The information of this study is based on a database of online survey studies during the COVID-19 pandemic. Some results of this survey study are reported elsewhere (7,18,19). For this study, some methodological aspects, mainly concerning participants’ recruitment, the description of the online survey, and the questionnaires (especially CPDI), are also included.
a. Study design and participation criteria
Since one of the main objectives of this study was the construct validation of the CPDI in its Spanish version (Latin America), we divided the following study into two phases with different recruited participants, as recommended in the literature for construct validation studies and factorial analysis (20–23): A phase for the exploratory factor analysis (EFA) and another phase for the confirmatory factor analysis (CFA). With the sample information of the second phase, we performed a structural equation model (SEM) to explain depression through CPDI values, using anxiety symptoms as a mediator and sociodemographic variables (age, sex, completed education, and presence of medical conditions) as moderators. Finally, we carried out a hierarchical regression model (HRM) to evaluate the amount of explained variance of the CPDI factors above depression and anxiety, correcting for possible interaction factors (sex, age, completed education, and presence of medical conditions).
For this purpose, we recruited, firstly, for the EFA phase, 300 voluntary participants from Lima (Peru) between March 27th and July 22nd, 2020. Participants were only included if they were older than 18 years, had sufficient knowledge of the Spanish language (CEFR B2), and signed the informed consent to participate. Furthermore, participants were excluded if they did not fulfill these criteria.
After finishing the first phase of this study, we recruited for the CFA 1135 voluntary participants from Lima, Peru, between July 23rd, 2020 and September 21st, 2021. Inclusion and exclusion criteria for the first phase were also considered in the second phase.
Finally, both phases were approved by the ethics committee of the Peruvian University “Cayetano Heredia” (UPCH). The ethical procedures of the study were carried out according to the Helsinki Declaration and the standards of the American Psychology Association (APA). This study was not financed by any industry, society, company, or educational institution. Moreover, no outside institution influenced the study design (7,18,19).
b. Online survey
Online electronic surveys were used to recollect the participants’ information since the Peruvian sanitary restrictions did not allow to perform personal contact for recollecting the data. These online surveys were carried out using an open-access internet-based program (Google Forms, Google Inc, USA). Questions concerning socio-economic status (i.e., age, gender, education), past medical history (i.e., presence of medical conditions, number of medical conditions), and psychometric data (CPDI scale, anxiety and depression symptoms) were included (7,18,19).
c. COVID-19 Peritraumatic distress index (CPDI)
The COVID-19 peritraumatic distress index (CPDI) consists of 24 items. Each item was evaluated using a Likert scale (from 0 to 4: never, occasionally, sometimes, often, and most of the time). The sum of each value per question results in the raw score. The displayed score is obtained by adding 4 to the raw score and calculating the CPDI severity degrees. For this reason, this instrument defines different categories for peritraumatic stress due to the COVID-19 pandemic: normal (0 to 28 display points), mild (29 to 52 display points), and severe (53 to 100 display points) (7,18).
d. Depressive and anxious symptomatology
Another main objective of this study is to evaluate the relationship between CPDI, depression, and anxiety scores, including moderators (i.e., socio-economic variables). For this purpose, we evaluated depressive symptoms using the Patient health questionnaire - 9 items (PHQ-9), validated in Peru (24). PHQ-9 scores ranged between 0 and 27 points, showing a significant internal consistency (Cronbach’s alpha = 0.87) and defining different severity grades (minimal -1 to 4 points-, mild -5 to 9 points-, moderate -10 to 14 points-, and severe -15 to 27 points-) (7,18,19,24). In addition, we applied the Generalized anxiety disorder – 7 items (GAD-7) to assess the anxiety symptoms, which is also validated in Peru (25). The GAD-7 shows a score ranging from 0 to 21 points, also showing a significant internal consistency (Cronbach’s alpha = 0.89) and defining different severity categories, such as minimal (0 to 4 points), mild (5 to 10 points), moderate (11 to 15 points), and severe (16 to 21 points) (7,18,19,25).
e. Statistical analyses
i. General aspects and descriptive statistics
General characteristics of the sample, including descriptive data of the instruments (CPDI, PHQ-9, and GAD-7), were represented in tables and described in the text. For quantitative data, we used the mean and standard deviation as measures of central tendency if the numerical variables were normally distributed. For variables with non-gaussian distribution, median and interquartile ranges (IQR), including 75- and 25-percentiles, were used to describe the variable. Regarding decimal data, we rounded the descriptive information by two decimals. In addition, descriptive data greater than a million was expressed by using scientific notation. Finally, qualitative data, including count data, was characterized using percentage and absolute numerical values. These procedures were performed for both subsamples, as shown in Table 1. Since the two subsamples (i.e., first and second phase) corresponded to two subprojects with different objectives and hypotheses, we did not perform statistical tests to evaluate their differences.
ii. Exploratory factor analysis
An exploratory factor analysis (EFA) was conducted with the items of the CPDI in the first subsample (n = 300). In determining factor adequacy, we took into consideration the following literature recommendations (26): all factors should be theoretically meaningful, at least three variables should saliently load on a factor (overdetermined, i.e., factor loadings ≥ 0.30), and variables should load significantly on only one factor (no cross-loadings), and each factor should have an internal consistency of α ≥ 0.70 (reliability).
