Quantitative methods to improve bivalirudin dosing in pediatric cardiac ICU patients.

Abstract

Background: A gap in knowledge exists related to optimal bivalirudin dosing in children. The purpose of our analysis is to use quantitative methods and baseline data to quickly predict the optimal therapeutic bivalirudin dose for children.

Methods: We developed an internal database of pediatric patients on ECMO or VAD, including baseline patient information, bivalirudin doses, and partial thromboplastin time (PTT) measurements throughout the treatment period. We fit an analysis of covariance (ANCOVA) model to the baseline data to determine the best predictors of therapeutic bivalirudin dose. We used five-fold cross-validation to ensure the model was not overfitting to any specific data subset.

Results: The most notable variables that were statistically significant (p < .05) were: the primary use of bivalirudin for heart failure prophylaxis, no complications before bivalirudin administration, other reasons for bivalirudin use, other race (including Asian, pacific islander, and native American), Hispanic or Latinx ethnicity, primary diagnosis of heart failure, and primary diagnosis of myocarditis. To compare our model-predicted dose and the actual starting dose administered to the patients, we looked at how far off each of those was from the therapeutic dose. The mean of absolute differences was 0.28 mg/kg/hr for the administered starting dose and 0.23 mg/kg/hr for the model-predicted dose; therefore, the model results in an improvement of 18% in the difference from the therapeutic dose.

Conclusion: Our model provides an initial framework for determining a starting bivalirudin dose that takes into account patient demographic information and baseline admission data.