基于机器学习模型预测经皮肾镜取石术后全身炎症反应综合征
Machine Learning Models to Predict Systemic Inflammatory Response Syndrome after Percutaneous Nephrolithotomy
摘要: 目的:本研究旨在开发和评估机器学习模型在预测经皮肾镜取石术(Percutaneous nephrolithotomy,PCNL)术后发生全身炎症反应综合征(Systemic inflammatory response syndrome,SIRS)的性能。方法:回顾性分析2021年10月至2022年12月接受PCNL治疗的218例患者的临床资料。在我们的研究中,按照8:2划分数据集为训练集和测试集。基于Light Gradient Boosting Machine (LightGBM) 机器学习算法在训练集构建预测模型。LightGBM机器学习模型的预测性能由使用测试集的受试者工作特征曲线下面积(Area under the receiver operating characteristic curve,AUC)、准确性、灵敏度和特异性决定。我们使用权重系数来解释每个变量对预测模型性能的贡献。结果:LightGBM模型在测试集中的准确率为0.837,AUC为0.918 (95%CI 0.827-1.000),灵敏度为0.875,特异性为0.829。对LightGBM模型的进一步分析显示,全身免疫炎症(systemic immune-inflammation,SII)指标对结局预测的贡献最大,其次是术前尿培养、前白蛋白、结石负荷、中性粒细胞与淋巴细胞比值(neutrophil to lymphocyte ratio,NLR)、肾盂积水、纤维蛋白原、淋巴细胞与单核细胞比值(Lymphocyte to monocyte ratio,LMR)。结论:基于LightGBM模型学习患者临床数据,能够提前准确预测PCNL术后SIRS发生的可能性,并用于指导外科医生的临床决策。
Abstract: Objective: The objective of this study was to develop and evaluate the performance of machine learning model for predicting the possibility of systemic inflammatory response syndrome (SIRS) following percutaneous nephrolithotomy (PCNL). Methods: We retrospectively reviewed the clinical data of 218 patients who received PCNL between October 2021 and December 2022. In our study, the dataset is divided into the training set and the testing set according to 8:2. The prediction mod-el based on the Light Gradient Boosting Machine (LightGBM) algorithms was created using the training set. The predictive performance of the LightGBM machine learning model was determined by the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and spec-ificity using the testing set. We used coefficients to interpret the contribution of each variable to the predictive performance. Results: The LightGBM model delivered a good performance with an accu-racy of 0.837, AUC of 0.918 (95% CI 0.827~1.000), sensitivity of 0.875, specificity of 0.829 in the testing set. Further analysis using the LightGBM model showed that systemic immune inflammation (SII) contributed the most to the prediction of the outcome, followed by preoperative urine culture, prealbumin, stone burden, neutrophil to lymphocyte ratio (NLR), hydronephrosis, fibrinogen, and Lymphocyte to monocyte ratio (LMR). Conclusion: The LightGBM models can accurately predict the possibility of SIRS after PCNL in advance by learning patient clinical data, and should be used to guide surgeons in clinical decision-making.
文章引用:张天伟, 焦伟. 基于机器学习模型预测经皮肾镜取石术后全身炎症反应综合征[J]. 临床医学进展, 2023, 13(11): 16968-16976. https://doi.org/10.12677/ACM.2023.13112376

1. 引言

肾结石是最常见的泌尿系统疾病之一,肾结石在世界范围内的患病率正在上升 [1] 。据报道,中国成人肾结石发病率为5.8%,目前诊断出的成人中约有1/17 [2] 。自1976年首次报道PCNL以来,PCNL逐渐成为结石大于2 cm、多发性或鹿角形结石患者的标准治疗 [3] 。

PCNL具有许多优点,包括创伤最小,结石清除率高,住院时间短和恢复快 [4] 。然而,手术后也有几种并发症。全身炎症反应综合征(systemic inflammatory response syndrome, SIRS)是PCNL相关的常见严重并发症,发病率为16.7%~27.4% [5] 。如果不及早诊断和治疗,术后SIRS可进展成为尿源性脓毒症,其发病率在0.3%~4.7%之间 [6] 。当脓毒症进展为脓毒性休克或多器官衰竭时,会导致死亡率和治疗费用增加。

