北京大学学报(医学版) ›› 2019, Vol. 51 ›› Issue (3): 602-608. doi: 10.19723/j.issn.1671-167X.2019.03.034

• 技术方法 • 上一篇    

循环神经网络模型在腹膜透析临床预后预测中的初步应用

唐雯1△(),高峻逸2,马辛宇2,3,张超贺2,马连韬2,3,王亚沙2,4△()   

  • 收稿日期:2019-03-18 出版日期:2019-05-22 发布日期:2019-06-26
  • 作者简介:唐雯,北京大学第三医院肾内科主任医师、副教授,北京大学第三医院肾内科副主任,北京医学会肾脏病学会青年委员、北京医学会血液净化学会分会腹膜透析专家组医疗专家。主要研究方向为慢性肾脏病和腹膜透析的临床及科研。主持国家自然科学基金、中华医学会临床科研基金及北大医学交叉研究(医学-信息科学)种子基金,参与多项国家自然科学基金和北京市科学技术委员会项目。以第一作者或通信作者发表中英文论文共20余篇,包括SCI论文15篇(最高影响因子8.52)。曾获北京大学医学部优秀教师奖,工作所在的腹膜透析临床医护团队获得北京护理学会科技进步奖二等奖。|王亚沙,北京大学软件工程国家工程研究中心副主任、教授、博士生导师,中国计算机学会高级会员、普适计算专业委员会常务委员、国家大数据标准委员会技术专题组组长、中国智慧城市产业与技术创新战略联盟副理事长。长期从事数据分析、普适计算、城市计算等领域的研究工作,在ACM International Joint Con-ference on Pervasive and Ubiquitous Computing、ACM Conference on Computer Supported Cooperative Work、IEEE Transactions on Mobile ComputingIEEE Internet of Things JournalIEEE Computer等国际高水平学术会议与期刊发表论文60余篇。承担国家自然科学基金、国家高技术研究发展计划(863计划)、“核高基”重大科技专项等7项国家和省部级科研课题的研究,并牵头制定了“智慧城市领域知识型”等国家标准。研究成果获国家科学技术进步奖二等奖、北京市科学技术奖二等奖、教育部科学技术进步奖一等奖。
  • 基金资助:
    北大医学交叉研究种子基金(BMU20160584)-中央高校基本科研业务费

Application of recurrent neural network in prognosis of peritoneal dialysis

Wen TANG1△(),Jun-yi GAO2,Xin-yu MA2,3,Chao-he ZHANG2,Lian-tao MA2,3,Ya-sha WANG2,4△()   

  • Received:2019-03-18 Online:2019-05-22 Published:2019-06-26
  • Supported by:
    Supported by the Fundamental Research Funds for the Central Universities: Peking University Medicine Seed Fund for Interdisciplinary Research (BMU20160584)

摘要: 目的 应用深度学习模型循环神经网络(recurrent neural network,RNN)及其变体门控循环单元(gated recurrent unit,GRU),基于临床真实数据,构建腹膜透析临床预后预测模型,并比较其与医学研究中常用的逻辑回归(logistic regression, LR)模型的预测性能,探索预测结果中可能的医学意义。方法 使用北京大学第三医院腹膜透析门诊的常规诊疗数据,基于患者在开始透析时的基线数据、随访数据和预后数据构建RNN和GRU预测模型。使用受试者工作特征曲线下面积(area under the ROC curve,AUROC)、召回率(recall)、F1分数(F1-score)三个指标在测试集上评价比较模型对患者死亡风险的预测效果。结果 共纳入656例患者,其中死亡患者261例,共计13 091条诊断记录。经过十折交叉验证调整超参数并在单独的测试集测试结果显示,LR模型、RNN模型、GRU模型的AUROC分别为0.701 4、0.786 0、0.814 7,RNN和GRU模型的预测性能显著优于传统的LR模型。在召回率和F1分数方面,RNN和GRU模型的性能也均显著优于LR模型,且GRU模型表现最好。进一步分析显示GRU模型在不同预测窗口下对于不同死因或相同死因的召回率不尽相同。结论 RNN模型(尤其是GRU模型)相比于传统医学研究所使用的LR模型,对于腹膜透析临床预后预测具有更佳效果,可能有助于医生早期干预,提高医疗质量,具有很强的临床应用价值。

