Dr. Breslau and Colleagues Reply

Dr. Mossman correctly interprets the positive and negative predictive values in Table 3 of our article—namely, the values of those who scored four or more on our scale (defined as cases). A total of 29% did not have PTSD according to the clinical assessment, whereas of those who scored four or more (defined as noncases), 2% did have PTSD according to the clinical assessment. Despite our awareness of the common tendency to confuse positive and negative predictive values with sensitivity and specificity, we regret that we did not avoid this pitfall and that we misstated the positive and negative predictive values as if they were sensitivity and specificity (i.e., using as denominators the number of cases and noncases defined by the clinical assessment instead of by a score of four or more on the screening scale).

However, we disagree with Dr. Mossman in his emphasis on sensitivity and specificity instead of on positive and negative predictive values. The latter constructs are of considerable utility for potential users, who wish to know what proportion of persons identified as cases by the scale would be confirmed by clinical examination.

We wish to emphasize that the prevalence of PTSD in our study is typical of that in recent general population studies, such as the National Comorbidity Survey. Therefore, estimates of the performance of the scale based on this prevalence estimate are useful.

Our recommended cutoff of four or more symptoms was based on the data for both positive and negative predictive values in Table 3. We considered the fact that positive predictive value diminishes rapidly with lower cutoffs (i.e., two or three), whereas the accompanying improvement in negative predictive value is slight. If we were to use a cutoff of two or three instead of four, the number of identified cases that would not be confirmed would double or triple.