External Validation of a Risk Score for Major Toxicity Among Nonsteroidal Anti‐Inflammatory Drug Users: Real‐World Application

• Main Authors: Daniel H. Solomon, Nina P. Paynter, Hongshu Guan, Joel M. Kremer
• Format: Article
• Language: English
• Published: Wiley 2020-05-01
• Series: ACR Open Rheumatology
• Online Access: https://doi.org/10.1002/acr2.11134

Objective We previously derived and validated a risk score for major nonsteroidal anti‐inflammatory drug (NSAID) toxicity over 1 year among NSAID users in a randomized controlled trial. This work was extended to examine the risk score's performance in an external population using real‐world data. Methods Patients enrolled in the Corrona Rheumatoid Arthritis (RA) Registry were included if they initiated use of an NSAID. We defined the original risk factors previously identified in the risk score for major NSAID toxicity: age; male sex; history of cardiovascular disease, hypertension, and diabetes; tobacco use; statin use; elevated serum creatinine and hematocrit values; and RA. Additionally, we defined the occurrence of major toxicity, including major adverse cardiovascular events, acute kidney injury, significant gastrointestinal events, and mortality. The original risk factors were assessed in Cox regression examining discrimination and calibration. Low (less than 1%), intermediate (1%‐4%), and high (more than 4%) risk categories for 1‐year risk were applied to the population. Results A total of 5231 patients from Corrona who had a new NSAID exposure period were included. The original risk score model showed good discrimination (C‐index 0.70). Not all of the original variables were statistically significant in real‐world data. Using the original risk score weights, 1363 (26.1%) patients had predicted risk of less than 1%, 3571 (68.3%) had predicted risk of 1% to 4%, and 297 (5.7%) had predicted risk of more than 4%. Conclusion The original NSAID major toxicity risk score demonstrated good model fit characteristics in this external real‐world cohort. These results suggest that such a risk score is valid in typical practice and could be considered for clinical care.