|
Abstract:
|
In Psychiatry, clinicians use criteria sets from the Diagnostic and Statistical Manual of Mental Disorders (DSM) to diagnose mental disorders. Most criteria sets have several symptom domains and in order to be diagnosed, an individual must meet the minimum number of symptoms required by each domain. In this scenario, the overall classification rule is the intersection of these domain-specific rules. One criticism of this approach is that all symptoms within a domain are treated as equally important in the sum score. In previous work, we proposed an iterative algorithm, built on either support vector machines or logistic regression, which fits decisions rules consistent with this structure. This algorithm is flexible and allows for item weights to be estimated. In this talk, we will compare the performance of our method without estimating item weights (consistent with current DSM criteria sets) to our method when we allow item weights to be estimated and used in the decision rule. The data for our application comes from a study of Complicated Grief, a relatively new psychiatric disorder, for which consensus criteria have not yet been established.
|
