Real-world interpretation of procalcitonin to guide antibiotic prescribing: a retrospective cohort study with regression discontinuity analysis.

Objective: It is unknown how providers are utilizing procalcitonin in the real world to make antibiotic prescribing decisions, and whether procalcitonin can limit harms related to antibiotic misuse. We examined how the probability of receiving antibiotics changed just below and above the pre-specified procalcitonin cut-points. We sought to understand whether providers interpret procalcitonin as a dichotomous or continuous diagnostic test. Design: Retrospective cohort study. Participants: We included adult inpatients who had procalcitonin collected as part of routine care. Patients with procalcitonin collected more than 48 hours after the first antibiotic dose, those discharged from the emergency department, or those on an obstetrics service were excluded. Methods: We used administrative data from the Health Data Compass database (2018-2019) at the University of Colorado Hospital to examine the correlation of pre-specified procalcitonin cut-points (0.1, 0.25, and 0.5 ng/mL) with the antibiotic treatment decisions using a regression discontinuity analysis (RDA), stratified by level of care. We constructed receiver operating characteristic (ROC) curves depicting the relationship between procalcitonin level and antibiotic prescribing. We performed sensitivity analyses by varying bandwidth for RDA. Results: The study included 4383 patients. A total of 68.9% received a full antibiotic course. RDA did not demonstrate any discontinuity at the pre-specified cut-points. However, sensitivity analyses showed a potential discontinuity at the 0.25 ng/mL cut-point in the ICU subgroup. The ROC curves were consistent with the RDA findings. Conclusions: This study suggests that most clinicians in real-world settings interpret procalcitonin as a continuous diagnostic test when prescribing antibiotics.