关键词: 腹膜透析, 预后, 死亡风险预测, 循环神经网络, 门控循环单元

Abstract: Objective: Deep learning models, including recurrent neural network (RNN) and gated recurrent unit (GRU), were used to construct the clinical prognostic prediction models for peritoneal dialysis (PD) patients based on routine clinical data. The performance of the RNN and GRU were compared with logistic regression (LR), which is commonly used in medical researches. The possible underlining clinical implications based on the result from the GRU model were also investigated.Methods: We used the clinical data from the PD center of Peking University Third Hospital as the data source. Both the baseline data at the beginning of dialysis, and the follow-up and prognostic data of the patients were used by the RNN and GRU prediction models. The hyper-parameters were tuned based on the 10-fold cross-validation. The risk prediction performance of each model was evaluated via area under the receiver ope-ration characteristic curve (AUROC), recall rate and F1-score on the testset. Results: A total of 656 patients with the 261 occurrences of death were included in the experiment. The total number of all diagnostic records were 13 091. The results on the testset showed that the AUROC of the LR model, RNN mo-del, and GRU model was 0.701 4, 0.786 0, and 0.814 7, respectively. The predictive performances of the GRU and RNN models were significantly better than that of the LR model. The performances of the GRU and RNN models assessed by recall rate and F1-score were also significantly better than that of the LR model, in which the GRU model reached the best performance. In addition, the recall rates were different among different causes of death or by different prediction time windows.Conclusion: The recurrent neural network model, especially the GRU model, is more effective in predicting PD patients’ prognosis as compared with the LR model. This new model may be helpful for clinicians to provide timely intervention, thus improving the quality of care of PD.

Key words: Peritoneal dialysis, Prognosis, Mortality risk prediction, Recurrent neural network, Gated recurrent unit

中图分类号: 

  • R459.51

表1

腹膜透析相关医疗特征"

Categories of Features
Dynamic features
Blood test
Blood routine: WBC counts, Hb
ALB
Renal function: urea, Cr, uric acid
Electrolyte: K, Na, Cl, CO2CP
Others: Glu, hs-CRP, Ca, P, iPTH
Complications: peritonitis, catheter-related infections, other infections, cardiovascular disease, diabetic foot, cerebrovascular disease, cancer,
hernia, fracture, upper respiratory tract infection, other complications
Nutrition & physical tests: weight, systolic pressure, diastolic pressure
Static baseline features
Demographic features: age, gender
Underlying diseases: baseline comorbidities
Physical tests: height

图1

健康风险状态标签的定义"

图2

循环神经网络模型"

表2

数据集中腹膜透析相关医疗特征统计"

Feature Mean Std Median
Cl/(mmol/L) 98.01 4.84 98.00
CO2CP 27.39 3.64 27.40
Blood uric acid/(mmol/L) 391.61 148.41 380.00
WBC counts/(×109/L) 8.30 55.51 7.49
Hb/(g/L) 114.77 16.83 115.00
Urea/(mmol/L) 19.96 5.46 19.60
Ca/(mmol/L) 2.40 0.37 2.38
K/(mmol/L) 4.32 0.72 4.24
Na/(mmol/L) 138.54 3.88 139.00
Cr/(μmol/L) 857.34 283.70 839.00
P/(mmol/L) 1.60 0.43 1.56
ALB/(g/L) 37.65 4.44 37.90
iPTH/(pg/mL) 214.63 237.93 141.60
hs-CRP/(mg/L) 8.68 15.92 3.34
Glu/(mmol/L) 6.67 3.05 5.70

表3

数据集中腹膜透析相关合并症统计"

Disease Percentage of diagnosed patients
Peritonitis 60.82%
Upper respiratory tract infection 26.98%
Catheter-related infections 9.76%
Other infections 22.10%
Cardiovascular disease 11.74%
Cerebrovascular disease 2.59%
Diabetic foot 3.05%
Cancer 4.12%
Hernia 2.13%
Fracture 1.37%
Other complications 19.21%

图3

基于时序数据进行分类的门控循环单元"

表4

数据集基本统计特征"

