人工神经网络模型预测药物性肝损伤的临床转归及影响因素MIV值评价  

Artificial Neural Network Prediction Model Predict the Clinical Prognosis of Drug-induced Liver Injury And Evaluate the Importance of MIV Values for the Relevant Influencing Factors

作　　者：张琦[1] 李青[2] 冷光[2] 赵伟[2] Zhang Qi;Li Qing;Leng Guang;Zhao Wei(School of Pharmacy,Shanxi Medical University,Taiyuan 030001,China;Department of Pharmacy,the First Hospital of Shanxi Medical University)

机构地区：[1]山西医科大学药学院,太原030001 [2]山西医科大学附属第一医院药学部

出　　处：《药物流行病学杂志》2018年第7期444-449,共6页Chinese Journal of Pharmacoepidemiology

摘　　要：目的:利用人工神经网络方法,对药物性肝损伤(DILI)的转归建立人工神经网络反向传播(Back propagation-Artificial Nerual Network,BP-ANN)预测模型,预测DILI患者的临床转归并对相关影响因素进行平均影响值(mean impact value,MIV)评价。方法:收集某三甲医院全部科室2014年6月1日~2017年6月1日以"药物性肝损害"、"药物性肝损伤"、"药物性肝炎"、"药物性肝病"、"药物性肝衰竭"、"药物性肝硬化"其中之一为主要诊断的266例住院患者资料。符合纳排标准的,对其临床各项指标与转归情况之间的关联性进行Spearman相关性分析,筛选出具有相关性的指标将其作为输入神经元,将转归情况作为输出神经元,构建BP-ANN模型。模型构建训练完毕后,从2017年7月1日起收集70例符合纳入排除标准的DILI住院患者进行临床转归预测,并监测其实际转归情况,将预测结果与实际结果进行对比。进一步用已经训练好的BP-ANN预测模型对有相关性的指标即影响因素进行MIV评价,分析各影响因素对于DILI影响的重要性大小。结果:266例住院患者中,最终符合纳入排除标准的有190例。Spearman相关性分析结果显示共有17项指标有统计学意义,提示有相关性。预测结果显示70例患者中有64例的预测转归与实际转归相符,模型预测的符合率为91.43%。经BP-ANN分析,根据MIV值,直接胆红素首次异常值、血清白蛋白含量、γ-谷氨酰转肽酶首次异常值、体重指数、天冬氨酸转氨酶首次异常值是对DILI患者临床转归影响最大的5个相关指标。结论:人工神经网络模型预测药物性肝损伤临床转归符合率较高,药物性肝损伤的临床转归大部分趋于痊愈或好转。Objective：The artificial neural network method was used to establish the back propagation artificial neural network prediction model for predicting the clinical prognosis of drug-induced liver injury（ DILI）, and evaluating the importance of MIV values for the relevant influencing factors. Methods ：266 hospitalized patients whose main diagnosis was ＂drug-induced liver damage＂ or ＂drug-induced liver injury＂ or ＂drug-induced hepatitis＂ or ＂drug-induced liver disease＂ or ＂drug-induced liver failure＂ or ＂drug-induced cirrhosis＂ were selected from all departments in a top three hospital in June 1, 2014 - June 1, 2017. According to the inclusion exclusion criteria, the correlation between clinical indicators and prognosis was analyzed by Spearman correlation. We screened out the relevant index as input neurons, the prognosis as out- put neurons, and the artificial neural network back propagation （BP-ANN） model was constructed. After the completion of the model construction training, from July 1, 2017, 70 inpatients with DILI who meet the inclusion criteria were collected to predict the clinical outcome. We monitored their actual outcome, then the forecast results and the actual results were com~. Furthermore, MIV evaluation was made on the related indexes and the influencing factors by the trained BP neu- nil network model, and the importance of each influencing factor on the DILI was analyzed. Results： Of the 266 hospitalized patients, 190 eventually met the inclusion criteria. The results of the Spearman correlation analysis showed that there were 17 indicators with statistical significance, suggesting a correlation. The prediction results showed that the prediction outcome of 64 out of 70 patients was consistent with the actual outcome, and the coincidence rate of the model prediction was 91.43%. By BP-ANN analysis, according to MIV value, the first abnormal value of direct bilirubin, serum albumin, the first abnormal value of 7-glutamyl transferase, body mass index and the first

关 键 词：人工神经网络 药物性肝损伤 转归预测模型 平均影响值 

分 类 号：R595.3[医药卫生—内科学]