Category: Adult Depression / Dysthymia
The anxious distress (AD) specifier was added to the DSM-5 depressive disorders to capture the presence and severity of co-occurring anxiety that is associated with poorer clinical outcomes. To assign the specifier, 2 of 5 AD symptoms (e.g., feeling keyed up or tense, feeling unusually restless, difficulty concentrating because of worry, fear that something awful may happen, and feeling that you may lose control of yourself) must be present during most days of the depressive episode. Given the novelty of this specifier, research on AD has relied on data collected prior to DSM-5. Researchers have thus approximated the specifier using questionnaires items worded similarly to the DSM-5 AD symptoms (Gaspersz et al., 2016, 2017; McIntyre et al., 2016). In these studies, cases of AD were identified by applying a pre-determined cut-off to 5 questionnaire item proxies to ascertain the presence of 2 or more AD symptoms. However, these studies have not investigated the validity of this approach, or determined if a model-based approximation using ROC curve analysis results in more accurate identification of AD than a pre-determined cutoff score. The goal of this study was to determine how accurately AD symptoms and the AD specifier could be approximated using items from popular questionnaires.
The sample consisted of 237 outpatients who were diagnosed with unipolar depression with (n = 157) or without AD n = 80). Diagnoses and the AD specifier were assigned by trained clinicians using the Anxiety and Related Disorders Interview Schedule for DSM-5. The 5 AD symptoms were rated on a 0-8 scale and were considered present if they occurred most days of the depressive episode. Patients also completed several well-validated questionnaires. Each AD symptom was represented by items from the NEO-Five Factor Inventory, Beck Depression Inventory-II, and Beck Anxiety Inventory. The AD symptom was considered present if the corresponding questionnaire item was endorsed by the patient with either of the top two response choices (indicating the symptom was frequently present).
All phi correlations between clinical ratings and questionnaire proxies of AD symptoms were statistically significant (rs = .31-.51, ps < .001). Using the pre-determined cutoff method (from individual item responses), sensitivity was 72.6% and specificity was 41.3% (Φ = .140). A model-based approximation was conducted by regressing the DSM-5 AD specifier onto the self-report items, followed by an ROC curve analysis. The area under the curve was .741 (fair prediction). Maximum value of Youden’s index achieved a sensitivity of 57.8% (∆ = -14.8) and a specificity of 84.7% (∆ = +43.3; Φ = .409). When sensitivity of the model-based approximation was set to that achieved by the item-response approximation (i.e., 72.6% or higher), specificity was 50% (Φ = .233).Given the poor specificity associated with the use of predetermined self-report item cut-offs, the findings of extant studies that approximated AD by this method should be interpreted cautiously. Methods for fostering the accuracy of the model-based approach to approximating AD will be discussed (e.g., inclusion of covariates).