Identifying personality disorders: towards the development of a clinical screening instrument

The study objective was to identify a set of personality disorder (PD) criteria from the DSM PD diagnostic sets that can be used to detect subjects with an increased likelihood of having a PD diagnosis. In a series of outpatients evaluated systematically in two waves for every criteria item for 12 DSM-III-R PDs, stepwise logistic regression identified 45 criteria as discriminative for their specific PDs, which are selected for further analysis to assess their ability to discriminate for any PD. Receiver operating characteristic (ROC) analysis is used to evaluate their discriminative power in an independent conjoined sample (N = 1,342) from six centers that assessed every PD criteria item by structured instrument (Structured Clinical Interview for DSM-III-R PDs [SCID-II, Personality Disorder Examination [PDE], and Structured Interview for DSM-III-R PDs [SIDP-R]). The cutoff that maximizes information gain is used to determine the diagnostic threshold (DT). Initially, 15 of 45 criteria are identified. At the 0.43 PD prevalence, a DT of 2 or more of the 15 PD criteria across samples is optimal. The maximum information gain (MIG) is .42 bits, and the AUR is 0.94+/-.007. Other performance indices at this cutoff are .90 sensitivity, .84 specificity, .81 positive predictive power (PPP), .91 negative predictive power (NPP), and .86 hit rate (HR). Taken collectively, the 15 PD criteria selected by the data reduction techniques suggest a narrowed set to be assessed in screening for the presence or absence of any PD with comparable or better psychometric properties than other tests routinely used for diagnosing medical and psychiatric disorders. If specific PD categorization is needed, a second-step comprehensive assessment should follow.