Statistics Mean Median Maximum Minimum Std
Age of the patients 58.55 60.70 97.45 16.79 15.81
Number of visits 19.95 16 69 1 13.53
Number of visits that are one year before the death 2 0 29 0 2.95

图4

数据集中年龄与就诊频次的分布"

表5

测试集各预测模型性能的平均值"

Metric LR RNN GRU
AUROC 70.14% 78.60% 81.47%
F1-score 79.32% 86.84% 88.01%
Recall 26.13% 39.42% 48.19%

图5

预测模型的ROC曲线"

图6

不同预测窗口下对不同死因的召回率"

[1] Li KT, Chow KM , Van de Luijtgaarden MW, et al. Changes in the worldwide epidemiology of peritoneal dialysis[J]. Nat Rev Nephrol, 2017,13(2):90-103.
doi: 10.1038/nrneph.2016.181
[2] Lee C, Luo Z, Ngiam KY , et al. Big healthcare data analytics: Challenges and applications[M] //Handbook of large-scale distributed computing in smart healthcare. German: Springer, 2017: 11-41.
[3] Schalkoff RJ . Artificial neural networks[M]. New York: McGraw-Hill, 1997.
[4] Ma F, Chitta R, Zhou J, et al. Dipole: diagnosis prediction in healthcare via attention-based bidirectional recurrent neural networks [C]. Proceedings of the 23rd ACM SIGKDD international conference on knowledge discovery and data mining, 2017: 1903-1911.
[5] Khitan Z, Shapiro AP, Shah PT , et al. Predicting adverse outcomes in chronic kidney disease using machine learning methods: data from the modification of diet in renal disease[J]. Marshall J Med, 2017,3(4):67.
[6] Korchiyne R, Farssi SM, Sbihi A , et al. A combined method of fractal and GLCM features for MRI and CT scan images classification[J]. Signal & Image Processing: An International Journal, 2014,5(4):85.
[7] Lipton ZC, Berkowitz J , Elkan C. A critical review of recurrent neural networks for sequence learning [J/OL]. ( 2015 -10-17)[2019-01-10]. https://arxiv.org/pdf/1506.00019.pdf.
[8] Chung J, Gulcehre C, Cho KH , et al. Empirical evaluation of gated recurrent neural networks on sequence modeling [J/OL]. ( 2014 -12-11)[2019-01-10]. https://arxiv.org/pdf/1412.3555.pdf.
[1] 王飞,朱翔,贺蓓,朱红,沈宁. 自发缓解的滤泡性细支气管炎伴非特异性间质性肺炎1例报道并文献复习[J]. 北京大学学报(医学版), 2021, 53(6): 1196-1200.
[2] 高伟波,石茂静,张海燕,吴春波,朱继红. 显著高铁蛋白血症与噬血细胞性淋巴组织细胞增多症的相互关系[J]. 北京大学学报(医学版), 2021, 53(5): 921-927.
