Theses Doctoral

Measuring Psychopathology: Exploring Construct Validity Evidence for PTSD A 2010 Haitian Earthquake Example

Hermosilla, Sabrina

Measurement is the foundation of epidemiologic thought and practice. The appropriate measurement of exposures and outcomes of interest is the underlying assumption to all causal investigations. Poor quality measurement, be it through inappropriate data collection methods or changing diagnostic criteria, which can result in erroneous estimates, has a deleterious impact on scientists, policy makers, and the public.

Mental health disorders particularly suffer from a lack of diagnostic clarity as diagnosis is often based on self-report of overlapping symptoms with no clear measureable biomarkers. The release of the Diagnostic and Statistical Manual of Mental Disorders, fifth edition (DSM-5) in May of 2013 is the most recent attempt to codify existing diagnostic criteria for psychiatric disorders. Posttraumatic stress disorder (PTSD) diagnostic criteria shifted from a three-cluster approach of avoidance, hyper-arousal, and re- experiencing to a four-cluster approach of avoidance, arousal, negative cognitions and mood, and re- experiencing. The very existence of multiple diagnostic frameworks for the same psychiatric disorder is proof that accurate diagnosis is a complex and unresolved issue that warrants investigation.

This complexity in posttraumatic stress disorder (PTSD) symptom presentation, limits our ability to develop appropriate responses. In this dissertation I conducted four independent but related studies to explore the construct validity of PTSD. In Chapter 1 I systematically reviewed the extant empiric literature from PubMed and PsychINFO on PTSD symptom structure to identify a universal PTSD factor structure. I found 40 (3%) of 1,302 citations published between 1980-2014 provided empiric PTSD factor structure estimates forming the basis of my review. While consensus exists with respect to the general multifactorial make-up of PTSD, a universal understanding of the specific operationalization of this structure, supported by the empiric literature, is absent.

In Chapter 2, I used population-based, cross-sectional data from adult survivors of the 2010 earthquake in Haiti, to assesses model fit of six theoretical factor structures of PTSD: one-factor Diagnostic and Statistical Manual of Mental Disorders (DSM)-IV; three-factor DSM-IV-TR (arousal, avoidance, and intrusion); three-main factor (arousal, avoidance, and intrusion) and one-hierarchical factor DSM-IV-TR; four-factor King 1998 (avoidance, hypervigilance, emotional numbing, re-experiencing); four-factor Simms 2002 (avoidance, dysphoria, hyperarousal, intrusion); and four-factor DSM-5 (arousal, avoidance, intrusion, negative mood and cognition) models through confirmatory factor analyses (CFA). While all models adequately fit the data, the three-factor DSM-IV-TR (arousal, avoidance, and intrusion) model best fit the sample (χ2=593.257, 116 degrees of freedom; RMSEA=0.056; CFI=0.927; TLI=0.915, WRMR=1.769; AIC=24,760.459; and BIC=24,952.178).

Again drawing on the cross-sectional, population data from Haitian earthquake survivors, in Chapter 3 I used multiple linear regressions to model pre-, peri-, and post-earthquake factor associations with mean PTSD symptom cluster (arousal, intrusion, and avoidance, validated in Chapter 2) endorsement. I found that mean PTSD symptom factor endorsement is heterogeneously associated with pre-, peri-, and post- earthquake factors, consistent with dimensional theoretical foundations: arousal endorsement more likely to be associated with pre-earthquake factors, intrusion endorsement more likely to be associated with factors across the temporal field, and avoidance endorsement more likely to be associated with post- earthquake factors.

In Chapter 4, I used exploratory factor analysis (EFA) to assess the factor stability of the DSM-IV-TR (arousal, intrusion, avoidance) defined PTSD structure when major depressive disorder (MDD) items are introduced, in the same Haitian post-earthquake population-based study. A six-factor, 25-item model was estimated and fit the data (χ2=253.427, 165 degrees of freedom, p<0.001; RMSEA=0.021, 90% CI:0.016, 0.026; CFI=0.987; TLI=0.976) better than the PTSD-only model specified in Chapter 2. PTSD-specific items did not load on the original PTSD factors or with the original factor items (new factors included items from 0-3 different original PTSD factors), in the presence of MDD items. PTSD dimensionality was not stable in the presence of MDD items, thus challenging the discriminant validity of PTSD.

This exploration into PTSD construct validity found that while consensus exists with respect to the general multifactorial make-up of PTSD, a universal understanding of the specific operationalization of this structure, supported by the empiric literature, is absent. The tight range in model fit statistics documented in the CFA provides additional evidence of this, suggesting that empirical-based model selection is insufficient to universally characterize PTSD. Given the overall consensus of general factors, the significant and heterogeneous pre-, peri-, and post-earthquake factor associations with the unique PTSD symptom clusters provides additional evidence of the multidimensional theoretical mechanisms behind PTSD psychopathology. PTSD model stability, an indication of discriminant validity, failed to hold when challenged by MDD items, further challenging PTSD construct validity.

There are several important implications of this work. First, based on the systematic review and CFA findings, adjudication of PTSD model selection based on empiric findings is insufficient and should be theoretically driven. Future investigations should always include the most commonly supported models as they develop and refine additional models, thus enabling rigorous cross-context, cross-potentially traumatic event, and cross-study comparisons that are currently not possible. Second, the multidimensional modeling of PTSD factors provided valuable insight into the psychopathology of PTSD without additional data collection burden and should be widely adopted. Researchers should look to model PTSD both as a dichotomous variable and on a continuous scale, both as a complete construct and by each dimensional component. Third, while the exploration into discriminant validity builds on another study that found PTSD factor structure unstable in the presence of MDD item challenges, more research is needed here to understand the theoretic and empiric utility of the specified six-factor model across settings and diagnostic criteria. Fourth, while endeavoring to explore construct validity, exploratory qualitative methods with populations beyond the highly studied U.S. military populations are needed to propose additional items that could, provide valuable missing empiric evidence for PTSD factor dimensionality.

Geographic Areas


  • thumnail for Hermosilla_columbia_0054D_12691.pdf Hermosilla_columbia_0054D_12691.pdf application/pdf 2.41 MB Download File

More About This Work

Academic Units
Thesis Advisors
Galea, Sandro
Ph.D., Columbia University
Published Here
May 7, 2015