Multilayer perceptrons can predict cognitive status during recovery from chronic substance misuse: Implications for individualized treatment planning

Frank R. Funderburk, Karen I. Bolla, Jean Lud Cadet

Research output: Contribution to conferencePaperpeer-review

Abstract

We investigated the feasibility of using feedforward multilayer perceptrons to predict the performance of chronic users of cocaine on the interference component of the Stroop Color-Word Task after a one-month period of enforced abstinence. A total of 77 individuals who met the diagnostic criteria for cocaine dependence or polydrug dependence volunteered to be admitted to a special research unit where they remained abstinent from drugs of abuse for at least one month. On two separate occasions, once within 3 days of admission and again at least 28 days following admission, participants completed a battery of neuropsychological tests. The output variable to be predicted by the model was performance on the interference component of the Stroop Color-Word Task, a cognitive task that reflects the ability of the individual to use contextual information to guide behavioral responding. Inputs to the model included factor scores derived from a principal components analysis of the performance on the neuropsychological test battery administered within 3 days of admission to the unit as well as reported levels of cocaine and alcohol consumption during the month prior to the period of enforced abstinence. The network implemented contained two hidden layers and was trained using a standard error backpropagation algorithm. Several versions of the model using this same general architecture were evaluated. Training was very successful, routinely reducing average RMS error to <.05, with median correlation between predicted and criterion values of over r = .86. Generalization of the training to randomly selected sets of 15 test observations not included in the training sample showed a median r = .64. The networks were moderately successful in predicting outcomes for novel exemplars. The pattern of weights observed in the neural networks following successful training suggested that cognitive status at one month of abstinence, as measured by the Stroop task, is strongly influenced by various person-specific states reflecting that individual's `cognitive reserve,' in interaction with the levels of recent prior drug consumption. These preliminary results suggest that neural network approaches can provide a useful characterization of the dynamics of cognitive recovery during abstinence from drugs of abuse that could lead to the development of more efficacious therapeutic interventions aimed at reducing the use of illicit and/or harmful substances.

Original languageEnglish (US)
Pages663-668
Number of pages6
StatePublished - Jan 1 2000
EventInternational Joint Conference on Neural Networks (IJCNN'2000) - Como, Italy
Duration: Jul 24 2000Jul 27 2000

Other

OtherInternational Joint Conference on Neural Networks (IJCNN'2000)
CityComo, Italy
Period7/24/007/27/00

ASJC Scopus subject areas

  • Software
  • Artificial Intelligence

Fingerprint

Dive into the research topics of 'Multilayer perceptrons can predict cognitive status during recovery from chronic substance misuse: Implications for individualized treatment planning'. Together they form a unique fingerprint.

Cite this