The use of natural language processing to assess patients with bleeds in their notes in the optum EHR database

Capturing bleeding outcomes in hemophilia patients is challenging. However, doctors do ask patients about bleeds at appointments and assess annualized bleed rates (ABR) in patients. The prevalence of bleeding documented in doctors’ notes in the Electronic Health Records (EHR) is currently largely unknown. Using the Optum© EHR data acquisition model, which aggregates deidentified data from providers across the continuum of care, patients with eight Natural Language Processing (NLP) terms were searched by using deidentified snippets of phrases through the last five years of the Optum© EHR deidentified database (2012-2017). The terms were searched among males with ICD codes for hemophilia A and B (286.0/D66 & 286.1/D67) receiving factor replacement product. The phrases were: ABR, annual bleed rate, annualized bleeding rate, annualized bleed rate, ARB, bleeds per year, spontaneous bleeding, and spontaneous hematoma. The terms were determined with help from medical doctors and prior hemophilia literature. Patients could appear in more than one NLP term. A total of 2283 patients fit the criteria; 293 hemophilia A patients and 91 hemophilia B patients had NLP associated with the above criteria, with the term ABR (166 [57%] and 55 [60%], respectively) being the most common term appearing in the notes. Spontaneous bleeding was the second most common term in hemophilia A (151) and annualized bleeding rate was second in hemophilia B (47), both accounting for 52% of the total patients with NLP. Terms such as annual bleed rate, annualized bleed rate, and spontaneous hematoma were all sparsely populated in both disease conditions. This real-world data NLP analysis shows that there is a lot to be learned about NLP outcomes in hemophilia. The use of specific terms, and the refinement of said terms, will help to make future NLP projects easier as terms appearing in charts are pinpointed.