随着统计理论和计算机技术的发展,机器学习算法越来越多地用于辅助医学中的诊断、治疗和自动分类 [7] 。基于机器学习的影像组学用于泌尿系统肿瘤的早期诊断 [8] 、疗效评价 [9] 、预后评估 [10] 。基于机器学习模型识别非小细胞肺癌患者的病理亚型 [11] 。我们必须开发基于机器学习算法的高效预测模型,提前预测接受PCNL的患者术后发生SIRS的可能性,并在围手术期密切监测患者生命体征,这可以显著减轻患者的负担。

2. 资料与方法

2.1. 资料来源

回顾性收集了2021年10月至2022年12月在青岛大学附属医院泌尿科接受PCNL的218例患者的临床资料。根据以下标准,患者被排除在分析之外:(a) 双侧PCNL病史;(b) 存在患有肿瘤、血液系统或免疫系统疾病的患者;(c) 先天性畸形,如多囊肾、马蹄肾和孤立肾;(d) 缺失数据。这项研究符合《赫尔辛基宣言》的原则,并按照我们机构医学伦理委员会的道德标准进行。

2.2. 数据收集

患者的术前数据包括:年龄、性别、体重指数(BMI)、术前白细胞(WBC)、中性粒细胞(N)、淋巴细胞(L)、单核细胞(M)、血小板(PLT)、血红蛋白(HB)、中性粒细胞与淋巴细胞比值(NLR)、血小板与淋巴细胞比值(PLR)、淋巴细胞与单核细胞比值(LMR)、全身免疫炎症(SII、中性粒细胞计数 × 血小板计数/淋巴细胞计数)、术前血清肌酐、尿素氮、尿酸、胱抑素、白蛋白、前白蛋白、纤维蛋白原、结石负荷(长×宽 × π × 0.25)、尿白细胞、尿亚硝酸盐、尿培养和肾积水。术中信息包括手术时间。术后第二天早上6点统一测量的术后信息包括外周白细胞计数、血压、心率、氧合、呼吸频率和体温。如果患者符合以下四个标准中的两个或更多,则被诊断为SIRS:1) 白细胞计数 < 4 × 109/L或>12 × 109/L;2) 体温 > 38℃或<36℃;3) 心率 > 90/min;4) 呼吸频率 > 20/min或动脉血二氧化碳分压 < 32 mmHg [4] 。

2.3. 特征筛选

首先,对临床特征进行统计t检验、曼–惠特尼检验和卡方检验,以确定两组之间是否存在差异。其次,采用spearman相关分析降低特征间的共线性;为了降低过度拟合的风险,采用最小绝对值收敛和选择算子(Least absolute shrinkage and selection operator, LASSO)算法来选择具有非零系数值的特征。

2.4. 模型构建

采用LightGBM机器学习算法预测PCNL术后SIRS的发生,建立预测PCNL术后SIRS发生预测模型 [12] 。数据被随机分为训练集(80%)和测试集(20%)。训练集包括175名患者,而测试集43名患者。训练集用于使用五重交叉验证建立预测模型,而测试集用于使用接收器工作特征(Receiver operating characteristics, ROC)的曲线下面积(Area under the curve, AUC)来验证预测模型的性能。我们计算了筛选后用于建模的特征的相关系数,并且可视化每个特征对模型预测的贡献。

2.5. 统计分析

具有正态分布的连续变量表示为平均 ± 标准差(SD),并使用t检验进行比较。具有非正态分布的连续变量表示为具有四分位距和中位数,并使用曼–惠特尼检验进行比较。分类变量表示为具有比例的频率,并使用卡方检验进行比较。机器学习模型是用Python 3.7语言编写的。

3. 结果

3.1. 患者一般资料

该研究包括218名接受PCNL并拥有完整病历的患者。其中,42例患者术后发生了SIRS。根据PCNL术后是否发生SIRS将患者分为两组。两组患者的术前N、术前L、术前PLT、术前Hb、白蛋白、纤维蛋白原、血清前白蛋白、NLR、PLR、LMR、SII、手术时间、结石负荷、肾盂积水、尿培养的差异有统计学意义。纳入患者的基线数据(见表1)。

Table 1. Analysis of related factors of SIRS after percutaneous nephrolithotripsy

表1. 经皮肾镜碎石术后SIRS的相关因素分析

3.2. 特征选择与模型构建

使用Spearman相关分析和具有五重交叉验证的LASSO算法,27个变量最终减少为8个潜在预测因子(见表2),这些变量被纳入我们研究中预测模型的构建中。

Table 2. Feature selection results and coeffients for each feature.

