Short Screening Scale for Posttraumatic Stress Disorder

I offer these comments about the short screening scale for posttraumatic stress disorder (PTSD) described by Naomi Breslau, Ph.D., and colleagues (1), both to help those who will use this instrument and to promote better understanding of the roles of sensitivity, specificity, and prevalence in test interpretation (2).

From the authors’ Table 3, one sees that when four or more items on the seven-symptom scale are endorsed, sensitivity equals 0.803 and specificity equals 0.973. In the population evaluated, the proportion of individuals with PTSD, or disorder prevalence, was 142/1,830=0.078. Using this prevalence, Dr. Breslau and colleagues correctly calculated the positive predictive value as (sensitivity × prevalence) / (sensitivity × prevalence + [1 – prevalence] × [1 – specificity]) = 0.713 and the negative predictive value as (specificity × [1 – prevalence]) / (specificity × [1 – prevalence] + prevalence × [1 – sensitivity]) = 0.983. The authors should have pointed out, however, that these values are correct only for the prevalence in their group. The prevalence of PTSD has been reported to range from 1% in community-based samples to 75% among rape victims (3). If the cutoff of four or more symptoms is used in a population where the prevalence is 2%, the positive predictive value is 0.378 and the negative predictive value is 0.996; if the same cutoff is used in a population in which the prevalence is 50%, the positive predictive value is 0.967 and the negative predictive value is 0.832.

The authors say that with the cutoff of four or more symptoms, “less than 2% of ‘true’ cases of PTSD were missed, whereas 29% of subjects without PTSD were falsely identified as having PTSD” (p. 910). This is incorrect. A positive predictive value of 0.713 means that about 29% of the subjects identified as having PTSD do not actually have it, and a negative predictive value of 0.983 means that about 2% of the subjects identified as not having PTSD actually have the disorder. The scale’s sensitivity is the probability that it will detect PTSD if a subject has the disorder; the scale’s specificity is the probability that a subject without PTSD will be deemed not to have the disorder. So if sensitivity is 0.803, the fraction of “true” cases missed is (1 – sensitivity) = (1 – 0.803) = 0.197, or about 20% of the subjects. If specificity is 0.973, the chance of falsely identifying someone without PTSD as having the disorder is (1 – specificity) = (1 – 0.973) = 0.027, or about 3%.

Dr. Breslau and colleagues say that the cutoff of four or more symptoms is well suited “to maximize the number of true cases of PTSD” detected initially when a subsequent evaluation will be used “to reclassify those who were wrongly classified as having the disorder” (p. 911). However, better cutoffs for such a purpose might be two or more symptoms (sensitivity=0.993) or three or more symptoms (sensitivity=0.951); these cutoffs would miss only 1% or 5% of the actual cases of PTSD, respectively.