Algorithms to Identify Second Breast Cancer Events from Electronic Data

描述(申请人提供)：美国乳腺癌幸存者人数为250万，超过其他任何癌症的幸存者。研究如何在不断增长的人口中提高存活率和生活质量，对于减轻国家癌症负担至关重要。识别二次乳腺癌事件(即，乳腺癌复发和二次原发乳腺癌)的能力对于癌症生存研究至关重要。为了响应国家癌症研究所呼吁使用健康声明数据进行癌症监测研究，我们建议开发和验证算法，从自动医疗保健利用数据中识别第二次乳腺癌事件，以最大限度地减少昂贵和耗时的手动医疗记录审查的需要。自动化的医疗保健利用数据正变得越来越容易获得；然而，这些来源尚未根据黄金标准的医疗记录摘要进行验证，以获得有关第二次乳腺癌事件的信息。这项工作是重要的和必要的，因为州肿瘤登记机构不会定期收集癌症复发的信息。这项拟议的研究将使用癌症研究网络(CRN)内两个综合医疗保健提供系统的数据进行：团体健康合作社(华盛顿州西部)和亨利·福特健康系统(密歇根州底特律)。这些医疗保健系统拥有关于登记、诊断、程序和处方药填写的广泛自动化数据。这项拟议的研究是有效的，因为它将使用第二次乳腺癌事件的黄金标准数据，这些数据已经在约2500名女性身上抽象出来，作为先前资助的乳腺癌结果研究的一部分。妇女样本将被分为用于算法开发的训练数据集(60%)和用于验证的测试数据集(40%)。这项研究的主要目的是开发一个算法菜单，研究人员可以在不同的情况下进行选择，即当他们想要最大限度地提高灵敏度、特异度或阳性预测值时。二次分析将探索：1)在一个群体中开发的算法是否在另一个群体中有效，以及2)是否可以使用更有限的数据来源来开发有效的算法，这些数据来源可能在更多的医疗保健环境中可用。这个项目将使用创新的方法来开发算法菜单，并探索算法开发的普适性。 公共卫生相关性：随着乳腺癌幸存者数量的增加，关于乳腺癌预后和生活质量的研究对公共卫生变得越来越重要；然而，目前收集乳腺癌复发和第二原发乳腺癌数据的方法要么耗时又昂贵，要么尚未得到验证。能够从自动化的医疗保健数据中识别癌症乳腺癌的结果，对于进行大规模的基于人群的研究，以确定和修改影响乳腺癌女性患者预后和生活质量的因素是必要的。

DESCRIPTION (provided by applicant): U.S. breast cancer survivors number 2.5 million, more than the survivors of any other cancer. Studies on how to improve survival and quality of life in this ever-growing population are critical in reducing the national cancer burden. The ability to identify second breast cancer events (i.e., breast cancer recurrences and second primary breast cancers) is critical for cancer survivorship research. In response to the National Cancer Institute's call for studies of cancer surveillance using health claims data, we propose to develop and validate algorithms to identify second breast cancer events from automated healthcare utilization data in order to minimize the need for expensive and time-consuming manual medical record review. Automated healthcare utilization data are becoming increasingly accessible; however, these sources have yet to be validated against gold-standard medical record abstraction for obtaining information on second breast cancer events. This work is significant and necessary since state tumor registries do not routinely collect information on cancer recurrences. The proposed study will be conducted using data from two integrated healthcare delivery systems within the Cancer Research Network (CRN): Group Health Cooperative (in western Washington State) and the Henry Ford Health System (in Detroit, Michigan). These healthcare systems have extensive automated data on enrollment, diagnoses, procedures, and prescription medication fills. The proposed study is efficient because it will use gold-standard data on second breast cancer events that have already been abstracted on ~2500 women as part of previously funded studies of breast cancer outcomes. The sample of women will be divided into a training dataset (60%) for algorithm development and a testing dataset (40%) for validation. The primary aim of this study is to develop a "menu" of algorithms that researchers can select from under different circumstances; i.e., when they want to maximize sensitivity, specificity, or positive predictive value. Secondary analyses will explore: 1) whether algorithms developed in one population are valid in another, and 2) whether valid algorithms can be developed using more limited sources of data that are likely to be available in a larger number of healthcare settings. This project will use innovative approaches to develop the algorithm "menu" and to explore the generalizability of algorithm development. PUBLIC HEALTH RELEVANCE: As the number of breast cancer survivors grows, research on breast cancer prognosis and quality of life is becoming increasingly important to public health; however, current methods for collecting data on breast cancer recurrences and second primary breast cancers are either time-consuming and costly or have not yet been validated. Being able to identify cancer breast cancer outcomes from automated healthcare data is necessary for conducting large-scale, population-based studies to identify and modify factors that impact the prognosis and quality of life of women with breast cancer.