目的 构建首发脑出血患者并发卒中相关性肺炎的风险预测模型并验证模型的预测性能。方法 回顾性分析2012年1月—2022年12月广州市第一人民医院治的419例首发脑出血患者的临床资料，按照7︰3比例随机化分为训练列（293例）和验证队列（126例）。统计基于开发队列数据，采用Logistic回归模型分析首发脑出血患者并发卒中相关性肺炎的影响因素，并构建风险预测模型。基于开发队列和验证队列数据，采用校准曲线、受试者操作特征（ROC）曲线下面积和决策曲线分析模型的预测性能。结果 419例首发脑出血患者中有113例发生卒中相关性肺炎，发生率为26.97%。美国国立卫生研究院卒中量表（NIHSS）评分、吞咽困难、初始血肿体积、中性粒细胞百分比与白蛋白比值（NPAR）、中性粒细胞计数与淋巴细胞计数比值（NLR）、手术治疗、气管插管、留置胃管均是首发脑出血患者并发卒中相关性肺炎的影响因素（P＜0.05）。基于上述影响因素构建了首发脑出血患者并发卒中相关性肺炎的风险预警模型，校准曲线显示模型在开发队列和验证队列中预测卒中相关性肺炎发生率均与实际发生率相近；ROC曲线显示此模型在开发队列、验证队列中预测的曲线下面积分别为0.906（95%CI：0.867~0.937）、0.884（95%CI：0.815~0.934）；决策曲线分析显示当开发队列阈概率在3%~80%内、验证队列阈概率在2%~76%内使用此模型干预比全/无干预更有临床价值。结论 基于NIHSS评分、吞咽困难、初始血肿体积、NPAR、NLR、手术治疗、气管插管、留置胃管构建的首发脑出血患者并发卒中相关性肺炎的风险预测模型具有良好预测性能和临床应用价值。

      Objective To construct a risk prediction model for stroke associated pneumonia in patients with initial cerebral hemorrhage（ICH）and validate the predictive performance of the model．Methods A retrospective analysis was conducted on the clinical data of 419 patients with ICH admitted to our hospital from January 2012 to December 2022．They were randomly divided into a development cohort（293 cases）and a validation cohort（126 cases）according to a 7∶3 ratio．The Logistic regression model was used to analyze the influencing factors of stroke related pneumonia in patients with ICH based on the development cohort data，and a risk prediction model was constructed．Based on the development cohort data and validation cohort data，the predictive performance of the model was analyzed using calibration curves，receiver operating characteristic（ROC）curve，and decision curve analysis．Results Among 419 patients，113 developed stroke associated pneumonia，with a rate of 26.97%．The National Institutes of Health Stroke Scale（NIHSS）score，swallowing difficulties，initial hematoma volume，neutrophil percentage to albumin ratio（NPAR），neutrophil count to lymphocyte count ratio（NLR），surgical treatment，endotracheal intubation，and indwelling gastric tube were all independent influencing factors for stroke associated pneumonia in patients with ICH（P＜0.05）．Based on the above influencing factors，a risk prediction model for stroke associated pneumonia in patients with ICH was constructed．The calibration curve showed that the predicted incidence of stroke associated pneumonia by the model in both the development and validation cohorts was close to the actual incidence．The ROC curve showed that the predicted area under the curve for this model in the development cohort and validation cohort was 0.906（95%CI：0.867-0.937）and 0.884（95%CI：0.815-0.934），respectively．The decision curve analysis showed that when the threshold probability of the development cohort was between 3%-80%，and the threshold probability of the validation cohort was between 2%-76%，the intervention using this model was more clinically valuable than all/no intervention．

Conclusions The risk prediction model for stroke associated pneumonia in patients with ICH based on NIHSS score，swallowing difficulties，initial hematoma volume，NPAR，NLR，surgical treatment，tracheal intubation，and indwelling gastric tube has good predictive performance and clinical application value．

