多源三角互证法为临床实践指南设定决策阈值:加拿大胃肠病学会共识报告
该研究开发并测试了一种三角互证协议,用于在GRADE证据到决策框架中预先设定决策阈值。研究以加拿大胃肠病学会结直肠息肉切除术后监测指南为应用场景,整合四类证据流并经改良德尔菲共识(≥80%同意)确定阈值。结果为结直肠癌诊断、死亡率及主要出血和穿孔等结局设定了绝对风险差异阈值。
要点
- 提出四源三角互证协议:既往指南基准、试验最小重要差异与模型大效应、经验衍生通用系数、专家启发。
- 经改良德尔菲共识(≥80%同意)确定最终决策阈值。
- 为结直肠癌诊断设定小、中、大效应阈值分别为5、13、25/1000人10年。
核心结论
- 该三角互证协议为预先设定决策阈值提供了透明、可操作的方法学,弥补了指南制定中的关键空白。
- 该方法强化了GRADE证据到决策过程中价值敏感的关键步骤,但过程资源密集且阈值可能依赖背景和专家驱动。
研究设计与样本
- 方法学框架在加拿大胃肠病学会结直肠息肉切除术后监测指南中开发并应用,比较不同监测间隔,有多学科专家小组参与。
- 决策阈值通过整合四类证据流得出:既往指南基准、试验最小重要差异与模型大效应、经验衍生通用系数、专家启发。
- 采用改良德尔菲技术达成共识(≥80%同意)确定最终决策阈值。
主要发现
- 为比较不同监测间隔的结局生成了一组决策阈值(每1000人10年绝对风险差异)。
- 结直肠癌诊断的小、中、大效应阈值分别为5、13、25/1000。
- 结直肠癌死亡率的小、中、大效应阈值分别为2、5、10/1000。
- 主要出血和穿孔的小效应阈值分别为15/1000和7/1000。
- 过程导致将「诊断时结直肠癌分期」排除为结局,但因对其作为结局的适宜性存在争议,未就进展期腺瘤达成共识。
关键数值与推荐
- 结直肠癌诊断:小效应阈值5/1000,中效应13/1000,大效应25/1000(每1000人10年绝对风险差异)。
- 结直肠癌死亡率:小效应阈值2/1000,中效应5/1000,大效应10/1000。
- 主要出血小效应阈值15/1000,穿孔小效应阈值7/1000。
- 共识定义为改良德尔菲技术≥80%同意。
临床怎么用
- 指南小组可参照该三角互证协议,为优先结局预先设定决策阈值,减少主观判断偏倚。
- 应用时需整合本地背景证据流,并经多学科专家共识确认阈值。
局限性
- 过程资源密集。
- 阈值可能依赖具体背景且由专家驱动。
- 未就进展期腺瘤作为结局达成共识。
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 AND OBJECTIVES: Within the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) Evidence-to-Decision (EtD) framework, guideline panels must judge the magnitude of health effects (trivial, small, moderate, and large). Current implicit judgments are subjective and susceptible to bias. While GRADE advocates for predefined decision thresholds (DTs), a practical framework to derive them is lacking. We aimed to develop and test a triangulation protocol for establishing a priori DTs for guideline outcomes. STUDY DESIGN AND SETTING: The methodological framework was developed and applied within a Canadian Association of Gastroenterology guideline on postpolypectomy surveillance, comparing different surveillance intervals, with input from a multidisciplinary expert panel. DTs were derived by integrating 4 evidence streams for prioritized outcomes: (1) benchmarks from prior guidelines, (2) trial-based minimally important differences and model-based large effects, (3) empirically derived generic coefficients, and (4) expert elicitation. A consensus process using a modified Delphi technique (≥80% agreement) established the final DTs. RESULTS: A set of DTs (absolute risk differences per 1000 people over 10 years) was generated for outcomes comparing different surveillance intervals. For colorectal cancer (CRC) diagnosis, thresholds for small, moderate, and large effects were 5, 13, and 25 per 1000. For CRC mortality: 2, 5, and 10 per 1000. For major bleeding and perforation, the small effect thresholds were 15 and 7 per 1000, respectively. The process led to excluding "CRC stage at diagnosis" as an outcome but no consensus on advanced adenoma due to debate about its suitability as an outcome. CONCLUSION: This triangulation protocol provides a transparent, operational methodology for setting a priori DTs, addressing a key gap in guideline development and strengthening a critical, value-sensitive step in the GRADE EtD process. Although the process is resource intensive and the thresholds may be context dependent and expert driven, the framework offers a structured, replicable approach that enhances rigor and transparency in one of the most subjective steps of guideline development.
引用
- Tse F, Forbes N, Singh H, et al. A novel multi-stream triangulation method to define decision thresholds for clinical practice guidelines: a Canadian Association of Gastroenterology consensus report. J Clin Epidemiol. 2026. PMID: 42537857. doi:10.1016/j.jclinepi.2026.112437
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。