Testing the accuracy of eye tracking as a screening tool for ASD in the general population

ASD筛查领域正处于十字路口：最流行的筛查工具敏感度仅为33%-- 38%的1，3，儿科医生一致认为只有大约三分之一的儿童未能通过筛查工具进行检查 评价4、5--称对筛查结果缺乏信心是不转诊的主要原因5。在这个范围内 背景下，ASD诊断和最终治疗的平均年龄保持在~52岁并不完全令人惊讶 6个月--该疾病发生于产前7岁之后，超过了可以可靠诊断的年龄 在许多情况下。显然，需要测试新的方法。眼睛跟踪，它产生与生物相关的， 社交和非社交视觉注意模式的客观和可量化的衡量标准是一种能够 作为一种工具，极有希望极大地改变筛查的实施方式。在NIH的帮助下 资助，我们开发了6种新颖的眼球跟踪测试，针对ASD儿童的关键挑战领域，包括 视觉社交注意、视线转移和听觉社交注意。利用我们的大型传统眼球跟踪 从跨越多个诊断组的2,000名幼儿收集的数据集，包括ASD、非ASD延迟和 TD我们为每项测试确定了最优的眼球跟踪指标和截止值，结果是非常高的 特异性和PPV(~97%和~90%)，但敏感性不高(每次检测~20%)。结合所有6项测试 然而，显著提高了灵敏度(~90%)并导致了高分类精度(AUC.95)。这些 然而，这些发现是在实验室环境中证明的，而在真实世界的临床环境中的实用性尚不清楚。 因此，在目标1中，我们采取了大胆的步骤，测试是否对8,000名12、18和24岁的人实施眼球跟踪 每月良好的婴儿体检(从5,2000名独特的学步儿童开始)，为来自不同种族、民族、 当儿科办公室的医务人员实施SES回访时，可以改善ASD的早期筛查。 使用研究人员定义的标准未能通过眼睛跟踪的幼儿，以及通过的百分比，将由 对眼球跟踪分数和眼球跟踪诊断分类准确性视而不见的执业心理学家 计算出来的。我们还将研究眼球跟踪特征和临床表型之间的关系。为了 充分了解眼球跟踪作为筛查工具的准确性，对整个筛查结果进行诊断 需要确定队列。因此，在AIM 2中，将利用电子健康记录(EHR)使我们能够 不仅确定眼球跟踪检测ASD的真实灵敏度、特异度、PPV和NPV，而且还 将结果与使用CSBS的ASD检测率进行比较，CSBS是用作标准的父母报告筛查工具 在圣地亚哥的护理作为我们的GET早期模式的一部分9。最先进的生物信息学将使我们能够进一步 确定眼球跟踪与家长报告相结合是否比单独使用其中一种方法更好。统计 建模将揭示筛查年龄、性别、种族、种族或SES等因素是否会影响眼睛 追踪分数。最后，在AIM 3中，儿科医生和家长将对他们对眼球跟踪的满意度进行评分。

The field of ASD screening is at a crossroads: the sensitivity of the most popular screening tool is only 33%- 38%1,3, and pediatricians consistently refer only about a third of children who fail a screening tool for an evaluation4,5 - citing a lack of confidence in screening results as the primary reason for non-referral5. Within this context, it is not entirely surprising that the mean age of ASD diagnosis and eventual treatment remains at ~52 months6 - years beyond the disorder’s prenatal origins7, and beyond the age when it can be reliably diagnosed in many cases8. Clearly, new approaches need to be tested. Eye-tracking, which generates biologically-relevant, objective, and quantifiable metrics of social and non-social visual attention patterns, is a technology that holds considerable promise as a tool to dramatically change how screening is implemented. With the help of NIH funding, we developed 6 novel eye tracking tests that tap into key challenge areas for children with ASD including visual social attention, gaze shifting, and auditory social attention. Leveraging our large legacy eye tracking dataset collected from >2,000 toddlers spanning multiple diagnostic groups including ASD, non-ASD delay, and TD we determined optimal eye tracking metrics and cut-off values across for each test that result in very high specificity and PPV (~97% & ~90%) but modest sensitivity (~20% per test). Combining across all 6-tests however, dramatically improves sensitivity (~90%) and results in high classification accuracy (AUC .95). These findings, however, were demonstrated in a laboratory setting with utility in real-world clinical settings unknown. Thus, in AIM 1, we take the bold step of testing whether eye-tracking administered across 8,000 12, 18, & 24 month well-baby check-ups (from ~5,2000 unique toddlers) serving families from a wide range of racial, ethnic, and SES backrounds can improve ASD early screening when implemented by medical staff in pediatric offices. Toddlers who fail eye-tracking using resarcher defined criteria, and a percentage who pass, will be evaluated by a licensed psychologist blind to eye tracking scores, and diagnostic classification accuracy of eye-tracking computed. Relationships between eye tracking profiles and clinical phenotype will also be examined. In order to fully understand the accuracy of eye tracking as a screening tool, diagnostic outcomes of the entire screened cohort needs to be determined. Thus, in AIM 2 electronic health records (EHRs) will be leveraged to allow us to not only determine the true sensitivity, specificity, PPV and NPV of eye tracking for detecting ASD, but to also compare results to rates of ASD detection using the CSBS, a parent report screening tool used as standard of care in San Diego as part of our Get SET Early model9. State of-the-art bioinformatics will allow us to further determine if combining eye-tracking with parent report is superior relative to either approach alone. Statistical modeling will reveal whether or not factors such as age at screening, sex, race, ethnicity or SES impacts eye tracking scores. Finally in AIM 3, pediatricians and parents will rate their satisfaction with eye tracking.