A Validation Study of Red Blood Cell Transfusions in Patients with Sickle Cell Disease Using Optum Electronic Health Records and Administrative Claims Data

Introduction: Sickle cell disease (SCD) has complex and heterogeneous pathophysiology, with complications affecting many organs. Red blood cell (RBC) transfusion is common to mitigate vaso-occlusive crisis (VOCs), stroke risk, or other severe complications of SCD. Administrative claims databases and electronic health records (EHR) are valuable resources to study clinical outcomes in real-world evidence (RWE) studies. Data recorded in claims databases are mainly for billing purposes, and the quality of claim-based outcomes data relies on accurate assignment of treatment exposures and events, and correct diagnostic and procedural codes.

To understand the potential impact of outcome misclassification from ICD codes for RBC transfusion in real world data (RWD) in the US, we conducted a validation study of RBC transfusions in SCD using Optum-linked EHR and administrative claims data (Market Clarity). To date, there has been sparse information on the validity of blood transfusion from claims/EHR data, especially for patients (pts) with SCD, so this large RWD study provides more data to fill this gap.

Methods: This was a retrospective, non-interventional cohort study of pts with SCD using data from the US-based Optum de-identified Market Clarity database from January 1, 2016 to March 31, 2023. Pts aged 12-85 years with ≥1 RBC transfusion were included if they had ≥1 inpatient or 2 outpatient ICD-10-CM diagnosis codes D57.0, D57.1, D57.2, D57.4 or D57.8, separated by ≥30 days in claims/EHR. Data were captured by clinical notes review for RBC transfusions identified in structured EHR/claims data via procedure codes for 2 SCD population cohorts: Cohort 1: pts with a record of RBC transfusion in structured EHR data and medical claims and with free text clinical notes; Cohort 2: pts with a record of RBC transfusion in medical claims but not in EHR, although with other activities documented in structured EHR data and available free-text notes within 30 days of the claim. Cohort 3 (data not shown): pts with a record of RBC transfusion in structured EHR data but no record of transfusion in claims, although with other activities documented in claims within 30 days of transfusion documentation in EHR. The aim was to assess if medical claims data (ICD-10 procedure codes) with no record of transfusion in EHR (Cohort 2) are a reliable source for assessing RBC transfusions in pts with SCD, compared with Cohort 1, and informative as an endpoint in RWE studies.

Results: 13,327 pts with SCD with ≥1 RBC transfusion in either a medical claim or EHR data in the Optum Market Clarity database from January 1, 2016 to March 31, 2023 were included in the study. There were 46,381 RBC transfusions in total; 42,106 (91%) identified from a medical claim and 14,669 (32%) identified in EHR; 10,394 (22%) were identified in both claims and EHR. Of 200 randomly selected pts each from Cohorts 1 and 2, 159 (79.5%) and 40 (20%) in Cohort 1, and 135 (67.5%) and 46 (23%) in Cohort 2 had transfusion administration confirmed and transfusion scheduled/planned in clinical notes, respectively. One (0.5%) pt in Cohort 1 and 19 (9.5%) pts in Cohort 2 had no RBC transfusion administration confirmed in notes, resulting in a positive predictive value for RBC transfusion episodes based on ICD-10 codes from EHR/claims of 99.5% (95% CI: 98.5 - 100) (Cohort 1) and medical claims of 90.5% (95% CI: 86.4 - 94.6) (Cohort 2).

Mean (SD) annualized rates of RBC transfusions were 6.2 (6.7) and 6.7 (8.0) per patient per year, in Cohorts 1 and 2 respectively. Where the setting was recorded, transfusions occurred at inpatient stays (32% [n=64], 54.5% [n=109]), in emergency departments/ urgent care (17.5% [n=35],12.5% [n=25]), and at office visits/outpatient facilities (10% [n=20], 8% [n=16]). The most common reasons for transfusion were anemia (51% [n=102] and 64% [n=128]) and VOCs (18% [n=36] and 38.5% [n=77]).

Conclusions: Most transfusions (91%) were identified from medical claims. A high proportion of transfusions (90.5%) detected based on procedure codes from claims were confirmed in clinical notes for Cohort 2. Most transfusions were indicated for anemia and unscheduled as they occurred in an acute care setting. Overall, the results are reassuring that medical claims are a reliable source of data for RBC transfusions in pts with SCD and have utility as an endpoint for RWE studies.

Gu:Pfizer: Current Employment, Current equity holder in publicly-traded company. Purdie:Pfizer: Current Employment, Current equity holder in publicly-traded company. Shambhu:Pfizer: Current Employment, Current equity holder in publicly-traded company. Ho:Pfizer: Current Employment, Current equity holder in publicly-traded company. Colavecchia:Pfizer: Current Employment, Current equity holder in publicly-traded company. Jiang:Pfizer: Current Employment, Current equity holder in publicly-traded company. Meier:Pfizer: Current Employment, Current equity holder in publicly-traded company. Demas:Pfizer: Current Employment, Current equity holder in publicly-traded company. Chaudhari:Pfizer: Current Employment. Senerchia:Optum: Current Employment. Margolin-Katz:Optum: Current Employment. Branner:Optum: Current Employment.