表2. 特征筛选结果和每一个特征的权重系数

训练集和测试集分别由80%和20%的数据库组成。使用LightGBM算法在训练集中建立预测模型,并使用测试集评估模型的性能,并通过AUC,准确性,灵敏度和特异性表示。基于LightGBM的预测模型的受试者工作特性(ROC)和曲线下面积(AUC)如图1所示。LightGBM模型的展现良好的预测能力,准确率为0.837,AUC为0.918 (95%CI 0.827~1.000),灵敏度为0.875,特异性为0.829 (见表3)。

Table 3. The performance of machine learning model based on LightGBM algorithm

表3. 基于LightGBM算法的机器学习模型的性能

Figure 1. Performance for the LightGBM machine learning model

图1. LightGBM机器学习模型的性能

权重系数用于评估每个变量对预测模型的贡献,详见图2。如图2所示,全身免疫炎症(systemic immune-inflammation, SII)指标对结局预测的贡献最大,其次是术前尿培养、前白蛋白、结石负荷、NLR、肾盂积水、纤维蛋白原、LMR。

Figure 2. Top 8 selected features and the corresponding variable coefficients

图2. 8个筛选出的特征和相应的权重系数

4. 讨论

PCNL已成为治疗大于2 cm的多发性或鹿角形肾结石的首选 [13] 。尽管PCNL具有创伤小、结石去除率高的优点,但与其他微创结石手术技术相比,PCNL仍有许多并发症,尤其是术后出血和术后感染。SIRS初始阶段的典型临床症状不明显,难以在早期发现PCNL术后SIRS的发生。如果不及时治疗,SIRS可能导致败血症或者多器官功能衰竭 [14] 。因此,我们应该建立一个合适的基于机器学习算法的预测模型,提前预测SIRS的发生。

随着统计理论和计算机技术的发展,与传统的预测方法相比,新颖的机器学习技术提高了预测性能。以前的文献表明,机器学习算法被用于预测重症监护病房或急诊科患者中SIRS的发生 [15] 。Kijpaisalratana等建立了急诊科脓毒症早期风险预测预测模型 [16] 。Hou等人开发了一个使用XGboost算法的预测模型来预测重症监护病房的脓毒症患者的30天死亡率 [17] 。然而,关于基于机器学习算法的PCNL后SIRS预测模型的文章并没有报道。

据我们所知,这是第一次利用机器学习算法来预测PCNL术后SIRS的发生。在这项研究中,我们使用LightGBM机器学习算法来预测PCNL术后SIRS的发生。在我们的研究中,LightGBM模型展现了较高的AUC、准确性。LightGBM模型是基于决策树算法的快速、分布式和高性能梯度提升框架 [18] 。LightGBM解决了过度拟合问题,提高了模型的泛化能力和稳定性,尤其适用于小样本和非线性数据。

图2显示SII、术前尿培养、结石负荷、NLR、肾盂积水、纤维蛋白原是潜在的危险因素。以前的文献报告显示,SII是肝细胞癌 [19] ,前列腺癌 [20] 和PCNL后SIRS的有希望的预后指标 [21] 。结石的存在导致炎症介质如IL-6,IL-7,IL-8,TNF-α的释放和中性粒细胞数量的增加 [22] 。血小板富含促炎因子,可以释放活性炎症代谢物 [23] 。过度的炎症反应会抑制免疫反应,导致淋巴细胞数量减少,这与SII数值的增加有关。SII是一种廉价且易于获得的生物标志物,可以全面反映宿主的免疫状态,在我们的研究中对结果预测的贡献最大。在这项研究中,NLR也是一种易于获取且具有成本效益的预测PCNL术后SIRS发生的生物标志物 [24] 。如图2所示,术前LMR是一种保护因素。Jager等人强调淋巴细胞减少症是由于细胞凋亡加速引起的菌血症的标志物 [25] 。因此,应多关注NLR较高、SII较高或LMR较低的患者。

