2006 HSR&D National Meeting Abstract
3074 — Selecting an Optimal Cutscore for the PTSD Checklist (PCL) in VA Primary Care Settings
Yeager DE (Center for Health Care Research, Medical University of South Carolina)
Magruder KM (TREP)
Frueh BC (TREP)
The objectives of this secondary analysis are to: 1) assess the diagnostic performance of the PCL for screening for the presence of PTSD in primary care settings; 2) determine differential performance by gender and race.
These secondary analyses are based on a large database that was derived from a recent multi-site, cross-sectional study conducted at four Southeastern VA Medical Centers. A total of 1076 randomly sampled primary care patients were administered the PCL and queried about socio-demographic information; 858 were contacted by telephone and administered the Clinician Administered PTSD scale (CAPS) as a ‘gold-standard’ test to establish PTSD caseness. Receiver operator characteristic (ROC) curves were constructed to determine an optimal cutscore for the PCL relative to CAPS status. STATA statistical software’s roctab and roccomp commands were used to create and plot ROC curves, derive areas under the curve (AUC’s), and test for gender and race subgroup differences.
Since the development of the PCL, ‘50’ has remained the unchallenged cutscore (with few noteworthy exceptions) commonly used to indicate the presence of PTSD. The present study demonstrates that a cutscore of 50 is excessively high as evidenced by its low sensitivity (53.13%) and high specificity (94.88%). Our data suggest that a cutscore of 31 is more appropriate for certain applications; this is based on a nearly perfect balance of sensitivity and specificity (81.25% and 81.23% respectively). AUC analyses reveal that the PCL is adequate in terms of overall detection capability with an AUC of .883. Lastly, we compared ROC curves based on sex and race and found that no significant differences exist (x2 > .479; x2 > .944, respectively).
The results from the current analysis suggest that extant cutscores for the PCL should be modified in VA populations to reduce false positive and false negative forms of misclassification of PTSD. The PCL performs similarly for different race and gender group, thus interpretations of the PCL score do not need to be modified for race or gender.
More appropriate cutscores will enhance the utility of these screening assessments to the benefit of individuals, clinicians, and society at large.