Screening for posttraumatic stress disorder in ARDS survivors: validation of the Impact of Event Scale-6 (IES-6)

Data used for this analysis were collected as part of two multi-site prospective studies of ARDS survivors: (1) ARDS Network Long-term Outcomes Study (ALTOS) [12, 13] and (2) Improving Care of Acute Lung Injury (ALI) Patients (ICAP) [1]. ALTOS enrolled patients from > 40 hospitals across the USA who participated in four national ARDS Network trials [14‐17], with telephone-based follow-up at 6 and 12 months occurring between 2008 and 2014 [12, 13, 18]. The ICAP study evaluated ARDS survivors enrolled from 13 intensive care units (ICUs) at four hospitals in Baltimore, Maryland via in-person follow-up at 3, 6, 12, 24, 36, 48, and 60 months occurring between 2005 and 2012 [1, 19]. The Institutional Review Board of Johns Hopkins University School of Medicine and all participating institutions approved these studies, and patients provided an informed consent.

The IES-R is a 22-item screening instrument assessing PSTD symptoms [20]. Respondents are asked to report how distressed or bothered they are, over the past 7 days, by symptoms related to a specific trauma, using the following scale: “not at all” (item score 0), “a little bit” (score, 1), “moderately” (score, 2), “quite a bit” (score, 3), or “extremely” (score, 4). The IES-6 is an abbreviated 6-item version of the IES-R (Table 1, [11]) with its score calculated as the mean of the six items. The CAPS is a semi-structured interview conducted by a clinician for assessing clinically significant PTSD symptoms and serves as a reference standard for PTSD diagnosis [21].

We hypothesized, a priori, that the IES-6 would correlate with other clinically related constructs (i.e., anxiety symptoms and general mental health measures), as evaluated by the following instruments, administered at the same time as the IES-R in ALTOS and ICAP studies: (1) Short Form Health Survey-36 version 2 (SF-36, [22]) Mental Health Domain and Mental Component Summary scores, (2) Hospital Anxiety and Depression Scale (HADS) Anxiety and Depression Symptom Subscales [23], and (3) EQ-5D-3L [24] Anxiety/Depression Item.

In contrast to the prior related constructs, we hypothesized a priori that IES-6 scores would be more weakly correlated with other clinically unrelated constructs (i.e., physical function and healthcare utilization measures), as evaluated by the following instruments: (1) SF-36 Physical Function (PF) and Role Physical (RF) Domains; (2) EQ-5D-3L Mobility and Self-care items, (3) Functional Performance Inventory—Short Form (FPI-SF) [25] Body Care, Household Maintenance, and Physical Exercise Domains; and (4) healthcare utilization evaluated as ever vs. never using oxygen, x-ray imaging, magnetic resonance imaging, and hemodialysis. All of the aforementioned measures were previously evaluated in ARDS survivors [10, 25‐29].

Criterion validity of the IES-6 was assessed using the CAPS-based diagnosis of PTSD as previously described.

In both ALTOS and ICAP studies, trained research staff administered the IES-R at scheduled follow-up assessments [1, 12‐14, 19]. Respondents were asked about PTSD symptoms related to their ICU stay for ARDS. The IES-R, SF-36, HADS, EQ-5D-3L, and healthcare utilization survey were asked at follow-up in each of the studies. Only ALTOS participants completed the FPI-SF. As part of a previously reported sub-study [10], 60 of 77 (78%) participants from ALTOS and ICAP completed a telephone-based CAPS diagnostic interview within 1 week of completing the IES-R. Interviewers included a board-certified attending psychiatrist, a fourth-year psychiatry resident, and a medical student who were extensively trained in the administration of the structured interview.

