VALIDATION OF ALGORITHMS TO IDENTIFY CARDIAC-RELATED DEATHS IN POPULATION-BASED STUDIES USING A CLAIMS DATABASE
Author(s)
Ryan Jacobs, MD1, Alessandra Ferrajoli, MD2, Xiaoliang Wang, PhD3, QIANHONG FU, MS3, Andrea Barthel, MS4, Neil Lamarre, PhD4, Yajin Zhao, MS4, Jeffrey Anderson, ScD4, Derrick van Beuge, PharmD3, Ayad K. Ali, PhD3, Michael Fradley, MD5;
1Atrium Health Levine Cancer Institute, Wake Forest University School of Medicine, Charlotte, NC, USA, 2The University of Texas MD Anderson Cancer Center, Houston, TX, USA, 3BeOne Medicines, Ltd, San Carlos, CA, USA, 4Genesis Research Group, Hoboken, NJ, USA, 5University of Pennsylvania School of Medicine, Philadelphia, PA, USA
1Atrium Health Levine Cancer Institute, Wake Forest University School of Medicine, Charlotte, NC, USA, 2The University of Texas MD Anderson Cancer Center, Houston, TX, USA, 3BeOne Medicines, Ltd, San Carlos, CA, USA, 4Genesis Research Group, Hoboken, NJ, USA, 5University of Pennsylvania School of Medicine, Philadelphia, PA, USA
OBJECTIVES: Cardiac-related death (CRD) is a cause-specific outcome of interest for assessing safety and effectiveness of new treatments. While mortality data can be captured in claims-based studies via linkage to other sources, cause of death may not always be accessible. This study assessed claims-based algorithms to identify CRD.
METHODS: This retrospective cohort study utilizing the US Medicare Fee-For-Service database included beneficiaries with indolent B-cell malignancies who started ibrutinib in 2017-2021 and were followed until the earliest of ibrutinib discontinuation, death, disenrollment, or study end. Four algorithms were evaluated using pre-specified ICD-10 codes for cardiac-related events at primary diagnosis within 7 (A) or 30 days (B), any diagnosis (primary or secondary in the claim) within 7 (C) or 30 days (D) of death. National Death Index (NDI) was used as the reference. Accuracy, sensitivity, specificity, PPV, NPV, and Cohen’s kappa were calculated.
RESULTS: In total, 13,241 patients were included (male, 57.7%; median age, 77.2 years; Non-Hispanic White, 90.7%; CLL/SLL, 78.9%). Using NDI, 568 (4.3%) CRD events were identified, with an incidence rate of 35.2 per 1000 person-years. Algorithms using any diagnosis had higher sensitivity (C: 0.66 [95% CI, 0.62-0.70]; D: 0.70 [95% CI, 0.66-0.74]) than primary diagnosis only (A: 0.36 [95% CI, 0.32-0.40]; B: 0.49 [95% CI, 0.45-0.53]). Specificities and NPVs were similar across algorithms (specificity range, 0.94-0.97; NPV range, 0.97-0.99). PPVs were slightly higher using primary diagnosis (A: 0.39 [95% CI, 0.35-0.43]; B: 0.37 [95% CI, 0.34-0.41]) vs any diagnosis (C: 0.35 [95% CI, 0.33-0.38]; D: 0.33 [95% CI, 0.31-0.36]). Algorithm C had the highest Cohen’s kappa (0.43; 95% CI, 0.40-0.46).
CONCLUSIONS: With high specificity and NPV and moderate sensitivity and PPV, the claims-based CRD definition at any diagnosis within 7 days of death (algorithm C) had the highest Cohen’s kappa in identifying CRD. Increasing the identification window did not improve algorithm accuracy.
METHODS: This retrospective cohort study utilizing the US Medicare Fee-For-Service database included beneficiaries with indolent B-cell malignancies who started ibrutinib in 2017-2021 and were followed until the earliest of ibrutinib discontinuation, death, disenrollment, or study end. Four algorithms were evaluated using pre-specified ICD-10 codes for cardiac-related events at primary diagnosis within 7 (A) or 30 days (B), any diagnosis (primary or secondary in the claim) within 7 (C) or 30 days (D) of death. National Death Index (NDI) was used as the reference. Accuracy, sensitivity, specificity, PPV, NPV, and Cohen’s kappa were calculated.
RESULTS: In total, 13,241 patients were included (male, 57.7%; median age, 77.2 years; Non-Hispanic White, 90.7%; CLL/SLL, 78.9%). Using NDI, 568 (4.3%) CRD events were identified, with an incidence rate of 35.2 per 1000 person-years. Algorithms using any diagnosis had higher sensitivity (C: 0.66 [95% CI, 0.62-0.70]; D: 0.70 [95% CI, 0.66-0.74]) than primary diagnosis only (A: 0.36 [95% CI, 0.32-0.40]; B: 0.49 [95% CI, 0.45-0.53]). Specificities and NPVs were similar across algorithms (specificity range, 0.94-0.97; NPV range, 0.97-0.99). PPVs were slightly higher using primary diagnosis (A: 0.39 [95% CI, 0.35-0.43]; B: 0.37 [95% CI, 0.34-0.41]) vs any diagnosis (C: 0.35 [95% CI, 0.33-0.38]; D: 0.33 [95% CI, 0.31-0.36]). Algorithm C had the highest Cohen’s kappa (0.43; 95% CI, 0.40-0.46).
CONCLUSIONS: With high specificity and NPV and moderate sensitivity and PPV, the claims-based CRD definition at any diagnosis within 7 days of death (algorithm C) had the highest Cohen’s kappa in identifying CRD. Increasing the identification window did not improve algorithm accuracy.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
EPH115
Topic
Epidemiology & Public Health
Topic Subcategory
Safety & Pharmacoepidemiology
Disease
SDC: Oncology