目的:探索一种利用医疗大数据算法筛选临床数据库中能够用于评估老年肺炎患者预后的核心指标。方法:基于首都医科大学附属北京朝阳医院医联体朝阳急诊病房临床数据库，应用大数据检索技术，以数据库中老年肺炎患者为研究对象，根据出院时预后将患者分为死亡组和存活组。收集患者的一般资料，包括性别、年龄、血气、实验室指标集合数据，使用计算机语言Python批量计算出影响老年肺炎患者死亡的关键指标，并采用Logistic回归分析实验室指标与患者预后的相关性；绘制受试者工作特征曲线（ROC曲线），分析本研究使用的筛选方法对患者预后的预测价值。结果:最终入选265例患者，死亡64例，存活201例。取每例患者入院首次检测指标的数据，最终从472项指标中筛选出23项差异有统计学意义的关键指标，包括：血常规指标7项、血气指标3项、肿瘤标志物指标3项、凝血功能相关指标4项、营养及器官功能相关指标6项。①肺炎死亡患者血气关键指标：51.6%（33例）的患者Cln -浓度为97～111 mmol/L，81.2%（52例）的患者血乳酸（Lac）为0.5～2.5 mmol/L，87.5%（56例）的患者Hn +浓度为0～46 mmol/L。②肺炎死亡患者血常规关键指标：46.9%（30例）的患者血红蛋白（Hb）为80～109 g/L，67.2%（43例）的患者血中嗜酸粒细胞比例（EOS%）为0.000～0.009，51.6%（33例）的患者血中淋巴细胞比例（LYM%）为0.00～0.09，50.0%（32例）的患者血中红细胞计数（RBC）为（3.0～3.9）×10n 12/L，54.7%（35例）的患者血中白细胞计数（WBC）为（0.0～9.9）×10n 9/L，48.4%（31例）的患者血中红细胞分布宽度变异系数（RDW-CV）为10.0%～14.9%，48.4%（31例）的患者血中C-反应蛋白（CRP）为0.0～49.9 mg/L。③肺炎死亡患者肿瘤标志物关键指标：76.6%（49例）的患者血游离前列腺特异抗原/总前列腺特异抗原（FPSA/TPSA）为阴性（比值为0），92.2%（59例）的患者细胞角蛋白19片段（CYFRA21-1）为0.0～11.0 μg/L，75.0%（48例）的患者糖类抗原125（CA125）为0～104 kU/L。④肺炎死亡患者凝血功能关键指标：68.8%（44例）的患者活化部分凝血活酶时间（APTT）为57～96 s，73.4%（47例）的患者D-二聚体为0～6 mg/L，93.8%（60例）的患者凝血酶时间（TT）为14～22 s，89.1%（57例）的患者二磷酸腺苷（ADP）的抑制率为0%～53%。⑤肺炎死亡患者营养及器官功能关键指标：92.2%（59例）的患者B型脑钠肽（BNP）为0，46.9%（30例）的患者前白蛋白（PA）为71～140 mg/L，90.6%（58例）的患者尿酸（UA）为21～41 μmol/L，75.0%（48例）的患者白蛋白（Alb）为10～20 g/L，93.5%（60例）患者白蛋白/球蛋白比值（A/G比值）为0～0.9，84.4%（54例）的患者乳酸脱氢酶（LDH）为0～6.68 μmol/L·sn -1·Ln -1。⑥ Logistic回归和ROC曲线分析：Logistic回归分析表明，PA和Lac是影响患者预后的因素，PA可使死亡风险降低0.9%，Lac可使死亡风险增加69.4%；实验室指标与患者死亡预测模型预测效果的ROC曲线下面积（AUC）＝0.80，说明本研究使用的筛选方法效果较好，通过本研究模型能较好地预测老年肺炎患者预后。n 结论:运用大数据技术可从急诊病房临床数据库中筛选出23项用于评估老年肺炎患者预后的核心指标，为临床评估老年肺炎患者预后提供了新的角度和方法。“,”Objective:To explore a medical big data algorithm to screen the core indicators in clinical database that can be used to evaluate the prognosis of elderly patients with pneumonia.Methods:Based on the clinical database of a Beijing Chaoyang Hospital Consortium Chaoyang Emergency Ward in Beijing Chaoyang Hospital, Capital Medical University, patients with pulmonary infection were selected through the big data retrieval technology. According to the prognosis at the time of discharge, they were divided into death group and survival group. The general data of patients were collected, including gender, age, blood gas and laboratory indices. A computer language called Python was used to make batch calculations of key indicators that affect mortality in elderly patients with pneumonia. Logistic regression analysis was used to analyze the relationship between laboratory indicators and patients' prognosis. Receiver operating characteristic curve (ROC curve) was drawn to analyze the predictive value of screening method for patients' prognosis.Results:A total of 265 patients were included in the study, 64 died and 201 survived. The data of the first detection indexes of each patient after admission were collected, and 23 key indicators with significant differences were selected from 472 indicators: blood routine indicators (n n = 7), blood gas indicators (n n = 3), tumor markers indicators (n n = 3),coagulation related indicators (n n = 4), and nutrition