In this study we observed: 1) No socio-demographic or psychological variables studied by us are confounding the IPDS as a screener for PDs. 2) The sensitivity and specificity of the IPDS supported the values reported by Germans et al. [15
]. 3) The PLSPM analysis of the IPDS showed satisfactory coefficients (cf. standardized regression coefficient and factor loadings) and an adequate fit value.
We found that the GSI and the GAF-S as measures of psychopathology and the GAF-F as a measure of function as well as the IPDS were significantly associated with the presence of PDs in bivariate analysis. A new finding is that only the IPDS score remained significant in the multivariate analysis. Our interpretation of these results is that psychological and functional variables do not seem to interfere to any significant extent on the IPDS as a screener for PDs.
Among the previous studies of the screening properties of the IPDS, comparisons with the study of Germans et al. [15
] is the most relevant one since they also studied psychiatric outpatients and had a base rate of 50%. Our findings concerning the IPDS on sensitivity, specificity, positive and negative predictive value, and proportion correctly classified were close to those of Germans et al., and could be considered as a replication. In POC samples with a base rate of 50% for PDs, a sensitivity of 0.82 a specificity of 0.74, seem to the optimal screening ability reached by the IPDS using a brief 5 items version consisting of the IPDS items 4-8 with cut-off ≥ 2 positive items.
What do such figures mean in practical clinical work? In a sample of 100 patients admitted to the POC, 50 have PDs, when the PDs base rate is 50%. A sensitivity of 0.82 tells that 41 (50 * 0.82) of these 50 PDs patients are correctly identified, while 9 are missed as false negatives. Among the 50 patients without PDs 37 (50 * 0.74) are correctly identified without PDs, while 13 are rated as false positive for PDs. Taken together 78 of the 100 patients are correctly classified. Doing 54 (41+13) instead of 100 SCID-II interviews, will miss 9 PDs patients and have 13 negative SCID-II interviews. If this consequence of sparing 46 interviews is considered suboptimal, setting a lower cut-off with higher sensitivity will reduce the number of PDs patients missed, however at a price of performing more negative SCID-II interviews. Therefore the cut-off value of the items, as well as the item combination used should be considered when the price of false negatives and false positives are considered at the local POC.
The PLSPM analysis indicated a relatively strong relation between the IPDS and PDs, i.e. IPDS explains 41 percent of the variation in the PDs. The analysis also supports IPDS as a second order construct with five different sub dimensions. Both a set of satisfactory factor loadings and an adequate fit value support this conceptualization of IPDS. Two factor loadings were, however, relatively weak; cf. the concept histrionic PD in Figure and the low coefficients of 0.27 and 0.34. This may indicate that histrionic PD does not represent a valid dimension of IPDS, but it may as well be a result of setting specific conditions. Our sample was relatively low (N = 145) and it is legitimate to ask if this is large enough for the second order PLSPM analysis. PLSPM is categorized as a "soft modeling technique" if compared with covariance based structure equation modeling technique (such as LISREL). Soft modeling means an approach where no strong assumptions (with respect to the distributions, the sample size and the measurement scale) are required [27
], and we therefore conclude that our sample size is adequate for the second order PLSPM analysis. However, further research is clearly needed to address these issues.
The positive hit rate of the 11 item version of the IPDS with cut-off ≥4 varied from 76% for cluster A PDs to 94% for PD-NOS (Table ). These findings were in accordance with those of Germans et al. [15
]. When we compared the distribution of positive ratings of the 11 IPDS items, item #5 (social anxiety) and item #6 (unwilling to get involved) were significantly more common in our sample than in Germans et al., while the distribution of the other 9 items did not differ significantly. The most probable explanation is differences in the diagnostic distribution of the samples, since our sample contained significantly more cluster A and C PDs and significantly fewer cluster B PDs compared to the sample of Germans et al.
We also want to point out the considerable difference between the IPDS items concerning their proportions of correct classification. The two best items (item #5 and #6, with 78% and 74%, respectively) belonged to avoidant PD, while the two poorest ones (#2 and #3 with 52% and 55%, respectively) belonged to histrionic PD. This result confirms the finding from the path analysis, namely that the histrionic items are the weakest ones in relation to the PD concept of the IPDS.
Our results have to be considered in the light of some limitations. The reference diagnoses based on the SCID-II interviews were performed by 22 therapists, that each did from 1 to 15 interviews. In spite of the SCID-II training seminar, there is a definite risk for heterogeneity of the diagnostic practice concerning PDs. Further, we included 145 patients, which could be considered as suboptimal for the power of some of the statistical tests. The exclusion of patients referred with drug and alcohol dependence as main diagnosis might contribute to a selection bias, mostly decreasing the prevalence rate of cluster B PDs. A certain degree of consensus has emerged concerning prevalence rates of PDs in the general population [4
]. Seeking treatment is however related to a number of clinical and demographical factors [29
], and prevalence rates and distribution of PDs in clinical samples in vary considerably with methodological and diagnostic tools used in the assessments [1
]. In The Rhode Island Methods to Improve Diagnostic Assessment and Services (MIDAS) project [30
] patients referred to a community based POC were diagnosed with reliable and valid procedures. The project found a base rate of 45% for PDs and a 24% prevalence rate Cluster B among those having a PD. Despite our lower prevalence rate of 14% and Germans et al. [15
] higher prevalence rate of 48% of Cluster B the sensitivity and specificity of IPDS in the studies are fairly comparable.
Finally, the IPDS was developed using 11 DSM-III-R criteria for PDs. 10 of these criteria were retained in DSM-IV, and one (histrionic PD criterion 7) was omitted. This omission is a minor point in our view since we test to what extent a set of criteria function as a good screening for PDs in DSM-IV. Such a task does demand that the criteria are derived from DSM-IV, although that would have been to some advantage.
Since performing SCID-II interviews are extensive time consuming a screening instrument for PDs is needed in POC due to heavy work burdens and lack of qualified SCID-II interviewers. Taking the limitations of the study into account we regard the short and feasible IPDS in Norwegian as a useful screening instrument in a busy clinical setting until the revision of the DSM-IV is completed.