The first step was determining whether the items were suitable for an EFA. For this purpose, Bartlett’s test for sphericity (27) was used to ensure that the correlation matrix was not random. In addition, we applied the Kaiser-Meyer-Olkin (KMO) criterion (28) to determine that only items with a measure of sampling adequacy (MSA) value > 0.70 were included.
Additionally, we tested whether the 24 items of the CPDI had a good item-corrected item-scale correlation (i.e., r > 0.30) (29). If the items showed an item-corrected item-scale correlation lower than 0.30, we excluded them from further analyses.
Concerning the number of retained factors, several procedures should be used to determine the appropriate number of factors that are suitable to keep (30). For this reason, we performed the parallel analysis (31), the minimum average partial (MAP) (32), and the visual Scree test (33) in this study to determine the number of retained factors.
Further, we checked for multivariate normal distribution of the items using the Mardia test for skewness and excess (34). In this case, Mardia tests indicated a non-parametrical distribution of the items. Since our data violates the principles of the standard distribution assumptions and the CPDI items have an ordinal nature, we carried out a polychoric correlation matrix as an input method for the EFA and a principal axis as a factor extraction method. This method followed the recommendations of the robustness of the principal axis method towards the violation of standard distribution assumptions published elsewhere (35).
Finally, we performed a reliability analysis for each extracted factor using Cronbach’s alpha to define the factors’ internal consistency. In this case, we considered only alpha values greater than 0.70 (acceptable), following the literature recommendations for the grading of Cronbach’s alpha (36). The interpretation of the alpha values is defined in previous studies as follows: excellent (α ≥ 0.90), good (0.90 > α ≥ 0.80), acceptable (0.80 > α ≥ 0.70), questionable (0.70 > α ≥ 0.60), poor (α < 0.60) (36).
iii. Confirmatory factor analysis
To confirm the hypothetical factor model obtained from the EFA, we performed a confirmatory factor analysis (CFA) in an independent sample (subsample of the second phase) of 1135 participants. For this purpose, we used Diagonal Weighted Least Square (DWLS) as an estimation method, recommended and published elsewhere (37). As model fit measurements, we used the comparative fit index (CFI), the Tucker-Lewis index (TLI), the Root-mean-square-error of approximation (RMSEA), and the standardized Root-mean-square-residual (SRMR). Concerning the TLI and CFI, a good model fit was given if the values of the CFI and TLI were greater or equal to 0.95 (15). Finally, regarding RMSEA and SRMR, a good model fit was given if the values of both Root-mean-square indicators were below or equal to 0.05 (15). Additionally, we calculated the 90-percent confidence intervals (90CI) for the RMSEA (38). Finally, we presented the results of the CFA in a path diagram indicating the standardized regression coefficients of the items and the factors.
iv. Statistical modeling: structural equation modeling and hierarchical regression model
In addition, we performed a structural equation modeling (SEM) to establish a model explaining depression during the COVID-19 pandemic through CPDI, anxiety (mediator of the relationship), and socio-economic parameters (moderator of the relationship), such as age, gender, educational status, and medical conditions. For this purpose, we carried out this model using the diagonal weighted least squares (DWLS) as an estimator since it is robust to the non-normal distribution of scales. Following Hu and Bentler (37), a good model fit is given when CFI and TLI are greater than 0.90 and SRMR and RMSEA are lower than 0.08.
Finally, to establish the predictive values of the CPDI factors obtained from the construct validation analysis for anxiety and depression symptoms (GAD-7 and PHQ-9 respectively), we carried out a hierarchical regression analysis (HRA) with the sample size of this study’s second phase (n = 1135 participants). For the first step, we carried out a multiple regression model with well-studied predictors (i.e., age, sex, completed education, and presence of medical conditions) for anxiety and depression as dependent variables (DV). In a second step, factors were separately added to the multiple regression analysis models to examine if the factor could explain variance above the previously mentioned predictors in the first step. In the end, both regression models were compared with ANOVA to test if the additional variance that could be explained due to the CPDI factors is statistically significant. P-values are considered for this analysis as significant if the two-tailed p < 0.05. Moreover, we used Cohen´s f 2 (small = 0.02, medium = 0.15, and large = 0.35) for effect sizes (39,40).
Descriptive information was performed using JASP version 0.11.1 (Jeffreys’s Amazing Statistic Program, The University of Amsterdam, Amsterdam, The Netherlands) (41).
Statistical analyses of the EFA, CFA, SEM, and HRA were performed under the R-software version 4.1.2 (R Core Team, 2021, R Foundation for Statistical Computing, Vienna, Austria) (42). Finally, we computed SEM analysis using the lavaan package (version 0.6-7) under the R-software.