文献速递贫血与红细胞疾病

人工智能学习与粪便免疫化学检测在贫血患者结肠镜检查优先级排序中的潜在作用

Potential roles of artificial intelligence learning and faecal immunochemical testing for prioritisation of colonoscopy in anaemia.
PubMed 文献 · Br J Haematol · 2019-02-03

研究在普利茅斯胃肠科门诊的592例贫血患者中评估粪便免疫化学检测(FIT)与人工智能学习算法(ColonFlag™)对结肠镜检查优先级排序的作用,并与伦敦因缺铁性贫血行结肠镜检查的队列结合分析。592例中21例(3.5%)检出结直肠癌;使用ColonFlag™而非血红蛋白浓度结合症状可将紧急转诊优先级人数从592例降至304例,可能漏诊1例结直肠癌但或可通过FIT检出。在<55岁人群中,紧急转诊优先级人数从109例降至15例且无癌症漏诊。

要点

  • 592例贫血患者中21例(3.5%)检出结直肠癌。
  • 用ColonFlag™替代血红蛋白浓度结合症状,紧急转诊优先级从592例降至304例,可能漏诊1例结直肠癌。
  • 在<55岁人群中,紧急转诊优先级从109例降至15例,无癌症漏诊。

核心结论

  • FIT在缺铁性贫血中具有高阴性预测值,其使用可能使部分患者免于检查。
  • 人工智能学习可能比血红蛋白浓度更适合作为紧急结肠镜检查转诊的触发因素。

研究设计与样本

  • 研究在普利茅斯一家胃肠科门诊对贫血患者进行FIT检测并评估人工智能学习算法(ColonFlag™),同时结合伦敦因缺铁性贫血行结肠镜检查的队列进行分析。
  • 普利茅斯队列共592例因血红蛋白浓度转诊的患者。

主要发现

  • 592例因血红蛋白浓度转诊的患者中,21例(3.5%)检出结直肠癌。
  • 使用ColonFlag™而非血红蛋白浓度结合症状,紧急转诊优先级为304例而非592例,可能漏诊1例结直肠癌,但该例或可通过FIT检出,而此例中FIT不可用。
  • 在<55岁且结直肠癌发病率低的人群中,紧急转诊优先级为15例而非109例,无癌症漏诊。

关键数值与推荐

  • 592例中21例(3.5%)检出结直肠癌。
  • 使用ColonFlag™结合症状时紧急转诊优先级为304例,使用血红蛋白浓度结合症状时为592例。
  • 在<55岁人群中,使用ColonFlag™时紧急转诊优先级为15例,使用血红蛋白浓度时为109例,无癌症漏诊。

临床怎么用

  • FIT在缺铁性贫血中具有高阴性预测值,其使用可能使部分患者免于检查。
  • 人工智能学习可能比血红蛋白浓度更适合作为紧急结肠镜检查转诊的触发因素。

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

Iron deficiency anaemia (IDA) is the most common cause of anaemia and a frequent indication for colonoscopy, although the prevalence of colorectal cancer (CRC) in IDA is low. Measurement of faecal haemoglobin by immunochemical techniques (FIT) is used to detect symptomatic patients. We studied FIT in patients with anaemia attending a gastroenterology clinic in Plymouth and looked at an artificial intelligence (AI) learning algorithm (ColonFlag™) in these patients, together with a cohort who had undergone colonoscopy for IDA in London. Of 592 patients referred on the basis of haemoglobin concentration, 21 (3.5%) had CRC. Using ColonFlag™, rather than haemoglobin concentration, in combination with symptoms, would have resulted in prioritisation of 304 patients for urgent referral rather than 592. One CRC would have been missed but might have been detected by FIT, which was not available in this case. In patients aged <55 years in whom the incidence of CRC is low, 15 rather than 109 patients would have been prioritised for urgent referral with no cancers missed. FIT has a high negative predictive value in IDA so its use may enable some patients to avoid investigation and AI learning may be a more useful trigger than haemoglobin concentration for urgent referral for colonoscopy.

来源:Br J Haematol,查看原文。版权归原作者所有。

引用

  1. Ayling RM, Lewis SJ, Cotter F, et al. Potential roles of artificial intelligence learning and faecal immunochemical testing for prioritisation of colonoscopy in anaemia. Br J Haematol. 2019. PMID: 30714125. doi:10.1111/bjh.15776

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。