Plumb Andrew A, Nickerson Claire, Wooldrage Katherine, Bassett Paul, Taylor Stuart A, Altman Douglas, Atkin Wendy, Halligan Steve
Centre for Medical Imaging, Division of Medicine, University College London, London, United Kingdom.
NHS Cancer Screening Programmes, Fulwood House, Sheffield, United Kingdom.
Endoscopy. 2016 Oct;48(10):899-908. doi: 10.1055/s-0042-108727. Epub 2016 Jul 21.
Terminal digit preference bias for "pleasing" numbers has been described in many areas of medicine. The aim of this study was to determine whether endoscopists, radiologists, and pathologists exhibit such bias when measuring colorectal polyp diameters.
Colorectal polyp diameters measured at endoscopy, computed tomographic colonography (CTC), and histopathology were collated from a colorectal cancer screening program and two parallel multicenter randomized trials. Smoothing models were fitted to the data to estimate the expected number of polyps at 1-mm increments, assuming no systematic measurement bias. The difference between the expected and observed numbers of polyps was calculated for each terminal digit using statistical modeling. The impact of measurement bias on per-patient detection rates of polyps ≥ 10 mm was estimated for each modality.
A total of 92 124 individual polyps were measured by endoscopy (91 670 screening and 454 from trials), 2385 polyps were measured by CTC (1664 screening, 721 trials), and 79 272 were measured by histopathology (78 783 screening, 489 trials). Clustering of polyp diameter measurements at 5-mm intervals was demonstrated for all modalities, both in the screening program and the trials. The statistical models estimated that per-patient detection rates of polyps ≥ 10 mm were over-inflated by 2.4 % for endoscopy, 3.1 % for CTC, and 3.3 % for histopathology in the screening program, with similar trends in the randomized trials.
Endoscopists, radiologists, and pathologists all exhibit terminal digit preference when measuring colorectal polyps. This will bias trial data, referral rates for further testing, adenoma surveillance regimens, and comparisons between tests.
在医学的许多领域都已描述了对“讨人喜欢”数字的末位数字偏好偏差。本研究的目的是确定内镜医师、放射科医师和病理科医师在测量结直肠息肉直径时是否存在这种偏差。
从一项结直肠癌筛查项目以及两项平行的多中心随机试验中整理出在内镜检查、计算机断层结肠成像（CTC）和组织病理学检查中测量的结直肠息肉直径。对数据拟合平滑模型，以估计息肉以1毫米增量增加时的预期数量，假设不存在系统测量偏差。使用统计模型计算每个末位数字的息肉预期数量与观察数量之间的差异。估计了测量偏差对每种检查方式下息肉≥10毫米的患者检出率的影响。
通过内镜检查共测量了92124个单个息肉（91670个来自筛查，454个来自试验），通过CTC测量了2385个息肉（1664个来自筛查，721个来自试验），通过组织病理学测量了79272个息肉（78783个来自筛查，489个来自试验）。在筛查项目和试验中，所有检查方式下息肉直径测量值均以5毫米间隔聚类。统计模型估计，在筛查项目中，内镜检查下息肉≥10毫米的患者检出率高估了2.4%，CTC高估了3.1%，组织病理学高估了3.3%，在随机试验中也有类似趋势。
内镜医师、放射科医师和病理科医师在测量结直肠息肉时均表现出末位数字偏好。这将使试验数据、进一步检查的转诊率、腺瘤监测方案以及不同检查之间的比较产生偏差。
