A comparison of human and machine-based predictions of successful weaning from mechanical ventilation

Allan Gottschalk, M. Chris Hyzer, Ralph T. Geer

Research output: Contribution to journalArticle

Abstract

Purpose. To evaluate the ability of an appropriately trained neural network to correctly interpret a set of weaning parameters to predict the liberation of a patient from mechanical ventilation, and to contrast these predictions with those of human experts restricted to the same limited set of physiologic data. Methods. For each set of weaning parameters, a prediction was made by multiple realizations of a neural network and six expert volunteers. Results. The percentage of correct predictions made by the neural network when the decision threshold was set to 0.5 (range 0-1) was 83.3 ± 4.2 (mean ± SD) and that for the experts was 83.3 ± 4.7. Predictions by the network when the threshold was 0.5 had a sensitivity of 0.83 and a specificity of 0.84, compared with 0.90 and 0.77, respectively, for the experts. However, sensitivity and specificity comparable to those of the human experts could be obtained by adjusting the decision threshold of the network predictor so that only the most clearly ventilator-dependent patients would not be given a trial of extubation. Conclusion. When both are restricted to the same limited set of patient data, appropriately trained neural networks can be as effective as human experts in predicting whether weaning from mechanical ventilation will be successful.

Original languageEnglish (US)
Pages (from-to)160-169
Number of pages10
JournalMedical Decision Making
Volume20
Issue number2
DOIs
StatePublished - Jan 1 2000

Keywords

  • Decision analysis
  • Mechanical ventilation
  • Neural networks
  • Respiratory failure
  • Weaning

ASJC Scopus subject areas

  • Health Policy

Fingerprint Dive into the research topics of 'A comparison of human and machine-based predictions of successful weaning from mechanical ventilation'. Together they form a unique fingerprint.

  • Cite this