基于随机森林算法构建多发性骨髓瘤患者急性肾损伤的预测模型及其应用价值验证-东南大学学报医学版

基于随机森林算法构建多发性骨髓瘤患者急性肾损伤的预测模型及其应用价值验证

单位：1. 克拉玛依市中心医院血液肿瘤科, 新疆克拉玛依 834000;
2. 克拉玛依市中心医院护理部, 新疆克拉玛依 834000

分类号：R733.3;R692

出版年·卷·期（页码）：2025·44·第六期（924-931）

摘要：

目的: 开发并验证一个基于随机森林算法的预测模型,用于评估多发性骨髓瘤(MM)患者发生急性肾损伤的风险,并探究其在临床实践中的应用价值。方法: 本研究回顾性纳入2021年7月至2024年9月期间于本院确诊并收治的217例初发MM患者作为研究对象,并收集其临床资料和患者住院期间的肾功能检测结果。根据患者住院期间的首次肾功能监测结果,以是否发生急性肾损伤(AKI)作为分组标准,将全部患者分为AKI组(n=87)和非AKI组(n=130)。采用多因素Logistic回归分析和随机森林算法分别构建预测MM患者发生AKI的两个预测模型;采用受试者工作特征(ROC)曲线和决策曲线分析(DCA)评估两个预测模型的预测效能和临床适用性。结果: 单因素分析结果显示,两组患者身体质量指数(BMI)、饮酒史、吸烟史、高血压史、脑血管病、慢性肝病、MM分型、总胆固醇(TC)、血钙(Ca)、血清C反应蛋白(CRP)水平差异均无统计学意义(均P＞0.05)。性别、年龄、糖尿病史、血免疫球蛋白轻链κ/λ值、ISS分期、化疗方案以及血红蛋白(Hb)、血清白蛋白(ALB)、乳酸脱氢酶(LDH)、血尿酸(UA)、血清β2微球蛋白(β2-MG)、尿轻链(ULC)水平差异均具有统计学意义(P＜0.05)。多因素Logistic回归结果显示,年龄、化疗方案、LDH、UA、ULC均为MM患者发生AKI的独立危险因素(P＜0.05),ALB则为MM患者发生AKI的独立保护因素(P＜0.05)。基于Logistic回归分析结果构建随机森林模型,其预测MM患者发生AKI的曲线下面积(AUC)为0.784,Logistic回归模型的AUC为0.613,随机森林模型的诊断效能明显优于Logistic回归模型的诊断效能(Z=3.375,P＜0.05)。此外,DCA分析结果证实,随机森林模型相较于Logistic回归模型能够提供更高的临床净获益。结论: 本研究开发的基于随机森林算法的预测模型在预测MM患者AKI风险方面显示出比传统Logistic回归模型更高的诊断效能和临床净获益,具有重要的临床应用价值。

Objective: To develop and validate a prediction model based on the random forest algorithm for assessing the risk of acute kidney injury(AKI) in patients with multiple myeloma(MM) and to investigate its value in clinical practice. Methods: 217 patients with first-episode MM who were diagnosed and admitted to our hospital between July 2021 and September 2024 were retrospectively included as the study subjects, and their clinical data and the results of renal function tests during the patients' hospitalization were collected. In this study, all patients were divided into the AKI group(n=87) and the non-AKI group(n=130) based on their first renal function monitoring results during hospitalization, using the occurrence of AKI as a grouping criterion. Two prediction models for predicting the occurrence of AKI in MM patients were constructed using multifactorial Logistic regression analysis and random forest algorithm, respectively; the predictive efficacy and clinical applicability of the two prediction models were assessed using the receiver operating characteristic(ROC) curve and decision curve analysis(DCA). Results: The results of univariate analysis showed that there were no statistically significant differences between the two groups of patients in terms of body mass index(BMI), history of alcohol consumption, history of smoking, history of hypertension, cerebrovascular disease, chronic liver disease, MM classification, total cholesterol(TC), blood calcium, serum C-reactive protein(CRP) levels(P>0.05). However, there were statistically significant differences in terms of gender, age, history of diabetes, blood immunoglobulin light chain κ/λ ratio, ISS stage, chemotherapy regimen, and hemoglobin(Hb), serum albumin(ALB), lactate dehydrogenase(LDH), blood uric acid(UA), serum β2-microglobulin(β2-MG), and urine light chain(ULC) levels(P<0.05). The results of multifactorial Logistic regression showed that age, chemotherapeutic regimen, LDH, UA, and ULC were independent risk factors for AKI in MM patients(P<0.05), while ALB was an independent protective factor for AKI in MM patients(P<0.05). A random forest model was constructed based on the results of Logistic regression analysis, and its area under the curve(AUC) for predicting the occurrence of AKI in MM patients was 0.784, and that of the Logistic regression model was 0.613, and the diagnostic efficacy of the random forest model was significantly better than that of the Logistic regression model(Z=3.375, P<0.05). In addition, the results of DCA confirmed that the random forest model provided higher net clinical benefit compared with the Logistic regression model. Conclusion: The prediction model based on the random forest algorithm developed in this study showed higher diagnostic efficacy and net clinical benefit than the traditional Logistic regression model in predicting the risk of AKI in patients with MM, which is of significant clinical application value.

