迈向无阿片类日间手术:机器学习预测术后阿片类药物使用的前瞻性研究
本研究开发了以布洛芬600mg与对乙酰氨基酚650mg交替每3小时一次、羟考酮5mg(10剂)为补救的保阿片方案,前瞻性纳入223例日间手术成人,最常见为腹股沟疝修补(49%)。机器学习模型10次运行平均测试AUC为0.674(范围0.634-0.732),平均敏感度0.70、特异度0.68。
要点
- ML模型平均测试AUC为0.674,敏感度0.70,特异度0.68。
- 年龄与疼痛强度为影响模型的重要特征。
核心结论
- 机器学习模型可识别日间手术后可能使用阿片类药物的高风险个体,支持个体化保阿片策略。
研究设计与样本
- 前瞻性研究纳入223例日间手术成人,中位年龄50岁,69%男性,91%白人,49%为腹股沟疝修补。
- 保阿片方案为布洛芬600mg与对乙酰氨基酚650mg交替每3小时一次,补救羟考酮5mg共10剂。
- 机器学习模型采用80/20训练-测试划分并重复10个随机种子评估稳定性。
主要发现
- 模型10次运行平均测试AUC为0.674(范围0.634-0.732),平均敏感度0.70,平均特异度0.68。
- 筛选特征包括活动性癌症、年龄、麻醉类型、种族与族裔、COPD病史、术中并发症、术前对乙酰氨基酚使用及疼痛强度。
- 在种子3(AUC 0.67)中,影响最大的特征为年龄(模型增益17.8%)和疼痛强度(模型增益11%)。
临床怎么用
- 可利用模型识别术后可能需阿片类药物的患者,以制定个体化疼痛管理计划。
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
BACKGROUND: Postoperative opioid use has the risk of dependence and diversion. We developed an opioid-sparing regimen and identified factors associated with postoperative opioid use. STUDY DESIGN: The Toward Opioid-Free Ambulatory Surgery program was developed by establishing a regimen of ibuprofen 600 mg and acetaminophen 650 mg, alternating every 3 hours, with a rescue prescription of oxycodone 5 mg (10 doses). The study included adults undergoing ambulatory operations. A machine learning (ML) model was then developed to predict postoperative opioid use. Performance was evaluated using the area under the receiver operating characteristic curve (AUC) with an 80 and 20 train-test split and repeated across 10 random seeds to assess stability. Feature selection was performed iteratively using training data, whereas model performance was evaluated on test sets. RESULTS: A total of 223 patients were prospectively enrolled (median age 50 years, 69% men, 91% White race). The most common procedure was inguinal hernia repair (49%). Forty-two percent of patients filled their opioid prescription with a median of 4 doses used. The ML model achieved a mean test AUC of 0.674 (range 0.634 to 0.732) across 10 runs. The mean sensitivity was 0.70, and the mean specificity was 0.68. Most selected factors included active cancer, age, anesthesia type, race and ethnicity, COPD history, intraoperative complications, preoperative acetaminophen use, and pain intensity. Specifically, in seed 3 (AUC 0.67), the most influential features were age (model gain 17.8%) and pain intensity (model gain 11%). CONCLUSIONS: The ML model reliably identified high-risk individuals, supporting the potential for personalized opioid-sparing strategies in outpatient surgery. This model may help identify patients likely to require opioids, enabling tailored pain management planning.
引用
- Renshaw S, Satija D, Aly A, et al. Toward Opioid-Free Ambulatory Surgery: A Prospective Study Using Machine Learning to Predict Postoperative Opioid Use. J Am Coll Surg. 2026. PMID: 41589853. doi:10.1097/XCS.0000000000001803
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。