       目的  探讨结肠镜下息肉切除术后复发的危险因素，并基于机器学习算法构建复发风险预警模型，为防治对策提供依据。方法  回顾性收集2018年9月—2023年9月六安市人民医院1 058例初次行无痛结肠镜下息肉切除术患者的临床资料，使用单因素和多因素Logistic回归分析筛选复发危险因素。采用7∶3随机抽样分为训练集和验证集，分别通过决策树、贝叶斯及Logistic回归算法构建预测模型，并以受试者工作特征曲线（ROC）曲线下面积（AUC）、灵敏度、特异度等指标来评估模型效能。结果  单因素分析显示，性别、吸烟、代谢综合征、息肉数量、息肉位置、山田分型、组织病理学类型、切除方式、复查时间、肠息肉直径、手术时间是复发的危险因素（P＜0.05）。多因素分析显示，性别、代谢综合征、息肉数量、息肉直径、肠息肉位置、山田分型、组织学病理类型、切除方式、手术时间均是结肠息肉内镜下切除术后复发的危险因素。模型评估显示，决策树算法、贝叶斯算法、Logistic回归算法的ROC曲线下面积（AUC）分别为0.849、0.818、0.811；灵敏度分别为85.14%、81.62%、79.43%；特异度分别为81.69%、79.45%、74.18%；约登指数分别为0.534、0.551、0.573；95%CI分别为0.810~0.876、0.794~0.860、0.782~0.850；决策树算法模型效能最佳，Logistic回归算法的性能最差。结论  性别、代谢综合征、肠息肉特征（数量、直径、位置等）是术后复发的关键危险因素。决策树模型在风险预测中表现最优，可为临床制定个体化随访策略提供参考。

       Objective  To explore the  risk factors for  recurrence after painless colonoscopic polypectomy and construct a recurrence risk warning model based on machine learning algorithms to provide evidence for prevention and treatment strategies．Methods  A retrospective analysis was conducted on clinical data from 1 058 patients who underwent their first painless colonoscopy-guided polypectomy at our hospital between September 2018 and September 2023．Univariate and multivariate Logistic  regression analyses were performed to identify recurrence risk factors．The dataset was randomly divided into training and validation sets using a 7∶3 ratio．Prediction models were constructed using decision tree，Bayesian，and Logistic regression algorithms，and their performance was evaluated using metrics such as the area under the receiver operating characteristic curve（AUC），sensitivity，specificity，and others．Results  Univariate analysis revealed that gender，smoking，metabolic syndrome，number of polyps，polyp location，Yamada classification，histopathological type，resection method，follow-up time，polyp diameter，and operation duration were risk factors for recurrence（P＜0.05）．Multivariate analysis identified gender，metabolic syndrome，number of polyps，polyp diameter，polyp location，Yamada classification，histopathological type，resection method，and operation duration as independent risk factors for recurrence after endoscopic polypectomy．Model evaluation showed AUC values of 0.849，0.818，and 0.811 for the decision tree，Bayesian，and Logistic regression algorithms，respectively．Sensitivity values were 85.14%，81.62%，and 79.43%；specificity values were 81.69%，79.45%，and 74.18%；Youden’s indices were 0.534，0.551，and 0.573；and 95% confidence intervals（CIs）were 0.810–0.876，0.794–0.860，and 0.782–0.850，respectively．The  decision tree algorithm demonstrated the best predictive performance，while the Logistic regression algorithm performed the least favorably．Conclusions  Gender，metabolic syndrome，and polyp characteristics（number，diameter，location，etc．）are key  risk factors for recurrence after polypectomy．The decision tree algorithm exhibited optimal predictive efficacy，offering valuable insights for developing individualized follow-up strategies in clinical practice．

      目的   基于决策树构建老年患者吞咽障碍预警模型。方法  采用便利取样法对宁夏银川市宁夏回族自治区人民医院老年科住院的200例老年患者进行调查。结果  200例老年患者中，吞咽障碍发生率为40.5%。依据是否发生吞咽障碍将其患者分为两组，两组患者在性别、年龄、文化程度、职业、医保类型、家庭年收入、日常生活能力、衰弱、抑郁、营养、体质指数（BMI）比较（χ ² 值分别为13.321、4.064、31.944、36.695、18.230、19.681、52.509、10.253、20.456、9.070、9.483），差异均有统计学意义（均P＜0.05）。决策树模型筛选出老年患者吞咽障碍的影响因素主要有自理能力、职业、文化程度和抑郁，决策树模型受试者工作特征曲线下面积为0.862，灵敏度为79.8%，特异度为79.0%，P＜0.001。结论  基于自理能力、职业、文化程度和抑郁构建的决策树模型，能有效预测老年患者吞咽障碍风险。

