Using Measurement Invariance and Predictive Logistic Regression to Understand PHQ-9 Response Patterns Among English and Spanish Speakers Admitted to Emergency Departments

Background: In 2016, the U.S. Preventative Services Task Force recommended universal screening for depression across medical settings (Jackson et Machen, 2019). The Patient Health Questionnaire-9 (PHQ-9), a self-administered screening tool for depression, has gained popularity in various medical settings and has been adopted to adhere to this recommendation. The PHQ-9 has been translated to many languages, including Spanish; however, little is known about the measurement invariance between Spanish and English-speaking populations. As Hispanic Americans are more likely to receive medical care in acute medical settings (Parast et al., 2021), understanding how they self-report mental health symptoms can help enhance the detection and treatment of depression within this underserved population.

Methods: Adult patients (aged 25 and older) were admitted to emergency departments in a southeastern state for medical care. As part of routine screening, nursing staff administered the PHQ-9, either in English (n = 6,253) or in Spanish (n = 907) at intake. Confirmatory factor analysis (CFA) was performed to establish a two-factor model and test measurement invariance between language groups. Multi-group CFA tested configural, metric, and scalar invariance. Ordinal logistic regression (OLR) and item response theory (IRT) analyses were conducted to determine the differential item functioning (DIF) between language groups. Finally, logistic regression analyses examined if individual PHQ-9 items predicted suicidal ideation for both language groups.

Results: Multi-group CFA analysis confirmed configural invariance (CFI=0.936, TLI=0.911, RMSEA = 0.072, SRMR=0.041), showed mixed support for metric invariance (Metric: CFI = 0.934, TLI = 0.919, RMSEA = 0.134, SRMR = 0.044), but did not support scalar invariance. OLR analysis found that Spanish-speaking participants had a higher mean level of depressive symptomatology. IRT DIF analysis showed statistically significant differences in response patterns for all items between groups (p<0.001), however, low McFadden’s pseudo-R² values (all<0.03) indicated these differences had minimal clinical significance. Logistic regression analyses determined that distinct PHQ-9 items could be predictive of suicide risk across both language groups.

Conclusions and Implications: The CFA, DIF, and logistic regression analyses provide insight into how Spanish-speaking patients respond to each item of the PHQ-9. Understanding these differences can help clinicians better diagnose and treat depressive symptoms as they present in Spanish-speaking populations. Despite these nuances, McFadden’s pseudo-R² values were low, supporting the validity and clinical utility of the Spanish version of the PHQ-9. These findings support the use of this tool to screen for depression in Spanish-speaking populations in healthcare settings, while contributing to culturally competent mental health care.