通讯作者: 陈林丽, chenlinli2008@126.com

DOI：10.12201/bmr.202603.00060

Interpretable Machine Learning for NLR-Based Bleeding Prediction After Renal Biopsy in Young Adults

• 摘要：目的　探讨术前中性粒细胞与淋巴细胞比值（neutrophil-to-lymphocyte ratio, NLR）这一炎症指标预测青年人群（18~44岁）超声引导下经皮肾穿刺活检术后出血的价值，并基于可解释机器学习构建个体化风险预测模型。 方法　回顾性分析527例行超声引导下经皮肾穿刺活检的青年患者。构建极端梯度提升 (eXtreme Gradient Boosting, XGBoost)等5种预测模型，以沙普利加和解释 (Shapley Additive exPlanations, SHAP)方法解释最优模型。通过增量分析评估NLR的附加预测价值，以全身免疫炎症指数（Systemic Immune-inflammation Index, SII）和血小板与淋巴细胞比值（Platelet-to-Lymphocyte Ratio, PLR）进行敏感性比较。 结果　术后出血116例，出血组NLR显著高于非出血组（2.60±1.12 vs 2.33±0.73，P<0.05）。5种模型受试者工作特征曲线下面积 (Area Under the ROC Curve, AUC)范围为0.665~0.771，XGBoost综合指标最优且与SHAP天然兼容（验证集AUC=0.737）。SHAP揭示NLR存在非线性阈值效应：NLR<2.0呈保护作用，>3.0后风险急剧升高。加入NLR后模型AUC从0.696提升至0.735（ΔAUC=0.040，P=0.441），进一步联合SII和PLR仅带来边际改善。 结论　术前NLR是青年肾穿刺活检术后出血的重要预测特征，存在约3.0的阈值效应，对传统模型具有一定的增量预测价值。NLR仅需常规血常规即可获得，可作为术前出血风险评估的便捷标志物。

  关键词： 经皮肾穿刺活检; NLR; 可解释机器学习

   

  Abstract: Objective To evaluate the predictive value of preoperative neutrophil-to-lymphocyte ratio (NLR) for bleeding after ultrasound-guided percutaneous renal biopsy in young adults (18~44 years) using interpretable machine learning. Methods This retrospective study included 527 young adults who underwent ultrasound-guided renal biopsy. Five prediction models, including eXtreme Gradient Boosting (XGBoost), were constructed. The optimal model was interpreted using the Shapley Additive exPlanations (SHAP) method. The additional predictive value of NLR was assessed through incremental analysis, with sensitivity comparisons made against the Systemic Immune-inflammation Index (SII) and the Platelet-to-Lymphocyte Ratio (PLR). Results Post-biopsy bleeding occurred in 116 patients. NLR was significantly higher in the bleeding group (2.60±1.12 vs 2.33±0.73, P<0.05). The AUC of the five models ranged from 0.665 to 0.771. XGBoost was selected as the final model for its balanced performance and native SHAP compatibility (test AUC=0.737). SHAP revealed a non-linear threshold effect: NLR<2.0 was protective, while NLR>3.0 was associated with sharply increased risk. Adding NLR improved AUC from 0.696 to 0.735 (ΔAUC=0.040, P=0.441); further addition of SII and PLR provided only marginal gain. Conclusion Preoperative NLR is an important predictive feature of post-biopsy bleeding in young adults, with a threshold effect at approximately 3.0 and additive predictive value over traditional models. NLR offers a convenient, zero-cost biomarker derived from routine complete blood count for preoperative risk stratification.

  Key words: percutaneous renal biopsy; neutrophil-to-lymphocyte ratio; interpretable machine learning

  提交时间：2026-03-17

  版权声明：作者本人独立拥有该论文的版权，预印本系统仅拥有论文的永久保存权利。任何人未经允许不得重复使用。
• html

• 图表

• 徐鹏, 沈思, 朱佳辉, 夏森林, 徐华. 基于可解释机器学习的老年严重创伤并发急性呼吸窘迫综合征 预测模型构建及效能评估. 2026. doi: 10.12201/bmr.202602.00094

  梁登耀, 刘师伟. NLR、PLR对2型糖尿病肾病的预测价值. 2024. doi: 10.12201/bmr.202408.00066

  苏燕凤, 洪素茹, 陈钰霜, 吴夏阳. 结合机器学习模型的早产儿ICU死亡风险评估与可解释性分析. 2025. doi: 10.12201/bmr.202503.00066

  刘娇娇, 郭连梅, 孙晓红, 张喆, 田野. 基于机器学习的出血性卒中患者下肢深静脉血栓预测模型构建. 2025. doi: 10.12201/bmr.202510.00023

  成路平, 吴思扬, 路波. 基于机器学习的2型糖尿病发病因素分析与预测. 2025. doi: 10.12201/bmr.202509.00003

  芮晨. 基于机器学习的2型糖尿病合并高血压影响因素模型对比与解释性分析. 2025. doi: 10.12201/bmr.202511.00058

  兰雨姗, 郑思, 李姣. 机器学习方法在因果推断中混杂因素控制的应用. 2022. doi: 10.12201/bmr.202203.00015

  苏飞宇, 冯起顺, 曾庆嘉, 高书尧, 奚慧楠, 吴文欣, 王颖帅, 胡红濮. 融合健康画像与机器学习的糖尿病眼底病变风险预测模型研究. 2026. doi: 10.12201/bmr.202601.00077

  唐诗诗, 李宇轩, 唐圣晟, 刘庆华, 周毅. 基于SARIMA-LSTM模型的肾综合征出血热发病率预测研究. 2024. doi: 10.12201/bmr.202407.00046

  康宏宇, 徐晓巍, 郑思, 郝洁, 杨林, 王序文, 侯丽, 李姣. 新医科背景下R语言与机器学习产教研协同教学模式探索. 2025. doi: 10.12201/bmr.202509.00006

• 序号	提交日期	编号	操作
  1	2026-03-04	10.12201/bmr.202603.00060V1	下载

• 公开评论  匿名评论  仅发给作者 提交评论