在本研究中,尿培养阳性、肾盂积水、结石负荷和手术时间是PCNL术后发生SIRS的关键因素,这与以往文献报道相似 [26] 。尽管新的研究表明,肾盂尿培养比中段尿培养更能预测尿脓毒症,但培养肾盂尿需要很长时间。因此,中段尿尿培养仍然是一个可靠的指标,因为它很容易获得。较高的结石负荷会增加手术难度,延长手术时间,可能延长肾盂压,增加内毒素吸收到血液中的概率 [21] 。因此,有必要在PCNL前完成尿常规检查和尿培养检查,以评估尿路感染的严重程度。

一些文献报道描述了纤维蛋白原和前白蛋白是术后SIRS的相关因素 [27] 。纤维蛋白原是肺癌 [28] 、肝细胞癌 [29] 、结直肠癌 [30] 和泌尿系统癌症 [31] 的关键调节因子。图2显示前白蛋白是一种保护性预测因子。前白蛋白的水平反映了患者最近的营养状况,并对疾病的预后有积极影响 [32] 。因此,应多关注纤维蛋白原水平高、前白蛋白水平低的高危患者。

这项研究有几个局限性。首先,这些机器学习模型是基于单中心数据训练和开发的。因此,需要进一步的外部验证来解释这一模式的普遍性。其次,我们研究的操作都是单侧和单通道PCNL,因此我们无法研究束数对PCNL后SIRS的影响。

5. 结论

本研究确定了PCNL术后发生SIRS的相关影响因素,并构建了机器学习模型来预测PCNL后SIRS的可能性。基于LightGBM的机器学习预测模型具有较强数据分析能力、同时可以辅助临床决策。

