Analyzing Patient-Level Data in a Breast Cancer Clinical Trial

摘要 大多数接受乳腺癌治疗的女性将经历某种形式的药物相关毒性和随后的 健康相关生活质量(HRQOL)的损害，但在肿瘤学试验中毒性评估不一致。 尽管治疗的潜在副作用对患者来说是非常重要的 关于他们的治疗选择，毒性往往被低估。在评估试验的症状时 参与者、患者和提供者并不总是将症状归因于研究药物，这可能导致 对最大耐受剂量的错误分类。此外，许多药物毒性，如神经病， 疲劳和腹泻在试验中经常被提供者低估，因此是以患者为中心的评估 可能导致更早认识到可逆转的副作用。 知识方面的一个主要差距是如何实时分析和利用患者水平的毒性数据，以及如何 以一种可以及早减轻毒性的格式向提供者提交数据。而较低的- 根据这些较低级别的毒性，在报告患者结局的情况下，级别毒性可能会增加 可以减轻严重不良事件(SAE)。 我们最近实例化了一个电子患者报告结果(EPRO)平台，覆盖了i- SPY2，我们收集不良事件和撒谎信息的质量。I-SPY2是一种自适应平台试验 持续评估新辅助乳腺癌疗效的高危早期乳腺癌 治疗。这项提案的总体目标是改进和实施新的方法，使用 可解释的机器学习，可用于支持框架以重定向治疗并避免更多 严重的疾病。这种方法在今天的临床试验中还不存在，可以极大地造福于患者，他们的 提供者和临床护理团队通过跟踪患者痛苦的拐点，否则可能会 错过了，但可能需要更多的立即干预。这些方法将通过计算来开发 框架在与提供者的讨论中，在不同的治疗阶段，如当严重程度单一 症状真正影响身体功能(主要结果)，或者当一系列症状预示着 总体健康状况显著恶化。这一提议的中心假设是， 我们正在研究谁会患上可能影响生活质量的慢性病和症状 减轻严重不良反应的发生，提高整体生活质量，特别是身体方面 功能正常。我们将在一组I-SPY患者和乳房护理中心的早期阶段测试我们的方法 加州大学旧金山分校的参与者。

ABSTRACT Most women treated for breast cancer will experience some form of drug-related toxicity and subsequent impairments in Health-related Quality of Life (HRQOL), yet toxicity is assessed inconsistently in oncology trials. Although the potential for side effects of treatments is of great importance to patients in making informed choices about their treatment, the toxicities are often under-reported. When assessing symptoms of trial participants, patients and providers do not always attribute symptoms to the study drug, which can result in misclassification of the maximum tolerated dose. Furthermore, many drug toxicities such as neuropathy, fatigue and diarrhea are often underreported by providers in trials, and thus a patient-centered assessment may lead to earlier recognition of reversable side effects. A major gap in knowledge is how to analyze and utilize patient level toxicity data in real time, and how to present the data to providers in a format that can result in early toxicity mitigation. While the number of lower- grade toxicities may increase given the reporting of patient outcomes, acting on these lower grade toxicities can mitigate serious adverse events (SAEs). We have recently instantiated an electronic patient reported outcomes (ePRO) platform across 26 sites in I- SPY2 where we collect adverse events and quality of lie information. I-SPY2 is an adaptive platform trial for high risk, early-stage breast cancer that continuously evaluates the efficacy of new neoadjuvant breast cancer therapies. The overall objective of this proposal is to refine and implement new methodology using interpretable machine learning that can be used to underpin a framework to redirect treatment and avoid more serious illnesses. Such methodology does not exist in clinical trials today and can hugely benefit patients, their providers and the clinical care team by tracking the inflection points of patient distress that could otherwise be missed but may require more immediate intervention. The methods will be developed through a computational framework in discussion with providers, at different stages of treatment, such as when the severity of a single symptom really impacts physical functioning (primary outcome), or when constellation of symptoms herald a significant deterioration in overall health. The central hypothesis of this proposal is that the methodology that we are developing on who will develop chronic conditions and symptoms that may affect quality of life will mitigate the event of a serious adverse reaction and improve overall quality of life, particularly physical functioning. We will test our methodology in a group of I-SPY patients and Breast Care Center early-stage participants at UCSF.