Abstract
ABSTRACTBackgroundThere are concerns that the measurement of depression by the Patient Health Questionnaire-9 (PHQ-9), a self-report screening questionnaire, is biased by comorbid stroke sequelae. We, therefore, aimed to investigate these concerns in stroke, benchmarked against a non-stroke comparison sample, using factor analysis.MethodsThe secondary data sample constituted 787 stroke and 12,016 non-stroke participants, in a cross-sectional design. A subsample of 1,574 non-stroke participants was selected via propensity score matching. Dimensionality was assessed by comparing fit statistics of one-factor, two-factor, and bi-factor models. Between-group differences in factor structure were identified using measurement invariance.ResultsA two-factor model, consisting of somatic and cognitive-affect factors, had a superior fit to a unidimensional model (CFI = .984 versus CFI =.974, p<.001), but the high correlation between the factors indicated unidimensionality (r = .866). Configural invariance between stroke and non-stroke was supported (CFI = .983, RMSEA = .080), as were invariant thresholds (p = .092) and loadings (p = .103) for all items. Strong invariance was violated (p < .001, ΔCFI = -.003), indicating non-invariant item intercepts. Partially invariant models indicated responsibility of the tiredness and appetite intercepts, and latent depression severity was significantly overestimated in stroke, relative to the general population, using a summed score approach (Cohen’s d=.434).ConclusionsThe findings suggest that the PHQ-9 measures a single latent factor in stroke. However, the presence of non-invariant intercepts means that PHQ-9 total scores may be disproportionately influenced by fatigue in post-stroke vs. non-stroke patients and that total scores are incomparable between groups.
Publisher
Cold Spring Harbor Laboratory