Font Size: a A A

Neural network modeling of unipolar depression: Patterns of recovery and prediction of outcome

Posted on:1997-12-01Degree:Ph.DType:Dissertation
University:Boston UniversityCandidate:Luciano, Joanne Sylvia, JrFull Text:PDF
GTID:1464390014981288Subject:Mental Health
Abstract/Summary:
Neural network methods were used in two studies of unipolar depression. Study 1 explored the dynamics of recovery and revealed different clinical symptom recovery patterns for desipramine (DMI), a tricyclic antidepressant drug therapy, and cognitive behavioral therapy (CBT), a psychotherapy. The methods included statistical tests of patient's response times and parameter fits based on methods derived from optimal control theory. Study 2 predicted therapeutic outcome at highly significant levels (;Study 1. To explore recovery patterns, a linear second order system and a nonlinear shunting neural network were used. The linear second order model gave statistically superior results. We modeled changes in overall severity (HDRS total) and severity of, and interactions among, seven symptoms derived from the Hamilton Depression Rating Scale during the initial six weeks of treatment in two patient groups. The two groups were six patients who responded to CBT and six patients who responded to DMI. There was no difference in response time for overall severity. In both groups mood was the first symptom to improve and middle/late sleep was the last. Symptom improvements clustered differently by treatment. Mood and cognitions (sad mood, anxious mood, thoughts of guilt or suicide) improved significantly earlier (;Study 2. To look for nonlinear predictive relationships among pre-treatment symptoms, treatment, and outcome, several backpropagation studies were performed on raw and transformed data from 99 patients. This study investigated whether linear and nonlinear methodologies could reliably predict percent improvement of clinically depressed individuals exposed to fluoxetine, desipramine, or cognitive behavioral therapy. The linear model performed at chance levels with no factor statistically significant. However, both nonlinear models, backpropagation and quadratic regression, predicted outcome at statistically significant levels. This suggests that symptoms can significantly predict treatment outcome if nonlinear effects are included.
Keywords/Search Tags:Recovery, Outcome, Network, Depression, Nonlinear, Patterns
Related items