Abstract
In breast cancer, clinicians add data on social support to patient electronic health records (EHRs) often in free text notes, but those data may be challenging to use for population health initiatives or research purposes. We evaluated the EHR-Support score designed to summarize need for social support using data from the EHR. This study included 996 women from the Pathways study, a Kaiser Permanente Northern California cohort of women diagnosed in 2005-2013 with breast cancer. This unique data resource included both EHR data and questionnaire data on patient-reported social support. Using unstructured EHR data and natural language processing, we developed 11 concept groups (items) characterizing social support. We also used structured data to create two additional concept groups. EHR-Support scores reflecting the lack of social support were generated three ways: counting the number of negative concept groups (count score), using item response theory (IRT), and converting counts to the IRT metric (converted count scores). The count scores were only associated with two of six patient-reported measures (r's: -0.004 to -0.073). The IRT score (r's: -0.038 to -0.179) and converted count score (r's: -0.032 to -0.195) were associated with five of six patient-reported measures, indicating more need for support was associated with less patient-reported social support. The EHR-Support score is a valid and feasible measure of social support that can be used for health services research and managing population health. The converted count score may provide the best balance of validity, precision from IRT and feasibility.