Premium
PII‐76
Author(s) -
Roy J. J.,
Dabbagh N.,
Nguyen A.,
Hildgen P.
Publication year - 2006
Publication title -
clinical pharmacology and therapeutics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.941
H-Index - 188
eISSN - 1532-6535
pISSN - 0009-9236
DOI - 10.1016/j.clpt.2005.12.201
Subject(s) - vancomycin , mean squared error , artificial neural network , statistics , mathematics , coefficient of determination , correlation coefficient , medicine , population , trough (economics) , artificial intelligence , computer science , biology , environmental health , bacteria , genetics , staphylococcus aureus , macroeconomics , economics
AIMS To evaluate whether ANNs could predict accurate vancomycin trough concentrations in the intensive care unit (ICU) population compared to those predicted with a population pharmacokinetic model (pop‐PK) and to assess the most significant inputs used in the ANN model. METHODS Data of this pilot study, collected from 110 records of 41 patients who received vancomycin at the ICU, was used to train, validate and test various ANN models. A two hidden layers General Feed Foward network using the hyperbolic tangent function with four processing elements in each layer and a step learning rule was chosen to predict vancomycin trough concentrations. The proposed model was built with the Neurosolutions software (NeuroDimension, Inc). The criterion for judging the best model was the correlation coefficient (r test ) and the root mean square error (RMSE) between predicted and observed troughs in the testing set. Predictions from the ANN were compared to the pop‐PK model using one way ANOVA and RMSE. RESULTS The proposed ANN model predicted accurate vancomycin troughs with a r test of 0.93 and a RMSE of 10.72 compared to a RMSE of 41.01 for the pop‐PK model. ANN predictions were not statistically different to the observed trough levels (p = 0.493) while the calculated pop‐PK levels were (p < 0.001). The most important inputs in the proposed model were delay, serum creatinine and age. CONCLUSION The ANN model can accurately predict vancomycin trough concentrations and may serve as a valuable tool to monitor vancomycin in the ICU. Clinical Pharmacology & Therapeutics (2005) 79 , P56–P56; doi: 10.1016/j.clpt.2005.12.201