文献速递关节外科

基于院前采集病史数据训练的机器学习算法可准确预测因髋部主诉就诊骨科医生的患者接受手术或非手术治疗

Operative or Nonoperative Treatment is Predicted Accurately for Patients Who Have Hip Complaints Consulting an Orthopedic Surgeon Using Machine Learning Algorithms Trained With Prehospital Acquired History-Taking Data.
PubMed 文献 · J Arthroplasty · 2023-11-24

这项前瞻性临床队列研究纳入2023年1月20日至2月20日转诊的成年髋部主诉患者,用计算机辅助病史采集表数据训练机器学习算法,在就诊前预测非手术或手术治疗。分析119例咨询,总体预测准确率85%(P<.0001),非手术预测准确率97%(n=71),手术预测准确率67%(n=48);盲态与非盲态咨询准确率无统计学差异。

要点

  • 机器学习算法预测髋部主诉患者治疗策略总体准确率为85%(101/119例,P<.0001)。
  • 非手术治疗预测准确率97%(n=71),手术治疗预测准确率67%(n=48)。
  • 非盲态咨询准确率86.2%与盲态咨询84.4%相比无统计学差异(P>.05)。

核心结论

  • 机器学习算法能以较高准确率预测髋部主诉患者接受非手术或手术治疗。
  • 该预测或有助于将非手术患者安排给医师助理、手术患者安排给骨科医生并直接进入术前筛查,从而优化医疗资源利用。

研究设计与样本

  • 前瞻性临床队列研究,纳入2023年1月20日至2月20日因髋部主诉转诊的成年患者。
  • 患者完成计算机辅助病史采集表,机器学习算法据此在院内咨询前预测非手术或手术治疗结果。
  • 咨询时骨科医生和医师助理对预测结果设盲90例、非盲29例,共分析119例咨询。

主要发现

  • 总体治疗策略预测正确101例,准确率85%(P<.0001)。
  • 非手术治疗预测(n=71)正确率97%,手术治疗预测(n=48)正确率67%。
  • 非盲态咨询正确预测率86.2%,盲态咨询84.4%,两者无统计学差异(P>.05)。

临床怎么用

  • 作者认为机器学习预测可促进将非手术患者安排给医师助理、手术患者安排给骨科医生,并包括直接进入术前筛查,从而优化医疗资源利用。

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

BACKGROUND: Increasing numbers of patients suffering from hip osteoartritis will lead to increased orthopaedic health care consumption. Artificial intelligence might alleviate this problem, using Machine learning (ML) to optimize orthopaedic consultation workflow by predicting treatment strategy (non-operative or operative) prior to consultation. The purpose of this study was to assess ML accuracy in clinical practice, by comparing ML predictions to the outcome of clinical consultations. METHODS: In this prospective clinical cohort study, adult patients referred for hip complaints between January 20th to February 20th 2023 were included. Patients completed a computer-assisted history taking (CAHT) form and using these CAHT answers, a ML-algorithm predicted non-operative or operative treatment outcome prior to in-hospital consultation. During consultation, orthopaedic surgeons and physician assistants were blinded to the prediction in 90 and unblinded in 29 cases. Consultation outcome (non-operative or operative) was compared to ML treatment prediction for all cases, and for blinded and unblinded conditions separately. Analysis was done on 119 consultations. RESULTS: Overall treatment strategy prediction was correct in 101 cases (accuracy 85%, P < .0001). Non-operative treatment prediction (n = 71) was 97% correct versus 67% for operative treatment prediction (n = 48). Results from unblinded consultations (86.2% correct predictions,) were not statistically different from blinded consultations (84.4% correct, P > .05). CONCLUSIONS: Machine Learning algorithms can predict non-operative or operative treatment for patients with hip complaints with high accuracy. This could facilitate scheduling of non-operative patients with physician assistants, and operative patients with orthopaedic surgeons including direct access to pre-operative screening, thereby optimizing usage of health care resources.

来源:J Arthroplasty,查看原文。版权归原作者所有。

引用

  1. van der Weegen W, Warren T, Das D, et al. Operative or Nonoperative Treatment is Predicted Accurately for Patients Who Have Hip Complaints Consulting an Orthopedic Surgeon Using Machine Learning Algorithms Trained With Prehospital Acquired History-Taking Data. J Arthroplasty. 2023. PMID: 38007205. doi:10.1016/j.arth.2023.11.022

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。