treatment. We aimed to develop a risk model for the near-term prognosis of people with HIV/AIDS and PCP and verify its effectiveness.
This single-center, retrospective observational study was conducted at Beijing Youan Hospital from January 2012 to October 2022. 972 AIDS patients with Pneumocystis pneumonia met our criteria were recruited. The patients were divided into death group and survival group according to clinical outcome during hospitalization. Data of the two groups were collected including general information and laboratory test results. 53 medical characteristics of the two groups were collected. Prediction variables were screened with Multivariate logistic regression analysis and Lasso regression model. We used ROC curve to identify the discrimination of training and testing data sets. The Shapley Additive exPlanation (SHAP) method was applied to explain the final model and the weights of features.
The overall mortality rate among hospitalized patients was 17.8%. We found that the best prediction effect can be obtained when ALB, PO, TBIL, LDH, CD4 T lymphocyte counts are incorporated into the PCP risk prediction model. The model had a perfect discrimination with AUC of 0.994 and 0.947 in training and validation cohorts. The prognosis risk grade was divided into three grades: low-risk group (0-25 points with mortality of 5.9%), moderate-risk group (25-50 points with mortality of 45.1%) and high-risk group (above 50 points with mortality of 80%). There is a statistically significant difference in mortality among these three grades (χ = 419.271, P<0.001).
We developed and validated a model of the prognostic risk level of PCP in patients of AIDS with the results of blood tests reviewed by patients at routine visits. The model is more convenient to use, allowing clinicians to obtain a determined probability value of PCP mortality with simple calculations within the first 72 hours of the patient's admission.
肺孢子菌肺炎（PCP）是艾滋病毒/艾滋病常见且严重的并发症，在未接受抗逆转录病毒治疗的患者中患病率较高。由于PCP死亡率高，准确预测其病死率对临床治疗非常重要。我们旨在建立一个艾滋病毒/艾滋病合并PCP患者近期预后的风险模型并验证其有效性。
本单中心回顾性观察研究于2012年1月至2022年10月在北京佑安医院进行。招募了972例符合标准的艾滋病合并肺孢子菌肺炎患者。根据住院期间的临床结局将患者分为死亡组和存活组。收集两组患者的一般信息和实验室检查结果等数据。收集两组的53项医学特征。采用多因素逻辑回归分析和Lasso回归模型筛选预测变量。使用ROC曲线确定训练和测试数据集的辨别力。应用Shapley值法解释最终模型和特征权重。
住院患者的总死亡率为17.8%。我们发现，将白蛋白（ALB）、氧分压（PO）、总胆红素（TBIL）、乳酸脱氢酶（LDH）、CD4 T淋巴细胞计数纳入PCP风险预测模型时可获得最佳预测效果。该模型在训练队列和验证队列中的辨别力良好，AUC分别为0.994和0.947。预后风险等级分为三级：低风险组（0 - 25分，死亡率为5.9%）、中风险组（25 - 50分，死亡率为45.1%）和高风险组（50分以上，死亡率为80%）。这三个等级的死亡率有统计学显著差异（χ = 419.271，P<0.001）。
我们建立并验证了一个基于艾滋病合并PCP患者常规复诊血液检查结果的PCP预后风险水平模型。该模型使用更便捷，使临床医生能够在患者入院的前72小时内通过简单计算获得PCP死亡率的确定概率值。