NOTES

*第一作者。

#通讯作者Email: jiaowei3929@163.com

参考文献

[1] Wang, Z., Zhang, Y., Zhang, J., Deng, Q. and Liang, H. (2021) Recent Advances on the Mechanisms of Kidney Stone Formation (Review). International Journal of Molecular Medicine, 48, Article No. 149.
https://doi.org/10.3892/ijmm.2021.4982
[2] Zeng, G., Mai, Z., Xia, S., Wang, Z., Zhang, K., Wang, L., et al. (2017) Prevalence of Kidney Stones in China: An Ultrasonography Based Cross-Sectional Study. BJU International, 120, 109-116.
https://doi.org/10.1111/bju.13828
[3] 吴佳成, 徐海飞, 陈志刚, 王小林. 经皮肾镜取石术所致尿源性脓毒血症的预测因子临床分析[J]. 国际泌尿系统杂志, 2020, 40(1): 89-93.
[4] 张剑波. 泌尿系统结石经皮肾镜手术后发生尿源性脓毒血症的影响因素[J]. 国际泌尿系统杂志, 2016, 36(2): 187-190.
[5] Koras, O., Bozkurt, I.H., Yonguc, T., Degirmenci, T., Arslan, B., Gunlusoy, B., et al. (2015) Risk Factors for Postoperative Infectious Com-plications Following Percutaneous Nephrolithotomy: A Prospective Clinical Study. Urolithiasis, 43, 55-60.
https://doi.org/10.1007/s00240-014-0730-8
[6] Knoll, T., Daels, F., Desai, J., Hoznek, A., Knudsen, B., Montanari, E., et al. (2017) Percutaneous Nephrolithotomy: Technique. World Journal of Urology, 35, 1361-1368.
https://doi.org/10.1007/s00345-017-2001-0
[7] Jiang, H., Liu, L., Wang, Y., Ji, H., Ma, X., Wu, J., et al. (2021) Machine Learning for the Prediction of Complications in Patients after Mitral Valve Surgery. Frontiers in Cardiovascu-lar Medicine, 8, Article ID: 771246.
https://doi.org/10.3389/fcvm.2021.771246
[8] Li, Y., Huang, X., Xia, Y. and Long, L. (2020) Value of Radi-omics in Differential Diagnosis of Chromophobe Renal Cell Carcinoma and Renal Oncocytoma. Abdominal Radiology (NY), 45, 3193-3201.
https://doi.org/10.1007/s00261-019-02269-9
[9] Choi, S.J., Park, K.J., Heo, C., Park, B.W., Kim, M. and Kim, J.K. (2021) Radiomics-Based Model for Predicting Pathological Complete Response to Neoadjuvant Chemotherapy in Muscle-Invasive Bladder Cancer. Clinical Radiology, 76, 627.e13-.e21.
https://doi.org/10.1016/j.crad.2021.03.001
[10] Rallis, K.S., Kleeman, S.O., Grant, M., Ordidge, K.L., Sahdev, A. and Powles, T. (2021) Radiomics for Renal Cell Carcinoma: Predicting Outcomes from Immunotherapy and Targeted Therapies—A Narrative Review. European Urology Focus, 7, 717-721.
https://doi.org/10.1016/j.euf.2021.04.024
[11] Zhao, H., Su, Y., Wang, M., Lyu, Z., Xu, P., Jiao, Y., et al. (2022) The Machine Learning Model for Distinguishing Pathological Subtypes of Non-Small Cell Lung Cancer. Frontiers in Oncology, 12, Article ID: 875761.
https://doi.org/10.3389/fonc.2022.875761
[12] Rufo, D.D., Debelee, T.G., Ibenthal, A. and Negera, W.G. (2021) Diagnosis of Diabetes Mellitus Using Gradient Boosting Machine (LightGBM). Diagnostics (Basel), 11, Article No. 1714.
https://doi.org/10.3390/diagnostics11091714
[13] Chen, D., Jiang, C., Liang, X., Zhong, F., Huang, J., Lin, Y., et al. (2019) Early and Rapid Prediction of Postoperative Infections Following Percutaneous Nephrolithotomy in Patients with Complex Kidney Stones. BJU International, 123, 1041-1047.
https://doi.org/10.1111/bju.14484
[14] Rivera, M., Viers, B., Cockerill, P., Agarwal, D., Mehta, R. and Krambeck, A. (2016) Pre- and Postoperative Predictors of In-fection-Related Complications in Patients Undergoing Percutaneous Nephrolithotomy. Journal of Endourology, 30, 982-986.
https://doi.org/10.1089/end.2016.0191
[15] Yuan, S., Sun, Y., Xiao, X., Long, Y. and He, H. (2021) Using Machine Learning Algorithms to Predict Candidaemia in ICU Patients with New-Onset Systemic Inflammatory Response Syndrome. Frontiers in Medicine (Lausanne), 8, Article ID: 720926.
https://doi.org/10.3389/fmed.2021.720926
[16] Kijpaisalratana, N., Sanglertsinlapachai, D., Techaratsami, S., Musikatavorn, K. and Saoraya, J. (2022) Machine Learning Algorithms for Early Sepsis Detection in the Emergency Department: A Retrospective Study. International Journal of Medical Informatics, 160, Article ID: 104689.
https://doi.org/10.1016/j.ijmedinf.2022.104689
[17] Hou, N., Li, M., He, L., Xie, B., Wang, L., Zhang, R., et al. (2020) Predicting 30-Days Mortality for MIMIC-III Patients with Sepsis-3: A Machine Learning Approach Using XGboost. Journal of Translational Medicine, 18, Article No. 462.
