Reducing false positives during interpretations of ultrasound examinations for breast cancer screening

 描述（由申请人提供）：每年筛查乳腺癌的早期发现是一种广泛接受的做法，但在美国仍然存在争议，因为其可归因的“收益”和“成本”都在不断审查。一个主要的问题是大量的假阳性解释和大量的良性活检正在进行。在美国，每年约有500万名妇女被召回进行诊断检查，只有大约二十分之一的人被发现患有癌症。目前针对尚不清楚是否处于高风险的女性的做法包括全视野数字乳腺X射线摄影（FFDM）和数字乳腺断层合成摄影（DBT）。这两种做法都需要电离辐射，并且在检测浸润性癌症的灵敏度方面不是最佳的，特别是在占所有筛查妇女约40%-45%的致密乳房妇女中。全乳腺超声（WBUS），无论是手持式还是自动化，都不需要辐射暴露，并且对发现早期浸润性癌症更敏感，但在今天的手术中，也会导致更高的假阳性率（~ 1.5倍）。尽管我们多次尝试降低所有模式的召回率，但在很大程度上都失败了。大部分问题可能源于对解释放射科医师没有参考信息的事实（例如，“CAD”的负面情况或“第二意见”），这将提高他/她的信心，什么“不记得”。欧洲人通过练习双重阅读来解决假阳性，这导致两个阅读者中的一个最初得分为阳性的回忆检查减少了大约50%。越大 单个阅片者的假阳性率为，观察者间变异性越大，降低幅度越大。然而，这种做法在美国并不可行。因此，我们建议评估一个简单的，具有成本效益的，修改后的方法是否会影响放射科医生在超声检查的解释。为了测试我们的概念，我们建议首先进行一项双模式完全平衡的回顾性观察者研究，其中经验丰富的放射科医生将解释在临床上实际召回的致密乳房（密度BIRADS 3或4更有可能被召回）女性的WBUS检查（证实癌症为阳性或阴性）。在第二阅读模式中，将从“低回忆”放射科医师向解读放射科医师给出独立的第二意见结果。迄今为止，这种可能的干预措施尚未得到研究，很容易证明这种方法是“具有成本效益的”（专业努力），并且可能会显着降低WBUS召回率，同时保持癌症检测率。因此，我们建议回顾性地检验这一假设，即在这种方法下，筛查超声检查将在统计学上显著降低被召回的可能性，从而也不太可能被推荐进行良性活检，无论潜在的性能水平的解释放射科医生的问题。我们的主要假设是，这种方法将导致召回率至少降低20%，而癌症检测没有（或最坏情况下只有最小的）损失。

 DESCRIPTION (provided by applicant): Annual screening for the early detection of breast cancer is a widely accepted practice, but it remains controversial in the United States in that both the attributable "benefit" and "cost" are continually scrutinized. One primary concern is the high number of false positive interpretations and the large number of benign biopsies being performed. There are approximately five million women being recalled for diagnostic workup in the United States each year and only approximately one in 20 are found to have cancer. Current practices for women not known to be at high risk include full field digital mammography (FFDM) and digital breast tomosynthesis (DBT). Both practices require ionizing radiation and are not optimal in terms of sensitivity for detecting invasive cancers, particularly in women with dense breasts constituting approximately 40%-45% of all screened women. Whole breast ultrasound (WBUS), whether hand held or automated, does not require radiation exposure and is significantly more sensitive to finding early invasive cancers, but also results, in today's practie, in even a higher false positive rate (~1.5X). Despite many attempts to reduce recall rates of all modalities, we have largely failed. Much of the problem may stem from the fact that there is no reference information to the interpreting radiologist (e.g., "CAD" for negative cases or a "second opinion") that would raise his/her confidence in what "not to recall". The Europeans address false positives by practicing double reading with consensus that results in approximately a 50% reduction in recalling examinations initially scored positive by one of the two readers. The larger the single reader false positive rate is and the larger the inter- observer variability, the largerthe reduction. However, this practice is not operationally feasible in the United States. Therefore, we propose to assess if a simple, cost effective, modified approach would affect radiologists during interpretations of ultrasound examinations. To test our concept, we propose to initially perform a two mode fully balanced retrospective observer study in which experienced radiologists will interpret WBUS examinations of women with dense breasts (density BIRADS 3 or 4 who are more likely to be recalled) that had been actually recalled in the clinic (positive an negative for verified cancers). In a second reading mode, the interpreting radiologists will be given independent second opinion results from "low recalling" radiologists. To date, this type of a possible intervention has not been investigated and it is easy to demonstrate that this approach is "cost effective" (professional effort) and could lead to a significant reduction in WBUS recall rates while maintaining cancer detection rates. Hence, we propose to retrospectively test the hypothesis that under this approach screening ultrasound examinations will be statistically significantly less likely to be recalled and thereby also less likely to be recommended to undergo benign biopsies, regardless of the underlying performance levels of the interpreting radiologists in question. Our primary hypothesis is that this approach, will resul in at least a 20% reduction in recall rates with no (or at worst a minimal) loss in cancer detection.