Advanced Prediction Models to Optimize Treatment and Access for Veterans with Hepatitis C

背景 慢性丙型肝炎（CHC）是一个主要的公共卫生问题，目前困扰着超过25万名退伍军人。的 美国食品和药物管理局（FDA）最近批准了几种直接作用的抗病毒药物（DAA）， CHC的治疗与先前的基于干扰素的CHC治疗方案不同，DAA是高度有效的， 有利的安全性，并且耐受性良好，引起了显著的消费者需求。然而，DAA也 这使得医疗保健系统难以满足患者日益增长的需求。 医疗保健系统也缺乏足够数量的训练有素的提供者，以在一个月内治疗所有CHC患者。 时间短，进一步限制了访问。但对所有感染患者的立即治疗可能不仅 昂贵得令人望而却步--这也是不必要的。 CHC患者分为两大类：（1）晚期肝病（肝硬化，或“疤痕”） 或CHC的其他肝外表现（例如，肾脏疾病）（约占 （2）无晚期肝病或肝外表现者（约占75%）， 本提案的相关人群）。虽然数据和指导方针清楚地表明了 前一组的治疗（即，肝硬化患者），他们不太清楚后一组的益处 （无肝硬化者）。事实上，大多数这些无症状的非阿尔茨海默病患者多年来进展缓慢 几十年（低风险患者），因此可能不需要立即治疗。其他人进步更快 并能从立即治疗中获益然而，临床医生往往不确定如何接近 这样的病人。目前，非炎性CHC的治疗方法在医疗保健领域存在很大差异 这在很大程度上是由于在何时治疗这类患者的指导方面存在差异。因此，治疗是 通常由患者偏好、临床医生判断和药物可用性的组合驱动。的老兵 处于疾病进展的高风险，但不积极寻求治疗，因此可能无法接受潜在的 拯救生命的治疗此外，指导方针继续根据新的 毒品一种系统的，严格的治疗方法，一种由最先进的预测模型提供信息的方法 风险分层的非中风退伍军人，可以帮助指导基于风险的治疗和减轻这一点， 保健服务方面的缺陷。 目标 本研究的目的是为基于风险的非社区慢性丙型肝炎治疗奠定基础。 退伍军人健康管理局（VHA）的肝硬化退伍军人通过：（1）制定准确的、临床上的 相关的、可实施的风险预测模型;（2）让退伍军人就如何 实施基于风险的治疗;（3）评估基于风险的治疗的临床和经济效果。 方法 在我们的初步工作中，我们证明了使用机器学习（ML）风险预测模型的可行性 在临床试验队列中识别疾病进展的高风险和低风险患者。我们提出了一个4- 我们将使用2004-2014年VA电子数据来调整、验证和完善该模型 在退伍军人中。然后，我们将与退伍军人接触，了解他们对基于风险的 通过应用共识技术治疗CHC（例如，协商民主）。最后，估计 基于风险的治疗相对于当前治疗的增量效益，我们将使用模拟建模。

Background Chronic hepatitis C (CHC) is a major public health problem that currently afflicts over 250,000 Veterans. The Food and Drug Administration (FDA) recently approved several direct acting antiviral agents (DAAs) for the treatment of CHC. Unlike prior interferon-based treatment regimens for CHC, DAAs are highly effective, have a favorable safety profile, and are well tolerated, eliciting significant consumer demand. However, DAAs are also extremely costly, making it difficult for healthcare systems to meet this growing demand from patients. Healthcare systems also lack a sufficient number of trained providers to treat all patients with CHC within a short time, further limiting access. But immediate treatment of all infected patients may be not only prohibitively expensive – it is also unnecessary. Patients with CHC fall into two broad categories: (1) those with advanced liver disease (cirrhosis, or “scarring” of the liver) or other extrahepatic manifestations of CHC (e.g., kidney disease) (accounting for ~ 25% of patients); (2) those without advanced liver disease or extrahepatic manifestations (accounting for ~ 75% -- the population of interest for this proposal). While data and guidelines are clear about the short-term benefit of treatment in the former group (i.e., those with cirrhosis), they are less clear about the benefit in the latter group (those without cirrhosis). In fact, most of these non-cirrhotic, asymptomatic patients progress slowly over years to decades (low-risk patients) and thus may not require immediate treatment. Others progress more rapidly and could benefit from immediate treatment. However, clinicians are often uncertain about how to approach such patients. Current treatment approaches for non-cirrhotic CHC vary substantially across healthcare systems, owing largely to discrepancies in guidance on when to treat such patients. As a result, treatment is often driven by a combination of patient preferences, clinician judgment, and drug availability. Veterans who are at high-risk for disease progression but do not actively seek care may therefore fail to receive potentially life-saving therapy. In addition, the guidelines continue to change rapidly based on the availability of new drugs. A systematic, rigorous approach to treatment, one informed by state-of-art prediction modeling to risk-stratify non-cirrhotic Veterans, could help guide risk-based treatment and mitigate this healthcare delivery shortcoming. Objectives The purpose of this study is to lay the groundwork for risk-based treatment of CHC among non- cirrhotic Veterans in the Veterans Health Administration (VHA) by: (1) developing accurate, clinically relevant, and implementable risk prediction models; (2) engaging Veterans to develop consensus on how to implement risk-based treatment; and (3) evaluating the clinical and economic effects of risk-based treatment. Methods In our preliminary work, we demonstrated the feasibility of using a machine-learning (ML) risk prediction model to identify patients at high risk and low risk for disease progression in a clinical trial cohort. We propose a 4- year study where we will use VA electronic data from 2004-2014 to adapt, validate and refine this model among Veterans. We will then engage Veterans, eliciting their preferences and values regarding risk-based treatment of CHC by applying consensus techniques (e.g., deliberative democracy). Finally, to estimate the incremental benefit of risk-based treatment over current treatment, we will use simulation modeling.