OBJECTIVES: Many United States (US) cost-effectiveness and budget impact models of oncology treatments source Grade 3/4 adverse event (AE) costs from HCUPnet, a public website query tool for the National Inpatient Sample (NIS) of the Healthcare Cost and Utilization Project (HCUP). HCUPnet has shifted from allowing users to identify hospitalization costs and length of stay (LOS) via detailed diagnosis codes (International Statistical Classification of Diseases 10th Revision [ICD-10]) to broader categories not useful for parameterizing costs in economic models. We sought to estimate Grade 3/4 AE costs.
METHODS: The most commonly reported Grade 3/4 AEs were identified from the prescribing information (PI) for all Food and Drug Administration (FDA) approved novel therapies for an oncology indication in 2023 through mid-2024. Selected AEs were mapped to ICD-10 codes. Accounting for survey and sampling design, weighted hospitalization costs and LOS were calculated for each code from the most recent NIS dataset available (2021) using R.
RESULTS: Sixteen novel therapies were approved by the FDA from 2023 to mid-2024 for oncology indications. Thirty-five Grade 3/4 AEs were reported in the PIs of ≥4 treatments. Fatigue, decreased lymphocyte count, and decreased hemoglobin were the most reported (14/16) followed by musculoskeletal pain, decreased neutrophil count, decreased potassium, and increased alanine aminotransaminase (all 13/16). Survey weighted means and standard errors were calculated for each mapped ICD-10 code. Of AEs with reliable estimates (i.e., adequate sample size), total hospitalization costs (2021 US$) ranged from $6,693 for upper respiratory tract infection to $25,251 for hemorrhage. LOS (days) ranged from 1.50 for cytokine release syndrome to 6.07 for decreased appetite. Compared with 2020 data, costs and LOS increased for 63% and 69% of AEs, respectively.
CONCLUSIONS: Many AEs of oncology treatments are common across recent approvals. Costs and LOS reported may be useful in parameterizing future economic models.