       Objective  To construct a swallowing disorder warning model for elderly patients based on decision tree．Methods  Convenience  sampling was  used to  study  200  elderly  patients  admitted to the  geriatric  department  of  a tertiary comprehensive hospital in Yinchuan，Ningxia．Results  Among 200 elderly patients，the incidence of swallowing disorders was 40.5%．The two groups of patients were compared in terms of gender，age，education level，occupation，medical insurance type，annual family income，daily living ability，frailty，depression，nutrition，and BMI（χ ²  values were 13.321，4.064，31.944，36.695，18.230，19.681，52.509，10.253，20.456，9.070，9.483，respectively），and the  differences were  statistically significant（all P＜0.05）．The decision tree model identified the main influencing factors of swallowing disorders in elderly patients as self-care ability，occupation，education level，and depression．The Receiver Operating Characteristic curve of the decision tree model had an area under the curve of 0.862，sensitivity of 79.8%，and specificity of 79.0%，P＜0.001．Conclusions  A decision tree model based on self-care ability，occupation，education level，and depression can effectively predict the  risk of swallowing disorders in elderly patients．

目的 构建并验证机械通气患儿肠内营养支持发生误吸的风险预测模型。方法 回顾性分析中山市博爱医院2021年3月—2023年3月儿童重症监护病房330例行机械通气并进行肠内营养的患儿临床资料,通过二元Logistic回归,获取机械通气患儿肠内营养支持发生误吸的预测因素,绘制列线图模型,并进行模型评价及验证。结果 330例机械通气患儿中,104例患儿发生误吸、226例未发生误吸。两组患儿在意识状态、机械通气方式、管饲量、胃残留量、胃管置入深度、促胃动力药、镇静剂等方面对比差异具有统计学意义（P<0.05）。二元Logistic结果显示,胃残留量、机械通气方式、管饲量、意识状态、胃管置入深度、促胃动力药、镇静剂是机械通气患儿肠内营养支持发生误吸的影响因素（P<0.05）。建模组AUC为0.810（95%CI：0.760~0.860）,Hosmer-Lemesh结果显示,χ²=3.245,P=0.846;外部验证组AUC为0.873（95%CI：0.831~0.914）,Hosmer-Lemesh结果显示,χ²=3.567,P=0.875。建模组和训练组DCA曲线大部分落于Y=0上方。建模组与外部验证组校准曲线均与参考曲线高度贴合,预测概率与实际概率接近,校准度良好。结论 基于胃残留量、机械通气方式、管饲量、意识状态、胃管置入深度、促胃动力药、镇静剂等7项指标构建的风险预测模型具有一定的临床价值,可作为医护人员识别肠内营养机械通气误吸高危患儿的工具。

Objective To establish and verify the risk prediction model of enteral nutritional aspiration in children with mechanical ventilation．Methods The clinical data of 330 children who underwent mechanical ventilation and enteral nutrition in the PICU of Zhongshan Boai Hospital from March 2021 to March 2023 were retrospectively analyzed．The independent predictive factors of enteral nutrition support aspiration in children with mechanical ventilation were obtained by binary Logistic regression,and the nomographic model was drawn,and the model was evaluated and verified. Results Among 330 children with mechanical ventilation,104 had aspiration and 226 did not．There were statistically significant differences between the two groups in consciousness state,mechanical ventilation mode,tube feeding amount,gastric residual amount,gastric tube insertion depth,gastric motivity drugs,sedatives,etc.（P<0.05）．Binary Logistic results showed that gastric residual amount,mechanical ventilation mode,tube feeding amount,state of consciousness,depth of gastric tube insertion,gastric motonics and sedatives were the influential factors of enteral nutritional aspiration in children with mechanical ventilation（P<0.05）．The AUC of the modeling group was 0.810（95%CI：0.760-0.860）,and the Hosmer-Lemesh result showed that χ²=3.245,P=0.846．The AUC of the external verification group was 0.873（95%CI：0.831-0.914）,and the Hosmer-Lemesh result showed that χ²=3.567,P=0.875．The DCA curves of modeling group and training group mostly were above Y=0．The calibration curves of the modeling group and the external verification group are highly fit to the reference curves,and the prediction probability was close to the actual probability,and the calibration degree was good．Conclusion sThe risk prediction model based on 7 indexes,including stomach residual amount,mechanical ventilation mode,tube feeding amount,state of consciousness,depth of gastric tube insertion,gastric motivity drug and sedative,with certain clinical value,and can be used as a tool for medical staff to identify children at high risk of enteral nutritional mechanical aspiration．

