Longitudinal Measurement Invariance of the PHQ-9 and GAD-7

Abstract

Intraindividual change over time is commonly used to estimate treatment effectiveness. However, patients may not respond similarly to a scale after treatment, rendering pre–post change an unreliable metric. The current objective was to investigate longitudinal measurement invariance of the Patient Health Questionnaire–9 and Generalized Anxiety Disorder Scale–7 among 4,323 patients completing a partial hospital program. We used confirmatory factor analysis to determine (1) factor structure at pretreatment and posttreatment and (2) longitudinal invariance, accounting for dependent observations, using both classical and approximate measurement invariance approaches. Results indicated a two-factor solution for both scales. Longitudinal invariance was not established for either scale, thus, using raw score differences from the Patient Health Questionnaire–9 and Generalized Anxiety Disorder Scale–7 for measuring symptom change over time may be problematic. The most longitudinally consistent items captured somatic as opposed to affective/cognitive symptoms. We discuss the potential use of these measures for diagnostic screening and between-group comparisons and suggest alternative ways to monitor client progress over time. Limitations included a majority White sample and uniqueness of a partial hospital setting.

Longitudinal invariance of the Patient Health Questionnaire–9 among patients receiving pharmacotherapy for major depressive disorder: A secondary analysis of clinical trial data.

“…a "Other/mixed" was not specified and was the term used during data collection. Ong et al, 2022). These results are also consistent with previous findings that pharmacotherapy for depression may not result in a response shift (i.e., a change in the way a person self-evaluates and reports their symptoms; Fokkema et al, 2013).…”

“…The present study extends prior research by incorporating both a longer time frame and more frequent measurements of depressive symptoms, thereby offering an assessment of PHQ-9 LMI under conditions more similar to those of clinical practice, as well as examining LMI within the context of pharmacotherapy for major depressive disorder. The use of the RMSEA D statistic, as compared to the chi-square difference test, which is known to be overly sensitive to large sample sizes (Savalei et al, 2023), to test for LMI may explain why the present findings differ from previous studies that were unable to establish PHQ-9 LMI (Downey et al, 2016; Ong et al, 2022). These results are also consistent with previous findings that pharmacotherapy for depression may not result in a response shift (i.e., a change in the way a person self-evaluates and reports their symptoms; Fokkema et al, 2013).…”

“…Several studies have previously evaluated LMI in the PHQ-9 across several factor models, although findings to date have been mixed. Ong et al (2022) assessed LMI in a two-factor PHQ-9 model in participants enrolled in a short-term, skill-based partial hospitalization program for a range of psychiatric disorders. The authors were not able to establish LMI, suggesting that PHQ-9 scores could not reliably be compared across the intervention time points in this clinically diverse sample (Ong et al, 2022).…”

Factorial structure, reliability, and construct validity of the Generalized Anxiety Disorder 7-item (GAD-7): Evidence from Malaysia

“…Specifically, we first examined the following five one-factor models: the original 7-item one-factor model (Model 1), 7-item one-factor model with a residual covariance between items 3 and 7 (Model 1a) [ 23 ], 7-item one-factor model with a residual covariance between items 1 and 2 and items 2 and 3 (Model 1b) [ 15 ], 7-item one-factor model with a residual covariance between items 4 and 5 and items 5 and 6 (Model 1c) [ 24 ], 7-item one-factor model with a residual covariance between items 4 and 5, items 5 and 6, items 4 and 6 (Model 1d) [ 7 ], and 7-item one-factor model with a residual covariance between items 1 and 2, items 4 and 5, items 5 and 6 (Model 1e) [ 25 ]. Next, we examined the following two-factor models: the 7-item two-factor model (Model 2) [ 26 ] and the 6-item two-factor model without item 2 suggested by EFA (Model 3). Then, instead of testing the 6-item three-factor model (without item 3) suggested by EFA, we tested the 6-item second-order model consisting of one general factor of anxiety and three specific first-order factors (Model 4) to account for the correlations among the three factors.…”

YESS: A feasibility study of a supported employment program for youths with mental health disorders

“…Therefore, it is more ideal to measure recovery and quality of life after participants have demonstrated better symptom management and/or a change in employment status. Lastly, there has been some evidence that both the PHQ-9 and the GAD-7 are not sensitive enough to detect change over time at an individual level (40,41). As such, modified versions of the PHQ-9 and the GAD-7 have been suggested through modern measurement analysis (41,42).…”