机器学习在预测移植肾衰竭方面并未优于传统统计模型
研究基于全球14个中心8422例肾移植受者,在法国4000例推导队列中构建多种机器学习模型,并与经验证的Cox基预后系统(CBPS)比较。推导队列7年时CBPS的C-index为0.808,各机器学习模型为0.527至0.788,验证队列中机器学习模型区分度与CBPS相近、校准相似或更差。作者认为机器学习模型未优于传统CBPS,支持继续使用传统统计方法进行移植肾预后评估。
要点
- 推导队列7年时CBPS的C-index为0.808,高于各机器学习模型的0.527至0.788。
- 验证队列中机器学习模型的区分度与CBPS相近,校准相似或更差。
- 作者认为机器学习未优于传统CBPS,支持继续使用传统统计方法进行移植肾预后评估。
核心结论
- 在使用透明的方法学流程并经国际队列验证时,机器学习模型虽然总体表现良好,但在预测移植肾衰竭方面并未优于传统CBPS。
- 作者认为本研究支持继续使用传统统计方法进行移植肾预后评估。
研究设计与样本
- 研究基于全球14个中心的大队列肾移植受者,开发基于机器学习的移植肾存活预测模型,并与经验证的Cox基预后系统(CBPS)比较预测性能。
- 在法国4000例患者的推导队列中,以供者、受者及移植相关参数作为候选预测因子,构建树基模型(RSF、RSF-ERT、CIF)、支持向量机模型(LK-SVM、AK-SVM)和梯度提升模型(XGBoost)。
- 模型在来自欧洲的2214例、北美的1537例和南美的671例患者队列中进行外部验证。
主要发现
- 8422例肾移植受者中,1081例(12.84%)在移植后中位随访6.25年(四分位距4.33-8.73)后失去移植物。
- 风险评估后7年,RSF、RSF-ERT、CIF、LK-SVM、AK-SVM和XGBoost的C-index分别为0.788(95% bootstrap百分位置信区间0.736-0.833)、0.779(0.724-0.825)、0.786(0.735-0.832)、0.527(0.456-0.602)、0.704(0.648-0.759)和0.767(0.711-0.815),CBPS为0.808(0.792-0.829)。
- 在验证队列中,机器学习模型的区分度与CBPS处于相似范围。
- 机器学习模型的校准与CBPS相似或不如CBPS准确。
关键数值与推荐
- 推导队列风险评估后7年C-index:RSF 0.788(95% bootstrap百分位置信区间0.736-0.833),RSF-ERT 0.779(0.724-0.825),CIF 0.786(0.735-0.832)。
- 推导队列风险评估后7年C-index:LK-SVM 0.527(0.456-0.602),AK-SVM 0.704(0.648-0.759),XGBoost 0.767(0.711-0.815)。
- 推导队列风险评估后7年CBPS的C-index为0.808(0.792-0.829)。
临床怎么用
- 原文结论认为机器学习模型未优于传统CBPS,支持继续使用传统统计方法进行移植肾预后评估。
局限性
- 原文未报告局限性
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
Machine learning (ML) models have recently shown potential for predicting kidney allograft outcomes. However, their ability to outperform traditional approaches remains poorly investigated. Therefore, using large cohorts of kidney transplant recipients from 14 centers worldwide, we developed ML-based prediction models for kidney allograft survival and compared their prediction performances to those achieved by a validated Cox-Based Prognostication System (CBPS). In a French derivation cohort of 4000 patients, candidate determinants of allograft failure including donor, recipient and transplant-related parameters were used as predictors to develop tree-based models (RSF, RSF-ERT, CIF), Support Vector Machine models (LK-SVM, AK-SVM) and a gradient boosting model (XGBoost). Models were externally validated with cohorts of 2214 patients from Europe, 1537 from North America, and 671 from South America. Among these 8422 kidney transplant recipients, 1081 (12.84%) lost their grafts after a median post-transplant follow-up time of 6.25 years (Inter Quartile Range 4.33-8.73). At seven years post-risk evaluation, the ML models achieved a C-index of 0.788 (95% bootstrap percentile confidence interval 0.736-0.833), 0.779 (0.724-0.825), 0.786 (0.735-0.832), 0.527 (0.456-0.602), 0.704 (0.648-0.759) and 0.767 (0.711-0.815) for RSF, RSF-ERT, CIF, LK-SVM, AK-SVM and XGBoost respectively, compared with 0.808 (0.792-0.829) for the CBPS. In validation cohorts, ML models' discrimination performances were in a similar range of those of the CBPS. Calibrations of the ML models were similar or less accurate than those of the CBPS. Thus, when using a transparent methodological pipeline in validated international cohorts, ML models, despite overall good performances, do not outperform a traditional CBPS in predicting kidney allograft failure. Hence, our current study supports the continued use of traditional statistical approaches for kidney graft prognostication.
引用
- Truchot A, Raynaud M, Kamar N, et al. Machine learning does not outperform traditional statistical modelling for kidney allograft failure prediction. Kidney Int. 2022. PMID: 36572246. doi:10.1016/j.kint.2022.12.011
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。