目的 基于随机森林方法构建甲状腺功能减退（简称甲减）患病风险预测模型。方法 从MIMIC-IV数据库纳入5 735名甲减患者为病例组,4 803名非甲减患者为对照组,基于随机森林模型进行建模。同时利用逻辑回归、贝叶斯正则化神经网络、XGBoost作为比较模型。最后用准确率、F1分数、精确率、召回率、特异性以及AUC值评价四个机器学习模型性能。结果 随机森林模型准确率为0.85,F1分数为0.84,精确率为0.84,召回率为0.84,特异性为0.86,AUC值为0.91。在该模型中,促甲状腺激素、年龄、绝对淋巴细胞计数、血液中红细胞数、中性白细胞、性别、碱性磷酸酶、丙氨酸氨基转移酶、嗜酸性粒细胞绝对计数、尿素氮为甲减患者诊断重要性排前10的指标。结论 采用随机森林方法构建的甲减患病预测模型为甲减的早期诊断有潜在应用价值。

Objective To construct a risk prediction model for hypothyroidism based on the random forest model．Methods A total of 5 735 hypothyroidism patients were included from the MIMIC-IV database as the case group, and 4 803 non-hypothyroidism patients were included as the control group．Random forest models were constructed for both groups, and logistic regression, Bayesian regularized neural network, and XGBoost were used as comparative models．The performance of the four machine learning models was evaluated using accuracy, F1 score, precision, recall, specificity, and AUC value．Results The random forest model had an accuracy of 0.85, an F1 score of 0.84, a precision of 0.84, a recall of 0.84, a specificity of 0.86, and an AUC value of 0.91．In this model, thyroid-stimulating hormone, age, absolute lymphocyte count, red blood cell count in blood, neutrophil, gender, alkaline phosphatase, aspartate aminotransferase, absolute eosinophil count, and blood urea nitrogen were the top 10 indicators for diagnosing hypothyroidism patients．Conclusions The hypothyroidism disease prediction model constructed using the random forest method has potential application value for the early diagnosis of hypothyroidism．

目的 探讨老年吸入性肺炎的危险因素,建立风险预测模型,以期降低老年吸入性肺炎的发病率。方法 选取2017年8月28日—2020年 10月30日广州市第一人民医院老年病科住院治疗的老年肺炎患者205例,按照是否发生吸入性肺炎分为吸入性肺炎组和非吸入性肺炎组,对比2组患者的各项指标,分析老年吸入性肺炎的危险因素,建立风险预测模型,采用ROC曲线对模型进行预测效果检验。结果 多因素Logistic回归分析结果显示,脑梗塞、帕金森、留置胃管、长期卧床为老年吸入性肺炎的危险因素(P<0.05)。模型公式为Logit(P)=-2.952+1.221X₂+2.417X₃+2.388X₈+1.683X₁₀。该模型ROC曲线下面积为0.894。结论 本研究中的模型预测效果良好,可为医护人员预测老年患者发生吸入性肺炎的概率,及时采取相应的预见性护理及干预性治疗。

Objective To explore the risk factors of aspiration pneumonia in the elderly and establish the risk prediction model, in order to reduce the incidence of aspiration pneumonia in the elderly. Methods A total of 205 elderly patients with pneumonia who were hospitalized in the department of geriatrics, Guangzhou First People's Hospital from August 28, 2017 to October 30, 2020, were divided into aspiration pneumonia group and non-aspiration pneumonia group according to whether aspiration pneumonia occurred. The indicators of the two groups of patients were compared, the risk factors of aspiration pneumonia in the elderly were analyzed, the risk prediction model was established, and the prediction effect of the model was tested by receiver operating characteristic curve. Results Multivariate Logistic regression analysis showed that cerebral infarction, Parkinson's disease, indwelling nasogastric tube, and being bedridden were risk factors for aspiration pneumonia in elderly patients (P<0.05). The model formula was Logit (P)=-2.952+1.221X₂+2.417X₃+2.388X₈+1.683X₁₀. The area under receiver operating characteristic curve of this model was 0.894. Conclusion The prediction effect of the model in this study was good, which could predict the probability of aspiration pneumonia in elderly patients for medical staff, and to timely take the corresponding predictive care and interventional treatment.