In the original study, a total of 221 participants were recruited in PC settings from the Spanish regions of Aragon, Andalusia, and the Balearic Islands. The present study included 56 participants (78.6% women) who completed the PA intervention. Ages ranged between 19 and 68 years, with a mean of 44.14 years (SD = 10.38). In addition, 55.4% of the participants were married, 80.4% lived with family or a partner, 28.6% had higher education, 44.6% were employed, and 23.2% had an income below the national minimum wage. Regarding depression severity at baseline, the average on the Spanish Patient Health Questionnaire-9 (PHQ-9) was 15.79 (SD = 6.21).
Inclusion criteria were: (a) having mild or moderately severe depressive symptoms according to the Spanish Patient Health Questionnaire-9 (PHQ-9; [30
]) (5–9 = Mild depression; 10–14 = moderate depression); (b) age between 18–65 years; (c) ability to use a computer; (d) having Internet and an email account; and (e) being able to read and understand Spanish. The Mini International Neuropsychiatric Interview (MINI) 5.0 [31
] was used to assess different mental disorders and establish the diagnosis. Patients were excluded from the study if they had (a) severe depression (score ≥14 on the PHQ-9); (b) a severe Axis I psychiatric disorder (e.g., psychotic disorders, presence of suicidal ideation or plan, alcohol/substance abuse or dependence); (c) any disease that can affect the central nervous system (e.g., brain pathology, traumatic brain injury, dementia); or (d) if they were currently receiving psychological treatment.
Full information on the participant flow of the PA intervention is shown in Figure 1
2.6. Data Analyses
All statistical analyses were performed using the SPSS v.26 (IBM Corp, Armonk, NY, USA). First, we implemented Multiple Imputation with Chained Equations (MICE) to replace the outcomes’ missing values, performing 100 imputation models with 100 iterations per model [43
Second, preliminary analyses were conducted to ensure that relevant assumptions of repeated-measures ANOVAs and multiple regression were met (i.e., normality, sphericity, and absence of multicollinearity). We tested normality, carrying out a visual inspection of the Q-Q (quantile-quantile) plots, verifying that the observed data was approximately closed to the expected data (i.e., the distance between the observed and expected data was not extreme). Hence, we assured that we met the normality assumption to carry out a repeated-measures ANOVA and multiple regression. Moreover, we tested the assumption of sphericity using Mauchly’s test. If Mauchly’s test statistic was significant (i.e., sphericity was not met), the degrees of freedom were adjusted using the Greenhouse–Geisser correction. Finally, we tested the absence of multicollinearity using the Variance Inflation Factor (VIF).
Third, seven repeated-measures ANOVA with time as within-factor—pre-treatment (Pre), post-treatment (Post), 6-month follow-up (FW6), and 12-month follow up (FW12)—were conducted to analyze the changes in each primary (i.e., PHQ-9) and secondary (i.e., PANAS-PA, PANAS-NA, PHI, SF-12 mental and physical health, EQ-5D) outcomes. Post-hoc analyses using Bonferroni corrections were carried out when significant effects were found. Within-group Cohen’s d effect sizes with a 95% Confidence Interval were calculated.
Fourth, seven hierarchical multiple regression analyses with a stepwise selection of predictors within each block were conducted to explore which sociodemographic variables (i.e., sex, age, marital status, living alone or with others, educational level, employment, and income level) and pre-treatment scores on each primary and secondary outcome predicted the pre-post treatment, pre-FW6, and pre-FW12 change. The seven sociodemographic variables were entered in the first block, and six pre-treatment scores were entered in the second block to test the relevance of the extra explained variance of these variables in the dependent variables once the effects of the sociodemographic variables were controlled for. Given the small sample size to the number of predictors, a consideration of all predictors simultaneously in each regression was not tenable. Therefore, within each block, a statistical inclusion criterion for relevant predictors (stepwise method) was used. Creating two blocks in the regression analyses allowed us to test a lower number of predictors in each regression (i.e., seven instead of thirteen predictors).
To carry out correlation and regression analyses, categorical sociodemographic variables were transformed into recoded binary variables; that is, the correlation and regression analyses included: dichotomous variables (i.e., sex), continuous variables (i.e., age and scores in PHQ-9, PANAS-PA, PANAS-NA, PHI, SF-12 mental and physical health, EQ-5D) and recoded binary variables (i.e., ordinal or categorical variables were recoded into dichotomous variables with 0 and 1). Regarding the recoded binary variables, the categories were recategorized as follows: (a) marital status: 0 = not married (i.e., single, divorced, widowed); 1 = married or in a relationship; (b) living alone or with others: 0 = living alone; 1 = living with others (i.e., partner, sons, relatives, friends); (c) educational level: 0 = lower level (i.e., no education or primary school); 1 = higher level (i.e., secondary school or university studies); (d) work status: 0 = “not-working” (i.e., unemployed, retired, housekeeper, disability, student); 1 = employed; (e) income level: 0 = “lower than the minimum income (i.e., <641.40€); 1 = higher than the minimum income. We recoded into binary variables (instead of doing dummy variables) because the pair comparisons we constructed were theoretically appropriate, and a smaller number of predictors should be tested in the regression analyses. That is, if binary variables are used, only one predictor should be tested (e.g., 0 = “lower than the minimum income (i.e., <641.40€) vs. 1 = higher than the minimum income); however, if dummy variables are used, more than one predictor should be included in the regression equation (e.g., dummy 1: 0 = “lower than the minimum income < 641.40€ vs. 1 = between 1–2 minimum incomes; dummy 2: 0 = “lower than the minimum income < 641.40€ vs. 1 = between 2–4 minimum incomes). Moreover, the change in each primary and secondary outcome was calculated as follows: pre-post (post-treatment scores—pre-treatment scores), pre-FW6 (6-month follow up—pre-treatment scores), and pre-FW12 (12-month follow-up—pre-treatment scores). Negative values for the changes in PHQ and PANAS-NA meant improvements in these measures, whereas negative values for the changes in PANAS-PA, PHI, SF-12, and 5Q-5D meant a deterioration in these measures. The pairwise deletion method was used (i.e., whenever the variables of interest are present, they are analyzed) to deal with missing values and preserve all the data available in the regression analyses. The pre-treatment score on the corresponding dependent variable was not introduced in the equation regression model (e.g., the PHQ-9 pre score was not introduced as a predictor when the change in PHQ-9 was tested as a dependent variable).
Finally, four stepwise multiple regression analyses were conducted to analyze whether the changes in PHQ, PANAS-PA, or PANAS-NA were predictors of the change in PHI, SF-12 (mental and physical health), and EQ-5D. Changes in pre-post predictors were introduced for changes in pre-post dependent variables; changes in pre-post and pre-FW6 predictors were introduced for changes in pre-FW6 dependent variables; and changes in pre-post, pre-FW6, and pre-FW-12 predictors were introduced for changes in pre-FW12 dependent variables. All predictor variables were entered in the same block using the stepwise method.
It should be noted that the stepwise approach was used because we did not have an a priori hypotheses of which specific independent variables would predict the dependent variables, and consequently, we decided to identify the predictor variables relying on a statistical criterion.