Characterizing Death in Sickle Cell Disease in a National EHR Database: A Population Analysis

Background

Sickle cell disease (SCD) is characterized by overt signs of end organ damage that lead to early death, previously estimated to be 30 years younger than the general population (Platt et.al., 1994). National estimates of SCD are extrapolated from studies performed in the 1980s before the widespread use of disease-modifying therapies. A modern national cohort is needed to understand the public health impact of SCD morbidity and to update mortality trends, which will help develop risk stratification tools aiding clinicians in identifying patients who need curative and transformative therapies. We hypothesize that using a validated automated phenotyping algorithm from Vanderbilt University Medical Center (VUMC) (Cronin et.al., 2023) on Oracle’s Cerner Learning Health Network (LHN), national real world data, we can characterize mortality among SCD patients.

Methods

We conducted a retrospective observational study at Indiana University Health (IUH) using Oracle’s Cerner LHN, a real-world data set which includes over 80 million patient records nationwide. Adults and children with SCD were identified using the VUMC automated phenotyping algorithm, which relies on ICD codes, laboratory data including hemoglobin fractionation, and transfusion data. A control group matched by age, sex, race, ethnicity, and hospital system was created using distance precision matching (Austin, 2011) The primary outcome was mortality, assessed by comparing median survival of SCD patients to controls and HbSS/SB⁰ and HbSC phenotypes. Follow up was based on the time from first lab encounter to death or censor date. Accurate death records were obtained through a partnership with Datavant (Wallace et.al., 2021), covering all SCD patients and their matched controls. The study received IRB approval from Indiana University School of Medicine and approval from Oracle’s Cerner LHN review committee. Data access was managed within Oracle’s Cerner LHN cloud platform. The VUMC automated SCD algorithm was performed in Python, and survival curve analyses were conducted using R (R 4.0.2, June 2020).

Results

The observational study assessed patients from January 2000 to November 2023. Using the VUMC automated phenotyping algorithm on the Cerner LHN, we identified 11,606 SCD patients and 49,993 matched controls. After excluding those with less than 6 months follow-up, 10,598 SCD patients and 45,542 controls remained. The median age at the start of observation was 7 years for both groups. The SCD cohort was 49.6% male, while the control group was 50.3% male. Racial/ethnic composition of the SCD cohort versus controls: Black/African American (85.7% vs. 85.8%), White (2.6% vs. 3.0%), Asian (0.2% vs. 1.3%), and Latino (5.4% vs. 5.2%). Survival analysis showed a median survival of 59 years (95% CI: 57-61) for the SCD cohort compared to 77 years (95% CI:75-77) for controls. For HbSS/HbSB⁰ patients, it was 55 years (95% CI: 52-59), and for HbSC patients, it was 71 years (95% CI: 66-77). Female SCD patients had a median of 66 years (95% CI: 62-68) compared to 54 years (95% CI: 52-56) for males.

Conclusion

Our data shows that using the VUMC automated phenotyping algorithm in the Cerner LHN effectively identifies SCD patients. This method supports national sampling and assessment of overall survival. Median survival for SCD patients appears to be increasing. However, life expectancy remains 18 years shorter than matched controls. As expected, patients with HbSS/HbSB0 phenotype have a shorter life expectancy than patients with HbSC. Using a national EHR-based real world dataset is an effective approach to evaluate mortality in SCD and easily lends itself toward development of risk stratification models, which are already underway.