Key points are not available for this paper at this time.
The Patient Health Questionnaire-9 (PHQ-9) is a commonly used depression inventory, but its psychometric properties have not been thoroughly examined in either people living with HIV (PLWH) or in developing nations. We examined the psychometric properties of the PHQ-9 in PLWH in Thailand. The RV254/SEARCH010 cohort enrolls adults with acute HIV in Bangkok, Thailand. Most study participants initiate antiretroviral therapy (ART) at enrollment. We analyzed responses to the PHQ-9 at weeks 0 (enrollment), 4, and 24. Confirmatory factor analysis (CFA) was used to compare a 1-factor model, three 2-factor models, and a bi-factor model. Longitudinal measurement invariance (LMI) was assessed for the 1-factor, optimal 2-factor, and bi-factor models. Internal consistency and test-retest reliability were calculated. Correlations between factors and item response theory (IRT) were used to further examine performance. There were 734 (97.8 % male) participants with a median (Q1-Q3) age of 27 (23–32) years. CFA confirmed that the bi-factor model had the best fit. Regardless, LMI was confirmed for the 1-, 2-, and bi-factor models. The PHQ-9 was reliable (coefficient alpha = 0.86–0.89; test-retest reliability, r = 0.43–0.65) in this sample. Correlation and IRT analyses produced evidence that the bi-factor model's subfactors lack practical utility. These results support the use of the PHQ-9 in Thai PLWH. Despite the superior fit indices of the bi-factor model, the subfactors were strongly correlated with the general factor, so we recommend that the 1-factor model be used in practice. The evidence for LMI support use of the PHQ-9 in longitudinal research involving Thai PLWH. • Bi-factor solution with 2 subfactors indicated using confirmatory factor analysis. • Bi-factor model's subfactors evidenced limited practical utility. • 1-factor solution was deemed sufficient. • Longitudinal measurement invariance was supported for 1-, 2-, and bi-factor models. • Evidence for internal consistency and test-retest reliability was strong.
Hamby et al. (2025) studied this question.