Predictive modelling methodology for longitudinal data in long-term care****

该提案确定了Daniel J. Lizotte博士和PointClickCare之间的新合作。PointClickCare （PCC）是一家成熟的加拿大公司，主要为高级和长期护理机构提供基于云的电子健康记录（EHR）为中心的产品和服务。与大多数其他卫生信息学数据来源不同，PCC的数据库包含高度纵向的住院和患者记录，并具有频繁的观察点。有了这些数据，PCC就有机会开发预测分析模块，利用居民长期护理历史的全部细节，做出更有效的护理和运营决策。要做到这一点，PCC的数据科学团队需要额外的研究和开发专业知识，使用临时依赖的数据进行预测建模。为了满足这些需求，我们建议开发和实施一套全面的预测模型，专门用于利用PCC数据库中住院和护理机构数据的纵向性质。我们将根据需要开发和应用各种建模方法，从简单的回归分析和分类器到复杂的聚类方法。根据需要，还将开发用于时间依赖性特征构建的新方法，以确定可用于改进护理和操作决策的重要模式。该项目产生的模型和分析将使PCC能够为其客户提供新的有效工具，并通过提供在预测模型中使用纵向医疗保健数据的新方法，更广泛地使加拿大公众受益。****

This proposal identifies a new collaboration between Dr. Daniel J. Lizotte and PointClickCare. PointClickCare (PCC) is an established Canadian company that provides cloud-based electronic health record (EHR) centred products and services primarily for senior and long-term care facilities. Unlike most other sources of health informatics data, PCC's databases contain highly longitudinal resident and patient records with frequent observation points. With these data, PCC has an opportunity to develop modules for predictive analytics that leverage the full details of residents' histories in long term care to make more effective care and operational decisions. To do this, PCC's data science team requires additional research and development expertise in predictive modelling using temporally-dependent data. To meet these needs, we propose to develop and implement a comprehensive set of predictive models specifically built to leverage the longitudinal nature of resident and care facility data across PCC's databases. We will develop and apply various modelling methods ranging from simple regression analyses and classifiers to complex clustering methods, as needed. New methods for temporally-dependent feature construction will also be developed as needed with the goal of identifying important patterns that can be used to improve care and operational decisions. The resulting models and analysis from this project will enable PCC to offer new and effective tools to its customers, and more broadly will benefit the Canadian public by providing new methods for using longitudinal healthcare data in predictive models.****