基于随机森林算法的重症肺炎进展为急性呼吸窘迫综合征风险预测模型的构建与评价
该回顾性观察研究纳入2020年1月至2024年5月遵义医科大学第二附属医院ICU收治的181例重症肺炎患者,其中73例进展为ARDS,发生率为40.3%。比较ARDS组与非ARDS组临床指标,采用单因素Logistic回归筛选危险因素并构建随机森林预测模型,经1000次重采样验证模型性能。结果显示pH、最低收缩压、白蛋白、APACHE评分、LIPS评分为进展为ARDS的危险因素,模型AUC为0.909(95%CI 0.870-0.943),敏感度0.823,特异度0.869。该模型可有效预测重症肺炎患者是否进展为ARDS,为疾病评估和早期干预提供依据。
要点
- 重症肺炎ICU患者ARDS发生率为40.3%(73/181),进展组最低收缩压、最低舒张压、最低氧合指数、pH值、白蛋白水平显著低于非进展组。
- 单因素Logistic回归显示pH(OR=0.04,95%CI 0.00-0.96,P=0.047,cut-off 7.34)、最低收缩压(OR=0.98,95%CI 0.97-1.00,P=0.044,cut-off 90 mmHg)、白蛋白(OR=0.94,95%CI 0.89-0.99,P=0.032,cut-off 28.05 g/L)为保护性因素。
- APACHE评分(OR=1.08,95%CI 1.02-1.14,P=0.008,cut-off 23)和LIPS评分(OR=1.37,95%CI 1.09-1.72,P=0.007,cut-off 5)为危险因素。
- 随机森林模型指标重要性从高到低为pH、最低收缩压、白蛋白、APACHE评分、LIPS(基尼指数分别为31.08、30.74、29.35、28.01、24.92)。
- 模型经1000次bootstrap重采样验证,AUC为0.909(95%CI 0.870-0.943),敏感度0.823(95%CI 0.699-0.932),特异度0.869(95%CI 0.741-0.963)。
适用人群与场景
- 2020年1月至2024年5月遵义医科大学第二附属医院ICU收治的重症肺炎患者。
- 根据是否进展为ARDS分为ARDS组和非ARDS组。
诊断标准与关键阈值
- ARDS进展预测阈值:pH<7.34、最低收缩压<90 mmHg、白蛋白<28.05 g/L、APACHE评分>23、LIPS评分>5。
研究设计与关键结果
- 回顾性观察研究,纳入181例重症肺炎患者,其中73例进展为ARDS,发生率为40.3%。
- 与非ARDS组相比,ARDS组最低收缩压、最低舒张压、最低氧合指数、pH值、白蛋白水平显著降低,最高APTT、APACHE评分、LIPS评分显著升高(均P<0.05)。
- 单因素Logistic回归:pH(OR=0.04,95%CI 0.00-0.96,P=0.047,cut-off 7.34)、最低收缩压(OR=0.98,95%CI 0.97-1.00,P=0.044,cut-off 90 mmHg)、白蛋白(OR=0.94,95%CI 0.89-0.99,P=0.032,cut-off 28.05 g/L)、APACHE评分(OR=1.08,95%CI 1.02-1.14,P=0.008,cut-off 23)、LIPS评分(OR=1.37,95%CI 1.09-1.72,P=0.007,cut-off 5)。
- 随机森林模型指标重要性从高到低:pH、最低收缩压、白蛋白、APACHE评分、LIPS(基尼指数分别为31.08、30.74、29.35、28.01、24.92)。
- 模型经1000次bootstrap重采样验证:AUC=0.909(95%CI 0.870-0.943),敏感度=0.823(95%CI 0.699-0.932),特异度=0.869(95%CI 0.741-0.963)。
证据类型、等级与局限
- 证据类型:回顾性观察研究。
- 局限:原文未报告局限性。
对床旁实践的改变
- 作者认为,基于pH、最低收缩压、白蛋白、APACHE评分、LIPS评分构建的随机森林模型可有效预测重症肺炎患者是否进展为ARDS。
- 作者认为,该模型可为重症肺炎的疾病评估、早期干预和预后改善提供依据。
提炼说明
- 本条依据论文摘要整理;摘要未涉及「处置流程与药物剂量」「红旗征与不能漏的情况」「指南或共识出处」,故未列出这些栏目;全文请点「阅读原文」。
英文原文(摘要)· 点击展开
OBJECTIVE: To identify the risk factors for the progression of severe pneumonia to acute respiratory distress syndrome (ARDS) and to construct a prediction model based on the random forest algorithm, providing a basis for disease assessment, early intervention, and prognosis improvement in severe pneumonia.
