Inflammatory and ADRD Biomarker Predictors of Perioperative Digital Clock Drawing

摘要 术前认知障碍在准备手术的老年人中很常见。尽管越来越多 有证据表明，术前认知/神经元完整性是围手术期损伤和术后 手术不良结果，医疗保健系统没有系统地术前筛查认知。 临床研究人员尚未确定一种实用的方法来进行术前认知筛查。我们的团队 成员们开发了数字时钟绘制测试（dCDT），这是一种捕捉微妙行为的工具。 快速（5分钟）时钟绘制评估期间的变量。该工具提供的数据和好处 还有待于在围手术期背景下加以考虑。我们将在大量的预处理中应用dCDT， 手术患者（n= 5，000/年）与新型机器学习算法相结合，以解决三个特定的 目标。目的1：探讨老年人术前神经认知功能障碍的范围和分布 术前患者相对于非手术老年人口统计学匹配的同龄人（可用n= 2，400 通过NIH/波士顿大学心脏研究）使用新的先前未观察到的dCDT图 和决策变量;目的2：检查术前dCDT变量对 术后，临床医生报告/医院记录的事件;目标3：检查术前至术后6周，3- 胸外科（n=70）、骨科（n = 70）、重大手术的dCDT和NIH PROMIS指标的1个月和1年变化 腹盆腔患者（n=70）和非手术同行（n=70）。对于观察性研究（目标1和2）， > 65岁的个人到UFHealth术前诊所将完成dCDT以及三个- 单词记忆测试和虚弱评估作为标准临床评估的一部分。手术和麻醉 将通过电子病历获取详细信息。临床相关结局将包括 并发症、住院时间、护理费用、功能能力和死亡率。成果将得到补充 由一个单独的临床研究亚组（Aim 3）在6周、3个月、 术后1年。分析将集中在分层分布和集群的dCDT特征 在众多的社会人口统计学、手术和麻醉因素中。dCDT的预测值为 相对于临床结果建模。将比较dCDT和基线NIH PROMIS域的变化 术前和术后，并检查与纵向围手术期事件的相互作用。Subaims：我们 将“深度学习”方法应用于绘画，以识别手术前患者相对于 通过Frachial Heart研究获得的人口统计学等同dCDT数据点的大样本。 基于符号聚合近似（SAX）的机器学习方法将表征交互 术前dCDT特征和术中麻醉敏感性之间的关系。

ABSTRACT Preoperative cognitive impairment is common among older adults preparing for surgery. Despite growing evidence that preoperative cognitive/neuronal integrity is a risk factor for perioperative insults and post- operative adverse outcomes, health care systems do not systematically pre-operatively screen for cognition. Clinical researchers have yet to identify a pragmatic approach to pre-operative cognitive screening. Our team members have developed the digital Clock Drawing Test (dCDT), a tool that captures subtle behavioral variables during a rapid (5-minute) clock drawing assessment. The data and benefit afforded by this tool have yet to be considered across perioperative contexts. We will apply the dCDT within a large number of pre- surgical patients (n=5,000 per year) coupled with novel machine learning algorithms to address three specific aims. Aim 1: examine range and distribution of preoperative neurocognitive impairment with older adult preoperative patients relative to non-surgical older adult demographically matched peers (available n=2,400 via NIH/Boston University Framingham Heart Study) using novel previously unobserved dCDT graphomotor and decision making variables; Aim 2: examine the predictive validity of presurgical dCDT variables on postoperative, clinician reported/hospital recorded events; Aim 3: examine pre to postoperative 6-week, 3- month, and 1-year change in dCDT and NIH PROMIS metrics for thoracic (n=70), orthopedic (n=70), major abdominal-pelvic patients (n=70), and non-surgery peers (n=70). For the observational studies (Aim 1 and 2), individuals > 65 years presenting to the UFHealth presurgical clinic will complete the dCDT as well as a three- word memory test and frailty assessment as part of the standard clinical evaluation. Surgical and anesthetic details will be acquired via the electronic medical record. Clinically-relevant outcomes will include complications, length of stay, cost of care, functional capacity, and mortality. Outcomes will be supplemented by a separate longitudinally-studied subgroup (Aim 3) completing NIH PROMIS metrics at 6 weeks, 3 months, and 1year after surgery. Analyses will focus on stratifying distributions and clusters of dCDT characteristics across numerous sociodemographic, surgical, and anesthetic factors. The predictive value of the dCDT will be modeled relative to clinical outcomes. Changes in dCDT and baseline NIH PROMIS domains will be compared pre- and post-operatively and examined for interactions with longitudinal perioperative events. Subaims: We will apply `deep learning' approaches to drawings to identify novel features of pre-surgical patients relative to a large sample of demographically equated dCDT data points available through the Framingham Heart Study. Symbolic aggregate approximation (SAX)-based machine learning approaches will characterize interactions between preoperative dCDT features and intraoperative anesthetic sensitivities.