An Electronic Health Record-Based Tool to Identify Newly Diagnosed Breast Cancer Patients at Risk of Low Social Support

Social support is a definitive predictor of breast cancer (BC) treatment and mortality outcomes. Because of the recognition that social support is critical to BC patient outcomes, clinicians within Kaiser Permanente Northern California (KPNC) have documented information on social support in the electronic health record (EHR) since the advent of Epic in 2006. However, no EHR-based social support measure currently exists to help clinicians identify patients at high risk of low social support. Such a measure has high relevance for improving BC treatment and outcomes. Therefore, we propose to develop an Electronic Health Record Social support Patient Risk Tool (EHR-SUPPORT) that could be used to identify women with BC at risk of low social support for referral to social support resources. We propose to: 1) Identify terms in the EHR, based on theory and prior literature, and informed by KPNC stakeholders in BC care, that reflect structural and/or functional social support, and have been associated with BC treatment and outcomes; 2) Develop EHR-SUPPORT, using structured, semi-structured, and unstructured data (to include natural language processing of text) that identifies patients at risk of low social support, and validate the measure against published social support measures; and 3) Evaluate associations of EHR-SUPPORT and its component variables with BC treatment (surgery and chemotherapy delays, nonadherence to hormonal therapy) and BC-specific and total mortality in 44,348 women diagnosed with stage I-IV BC within Kaiser Permanente Northern California between 2006-2023 including 6,571 Asian, 3,450 Black, 4,441 Hispanic, women, and 28,589 non-Latina White women. We will review 100 medical records within two months of diagnosis, informed by investigator expertise and clinician stakeholders, to develop terms used to describe patient support. In addition to developing structured data, we will use natural language processing of text fields to further develop social support indicators (Aim 1). EHR-SUPPORT will be computed from social support indicators; we will use linear and logistic regression to validate the developed measure against established social support measures in Pathways, a well-established cohort of 4,505 women with BC and use factor analytic and confirmatory factor analytic methods as well as ROC curves to further evaluate the score (Aim 2). We will use linear, logistic, and Cox proportional hazards regression to evaluate associations in Aim 3. In an exploratory aim, we will develop, with KP clinician stakeholders, steps to the implementation of EHR-SUPPORT. The unique convergence of EHR and cohort data provides the first opportunity to develop and validate an EHR-based social support measure in women with BC, adjusted for an extensive set of covariates. This work is central to identifying patients at elevated risk of low social support and to enhancing social support-cancer research needed to improve clinical care and reduce adverse BC