Beyond the total score: Factorial validity of the PHQ-9 in high-symptom contexts

Abstract

Background:

Depression is heterogeneous, with symptoms that organise differently across populations and clinical contexts. How the Patient Health Questionnaire-9 (PHQ-9) performs in clinically elevated populations, where measurement accuracy matters most, remains insufficiently understood.

Methods:

We examined the factor structure and measurement invariance of the PHQ-9 in two independent samples restricted to PHQ-9 $\ge 10$: the PROACTIVE trial of older Brazilian adults ($n=1216$) and the U.S. National Health and Nutrition Examination Survey (NHANES, 2007 to 2023; $n=3902$). Confirmatory factor analysis compared unidimensional, two-factor, and bifactor models, with invariance assessed across sociodemographic groups using multi-group analysis.

Results:

In both samples, a two-factor model fit better than a single-factor model, but the structure differed by context. In the older Brazilian cohort, fatigue grouped with anhedonia and depressed mood in a cognitive-affective domain; the domains were only moderately correlated ($r=0.45$), and a bifactor analysis found no dominant general factor. In NHANES, the conventional somatic and cognitive-affective split emerged, but the domains were strongly correlated ($r=0.76$), with a dominant general factor explaining most shared variance. Invariance was broadly supported, with minor item-level departures in PROACTIVE and a marginal race/ethnicity decrement in NHANES.

Conclusions:

Rather than being uniformly multidimensional, the PHQ-9’s structure is context-dependent: genuinely multidimensional in the high-burden older cohort, but dominated by a general factor in the broad community sample. The total score remains an appropriate severity summary in both settings, whereas attention to somatic versus cognitive-affective profiles is most useful in high-burden clinical populations.