P1477: CHARACTERIZING SICKLE CELL DISEASE AND BURDEN OF ILLNESS IN THE UNITED STATES OF AMERICA: A RETROSPECTIVE ANALYSIS OF REAL-WORLD DATA

Background: Despite advances in screening, management and treatment, gaps remain in our understanding of sickle cell disease (SCD) in the real-world setting. Most observational studies rely on structured data (ie predefined clinical concepts such as ICD9/10 codes) from electronic medical records (EMRs), but this neglects clinically relevant information stored in unstructured narrative sections of physician notes. Advances in data analytic methods allow for the combination of physician notes and structured data, potentially providing a more accurate and complete picture of disease manifestation. Aims: To evaluate disease manifestation in terms of vaso-occlusive crisis (VOC) and SCD-related organ damage, as well as current management approaches, and to evaluate the clinically relevant information that can be captured from physician notes. Methods: 403 patients with confirmed SCD, who had ≥1 hospitalization due to VOC (H-VOC) in the 12 months before recruitment, were enrolled in this retrospective, non-interventional observational study. Data were analyzed from 235 patients (172 women and 63 men) aged ≥16 years in the year of index date (ie the first day after baseline [the first 12 months of the observation period]) and who had ≥1 year of follow-up data after the index date. The observation period (baseline plus follow-up) required ≥2 clinical document records ≥30 days apart, and ≥1 clinical document every 6 months. Each patient’s longest observation period was used for this analysis. Event rates are adjusted for exposure time and reported as rates per person-year with 95% confidence intervals (CI). Human-curated natural language processing with assistance from machine learning algorithms was used to extract EMR data. Results: During the baseline period, 102 (43%) patients had ≥1 H-VOC (including emergency department visits), 88 (37%) had ≥1 outpatient VOC, and 32 (14%) had been hospitalized for SCD-related organ damage, primarily related to bone necrosis (9%) and pulmonary disease (8%). During follow-up (median 3.6 years), 218 (93%) patients experienced ≥1 VOC (H-VOC rate [95% CI] of 4.25 [3.64–4.95] per person-year; outpatient VOC rate of 2.39 [1.97–2.91]), and 143 (61%) were hospitalized for SCD-related organ damage (0.97 [0.79–1.21]). Of these, 764 (19%) H-VOCs, 1351 (52%) outpatient VOCs, and 423 (50%) hospitalizations due to SCD-related organ damage were extracted exclusively from physician notes. The overall diagnostic procedure rate was 20.36 (18.54–22.36), with blood tests being the most common (14.07 [12.17–16.25]) followed by X-rays (7.26 [6.50–8.10]) and urinalysis (2.43 [2.08–2.83]). The blood transfusion rate was 0.99 (0.77–1.27), and most patients received hydroxyurea (78%) and supportive treatments (95%). Patients mostly visited (ie inpatient and outpatient visits, and visits related to diagnostic procedures and treatment) specialist medical practitioners (15.46 [13.73–17.40]), while relatively few visited general medical practitioners (1.72 [1.46–2.02]). Summary/Conclusion: To our knowledge, this is the first documented use of physician notes complementing structured data to characterize SCD in a real-world setting. We report a high level of healthcare consumption, predominantly involving specialist care and diagnostic procedures; only a small proportion of patients utilized outpatient-only care for VOC management. Our findings emphasize the importance of preventing and minimizing VOCs. Importantly, we demonstrate the considerable contribution that physician notes can make to better characterizing disease manifestation in SCD.