A study published in Infection Control & Hospital Epidemiology compared the use of logistic regression-derived risk scores and machine learning-derived decision trees for predicting multidrug-resistant gram-negative infections.
Researchers used a dataset of 1,288 patients with Escherichia coli or Klebsiella spp bacteremia to generate a risk score that predicts the likelihood a patient was infected with an extended-spectrum beta-lactamase bacteremia producer.
Fifteen percent, or 194, patients were infected with ESBL-producing bacteremia. The clinical risk score included 14 variables while the decision tree included five variables. Both methods had similar positive and negative predictive values.
"The decision tree was more user-friendly, with fewer variables for the end user, whereas the risk score offered higher discrimination and greater flexibility for adjusting sensitivity and specificity," study authors concluded.