Leveraging Novel Sources of Data with Analytic Morphomics to Improve Delivery of Specialty Care to At-Risk Veterans with Liver Disease

背景：慢性肝病(CLD)是一组常见的、代价高昂的、临床上 退伍军人中日益普遍的随之而来的疾病。慢性阻塞性肺疾病的患者有 通常会转介到专科护理以进行进一步评估。这样的转诊可以改善临床 结果，但由于资源的限制，普遍转介是不可行的。此外，普遍的 特殊护理转介不仅不切实际，而且也没有必要。大多数慢性阻塞性肺病患者会 有几年到几十年的稳定疾病(预后良好)，而其他人将迅速进展 至肝纤维化和肝硬变(预后不良)。鉴于有大量退伍军人 CLD，我们需要一种系统和可靠的方法来识别预后不良的患者 临床结果不佳的风险增加(因此将受益于早期 专科护理转介)。识别预后不良的患者是具有挑战性的，因为 大多数慢性阻塞性肺病患者缺乏临床体征和症状；然而，体征通常是 在放射成像上可检测到。因为这些放射学发现本质上是定性的 它们被编码在成像数据中，传统上临床用途有限。我们建议 利用一种新技术--分析形态组学--来利用这种表型数据。分析 形态组学利用高吞吐量的计算图像处理算法来提供 对器官和身体组织进行精确和详细的测量。在计算机断层扫描中 (CT)扫描，大量关于患者表型的数据在很大程度上被忽略了， 未使用过的。通过从这些扫描中数字提取和量化分析结构数据，我们 假设我们可以确定预后不良的患者。 目标：(1)利用分析形态组学改进和验证退伍军人风险预测模型 慢性肝病；(2)提高分析的吞吐量和自动化 基于机器学习的形态组学图像处理算法；(3)临床量化 基于风险的专科护理诊所分诊策略(使用预测模型)与 标准分诊策略(通常护理)，使用模拟建模。 方法：本研究将使用先进的定量方法，包括分析形态组学和 深度学习(机器学习的一种)。目标1将改进和验证风险预测模型 分析形态组学在退伍军人CLD中的应用。目标2将考察深度学习在以下方面的作用 提高AIM 1中使用的图像处理算法的吞吐量和自动化程度。 最后，目标3将量化基于风险的特殊护理诊所分诊策略的临床影响。 (使用预测模型)与标准分诊策略(通常护理)进行比较。 影响：退伍军人健康管理局(VHA)处于一个独特的位置，可以定期联系- 收集重要的临床结果的影像数据。这项研究将为下一步的 在VHA中使用此当前未得到充分利用的数据源。形态数据的使用具有潜在的 不仅要改善对慢性阻塞性肺病患者的护理，还要改善对各种疾病的护理。 下一步：调查结果将告知我们的合作伙伴基于风险的分诊的潜在临床影响 并为今后实施这一办法奠定了基础。

Background: Chronic liver diseases (CLD) are a group of common, costly, and clinically consequential disorders that are increasingly prevalent in Veterans. Patients with CLD are typically referred to specialty care for further evaluation. Such referrals can improve clinical outcomes, but universal referral is not feasible due to resource limitations. Moreover, universal specialty care referral is not only impractical – it is also unnecessary. Most patients with CLD will have stable disease for years to decades (favorable prognosis), while others will rapidly progress to liver fibrosis and cirrhosis (unfavorable prognosis). Given the vast number of Veterans with CLD, we need a way to systematically and reliably identify patients with an unfavorable prognosis who are at increased risk for poor clinical outcomes (and would therefore benefit from early specialty care referral). Identifying patients with an unfavorable prognosis is challenging due to a lack of clinical signs and symptoms in most patients with CLD; however, signs are routinely detectable on radiological imaging. Because these radiologic findings are qualitative in nature and encoded within imaging data, they have traditionally been of limited clinical utility. We propose to utilize a novel technology, analytic morphomics, to leverage this phenotypic data. Analytic morphomics utilizes high throughput computational image processing algorithms to provide precise and detailed measurements of organs and body tissues. Within computed tomography (CT) scans, an immense amount of data about patient phenotypes has been largely ignored and unused. By digitally extracting and quantitatively analyzing structural data from these scans, we hypothesize that we can identify patients with an unfavorable prognosis. Objectives: (1) to refine and validate risk prediction models using analytic morphomics in Veterans with chronic liver disease; (2) to increase the throughput and automation of the analytic morphomics image processing algorithms using machine learning; (3) to quantify the clinical impact of a risk-based specialty care clinic triage strategy (using prediction models) versus a standard triage strategy (usual care), using simulation modeling. Methods: This study will use advanced quantitative methods including analytic morphomics and deep learning (a type of machine learning). Aim 1 will refine and validate risk prediction models using analytic morphomics in Veterans with CLD. Aim 2 will examine the role of deep learning to increase the throughput and automation of the image processing algorithms used in Aim 1. Finally, Aim 3 will quantify the clinical impact of a risk-based specialty care clinic triage strategy (using prediction models) versus a standard triage strategy (usual care). Impacts: The Veterans Health Administration (VHA) is in a unique position to link routinely- collected imaging data to important clinical outcomes. This study will lay the groundwork for the use of this currently underused data source in VHA. The use of morphomic data has the potential to improve patient care for not only CLD, but also a variety of conditions. Next Steps: Findings will inform our partners about the potential clinical impact of risk-based triage for specialty care and lay the groundwork for future efforts to implement such an approach.