Machine Learning to Predict Mortality and Improve End-of-Life Outcomes among Minorities with Advanced Cancer

项目摘要/摘要 多项研究表明，少数族裔晚期癌症患者对 有助于提高卫生保健利用率的治疗、预后和护理目标 处于生命末期的少数群体。预后和护理目标低的主要因素 讨论涉及肿瘤学家无法准确预测死亡率。临床决策支持系统 (CDS)旨在通过利用个别患者的特征来直接帮助临床决策 生成特定于患者的评估。有限的研究表明，CDSS可以减少在 关爱和护理标准化。然而，现有的工具并不能识别死亡风险最高的患者， 没有与患者的预后挂钩，也没有在少数族裔患者中进行常规评估。 机器学习(ML)预测模型通过对患者和疾病进行建模，实现了更准确的预后- 具体的互动，并有可能消除存在于使用肿瘤学家的种族偏见- 具体的预测。利用电子健康记录(EHR)数据的ML模型可以准确地预测 肿瘤科患者的短期死亡率。然而，几乎没有证据表明这些模型有助于 少数族裔癌症患者的临床决策或改善预后。我们将解决系统性问题 与种族/民族有关的障碍导致少数族裔癌症患者的死亡结局存在差异 患者：1)开发和验证用于识别晚期实体癌患者的预测模型 90天内有高死亡风险；2)创建CDSS系统干预，将死亡率纳入其中 预测性工具数据，促进高死亡风险实体癌症患者的护理对话目标 在90天内；3)进行阶梯楔形整群随机对照试验 是否为晚期实体癌患者实施临床决策支持系统 90天内死亡增加了护理讨论的目标，并减少了对积极护理的利用 少数族裔与非少数族裔之间的生命终结。预测模型将从癌症中创建 链接到电子病历的注册表数据。然后，我们将通过在 肿瘤临床医生的跨学科团队(医生、高级实践提供者、护士和社会工作人员 工人)。此外，我们将与肿瘤临床医生共同举办设计工作坊，向 CDSS的实施。接下来，我们将进行阶梯式楔形整群随机对照试验 评估CDSS的使用是否增加了护理讨论的目标，并降低了医疗保健 少数民族与非少数民族高危晚期实体癌患者临终时的利用度 90天内死亡的可能性。最后，我们将进行离职访谈，以完善干预和研究 程序。调查结果将为旨在实施CDSS的更大规模的多中心试验提供参考 那些预计死亡率较高的患者将改善少数族裔癌症患者的临终结局。

PROJECT SUMMARY/ABSTRACT Multiple studies show that minority patients with advanced cancer have inadequate discussions about treatment, prognosis and goals of care which contributes to higher utilization of health care among minorities at the end-of-life. A primary contributor to the low rates of prognosis and goals of care discussions relates to oncologists inability to accurately predict mortality. Clinical decision support systems (CDSS) are designed to directly aid clinical decision making by utilizing individual patient characteristics to generate patient-specific assessments. Limited studies indicate CDSS can reduce disparities in process of care and care standardization. However, existing tools do not identify patients at highest risk of mortality, have not been linked to patient outcomes and have not been routinely evaluated in minority patients. Machine learning (ML) predictive models allow more accurate prognoses by modeling patient and disease- specific interactions and has the potential to obviate the racial bias that exists in the use of oncologist- specific prognostication. ML models utilizing electronic health record (EHR) data can accurately predict short-term mortality among oncology patients. However, little evidence exists that these models assist with clinical decision making or improve outcomes for minority patients with cancer. We will address systemic race/ethnicity-related barriers that contribute to disparities in end-of-life outcomes among minority cancer patients by: 1) Developing and validating a predictive model to identify patients with advanced solid cancers at high risk of death within 90-days; 2) Creating a CDSS system intervention that incorporates mortality predictive tool data to prompt goals of care conversations for solid cancer patients at high risk of mortality within 90 days; and 3) Conducting a stepped-wedge cluster randomized controlled trial to evaluate whether implementing a clinical decision support system for patients with advanced solid cancer at risk of death within 90 days increases goals of care discussions and decreases utilization of aggressive care at the end-of-life among minorities versus non-minorities. The predictive model will be created from cancer registry data linked to the EHR. We will then create a CDSS by conducting focus groups among an interdisciplinary team of oncology clinicians (physicians, advance practice providers, nurses and social workers). Additionally, we will conduct co-design workshops with the oncology clinicians to inform the implementation of the CDSS. Next, we will conduct a stepped-wedge cluster randomized controlled trial to evaluate whether utilization of the CDSS increases goals of care discussions, and decreases healthcare utilization at the end-of-life among minority versus non-minority patients with advanced solid cancer at risk of death within 90 days. Finally, we will perform exit interviews to refine the intervention and study procedures. Findings will inform a larger multi-center trial aimed at implementation of the CDSS among those predicted to have high mortality to improve the end-of-life outcomes of minority patients with cancer.