The R statistical software package version 3.5.0 was used for statistical analysis. We calculated the correlation between IES-R and IES-6 scores across time points using a longitudinal random effects model with a random intercept. To allow interpretation of the regression coefficient as a correlation, we standardized both IES-R and IES-6 scores using the mean and standard deviation (SD) of scores at 6 months—the earliest time point shared by both datasets. To evaluate internal consistency as a measure of reliability, Cronbach’s alpha was calculated for IES-R and IES-6 at each follow-up visit. External construct validity was assessed by examining Pearson correlations (calculated via linear mixed effects) between IES-6 and measures of other constructs, expected to be related or unrelated. We used the same method as described above for correlation between IES-6 and IES-R, to calculate correlations over time between IES-6 and measures of these other constructs. We calculated an analogous set of correlations using IES-R to determine if shortening the scale resulted in attenuation of these expected relationships. Finally, criterion validity was assessed by constructing a receiver operating characteristic curve with a diagnosis of PTSD from the CAPS using the pROC R package. The optimal cutoff score for the IES-6 was obtained, and sensitivity, specificity, negative and positive predictive values were calculated for that cutoff score.

A total of 1001 ARDS survivors (ALTOS: n = 815; ICAP: n = 186) were included in this evaluation, with 51% female, 76% white, and a mean (SD) age of 49 (14) years (Table 2). Mean values for the IES-R and IES-6 instruments are reported at each of the follow-up assessments in Table 3. In addition, mean scores for measures of constructs expected to be either related or unrelated with PTSD were derived from all follow-up visits (Table 3).

The correlation (95% CI) of the IES-R and the IES-6 was 0.96 (0.94 to 0.97) overall assessments. Internal consistency for the IES-6 was good to excellent over time (Cronbach’s alpha of 0.86 to 0.91; Table 4).

The IES-6 was moderately correlated with related measures of mental health constructs, including SF-36 Mental Health Domain and Mental Component Summary (|r|, 0.42; 95% CI, 0.39 to 0.46; and 0.46; 95% CI, 0.42 to 0.49 respectively), the HADS Anxiety Subscale (|r|, 0.52; 95% CI, 0.49 to 0.55), the HADS Depression Subscale (|r|, 0.40; 95% CI, 0.37 to 0.44) and the EQ-5D-3L Anxiety/Depression Item (|r|, 0.32; 95% CI, 0.28 to 0.35). The IES-R 22-item version demonstrated the same pattern of associations with these measures, in both magnitude and direction (Table 5).

When compared to correlations with related constructs (above), the IES-6 demonstrated weaker correlations with constructs hypothesized to be unrelated to mental health (Table 5). Of these measures, healthcare utilization variables were least related to the IES-6 ranging from the utilization of MRI (|r|, 0.02; 95% CI 0.02 to 0.07) to utilization of X-rays (|r|, 0.05; 95% CI 0.01 to 0.10). The IES-6 also had a weak correlation with the EQ-5D-3L Mobility and Self Care items (|r|, 0.15; 95% CI, 0.11 to 0.19; and |r|, 0.12; 95% CI, 0.09 to 0.16, respectively), the SF-36 Role Physical and Physical Function Domains (|r|, 0.27; 95% CI, 0.24 to 0.32; and |r|, 0.21; 95% CI, 0.18 to 0.24, respectively), and the FPI Body Care, Maintain Household, and Physical Exercise domains (|r|, 0.22; 95% CI, 0.18 to 0.27; |r|, 0.26; 95% CI, 0.21 to 0.30; |r|, 0.24; 95% CI, 0.20 to 0.29, respectively). Again, these patterns of correlations were similar in magnitude and direction for the IES-R 22 item version (Table 5).

Comparison of the IES-6 to the clinician-based current CAP diagnosis of PTSD (13% with PTSD at time of interview and 28% ever experiencing PTSD post-ARDS) yielded an AUROC of 0.93 (95% CI, 0.86 to 1.00) (Fig. 1). The optimal cut-off point is 1.75, resulting in a sensitivity of 0.88, specificity of 0.85 and positive and negative predictive values of 0.47 and 0.98 respectively.