参考文献：

[1] 李红伟,司松环,杨靖,等.多发性骨髓瘤患者血清lncRNA PITPNA-AS1、lncRNA NORAD水平及其与患者预后的关系[J].东南大学学报(医学版),2024,43(3):450-455.
[2] REWA O,BAGSHAW S M.Acute kidney injury-epidemiology,outcomes and economics[J].Nat Rev Nephrol,2014,10(4):193-207.
[3] WANG L,LIU C,SONG H,et al.Update on kidney injury caused by multiple myeloma[J].Ann Hematol,2024,103(12):5007-5018.
[4] BARSASELLA D,GUPTA S,MALWADE S,et al.Predicting length of stay and mortality among hospitalized patients with type 2 diabetes mellitus and hypertension[J].Int J Med Inform,2021,154:104569.
[5] ORLENKO A,MOORE J H.A comparison of methods for interpreting random forest models of genetic association in the presence of non-additive interactions[J].BioData Min,2021,14(1):9.
[6] DONG Y,HU C,LIU J,et al.Construction of an auxiliary scoring model for myelosuppression in patients with lung cancer chemotherapy based on random forest algorithm[J].Am J Transl Res,2023,15(6):4155-4163.
[7] GURYLEVA M V,PENZAR D D,CHISTYAKOV D V,et al.Investigation of the role of PUFA metabolism in breast cancer using a rank-based random forest algorithm[J].Cancers,2022,14(19):4663.
[8] RAJKUMAR S V,DIMOPOULOS M A,PALUMBO A,et al.International Myeloma Working Group updated criteria for the diagnosis of multiple myeloma[J].Lancet Oncol,2014,15(12):e538-548.
[9] NO A.KDIGO 2024 clinical practice guideline for the evaluation and management of chronic kidney disease[J].Kidneys,2024,13(2):140-171.
[10] KUMAR S K,RAJKUMAR V,KYLE R A,et al.Multiple myeloma[J].Nat Rev Dis Primers,2017,3:17046.
[11] KYLE R A,GERTZ M A,WITZIG T E,et al.Review of 1027 patients with newly diagnosed multiple myeloma[J].Mayo Clin Proc,2003,78(1):21-33.
[12] PALUMBO A,ANDERSON K.Multiple myeloma[J].N Engl J Med,2011,364(11):1046-1060.
[13] COWAN A J,GREEN D J,KWOK M,et al.Diagnosis and management of multiple myeloma:a review[J].JAMA,2022,327(5):464-477.
[14] DIMOPOULOS M A,DELIMPASI S,KATODRITOU E,et al.Significant improvement in the survival of patients with multiple myeloma presenting with severe renal impairment after the introduction of novel agents[J].Ann Oncol,2014,25(1):195-200.
[15] KLEBER M,IHORST G,DESCHLER B,et al.Detection of renal impairment as one specific comorbidity factor in multiple myeloma:multicenter study in 198 consecutive patients[J].Eur J Haematol,2009,83(6):519-527.
[16] KHALAFALLAH A A,LOI S W,LOVE S,et al.Early application of high cut-off haemodialysis for de-novo myeloma nephropathy is associated with long-term dialysis-independency and renal recovery[J].Mediterr J Hematol Infect Dis,2013,5(1):e2013007.
[17] SZABO A G,THORSEN J,IVERSEN K F,et al.Clinically-suspected cast nephropathy:a retrospective,national,real-world study[J].Am J Hematol,2020,95(11):1352-1360.
[18] 杜晔,郭一丹,贾萌,等.多发性骨髓瘤合并急性肾损伤患者的临床特征及相关风险因素分析[J].临床肿瘤学杂志,2023,28(5):415-419.
[19] BRIDOUX F,LEUNG N,BELMOUAZ M,et al.Management of acute kidney injury in symptomatic multiple myeloma[J].Kidney Int,2021,99(3):570-580.
[20] REITER T,KNAFL D,AGIS H,et al.Structural analysis of urinary light chains and proteomic analysis of hyaline tubular casts in light chain associated kidney disorders[J].PeerJ,2019,7:e7819.
[21] 刘莹莹,崔莉,卜泉东,等.多发性骨髓瘤患者肾损害预测模型的建立与验证[J].精准医学杂志,2021,36(6):544-549,554.
[22] RYU E S,KIM M J,SHIN H S,et al.Uric acid-induced phenotypic transition of renal tubular cells as a novel mechanism of chronic kidney disease[J].Am J Physiol Renal Physiol,2013,304(5):F471-F480.
[23] KELLUM J A,SILEANU F E,BIHORAC A,et al.Recovery after acute kidney injury[J].Am J Respir Crit Care Med,2017,195(6):784-791.
[24] FAN Z,JIANG J,XIAO C,et al.Construction and validation of prognostic models in critically Ill patients with sepsis-associated acute kidney injury:interpretable machine learning approach[J].J Transl Med,2023,21(1):406.
[25] 蓝潞杭,蒋炫东,王茂峰,等.随机森林模型预测急性心肌梗死后急性肾损伤[J].中华急诊医学杂志,2021,30(4):491-495.
[26] HUANG C Y,GüIZA F,DE VLIEGER G,et al.Development and validation of clinical prediction models for acute kidney injury recovery at hospital discharge in critically ill adults[J].J Clin Monit Comput,2023,37(1):113-125.

服务与反馈：

【文章下载】【发表评论】【查看评论】【加入收藏】

提示：您还未登录，请登录！点此登录