Support Vector Machines to improve physiologic hot flash measures: application to the ambulatory setting

Psychophysiology. 2011 Jul;48(7):1015-21. doi: 10.1111/j.1469-8986.2010.01155.x. Epub 2010 Dec 10.

Abstract

Most midlife women have hot flashes. The conventional criterion (≥2 μmho rise/30 s) for classifying hot flashes physiologically has shown poor performance. We improved this performance in the laboratory with Support Vector Machines (SVMs), a pattern classification method. We aimed to compare conventional to SVM methods to classify hot flashes in the ambulatory setting. Thirty-one women with hot flashes underwent 24 h of ambulatory sternal skin conductance monitoring. Hot flashes were quantified with conventional (≥2 μmho/30 s) and SVM methods. Conventional methods had low sensitivity (sensitivity=.57, specificity=.98, positive predictive value (PPV)=.91, negative predictive value (NPV)=.90, F1=.60), with performance lower with higher body mass index (BMI). SVMs improved this performance (sensitivity=.87, specificity=.97, PPV=.90, NPV=.96, F1=.88) and reduced BMI variation. SVMs can improve ambulatory physiologic hot flash measures.

Publication types

  • Research Support, N.I.H., Extramural
  • Research Support, Non-U.S. Gov't

MeSH terms

  • Adult
  • Artificial Intelligence*
  • Female
  • Galvanic Skin Response / physiology*
  • Hot Flashes / physiopathology*
  • Humans
  • Middle Aged
  • Monitoring, Ambulatory / methods*
  • Perimenopause / physiology*
  • Sensitivity and Specificity
  • Surveys and Questionnaires