Abstract:

Objective To explore the screening of core risk features using machine learning and the SHAPley Additive exPlanations（SHAP） method，formulate an individualized nursing intervention protocol based on these features，and verify the intervention effect of chronic rhinosinusitis patients at high risk for poor postoperative quality of life，thereby providing a reference for constructing a clinical nursing workflow "from prediction to intervention". Methods A prospective，single-blind，randomized controlled trial design was adopted. High-risk patients with chronic rhinosinusitis were screened by the machine learning prediction model constructed in the early stage. A total of 102 patients were recruited from January to December 2024 and randomly divided into an intervention group and a control group，with 51 cases in each group. The intervention group received modular individualized nursing based on the core risk features of SHAP force plots，while the control group received routine nursing. The quality of life improvement rate，nasal endoscopy score，patient satisfaction and medication compliance at 6 months after operation were compared between the 2 groups. Results The improvement rates of Sino-nasal Outcome Test-22（SNOT-22） scores in the intervention group at 1，3 and 6 months after surgery were 56.9%，74.5% and 86.3%，respectively，while those in the control group were 35.3%，49.0% and 62.7%，respectively. The improvement rates in the intervention group were significantly higher than those in the control group at all time points（all P<0.05）. Additionally，the improvement value of Lund-Kennedy nasal endoscopy score，patient satisfaction score and medication compliance score in the intervention group at 3 months after surgery were all significantly superior to those in the control group，with statistically significant differences（all P<0.01）. Conclusion The individualized nursing program based on risk features screened by machine learning can significantly improve the postoperative quality of life，nasal endoscopy score and diagnosis and treatment compliance of high-risk patients with chronic rhinosinusitis.

Key words: Rhinosinusitis, Machine Learning, Nursing Care, SHAPley Additive exPlanations, Quality of Life

摘要：

目的 探讨基于机器学习及沙普利可加性解释方法筛选慢性鼻窦炎核心风险特征，据此制订个体化护理干预方案，并验证其对术后生活质量未改善高风险慢性鼻窦炎患者的干预效果，为构建“从预测到干预”的临床护理流程提供参考。方法 采用前瞻性单盲随机对照试验设计，依托前期构建的机器学习预测模型筛选慢性鼻窦炎生活质量高风险患者，选取2024年1—12月北京市某三级甲等综合医院耳鼻咽喉头颈外科住院治疗的102例慢性鼻窦炎患者，随机分为试验组与对照组各51例，试验组实施基于沙普利可加性解释值中贡献度前5的核心风险特征的模块化、个体化护理干预，对照组实施常规护理，比较两组术后生活质量改善率、鼻内镜评分、患者满意度及用药依从性的差异。结果 两组均无样本脱落。试验组术后1、3、6个月生活质量改善率分别为56.9%、74.5%和86.3%，对照组分别为35.3%、49.0%和62.7%，各时间点试验组改善率均高于对照组（均P<0.05），且试验组术后3个月鼻内镜评分、患者满意度评分、用药依从性评分均优于对照组，差异有统计学意义（均P<0.01）。结论 基于机器学习筛选风险特征的个体化护理方案，可改善慢性鼻窦炎生活质量高风险患者术后生活质量、鼻内镜评分及诊疗依从性。

关键词: 鼻窦炎, 机器学习, 护理, 沙普利可加性解释, 生活质量