METHODS: A retrospective observational study was conducted. Patients with severe pneumonia admitted to the intensive care unit (ICU) of the Second Affiliated Hospital of Zunyi Medical University from January 2020 to May 2024 were enrolled. Data including general patient information, vital signs, blood test results, disease assessment indicators within 24 hours of ICU admission, and outcome measures were collected. Patients were divided into ARDS group and non-ARDS group according to whether they progressed to ARDS. Univariate Logistic regression analysis was used to screen the risk factors for the progression of severe pneumonia to ARDS, and a random forest based prediction model was constructed. Model performance and stability were validated using 1 000 resampling iterations.
RESULTS: A total of 181 severe pneumonia patients were included, of whom 73 progressed to ARDS, with an incidence rate of 40.3%. Compared to the non-ARDS group, the ARDS group had significantly lower lowest systolic blood pressure, lowest diastolic blood pressure, lowest oxygenation index, pH value, and albumin level, while showing significantly higher maximum activated partial thromboplastin time (APTT), Acute Physiology and Chronic Health Evaluation (APACHE), and Lung Injury Prediction Score (LIPS; all P<0.05). There were no statistically significant differences in other baseline data comparisons (all P>0.05). Univariate Logistic regression analysis showed that pH, lowest systolic blood pressure, albumin, APACHE score, and LIPS score were risk factors for the progression to ARDS in severe pneumonia patients [pH: odds ratio (OR)=0.04, 95% confidence interval (95%CI) was 0.00-0.96, P=0.047, cut-off was 7.34; lowest systolic blood pressure: OR=0.98, 95%CI was 0.97-1.00, P=0.044, cut-off was 90 mmHg (1 mmHg=0.133 kPa); albumin: OR=0.94, 95%CI was 0.89-0.99, P=0.032, cut-off was 28.05 g/L; APACHE score: OR=1.08, 95%CI was 1.02-1.14, P=0.008, cut-off was 23; LIPS: OR=1.37, 95%CI was 1.09-1.72, P=0.007, cut-off was 5]. A random forest model constructed with these risk factors ranked the importance of the indicators from high to low as follows: pH, lowest systolic blood pressure, albumin, APACHE score, and LIPS (with Gini Index of 31.08, 30.74, 29.35, 28.01, and 24.92, respectively). Validation with 1 000 bootstrap resamplings showed that the model had an area under the receiver operator characteristic curve (AUC) of 0.909 (95%CI was 0.870-0.943), a sensitivity of 0.823 (95%CI was 0.699-0.932), and a specificity of 0.869 (95%CI was 0.741-0.963).
CONCLUSIONS: pH<7.34, lowest systolic blood pressure<90 mmHg, albumin<28.05 g/L, APACHE>23, and LIPS>5 are risk factors for the progression of severe pneumonia to ARDS. The model constructed based on these factors using the random forest algorithm can effectively predict whether severe pneumonia patients will progress to ARDS.
参考来源
- Jia G, Wang T, Song E, et al. [Construction and evaluation of a risk prediction model for the progression of severe pneumonia to acute respiratory distress syndrome based on the random forest algorithm]. Zhonghua Wei Zhong Bing Ji Jiu Yi Xue. 2026. PMID: 41876238. doi:10.3760/cma.j.cn121430-20250102-00004
供执业医师抢救与诊疗参考,不替代临床判断;剂量请结合体重、肝肾功能和本院药品规格复核,以指南原文为准。