https://doi.org/10.1186/s12967-020-02620-5
[18] Kobayashi, Y. and Yoshida, K. (2021) Quantitative Struc-ture-Property Relationships for the Calculation of the Soil Adsorption Coefficient Using Machine Learning Algorithms with Calculated Chemical Properties from Open-Source Software. Environmental Research, 196, Article ID: 110363.
https://doi.org/10.1016/j.envres.2020.110363
[19] Hu, B., Yang, X.R., Xu, Y., Sun, Y.F., Sun, C., Guo, W., et al. (2014) Systemic Immune-Inflammation Index Predicts Prognosis of Patients after Curative Resection for Hepatocellular Carcinoma. Clinical Cancer Research, 20, 6212-6222.
https://doi.org/10.1158/1078-0432.CCR-14-0442
[20] Lolli, C., Caffo, O., Scarpi, E., Aieta, M., Conteduca, V., Maines, F., et al. (2016) Systemic Immune-Inflammation Index Predicts the Clinical Outcome in Patients with mCRPC Treated with Abiraterone. Frontiers in Pharmacology, 7, Article No. 376.
https://doi.org/10.3389/fphar.2016.00376
[21] Peng, C., Li, J., Xu, G., Jin, J., Chen, J. and Pan, S. (2021) Signifi-cance of Preoperative Systemic Immune-Inflammation (SII) in Predicting Postoperative Systemic Inflammatory Response Syndrome after Percutaneous Nephrolithotomy. Urolithiasis, 49, 513-519.
https://doi.org/10.1007/s00240-021-01266-2
[22] Tang, K., Liu, H., Jiang, K., Ye, T., Yan, L., Liu, P., et al. (2017) Predictive Value of Preoperative Inflammatory Response Biomarkers for Metabolic Syndrome and post-PCNL Systemic Inflammatory Response Syndrome in Patients with Nephrolithiasis. Oncotarget, 8, 85612-85627.
https://doi.org/10.18632/oncotarget.20344
[23] Gasparyan, A.Y., Ayvazyan, L., Mukanova, U., Yessirkepov, M. and Kitas, G.D. (2019) The Platelet-to-Lymphocyte Ratio as an Inflammatory Marker in Rheumatic Diseases. Annals of Laboratory Medicine, 39, 345-357.
https://doi.org/10.3343/alm.2019.39.4.345
[24] Kriplani, A., Pandit, S., Chawla, A., de la Rosette, J., Laguna, P., Jayadeva Reddy, S., et al. (2022) Neutrophil-Lymphocyte Ratio (NLR), Platelet-Lymphocyte Ratio (PLR) and Lympho-cyte-Monocyte Ratio (LMR) in Predicting Systemic Inflammatory Response Syndrome (SIRS) and Sepsis after Percuta-neous Nephrolithotomy (PNL). Urolithiasis, 50, 341-348.
https://doi.org/10.1007/s00240-022-01319-0
[25] de Jager, C.P., van Wijk, P.T., Mathoera, R.B., de Jongh-Leuvenink, J., van der Poll, T. and Wever, P.C. (2010) Lympho-cytopenia and Neutrophil-Lymphocyte Count Ratio Predict Bacteremia Better than Conventional Infection Markers in an Emergency Care Unit. Critical Care, 14, R192.
https://doi.org/10.1186/cc9309
[26] Tang, Y., Zhang, C., Mo, C., Gui, C., Luo, J. and Wu, R. (2021) Predictive Model for Systemic Infection after Percutaneous Nephrolithotomy and Re-lated Factors Analysis. Frontiers in Surgery, 8, Article ID: 696463.
https://doi.org/10.3389/fsurg.2021.696463
[27] Qiao, W., Leng, F., Liu, T., Wang, X., Wang, Y., Chen, D., et al. (2020) Prognostic Value of Prealbumin in Liver Cancer: A Systematic Review and Meta-Analysis. Nutrition and Cancer, 72, 909-916.
https://doi.org/10.1080/01635581.2019.1661501
[28] Zhang, Y., Cao, J., Deng, Y., Huang, Y., Li, R., Lin, G., et al. (2020) Pretreatment Plasma Fibrinogen Level as a Prognostic Biomarker for Patients with Lung Cancer. Clinics (Sao Paulo), 75, e993.
https://doi.org/10.6061/clinics/2020/e993
[29] Zanetto, A., Campello, E., Spiezia, L., Burra, P., Simioni, P. and Russo, F.P. (2018) Cancer-Associated Thrombosis in Cirrhotic Patients with Hepatocellular Carcinoma. Cancers (Basel), 10, Article No. 450.
https://doi.org/10.3390/cancers10110450
[30] Lin, Y., Liu, Z., Qiu, Y., Zhang, J., Wu, H., Liang, R., et al. (2018) Clinical Significance of Plasma D-Dimer and Fibrinogen in Digestive Cancer: A Systematic Review and Meta-Analysis. European Journal of Surgical Oncology, 44, 1494-1503.
https://doi.org/10.1016/j.ejso.2018.07.052
[31] Song, H., Kuang, G., Zhang, Z., Ma, B., Jin, J. and Zhang, Q. (2019) The Prognostic Value of Pretreatment Plasma Fibrinogen in Urological Cancers: A Systematic Review and Meta-Analysis. Journal of Cancer, 10, 479-487.
https://doi.org/10.7150/jca.26989
[32] 沈婧, 陈良琼. 前白蛋白、胱抑素C、β2微球蛋白联合检测在泌尿系统术后细菌感染中的预测价值[J]. 国际泌尿系统杂志, 2022, 42(3): 521-524.