摘要目的:建立预测急性胰腺炎(AP)严重程度的可视化模型并进行内部验证。方法:纳入昆明医科大学第一附属医院2017年9月1日至2020年8月31日确诊的600例AP患者，根据2012年亚特兰大分类标准将600例患者分为重症急性胰腺炎(SAP)组(128例)和非重症急性胰腺炎(NSAP)组(472例)，比较两组患者的一般临床资料(年龄、性别、体重指数等)、实验室检查指标(空腹血糖、尿素氮、血肌酐等)，合并腹水或胸腔积液情况，以及急性生理学和慢性健康状况评价Ⅱ(APACHEⅡ)和急性胰腺炎严重程度床旁指数(BISAP)评分。采用最小绝对收缩与选择算子(LASSO)回归模型对SAP潜在的预测因子进行筛选，将筛选后的预测因子纳入多因素logistic回归分析并建立logistic回归模型。绘制该模型与APACHE Ⅱ评分和BISAP评分的受试者操作特征曲线，通过比较曲线下面积(AUC)值评估其鉴别能力；运用校准图和Hosmer-Lemesshow检验，以及决策曲线分析(DCA)分别评估模型的准确性和临床实用性。应用Bootstrap方法对模型进行内部验证。统计学方法采用独立样本 t检验、Wilcoxon秩和检验和卡方检验。 结果:SAP组与NSAP组的性别构成比差异有统计学意义( χ2＝4.092， P<0.05)。SAP组的住院时间长于NSAP组[(20.33±16.21) d比(8.42±4.26) d]，病死率、住院费用，空腹血糖、尿素氮、血肌酐、C反应蛋白(CRP)、D-二聚体和纤维蛋白原水平，白细胞计数、中性粒细胞占比、中性粒细胞与淋巴细胞比值、APACHE Ⅱ评分、BISAP评分、胸腔积液或腹水的发生率，以及酒精性病因构成比均高于NASP组[21%(27/128)比0(0/472)，44 837.58元(23 017.73元，102 579.77元)比12 301.46元(8 649.26元，18 823.88元)，(10.48±4.84) mmol/L比(8.45±4.80) mmol/L，(8.80±6.50) mmol/L比(4.90±2.33) mmol/L，(139.56±127.75) mmol/L比(80.05±38.54) mmol/L，(187.33±87.25) mg/L比(90.81±82.53) mg/L，5.19 mg/L(2.96 mg/L，8.52 mg/L)比1.29 mg/L(0.53 mg/L, 2.87 mg/L)，6.13 mg/L (4.64 mg/L，7.31 mg/L)比4.58 mg/L(3.50 mg/L，5.98 mg/L)，(14.87±5.82)×10 9/L比(11.79±4.86)×10 9/L，0.84±0.12比0.78±0.12，13.16±7.57比8.77±7.28，(9.80±6.09)分比(3.79±2.59)分，(2.12±0.89)分比(1.04±0.78)分，65.6%(84/128)比12.9%(61/472)，70.3%(90/128)比20.3%(96/472)，18.8% (24/128)比11.4% (54/472)]，血清白蛋白、血钙、血细胞比容均低于NSAP组[(30.86±4.95) g/L比(37.14±5.44) g/L, (1.98±0.31) mmol/L比(2.16±0.20) mmol/L，(42.40±8.67)%比(44.30±6.45)%]，差异均有统计学意义( t＝8.235， χ2＝99.403， Z＝-13.330， t＝4.239、10.759、5.207、11.227， Z＝-11.406、-6.234， t＝6.097、4.829、6.011、10.899、12.395, χ2＝152.604、117.563、4.757， t＝-11.788、-6.180、-2.310， P均<0.05)。LASSO回归分析筛选出CRP、尿素氮、D-二聚体、腹水4个预测因子，多因素logistic回归分析发现，CRP[比值比( OR)＝1.009，95%可信区间( CI)1.006～1.012]、尿素氮( OR＝1.185，95% CI 1.097～1.280)、D-二聚体( OR＝1.166，95% CI 1.082～1.256)、腹水( OR＝4.848，95% CI 2.829～8.307)均为SAP的独立预测因子( P均<0.01)。模型的AUC值为0.895(95% CI 0.865～0.926)，高于APACHEⅡ评分(AUC值为0.835，95% CI 0.791～0.878)和BISAP评分(AUC值为0.803，95% CI 0.760～0.846)，差异均有统计学意义( Z＝2.578、4.466， P均<0.01)。校准图和Hosmer-Lemesshow检验结果显示模型预测结果与实际临床观察结果一致性较高。当模型预测SAP发生的概率为10%～95%时，模型的DCA曲线均高于2条极端线，具有一定临床实用价值。经Bootstrap内部验证，该模型仍具有较高的鉴别能力(AUC值为0.892)，其预测AP严重程度曲线与实际临床的AP严重程度曲线仍有良好的一致性。 结论:基于CRP、尿素氮、D-二聚体和腹水构建的模型能够预测AP的严重程度，有助于临床医师做出更科学的临床决策。

abstractsObjective:To establish and internally validate a visualized model for predicting the severity of acute pancreatitis (AP).Methods:From September 1st 2017 to August 31st 2020, 600 patients with AP diagnosed in the First Affiliated Hospital of Kunming Medical University were enrolled. According to the Atlanta classification of AP, the 600 patients were divided into severe acute pancreatitis (SAP) group (128 cases) and non-severe acute pancreatitis (NSAP) group (472 cases). The general clinical data (age, gender, body mass index, etc), laboratory indicators (fasting blood glucose, urea nitrogen, creatinine, etc.), complicated with ascites or pleural effusion, acute physiology and chronic health evaluation Ⅱ (APACHE Ⅱ) scores and bedside index of severity in acute pancreatitis (BISAP) score between the two groups were compared. The potential predictors of SAP were screened with least absolute shrinkage and selection operator (LASSO). The screened predictors were included in the multivariate logistic regression analysis to establish the logistic regression model. The operation characteristic curves of the model, APACHE Ⅱ scores and BISAP were drawn, the discriminative capability of the model was evaluated by comparing the area under the curve (AUC). Calibration, Hosmer-Lemesshow test and decision curve analysis (DCA) were used to evaluate the accuracy and clinical practicability of the prediction model. Bootstrap was used for internally validation of the model. Independent sample t test, Wilcoxon test and chi-square test were used for statistical analysis. Results:The difference of gender composition ratio between SAP and NSAP group was statistically significant ( χ2=4.092, P<0.05). The fatality rate of SAP group was higher than that of NSAP group(21.1%, 27/128 vs. 0, 0/472); the length of hospital stay of SAP group was longer than that of NSAP group((20.33±16.21) d vs. (8.42±4.26) d); the hospitalization cost, fasting blood glucose level, urea nitrogen level, creatinine level, C-reactive protein(CRP) level, D-dimer level, fibrinogen level, white blood cell count, percentage of neutrophils, neutrophil-lymphocyte ratio, APACHEⅡ and BISAP scores, the incidence of complicated with pleural effusion or ascites and the constituent ratio of alcoholic etiology of SAP group were all higher than those of NASP group (44 837.58 yuan (23 017.73 yuan, 102 579.77 yuan) vs. 12 301.46 yuan (8 649.26 yuan, 18 823.88 yuan); (10.48±4.84) mmol/L vs. (8.45±4.80) mmol/L; (8.80±6.50) mmol/L vs. (4.90±2.33) mmol/L; (139.56±127.75) mmol/L vs. (80.05±38.54) mmol/L; (187.33±87.25) mg/L vs. (90.81±82.53) mg/L; 5.19 mg/L (2.96 mg/L, 8.52 mg/L) vs.1.29 mg/L (0.53 mg/L, 2.87 mg/L); 6.13 mg/L (4.64 mg/L, 7.31 mg/L) vs. 4.58 mg/L (3.50 mg/L, 5.98 mg/L); (14.87±5.82)×10 9/L vs. (11.79±4.86)×10 9/L; 0.84±0.12 vs.0.78±0.12; 13.16±7.57 vs. 8.77±7.28; 9.80±6.09 vs. 3.79±2.59; 2.12±0.89 vs. 1.04±0.78; 65.6%, 84/128 vs. 12.9%, 61/472; 70.3%, 90/128 vs. 20.3%, 96/472; 18.8%, 24/128 vs. 11.4%, 54/472); serum albumin level, blood calcium level, and hematocrit level of SAP group were all lower than those of NSAP group ((30.86±4.95) g/L vs. (37.14±5.44) g/L; (1.98±0.31) mmol/L vs. (2.16±0.20) mmol/L; (42.40±8.67)% vs.(44.30±6.45)%), and the differences were all statistically significant ( χ2=99.403, t=8.235, Z=-13.330, t=4.239, 10.759, 5.207 and 11.227, Z=-11.406 and -6.234, t=6.097, 4.829, 6.011, 10.899 and 12.395, χ2=152.604, 117.563 and 4.757, t=-11.788, -6.180 and -2.310, all P<0.05). LASSO regression analysis screened out four predictors of CRP, urea nitrogen, D-dimer and ascites. The results of multivariate logistic regression analysis showed that CRP (odds ratio ( OR)=1.009, 95% (confidence interval) CI 1.006 to 1.012), urea nitrogen( OR=1.185, 95% CI 1.097 to 1.280), D-dimer( OR=1.166 95% CI 1.082 to 1.256), ascites ( OR= 4.848, 95% CI 2.829 to 8.307) were the independent predictors of SAP (all P<0.01). The AUC of the model (0.895 , 95% CI 0.865 to 0.926) was higher than those of the APACHE Ⅱ(AUC=0.835, 95% CI 0.791 to 0.878)and BISAP score (AUC=0.803, 95% CI 0.760 to 0.846), and the differences were statistically significant ( Z=2.578 and 4.466, both P<0.05). The results predicted by the model in the calibration chart and the Hosmer-Lemesshow test were highly consistent with the results of actual clinical observation. When the probability of SAP in the model was 10% to 95%, the DCA curve of the model was higher than the two extreme lines, which had certain clinical practical value. After bootstrap internal validation, the model had a high discrimination ability (AUC=0.892), and its predicted AP severity curve was still in good agreement with the actual clinical AP severity curve. Conclusion:The prediction model established based on CRP, urea nitrogen, D-dimer and ascites can predict the severity of AP, and help doctors to make more scientific clinical decision.