and organ function indicators (n n = 6). ① The key indicators of blood gas in patients died of pneumonia: Cln - was 97-111 mmol/L in 51.6% (33 cases) of patients, lactic acid (Lac) was 0.5-2.5 mmol/L in 81.2% (52 cases) of patients, and Hn + was 0-46 mmol/L in 87.5% (56 cases) of patients. ② The key indicators of blood routine of patients died of pneumonia: hemoglobin count (Hb) of 46.9% (30 cases) patients was 80-109 g/L, the eosinophils proportions (EOS%) in 67.2% (43 cases) patients was 0.000-0.009, the lymphocytes proportions (LYM%) in 51.6% (33 cases) patients was 0.00-0.09, the red blood cell count (RBC) in 50.0% (32 cases) patients was (3.0-3.9)×10n 12/L, the white blood cell count (WBC) in 54.7% (35 cases) patients was (0.0-9.9)×10n 9/L, and the red blood cell volume distribution width coefficientof variability (RDW-CV) in 48.4% (31 cases) patients was 10.0%-14.9%, serum C-reactive protein (CRP) was 0.0-49.9 mg/L in 48.4% (31 cases) patients. ③ The key indicators of tumor markers in patients died of pneumonia: 76.6% (49 cases) of patients had negative free prostate specific antigen/total prostate specific antigen (FPSA/TPSA, the ratio was 0), 92.2% (59 cases) had cytokeratin 19 fragment (CYFRA21-1) between 0.0-11.0 μg/L, and 75.0% (48 cases) had carbohydrate antigen 125 (CA125) between 0-104 kU/L.④ The key coagulation indexes of patients died of pneumonia: 68.8% (44 cases) of patients had activated partial thromboplastin time (APTT) of 57-96 s, 73.4% (47 cases) of patients had D-dimer of 0-6 mg/L, 93.8% (60 cases) of patients had thrombin time (TT) of 14-22 s, and 89.1% (57 cases) of patients had adenosine diphosphate (ADP) inhibition rate of 0%-53%. ⑤ Nutrition and organ function key indicatorsin patients died of pneumonia: 92.2% (59 cases) of brain natriuretic peptide (BNP) in patients with 0, 46.9% (30 cases) of patients had prealbumin (PA) of 71-140 mg/L, 90.6% (58 cases) of the patients with uric acid (UA) for 21-41 μmol/L, 75.0% (48 cases) of the patients with albumin (Alb) to 10-20 g/L, 93.5% (60 cases) of patients had albumin/globulin ratio (A/G ratio) of 0-0.9, 84.4% (54 cases) of the patients with lactate dehydrogenase (LDH) from 0-6.68 μmol/L·s n -1·Ln -1. ⑥ Logistic regression analysis and ROC curve analysis: Logistic regression analysis showed that PA and Lac were the prognostic factors. PA could reduce the risk of death by 0.9%, Lac could increase the risk of death by 69.4%; the area under ROC curve (AUC) between laboratory indicators and the prediction effect of death prediction model for patients' prognosis was 0.80, which showed that the classification effect was better, and this study model could better predict the prognosis of elderly patients with pneumonia.n Conclusion:By using big data technology, 23 core indicators for evaluating the prognosis of elderly patients with pneumonia can be screened from the clinical database of emergency ward, which provides a new perspective and method for clinical evaluation of the prognosis of elderly patients with pneumonia.