Group Director, Real-World Value and Evidence Janssen Scientific Affairs, LLC, United States
Background: Cytokine release syndrome (CRS) is an acute systemic inflammatory condition caused by an immune response to a treatment or an infection, for which the International Classification of Diseases (ICD-10) code was introduced in October 2020 with grading-level granularity. However, given the novelty of the code and the presence of an unspecified grade option, other means to identify CRS in retrospective databases, particularly severe CRS, may be warranted.
Objectives: To assess the performance of a published administrative claims-based algorithm (Keating et al, 2022) to identify CRS in two retrospective databases.
Methods: This study separately analyzed the MarketScan Databases (administrative payer claims) and the Premier Healthcare Database (hospital chargemaster data). Adults with an ICD-10 diagnosis for CRS were identified. Using the published algorithm, diagnoses for common CRS symptoms (e.g., fever, hypoxia, hypotension) and treatments of interest (e.g., tocilizumab, vasopressors) within 14 days of the CRS diagnosis (or during a CRS-related hospitalization for Premier) were used to identify any grade CRS and severe CRS. The sensitivity of the algorithm to identify any grade CRS and CRS severity (mild = grade 1-2; severe = grade 3+) was calculated in those with specified CRS grades in ICD-10.
Results: In MarketScan, 674 patients were identified by ICD-10 for CRS (11% grade 1, 17% grade 2, 17% grade 3+, 55% unspecified). The algorithm’s overall sensitivity to detect any grade CRS was 91%. Among patients with ICD-10 for grade 1 CRS, the algorithm detected that 72% of the patients had any grade CRS and 66% had mild CRS. Among patients with ICD-10 for grade 2 CRS, the rates were 94% and 76%, respectively. Among patients with ICD-10 for grade 3+ CRS, the algorithm detected 98% with any grade CRS but only 15% were identified as severe CRS. In Premier, 7,388 patients were identified by ICD-10 for CRS (12% grade 1, 16% grade 2, 11% grade 3+, 61% unspecified). The overall sensitivity for any grade CRS was 86%. Among patients with ICD-10 for grade 1 CRS, the algorithm detected that 71% of the patients had any grade CRS and 69% had mild CRS. The rates were 91% and 88%, respectively, in patients with ICD-10 for grade 2 CRS. Among those with ICD-10 for grade 3+ CRS, the algorithm detected 86% with any grade CRS, while only 2.5% were identified as severe CRS.
Conclusions: The algorithm published by Keating et al. performed well in identifying any grade CRS in retrospective databases, but few cases of severe CRS were identified in the diagnosed patients. Given the high proportion of CRS diagnoses with an ICD-10 code of unspecified grade, a modified algorithm may be warranted to better identify severity of CRS, potentially by adding procedures associated with the management of severe CRS to the algorithm.
