Measurement Invariance Testing of the Patient Health Questionnaire-9 (PHQ-9) across People with and without Diabetes Mellitus from the NHANES, EHMS and UK Biobank datasets

Nouwen, Arie; Deschênes, Sonya S.; Balkhiyarova, Zhanna; Albertorio-Díaz, Juan R.; Prokopenko, Inga and Schmitz, Norbert (2021). Measurement Invariance Testing of the Patient Health Questionnaire-9 (PHQ-9) across People with and without Diabetes Mellitus from the NHANES, EHMS and UK Biobank datasets. Journal of Affective Disorders, 292 pp. 311–318.

DOI: https://doi.org/10.1016/j.jad.2021.05.031

Abstract

Background
The prevalence of depression is higher among those with diabetes than in the general population. The Patient Health Questionnaire (PHQ-9) is commonly used to assess depression in people with diabetes, but measurement invariance of the PHQ-9 across groups of people with and without diabetes has not yet been investigated.

Methods
Data from three independent cohorts from the USA (n=1,886 with diabetes, n=4,153 without diabetes), Quebec, Canada (n= 800 with diabetes, n= 2,411 without diabetes), and the UK (n=4,981 with diabetes, n=145,570 without diabetes), were used to examine measurement invariance between adults with and without diabetes. A series of multiple group confirmatory factor analyses were performed, with increasingly stringent model constraints applied to assess configural, equal thresholds, and equal thresholds and loadings invariance, respectively. One-factor and two-factor (somatic and cognitive-affective items) models were examined.

Results
Results demonstrated that the most stringent models, testing equal loadings and thresholds, had satisfactory model fit in the three cohorts for one-factor models (RMSEA = .063 or below and CFI = .978 or above) and two-factor models (RMSEA = .042 or below and CFI = .989 or above).

Limitations
Data were from Western countries only and we could not distinguish between type of diabetes.

Conclusions
Results provide support for measurement invariance between groups of people with and without diabetes, using either a one-factor or a two-factor model. While the two-factor solution has a slightly better fit, the one-factor solution is more parsimonious. Depending on research or clinical needs, both factor structures can be used.

Viewing alternatives

Download history

Metrics

Public Attention

Altmetrics from Altmetric

Number of Citations

Citations from Dimensions

Item Actions

Export

About

Recommendations