Abstract 14344: Pitfalls of ICD-10 Codes for Identifying Pulmonary Embolism in Electronic Records: Results From the Multicenter PE-EHR+ Study
Behnood Bikdeli, Antoine Bejjani, Candrika D Khairani, Ying-Chih Lo, David Jimenez, Stefano Barco, Shiwani Mahajan, Cesar Caraballo, Eric A Secemsky, Erik Klok, Andetta Hunsaker, Ayaz Aghayev, Alfonso Muriel, Mohamad A Hussain, Abena Appah-Sampong, HAMID MOJIBIAN, Yuan Lu, Zhenqiu Lin, Sanjay Aneja, Rohan Khera, Samuel Goldhaber, Li Zhou, Manuel Monreal, Gregory Piazza, Harlan M Krumholz- Physiology (medical)
- Cardiology and Cardiovascular Medicine
Background: Many research studies from electronic databases and even the American Heart Association annual statistics for pulmonary embolism (PE) rely on the International Classification of Disease, 10 th modification (ICD-10) codes. However, the validity of ICD-10 codes for PE remains uncertain.
Methods: Using a pre-specified protocol, we identified three groups of patients in the Mass General-Brigham Health system hospitals (MGB, 2016-2021) in equal distribution: Those with ICD-10 Principal discharge codes for PE, those with ICD-10 secondary codes for PE, and those without codes for PE (N=578 each). The accuracy of ICD-10 codes for identification of PE was assessed in reference to review of each chart by two independent physicians who used pre-specified criteria for presence of PE. Weighted estimates were obtained by considering the total of number of hospitalizations at MGB in each group.
Results: After excluding duplicates, 1712 entered the analysis (age: 60.6 years, 52.3% female). Using the ICD-10 PEs only in the Principal discharge diagnosis, sensitivity and specificity were 60.3% and 94.6%, respectively. Although use of Principal-or-secondary ICD-10 PE codes had a sensitivity of 99.8% in the study sample, the weighted sensitivity in the parent cohort was reduced to 83.2%. The positive predictive value of Principal ICD-10 discharge codes and Principal-or-secondary ICD-10 codes were 91.9% and 79.2%, respectively (Table).
Conclusions: Although the Principal discharge diagnosis codes for PE have an excellent positive predictive value, they miss nearly 40% of new PE events in hospitalized patients. Integration of secondary discharge codes only partially improves the sensitivity at the cost of reducing the positive predictive value, highlighting the need for modified strategies, such as use combination with procedure codes, or use of natural language processing -when feasible -for identification of patients with PE.