A New Generation Clinical Decision Support System

描述(由申请人提供)：关键的临床活动涉及决策。对于患者个人和整个社会来说，做出良好的医疗决策是一项最重要的任务。本研究的目的是开发一种新的决策支持系统，该系统利用乳腺癌患者的临床特征和基因组特征来辅助医生整合关于特定患者的信息(诊断亚型、肿瘤分期和分级、年龄、合并疾病)，以制定患者的治疗计划。 传统的临床数据越来越多地以电子形式提供。随着下一代测序等先进测序技术的发展，研究人员可以获得前所未有的丰富的基因组数据。在可预见的未来，患者特有的基因组数据很可能对大多数患者可用。这些数据来源为开发新一代临床决策支持系统提供了重要机会，这些系统可以在目前可能的基础上取得实质性进展。然而，这些数据中的变量数量(通常以数百万计)的巨大规模带来了巨大的计算和建模挑战。此外，集成多个临床数据集和基因组数据集中的异质信息也是一个艰巨的挑战。 乳腺癌是女性中最常见的癌症。不同的乳腺癌亚型已经被定义，与肿瘤分期一起预测治疗反应和生存，尽管不完美。例如，HER2扩增的乳腺癌是一种预后较差的亚型，而HER2抗体(赫赛汀)的治疗极大地提高了此类患者的存活率。尽管赫赛汀被用于治疗所有HER2扩增的肿瘤患者，但只有一些人有反应。此外，它价格昂贵，可能会导致心脏毒性。因此，重要的是只给那些从中受益的患者。研究表明，数千个基因与乳腺癌的亚型和预后有关，特定的等位基因组合可能有助于指导有效治疗的选择。拟议的系统将收集所有这些基因组信息，并将其与临床信息相结合，因此有望提供准确的分类和治疗选择。 我们将使用以下来源建立拟议系统的知识库：1)匹兹堡大学医学中心的医疗档案系统；2)西北纪念医院林恩·赛奇综合乳房中心使用的林恩·赛奇数据库；3)癌症基因组图谱项目中的乳腺癌数据集；以及4)梦想7乳腺癌挑战数据。拟议的系统将建立在调查人员使用贝叶斯网络从高维数据集中学习的先前结果的基础上。我们的多学科团队拥有良好的记录，包括NIH的资助，生物医学信息学和人工智能方面的出版物，以及开发尖端决策支持系统的经验。

DESCRIPTION (provided by applicant): Critical clinical activities involve decision making. For both individual patients and for society at large, making good healthcare decisions is a paramount task. The objective of this research is to develop a novel decision support system that utilizes both the clinical features and the genomic profile of a breast cancer patient to assit the physician in integrating information about a specific patient (diagnostic subtype, tumor stage and grade, age, comorbidities) to make therapeutic plans for the patient. Traditional clinical data are becoming increasingly available in electronic form. Unprecedentedly abundant genomic data are available to researchers as the results of advanced sequencing technologies such as next generation sequencing. Patient-specific genomic data are likely to become available for most patients in the foreseeable future. These sources of data provide significant opportunities for developing new generation clinical decision support systems that can achieve substantial progress over what is currently possible. However, the sheer magnitude of the number of variables in these data (often in the millions) presents formidable computational and modeling challenges. Also, integrating the heterogeneous information in multiple clinical datasets and genomic datasets presents an arduous challenge. Breast cancer is the commonest cancer among women. Various breast cancer subtypes have been defined which, along with tumor stage, predict response to therapy and survival, albeit imperfectly. For example, HER2-amplified breast cancer is a subtype with poor prognosis, and therapy with an antibody to HER2 (Herceptin) has vastly improved the survival of such patients. Although Herceptin is used in the therapy of all patients with HER2-amplified tumors, only some respond. Also, it is expensive and can cause cardiac toxicity. So, it is important to give it only o patients benefiting from it. Studies show thousands of genes are associated with subtype and prognosis of breast cancer, and particular allele combinations may usefully guide the selection of effective treatment. The proposed system will amass all this genomic information and combine it with clinical information and therefore holds promise to provide accurate classification and treatment choices. We will build the knowledge base of the proposed system using the following sources: 1) The Medical Archival Systems at the University of Pittsburgh Medical Center; 2) The Lynn Sage Database used by the Lynn Sage Comprehensive Breast Center at Northwestern Memorial Hospital; 3) The breast cancer data sets from The Cancer Genome Atlas project; and 4) Dream 7 Breast Cancer Challenge Data. The proposed system will build on previous results of the investigators in using Bayesian Network to learn from high-dimensional data sets. Our multidisciplinary team has a track record, including NIH funding, publications in biomedical informatics and artificial intelligence, and experience developing cutting-edge decision support systems.