Introduction: Cefepime is a broad-spectrum, fourth-generation cephalosporin associated with neurotoxicity in ~3.7% of patients exposed to the drug. A Neurotoxicity Assessment Tool (NAT) was recently developed and internally validated to identify high-risk patients, but external validation has not yet been reported. This study aimed to assess the NAT’s prognostic performance in a separate critically ill population. Methods: This single-center, retrospective, type-4 model validation study included adult patients admitted to intensive care units (ICUs) within a large academic medical center who were treated with cefepime between January 2018 and December 2021. Patients with alternative causes of altered mental status or prior beta-lactam exposure were excluded. The primary outcome was cefepime-induced neurotoxicity (CIN), determined using the Naranjo adverse drug reaction probability scale. NAT scores were calculated using the original model’s variables: age, weight, creatinine clearance, and Charlson Comorbidity Index. Logistic regression explored the relationship between NAT scores, the model’s independent variables, and the development of CIN. Model discrimination, calibration, and diagnostic performance metrics were evaluated using a NAT score cutoff of 35. Results: Among 2,049 screened patients, 517 met study criteria. CIN occurred in 7 patients (1.4%). Compared to the derivation cohort, our population had greater comorbidity burden, lower renal function, and a lower incidence of CIN. The NAT score demonstrated limited ability to distinguish between patients with and without CIN (AUC 0.48). Using the defined high-risk threshold cutoff of 35, the NAT score yielded a sensitivity of 42.9%, specificity of 48.2%, positive predictive value of 1.1%, and negative predictive value of 98.4%. Conclusions: In this external ICU population, the NAT score demonstrated strong negative predictive value but lower overall predictive performance. Differences in patient characteristics and the prevalence CIN may have influenced these findings. Further refinement and evaluation of the NAT may enhance its utility beyond the model’s derivation population.
Barnes et al. (Sun,) studied this question.