Positive Predictive Value of Applying an Electronic Health Record Toolkit (deciphEHR^TM) in Identifying Patients with Paroxysmal Nocturnal Hemoglobinuria in a Searchable Data Warehouse

Introduction:

Paroxysmal nocturnal hemoglobinuria (PNH) is a rare and potentially life-threatening acquired hemolytic anemia caused by the clonal expansion of a hematopoietic progenitor cell that has acquired a mutation in the X-linked PIGA gene (Brodsky et al, 2014). Patients with PNH suffer from thrombosis and complications of bone marrow failure. Due to the nature of the disease combined with the heterogeneous presentation and broad symptoms, diagnosis of PNH is often delayed or missed. DeciphEHR^TM is an electronic health record toolkit developed to help with early identification of patients with PNH. It identifies numerous clinical codes that may be associated with a diagnosis of PNH and identifies some of those codes as being high-priority. The test characteristics (e.g. sensitivity, specificity, and positive predictive value) of this toolkit are unknown.

Methods:

We designed a retrospective cohort study using the University of Mississippi Medical Center's de-identified electronic retrospective cohort explorer (University of Mississippi Medical Center, Center for Informatics and Analytics (2020): Patient Cohort Explorer. figshare. Software. https://doi.org/10.6084/m9.figshare.12252737) to evaluate the test characteristics, including sensitivity, specificity, and positive predictive value, of the deciphEHR toolkit high priority codes in identifying patients with PNH. DeciphEHR toolkit identifies high priority codes included ICD-10 and SNOMED codes associated with PNH (e.g. codes for idiopathic aplastic anemia, aplastic anemia due to external causes, Budd-Chiari syndrome) .. The true cases of PNH were taken to be those patients with a diagnosis of PNH in the electronic retrospective cohort explorer. The sensitivity, specificity, and positive predictive value (PPV) of all high priority codes in identifying those patients with a PNH diagnosis was calculated.

Results:

1,533,821 patients were included in the retrospective cohort explorer. 28 patients in the data warheous had a diagnosis of PNH. All test specificities were greater than 99%. . SNOMED codes for cytopenia, aplastic anemia, venous thrombosis, and acute embolism and thrombosis of unspecified vein had the highest sensitivity values of 89, 46, 46, and 25 percent. ICD- 10 codes for aplastic anemia, idiopathic anemia, other constitutional aplastic anemia, and other specified aplastic anemia had sensitivity of 46, 25, 10.7, and 10.7 percent respectively. ICD-10 codes and SNOMED codes for idiopathic aplastic anemia had the highest PPV of 18-19%. Two other sets of ICD-10 and SNOMED codes (i.e. aplastic anemia due to other external causes, other constitutional aplastic anemias) had PPV's of approximately 6%. All other high priority codes had PPV's of 3% or less.

Conclusions:

The high priority codes listed in the deciphEHR toolkit for identifying patients with PNH may, based on these findings, need to be reconsidered based on the differential PPV of the codes. A threshold for alert fatigue should be used to avoid receiving alerts too frequently if the deciphEHR toolkit is integrated into an electronic health system.

Henegan:Pfizer: Honoraria; Aveo: Honoraria; Targeted Oncology: Honoraria; Exelexis: Honoraria.