[3] 张梅香,史文芝,刘建新,王春键,李燕,王蔚,江滨. MLL-AF6融合基因阳性急性髓系白血病的临床特征及预后[J]. 北京大学学报(医学版), 2021, 53(5): 915-920.
[4] 蒋艳芳,王健,王永健,刘佳,裴殷,刘晓鹏,敖英芳,马勇. 前交叉韧带翻修重建术后中长期临床疗效及影响因素[J]. 北京大学学报(医学版), 2021, 53(5): 857-863.
[5] 肖若陶,刘承,徐楚潇,何为,马潞林. 术前血小板参数与局部进展期肾细胞癌预后[J]. 北京大学学报(医学版), 2021, 53(4): 647-652.
[6] 于妍斐,何世明,吴宇财,熊盛炜,沈棋,李妍妍,杨风,何群,李学松. 延胡索酸水合酶缺陷型肾细胞癌的临床病理特征及预后[J]. 北京大学学报(医学版), 2021, 53(4): 640-646.
[7] 赵勋,颜野,黄晓娟,董靖晗,刘茁,张洪宪,刘承,马潞林. 癌栓粘连血管壁对非转移性肾细胞癌合并下腔静脉癌栓患者手术及预后的影响[J]. 北京大学学报(医学版), 2021, 53(4): 665-670.
[8] 陈怀安,刘硕,李秀君,王哲,张潮,李凤岐,苗文隆. 炎症生物标志物对输尿管尿路上皮癌患者预后预测的临床价值[J]. 北京大学学报(医学版), 2021, 53(2): 302-307.
[9] 刘世博,高辉,冯元春,李静,张彤,万利,刘燕鹰,李胜光,罗成华,张学武. 腹膜后纤维化致肾盂积水的临床分析:附17例报道[J]. 北京大学学报(医学版), 2020, 52(6): 1069-1074.
[10] 陈伟钱,戴小娜,余叶,王沁,梁钧昱,柯旖旎,易彩虹,林进. 原发性干燥综合征合并自身免疫性肝病的临床特点及预后分析[J]. 北京大学学报(医学版), 2020, 52(5): 886-891.
[11] 姜妮,乔国梁,王小利,周心娜,周蕾,宋雨光,赵艳杰,任军. 中性粒细胞与淋巴细胞比例对评估接受过继性细胞免疫治疗的晚期胰腺癌患者预后的临床意义[J]. 北京大学学报(医学版), 2020, 52(3): 597-602.
[12] 马茹,李鑫宝,闫风彩,林育林,李雁. 肿瘤间质比评估阑尾来源腹膜假黏液瘤的临床价值[J]. 北京大学学报(医学版), 2020, 52(2): 240-246.
[13] 王文鹏,王捷夫,胡均,王俊锋,刘嘉,孔大陆,李健. 结直肠间质瘤临床病理特征及预后分析[J]. 北京大学学报(医学版), 2020, 52(2): 353-361.
[14] 王骁,李兆星,范焕芳,魏莉瑛,郭旭瑾,郭娜,王彤. 罕见小肠囊腺瘤1例报道[J]. 北京大学学报(医学版), 2020, 52(2): 382-384.
[15] 欧阳雨晴,倪莲芳,刘新民. 恶性孤立性肺结节患者预后因素分析[J]. 北京大学学报(医学版), 2020, 52(1): 158-162.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
[1] 赵磊, 王天龙 . 右心室舒张末期容量监测用于肝移植术中容量管理的临床研究[J]. 北京大学学报(医学版), 2009, 41(2): 188 -191 .
[2] 万有, , 韩济生, John E. Pintar. 孤啡肽基因敲除小鼠电针镇痛作用增强[J]. 北京大学学报(医学版), 2009, 41(3): 376 -379 .
[3] 张燕, 韩志慧, 钟延丰, 王盛兰, 李玲玲, 郑丹枫. 骨骼肌活组织检查病理诊断技术的改进及应用[J]. 北京大学学报(医学版), 2009, 41(4): 459 -462 .
[4] 赵奇, 薛世华, 刘志勇, 吴凌云. 同向施压测定自酸蚀与全酸蚀粘接系统粘接强度[J]. 北京大学学报(医学版), 2010, 42(1): 82 -84 .
[5] 林红, 王玉凤, 吴野平. 学校生活技能教育对小学三年级学生行为问题影响的对照研究[J]. 北京大学学报(医学版), 2007, 39(3): 319 -322 .
[6] 丰雷, 程嘉, 王玉凤. 注意缺陷多动障碍儿童的运动协调功能[J]. 北京大学学报(医学版), 2007, 39(3): 333 -336 .
[7] 李岳玲, 钱秋瑾, 王玉凤. 儿童注意缺陷多动障碍成人期预后及其预测因素[J]. 北京大学学报(医学版), 2007, 39(3): 337 -340 .
[8] 牟向东, 王广发, 刁小莉, 阙呈立. 肺黏膜相关淋巴组织型边缘区B细胞淋巴瘤一例[J]. 北京大学学报(医学版), 2007, 39(4): 346 -350 .
[9] 张宏文, 丁洁, 王芳, 杨惠霞. 一例X连锁Alport综合征女性妊娠期随访[J]. 北京大学学报(医学版), 2007, 39(4): 351 -354 .
[10] 韩金涛, 赵军, 栾景源, 张龙. 多发结核性腹主动脉瘤一例[J]. 北京大学学报(医学版), 2007, 39(4): 361 -364 .