Implicit Bias in the Evidence: An Evaluation of Female-Predominant Disease

本研究探讨了在诊断决策的条件下，没有使用诊断学 明确的诊断测试，并依赖于一套诊断标准。因为临床医生天生就是贝叶斯， 认知概率学是用来估计一个给定的条件的基础上的代表性的概率 在诊断决策中，这种情况属于“平均水平”。随着时间的推移，对这些捷径的依赖 病人是如何被诊断的，反过来，这些病人增加了越来越多的证据基础。这可能导致 循证医学实践中的多状态确认偏差。我们评估这个综合体的第一步 循环是质疑证据和偏离规范如何影响临床医生给出的诊断。 我们要问：当病人的性别、性别或种族不同时，他们得到的诊断是否也不同？我们最初 试点表明，临床医生在与性别/种族不同的患者接触时会犯更多的诊断错误， 不太常见，尽管狼疮的临床体征和症状相同。我们将继续我们的实验 评估狼疮诊断中的内隐偏见的方法，该方法建立在对 招聘。这项工作特别询问，当疾病表现相同时，患者是否偏离了 基于证据的“规范”，影响诊断？我们将考虑增加女性占主导地位的 疾病，并确定医生的诊断是否受到患者的性别/性别/种族持有的影响 其他一切都是不变的。首先，我们将在全国范围内进行一项基于互联网的随机实验， 专家评估性别/性别/种族在临床检查中的作用及其对观察到的 异质性和差异。其次，我们将在初级保健中进行额外的随机实验， 确定性别/种族如何影响他们的临床检查和转诊模式。第三，我们将使用混合 表征和提高我们对临床医生如何处理诊断问题的理解的方法，以及 查明造成健康差距的非生物性、可能改变的因素。我们将集中 小组与提供者使用实验结果作为讨论提示， 解决这种多状态确认偏差的策略。这项创新工作采用了一种新的方法， 结合社会学、实验设计和行为科学的方法。我们的发现会增加 临床医生对认知偏差如何导致医疗错误的认识，并将展示如何 不完整证据基础可能扩大健康差异，导致诊断延误，疾病恶化， 不可逆转的伤害。

This research explores the use of heuristics in diagnostic decision making for conditions that do not have definitive diagnostic tests and rely on a set of diagnostic criteria. Because clinicians are Bayesian by nature, cognitive heuristics are used to estimate the probability of a given condition based on the representativeness of that case among “the average” in diagnostic decision making. Over time, reliance on these shortcuts influences how patients are diagnosed, and in turn, these patients add to the growing evidence base. This may lead to a multi-state confirmation bias in the practice of evidence-based medicine. Our first step to evaluate this complex cycle is to question how the evidence and deviations from the norm influence what diagnosis a clinician gives. We ask: Does the diagnosis a patient receives vary when all that differs is their sex, gender, or race? Our initial pilot shows that clinicians make more diagnostic errors when presented with patients of sex/gender/race that is less common, despite identical clinical signs and symptoms of lupus. We will extend on our experimental approach evaluating implicit bias in the diagnosis of lupus, which builds off of research on discrimination in hiring. This work specifically asks whether when the disease presents the same, does a patient deviating from the “norm” based on the evidence, influence the diagnosis? We will consider additional female-predominant diseases and determine whether a physician's diagnosis is influenced by a patient's sex/gender/race holding everything else constant. First, we will conduct a national internet-based randomized experiment among specialists to assess the role of sex/gender/race in the clinical workup and its contribution to observed heterogeneity and disparities. Second, we will conduct additional randomized experiments in primary care to determine how sex/gender/race influences their clinical workup and referral patterns. Third, we will use mixed methods to characterize and improve our understanding of how clinicians approach diagnostic questions, and identify non-biological, potentially modifiable factors contributing to health disparities. We will conduct focus groups with providers using experimental results as discussion prompts to contextualize findings and inform strategies to address this type of multi-state confirmation bias. This innovative work adapts a novel method by combining methods from sociology, experimental design, and behavioral science. Our findings will increase clinicians' awareness of how cognitive biases that may lead to medical errors, and will demonstrate how an incomplete evidence base may propagate health disparities leading to delayed diagnosis, worsening disease, and irreversible damage.