文献速递医学人工智能与信息学临床试验与 GCP

决策算法与多学科团队会议对肝癌患者的一致性:一项随机对照试验的研究方案

Concordance of a decision algorithm and multidisciplinary team meetings for patients with liver cancer-a study protocol for a randomized controlled trial.
PubMed 文献 · Trials · 2023-09-09

该研究方案拟利用自然语言处理从电子病历自动提取患者信息,并基于回顾性多学科团队会议数据与临床指南训练机器学习模型,形成ADBoard治疗推荐算法。研究将肝癌患者随机分为ADBoard辅助的多学科团队会议组(A组)与常规多学科团队会议组(B组),以评估者间信度比较两组推荐的一致性,假设Cohen's kappa值≥0.75。

要点

  • ADBoard利用自然语言处理自动提取病历信息并基于指南与回顾数据训练机器学习模型。
  • 研究将患者随机分入ADBoard辅助组(A组)与常规组(B组),比较推荐一致性。
  • 主要假设为ADBoard与常规多学科团队会议的推荐Cohen's kappa值≥0.75。

核心结论

  • ADBoard旨在提高多学科团队会议决策所需数据的质量与完整性,并以透明可重复的方式提出考虑当前医学证据与指南的治疗推荐。

研究设计与样本

  • 研究为随机对照试验,将符合纳入标准的患者随机分为ADBoard辅助的多学科团队会议组(A组,MDM-AB)与常规多学科团队会议组(B组,MDM-C)。
  • 利用自然语言处理从电子病历自动提取相关患者信息以完成经典肿瘤会议方案,并基于回顾性多学科团队会议数据与临床指南训练机器学习模型。
  • 研究方案经柏林夏里特医学院伦理委员会批准,于2023年1月12日在ClinicalTrials.gov注册(试验识别号NCT05681949)。

主要发现

  • 研究假设ADBoard的治疗推荐与常规多学科团队会议的推荐高度一致,Cohen's kappa值≥0.75。
  • 次要假设为使用ADBoard时多学科团队会议呈现的患者信息完整性高于不使用ADBoard时。
  • 次要假设为以系统因果性量表(System Causability Scale)测量,MDM-AB组肿瘤会议方案的可解释性高于MDM-C组。

关键数值与推荐

  • 主要结局采用评估者间信度比较两组推荐的一致性,假设Cohen's kappa值≥0.75。
  • 次要结局包括多学科团队会议中患者信息完整性(使用ADBoard时更高)与肿瘤会议方案可解释性(以系统因果性量表测量,MDM-AB组更高)。

临床怎么用

  • 原文未据此给出推荐。

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

INTRODUCTION: Multidisciplinary team meetings (MDMs), also known as tumor conferences, are a cornerstone of cancer treatments. However, barriers such as incomplete patient information or logistical challenges can postpone tumor board decisions and delay patient treatment, potentially affecting clinical outcomes. Therapeutic Assistance and Decision algorithms for hepatobiliary tumor Boards (ADBoard) aims to reduce this delay by providing automated data extraction and high-quality, evidence-based treatment recommendations. METHODS AND ANALYSIS: With the help of natural language processing, relevant patient information will be automatically extracted from electronic medical records and used to complete a classic tumor conference protocol. A machine learning model is trained on retrospective MDM data and clinical guidelines to recommend treatment options for patients in our inclusion criteria. Study participants will be randomized to either MDM with ADBoard (Arm A: MDM-AB) or conventional MDM (Arm B: MDM-C). The concordance of recommendations of both groups will be compared using interrater reliability. We hypothesize that the therapy recommendations of ADBoard would be in high agreement with those of the MDM-C, with a Cohen's kappa value of ≥ 0.75. Furthermore, our secondary hypotheses state that the completeness of patient information presented in MDM is higher when using ADBoard than without, and the explainability of tumor board protocols in MDM-AB is higher compared to MDM-C as measured by the System Causability Scale. DISCUSSION: The implementation of ADBoard aims to improve the quality and completeness of the data required for MDM decision-making and to propose therapeutic recommendations that consider current medical evidence and guidelines in a transparent and reproducible manner. ETHICS AND DISSEMINATION: The project was approved by the Ethics Committee of the Charité - Universitätsmedizin Berlin. REGISTRATION DETAILS: The study was registered on ClinicalTrials.gov (trial identifying number: NCT05681949; https://clinicaltrials.gov/study/NCT05681949 ) on 12 January 2023.

来源:Trials,查看原文。版权归原作者所有。

引用

  1. Ng SST, Oehring R, Ramasetti N, et al. Concordance of a decision algorithm and multidisciplinary team meetings for patients with liver cancer-a study protocol for a randomized controlled trial. Trials. 2023. PMID: 37684688. doi:10.1186/s13063-023-07610-8

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。