开发并验证用于预测过敏性鼻炎儿童ADHD共病的可解释机器学习模型：基于SHAP的分析
Development and validation of an interpretable machine learning model for predicting ADHD comorbidity in children with allergic rhinitis: a SHAP-based analysis.
文献信息
| PMID | 42846167 |
|---|---|
| 原文 | 在 PubMed 查看原文 ↗ |
| 发表日期 | 2026 |
| 作者 | Hexiu Yuan |
| 作者单位 | Ganzhou Hospital-Nanfang Hospital, Southern Medical University (Ganzhou People's Hospital), Ganzhou, China. |
| 期刊 | Frontiers in public health |
| SCI 分区 | Q1 |
| IF | 4.3 |
| 研究类型 | AI/ML · 临床 |
| 所属专科 | 鼻科 |
中文摘要
背景： 过敏性鼻炎（AR）与注意缺陷多动障碍（ADHD）的共病在儿科中带来了重大的临床挑战，但目前缺乏能够捕捉非线性神经免疫相互作用的预测工具。
方法： 我们利用混合特征选择流程，在518名儿科AR患者（其中255名共病ADHD）的队列中开发了一个可解释的极端梯度提升（XGBoost）模型。通过独立测试集评估模型性能。使用SHapley加法解释（SHAP）解码全局和局部可解释性。
结果： 识别出一个由六个预测因子组成的简约子集。XGBoost模型相比传统线性基线表现出更优的区分能力（AUC = 0.878）。SHAP拓扑分析揭示了深刻的非线性阈值，并发现了一个关键的协同相互作用：ADHD风险升高与同时存在严重过敏性炎症（总IgE > 250 IU/mL）和营养缺乏（维生素D < 20 ng/mL）的患者强烈相关。此外，决策曲线分析（DCA）显示在分诊应用中具有显著的净临床获益。
结论： 我们建立了一个高度稳健、可解释的机器学习算法，能够准确地对AR儿童进行ADHD风险分层。所识别的神经免疫协同表型提供了关键的机制见解，并为及时进行儿科多学科干预提供了实用的决策支持工具。
英文摘要
BACKGROUND: The co-occurrence of allergic rhinitis (AR) and attention-deficit/hyperactivity disorder (ADHD) presents significant clinical challenges in pediatrics, yet predictive tools capturing non-linear neuro-immune interactions are lacking.
METHODS: We developed an interpretable Extreme Gradient Boosting (XGBoost) model utilizing a hybrid feature selection pipeline on a cohort of 518 pediatric AR patients (255 with comorbid ADHD). Model performance was evaluated via an independent test set. Global and local interpretability was decoded using SHapley Additive exPlanations (SHAP).
RESULTS: A parsimonious subset of six predictors was identified. The XGBoost model achieved superior discriminative capacity (AUC = 0.878) compared to conventional linear baselines. SHAP topographical analysis revealed profound non-linear thresholds and uncovered a critical synergistic interaction: elevated ADHD risk was strongly correlated with patients presenting simultaneously severe allergic inflammation (Total IgE > 250 IU/mL) and nutritional deficiency (Vitamin D < 20 ng/mL). Furthermore, Decision Curve Analysis (DCA) demonstrated substantial net clinical benefit for triage application.
CONCLUSION: We established a highly robust, interpretable machine learning algorithm capable of accurately stratifying ADHD risk among AR children. The identified neuro-immune synergistic phenotypes provide critical mechanistic insights and offer a practical decision-support tool for timely pediatric multidisciplinary interventions.