Impact of Adverse Event Definitions on Real-World Detection of Oncology Immune-Related Adverse Events
Author(s)
Zachary T. Rivers, PharmD, PhD1, Akash Mitra, PhD1, Victoria L Chiou, MD1, Halla Nimeiri, MD1, Rotem Ben-Shachar, PhD1, Charu Aggarwal, MD2;
1Tempus AI, Chicago, IL, USA, 2University of Pennsylvania, Philadelphia, PA, USA
1Tempus AI, Chicago, IL, USA, 2University of Pennsylvania, Philadelphia, PA, USA
OBJECTIVES: Immune checkpoint blockade (ICB) represents an advance in cancer care, and has introduced novel toxicities, called immune-related AEs (irAEs). Accurate irAEs identification in real-world data (RWD) is vital for assessing the long-term safety profile ICB. A barrier to this identification is a lack of consensus in the definitions of AEs. Here, we apply three different peer-reviewed irAE definitions in a RWD cohort of non-small cell lung cancer (NSCLC) patients.
METHODS: We evaluated RWD from Tempus clinicogenomic data linked to Komodo Health’s claims. Patients were included if they had a diagnosis of stage 3C+ NSCLC, and were treated with ICB therapy for >=60 days. We assessed irAEs for up to one-year of ICB treatment or the last clinical record or claim if one-year follow-up was not available. We applied three definitions of irAEs to this cohort which varied in the irAEs included, the ICD-10 codes used to define them, and the length of pre-treatment washout period to exclude prevalent diagnoses.
RESULTS: This cohort included 4,831 patients; 44.3% (n=2,141) treated with ICB monotherapy and 55.7% (n=2,690) treated with ICB and chemotherapy. Study A defined 9 irAEs with an overall prevalence of 41.0% (n=1,981) irAEs, while study B defined 10 irAEs with an overall prevalence of 75.4% (n=3,849) irAEs and study C defined 3 irAEs, with an overall prevalence of 5.4% (n=264) irAEs. A chi-squared test of these counts was significant with a p value of <2.2e-16.
CONCLUSIONS: We demonstrated that the identification of oncology-associated irAEs in RWD varies based on the definitions used. This variance can impact post-market surveillance, clinical practice guidelines, and patient care. Researchers using RWD should accurately communicate the definitions used, and utilize sensitivity analyses to understand how definitions may impact findings.
METHODS: We evaluated RWD from Tempus clinicogenomic data linked to Komodo Health’s claims. Patients were included if they had a diagnosis of stage 3C+ NSCLC, and were treated with ICB therapy for >=60 days. We assessed irAEs for up to one-year of ICB treatment or the last clinical record or claim if one-year follow-up was not available. We applied three definitions of irAEs to this cohort which varied in the irAEs included, the ICD-10 codes used to define them, and the length of pre-treatment washout period to exclude prevalent diagnoses.
RESULTS: This cohort included 4,831 patients; 44.3% (n=2,141) treated with ICB monotherapy and 55.7% (n=2,690) treated with ICB and chemotherapy. Study A defined 9 irAEs with an overall prevalence of 41.0% (n=1,981) irAEs, while study B defined 10 irAEs with an overall prevalence of 75.4% (n=3,849) irAEs and study C defined 3 irAEs, with an overall prevalence of 5.4% (n=264) irAEs. A chi-squared test of these counts was significant with a p value of <2.2e-16.
CONCLUSIONS: We demonstrated that the identification of oncology-associated irAEs in RWD varies based on the definitions used. This variance can impact post-market surveillance, clinical practice guidelines, and patient care. Researchers using RWD should accurately communicate the definitions used, and utilize sensitivity analyses to understand how definitions may impact findings.
Conference/Value in Health Info
2025-05, ISPOR 2025, Montréal, Quebec, CA
Value in Health, Volume 28, Issue S1
Code
RWD102
Topic
Real World Data & Information Systems
Topic Subcategory
Reproducibility & Replicability
Disease
SDC: Oncology