Factor structure of the 10-item CES-D Scale among patients with persistent COVID-19

Abstract

The presence of persistent coronavirus disease 2019 (COVID-19) might be associated with significant levels of psychological distress that would meet the threshold for clinical relevance. The Center for Epidemiologic Studies Depression Scale (CES-D) version 10 has been widely used in assessing psychological distress among general and clinical populations from different cultural backgrounds. To our knowledge, however, researchers have not yet validated these findings among patients with persistent COVID-19. A cross-sectional validation study was conducted with 100 patients from the EXER-COVID project (69.8% women; mean (±standard deviation) ages: 47.4 ± 9.5 years). Confirmatory factor analyses (CFAs) were performed on the 10-item CES-D to test four model fits: (a) unidimensional model, (b) two-factor correlated model, (c) three-factor correlated model, and (d) second-order factor model. The diagonal-weighted least-squares estimator was used, as it is commonly applied to latent variable models with ordered categorical variables. The reliability indices of the 10-item CES-D in patients with persistent COVID-19 were as follows: depressive affect factor ( ${\alpha }_{\mathrm{Ord}}=0.82$ ; ${\omega }_{u-\mathrm{cat}}=0.78$ ), somatic retardation factor ( ${\alpha }_{\mathrm{Ord}}=0.78$ ; ${\omega }_{u-\mathrm{cat}}=0.56$ ), and positive affect factor ( ${\alpha }_{\mathrm{Ord}}=0.56$ ; ${\omega }_{u-\mathrm{cat}}=0.55$ ). The second-order model fit showed good Omega reliability ( ${\omega }_{\mathrm{ho}}=0.87$ ). Regarding CFAs, the unidimensional-factor model shows poor goodness of fit, especially residuals analysis (root mean square error of approximation [RMSEA] = 0.081 [95% confidence interval, CI = 0.040-0.119]; standardized root mean square residual [SRMR] = 0.101). The two-factor correlated model, three-factor correlated model, and second-order factor model showed adequate goodness of fit, and the χ² difference test ( $∆X^2$ ) did not show significant differences between the goodness of fit for these models ( $∆X^2=4.1128$ ; p = 0.127). Several indices showed a good fit with the three-factor correlated model: goodness-of-fit index = 0.974, comparative fit index = 0.990, relative noncentrality index = 0.990, and incremental fit index = 0.990, which were all above 0.95, the traditional cut-off establishing adequate fit. On the other hand, RMSEA = 0.049 (95% CI = 0.000-0.095), where an RMSEA < 0.06-0.08 indicates an adequate fit. Item loadings on the factors were statistically significant ( ${\lambda }_j\ge 0.449$ ; p's < 0.001), indicating that the items loaded correctly on the corresponding factors and the relationship between factors ( $\varphi \ge 0.382$ ; p's ≤ 0.001. To our knowledge, this is the first study to provide validity and reliability to 10-item CES-D in a persistent COVID-19 Spanish patient sample. The validation and reliability of this short screening tool allow us to increase the chance of obtaining complete data in a particular patient profile with increased fatigue and brain fog that limit patients' capacity to complete questionnaires.

Keywords: COVID-19; depression; mental illness; statistical factor analyses.