Richie Megan, Josephson S Andrew
a Department of Neurology , University of California San Francisco , San Francisco , California , USA.
Teach Learn Med. 2018 Jan-Mar;30(1):67-75. doi: 10.1080/10401334.2017.1332631. Epub 2017 Jul 28.
Construct: Authors examined whether a new vignette-based instrument could isolate and quantify heuristic bias.
Heuristics are cognitive shortcuts that may introduce bias and contribute to error. There is no standardized instrument available to quantify heuristic bias in clinical decision making, limiting future study of educational interventions designed to improve calibration of medical decisions. This study presents validity data to support a vignette-based instrument quantifying bias due to the anchoring, availability, and representativeness heuristics.
Participants completed questionnaires requiring assignment of probabilities to potential outcomes of medical and nonmedical scenarios. The instrument randomly presented scenarios in one of two versions: Version A, encouraging heuristic bias, and Version B, worded neutrally. The primary outcome was the difference in probability judgments for Version A versus Version B scenario options.
Of 167 participants recruited, 139 enrolled. Participants assigned significantly higher mean probability values to Version A scenario options (M = 9.56, SD = 3.75) than Version B (M = 8.98, SD = 3.76), t(1801) = 3.27, p = .001. This result remained significant analyzing medical scenarios alone (Version A, M = 9.41, SD = 3.92; Version B, M = 8.86, SD = 4.09), t(1204) = 2.36, p = .02. Analyzing medical scenarios by heuristic revealed a significant difference between Version A and B for availability (Version A, M = 6.52, SD = 3.32; Version B, M = 5.52, SD = 3.05), t(404) = 3.04, p = .003, and representativeness (Version A, M = 11.45, SD = 3.12; Version B, M = 10.67, SD = 3.71), t(396) = 2.28, p = .02, but not anchoring. Stratifying by training level, students maintained a significant difference between Version A and B medical scenarios (Version A, M = 9.83, SD = 3.75; Version B, M = 9.00, SD = 3.98), t(465) = 2.29, p = .02, but not residents or attendings. Stratifying by heuristic and training level, availability maintained significance for students (Version A, M = 7.28, SD = 3.46; Version B, M = 5.82, SD = 3.22), t(153) = 2.67, p = .008, and residents (Version A, M = 7.19, SD = 3.24; Version B, M = 5.56, SD = 2.72), t(77) = 2.32, p = .02, but not attendings.
Authors developed an instrument to isolate and quantify bias produced by the availability and representativeness heuristics, and illustrated the utility of their instrument by demonstrating decreased heuristic bias within medical contexts at higher training levels.
构建：作者研究了一种基于病例 vignette 的新工具是否能够分离并量化启发式偏差。
启发式思维是认知捷径，可能会引入偏差并导致错误。目前没有标准化工具可用于量化临床决策中的启发式偏差，这限制了旨在改善医学决策校准的教育干预措施的未来研究。本研究提供了效度数据，以支持一种基于病例 vignette 的工具来量化因锚定、可得性和代表性启发式思维而产生的偏差。
参与者完成问卷，要求为医疗和非医疗场景的潜在结果分配概率。该工具以两种版本之一随机呈现场景：版本 A，鼓励启发式偏差；版本 B，措辞中立。主要结果是版本 A 与版本 B 场景选项的概率判断差异。
在招募的 167 名参与者中，139 名登记入组。参与者为版本 A 场景选项分配的平均概率值（M = 9.56，标准差 = 3.75）显著高于版本 B（M = 8.98，标准差 = 3.76），t(1801) = 3.27，p = 0.001。仅分析医疗场景时，该结果仍然显著（版本 A，M = 9.41，标准差 = 3.92；版本 B，M = 8.86，标准差 = 4.09），t(1204) = 2.36，p = 0.02。按启发式思维分析医疗场景时，发现版本 A 和版本 B 在可得性方面存在显著差异（版本 A，M = 6.52，标准差 = 3.32；版本 B，M = 5.52，标准差 = 3.05），t(404) = 3.04，p = 0.003，在代表性方面也存在显著差异（版本 A，M = 11.45，标准差 = 3.12；版本 B，M = 10.67，标准差 = 3.71），t(396) = 2.28，p = 0.02，但在锚定方面不存在显著差异。按培训水平分层，学生在版本 A 和版本 B 医疗场景之间保持显著差异（版本 A，M = 9.83，标准差 = 3.75；版本 B，M = 9.00，标准差 = 3.98），t(465) = 2.29，p = 0.02，但住院医师和主治医师不存在显著差异。按启发式思维和培训水平分层，可得性在学生（版本 A，M = 7.28，标准差 = 3.46；版本 B，M = 5.82，标准差 = 3.22）和住院医师（版本 A，M = 7.19，标准差 = 3.24；版本 B，M = 5.56，标准差 = 2.72）中保持显著差异，t(153) = 2.67，p = 0.008，t(77) = 2.32，p = 0.02，但主治医师不存在显著差异。
作者开发了一种工具来分离并量化由可得性和代表性启发式思维产生的偏差，并通过展示在更高培训水平的医疗环境中启发式偏差的减少来说明该工具的效用。
