Participants

Women diagnosed with BC, older than 18 years, and who were receiving medical care at the Instituto Nacional de Cancerología (INCan) in Mexico City, were included in this study by a convenience sampling. The calculation was done by including five patients for each of the 20 items of the survey’s original version (^{Nunnally & Bernstein, 1994}). The following inclusion criteria were considered: patients with BC diagnosis, who knew how to read and write in Spanish. We eliminated patients who did not respond to >10% of the items of the instrument.

Instrument

The Spanish-Argentine version of the MOS SSS (^{Rodríguez-Espínola & Enrique, 2007}) was used for the Spanish-Mexican adaptation, with 19 items grouped into three factors in terms of the SS functionality: (a) emotional/informational support (EIS), which integrates guideline and feedback on possible solutions to the individual’s problems, care, empathy, and love; (b) affection support (AS), which entails companionship, and (c) instrumental support (IS), which refers to a tangible support. The instrument contains an additional item that evaluates the network size.

The Spanish-Argentine version had an internal consistency of 0.919 Cronbach’s alpha coefficient and explained 59.86% of the total variance. This version was selected because it has the most similar factorial structure to the original, and their use of the Spanish language matched to Spanish-Mexican.

Procedure

After the protocol was approved by The Ethics and Research Review Board of the INCan (015/018IBI) (CEI/968/15), we proceeded to the survey’s adaptation to the Mexican population and language, according to standardized procedures (^{Cull, Sprangers, Bjordal, Aaronson, West, & Bottomley, 2002}). The adaptation process consisted of three phases: 1) two experts evaluated the original content for linguistically adaptation; 2) presentation of the corrected and integrated version to another expert, who examined the differences between versions and integrated them in one version; 3) The version obtained was pilot-tested in 10 patients with characteristics of the target population. The patients were surveyed about each item’s clarity, their understanding of words, and the complexity of the item. Based on the results, the final version was elaborated.

Potentially eligible patients were identified and given a verbal explanation of the study’s objectives and data confidentiality. Those who agreed to participate were asked to provide their written informed consent. Due to the low literacy of most Mexican cancer patients and to maximize standardization, the scale was administered by a trained research psychologist in a face-to-face interview in a private office (^{Doubova, Aguirre-Hernandez, Gutiérrez-de la Barrera, Infante-Castañeda, & Pérez-Cuevas, 2015}; Gálvez-Hernández et al., in press). The research psychologist recorded participants’ responses for each item of the final version of the survey and ensured that each application was completed. In case of missing data, the application was removed from the analysis. All personal data was kept private and confidential.

To validate this version, four analysis were performed: 1) an independent samples t-test to determine item discrimination property; 2) the reliability of the scale was evaluated by Cronbach’s alpha coefficient; 3) the factorial validity was evaluated using principal components factor analysis (Exploratory Factor Analysis --EFA--). The principal components factor analysis was selected in order to continue the statistical analyses used in previous validations (^{Baca, 2016}; ^{Rodríguez-Espínola & Enrique, 2007}; ^{Sherbourne & Stewart, 1991}), and according to ^{Costello and Osborne (2005)} this factor analysis is an appropriate procedure when variable is non-normal distributed. The criteria for factor validity was a group with less than three items for confirmation of a factor (^{Nunnally & Bernstein, 1994}); 4) a Confirmatory Factor Analysis (CFA) was performed to empirically review if the structure obtained in the EFA could be verified in the MOS-SSS among Mexican BC patients.

Latent variables’ variance loading was set at 1.0. Variances of error terms were specified as free parameters, and maximum likelihood estimation methods were used. Goodness-of-fit (GFI) was evaluated using the goodness-of-fit statistics of the chi-square test and fit indexes as follows: Root Mean Square Error of Approximation (RMSEA), Comparative Fit Index (CFI), and Tucker-Lewis Index-Non-Normed Fit Index (TL-INNFI). The greater the probability associated with the chi-square test, the closer the fit between the model and the perfect fit; values of 0.06 for RMSEA and of 0.90 for CFI and TLINNFI are indicative of good fit between the hypothesized model and the observed data (^{Jackson, Gillaspy, & Purc-Stephenson, 2009}). Additionality Average Variance Extracted (AVE) and McDonald´s Coefficient Omega were estimated in order to identify the instrument convergent validity. Adequate convergent validity was accepted when we observed a AVE >0.50 and a Coefficient Omega >0.70 (^{Borsboom, Mellenbergh, & van Heerden, 2004}).

Based on the level of literacy of the patients, it was tested as a short version of the instrument. Another EFA was performed, item’s elimination was established in three criteria: 1) factor loading needed to be <0.40; 2) or a group with less than three items for confirmation of a factor (Nunally & Berstain, 1994), and 3) items attributed to 2 factors were removed. In case all criteria were met, in terms of explained variance and CFA index, this version would be maintained.

A p value of 0.05 or less than was adopted as the significance level in all of the statistical analyses, and all p values reported are two-tailed. All statistical procedures were conducted using the SPSS 22.0 J version software for Windows.

Patients’ Characteristics

A total of 330 women with BC were invited from June 2015 to August 2016, 300 consented to participate; 20 (6%) refused to participate; 10 (3%) patients were removed because they were repeated (in database) or did not provide evaluable forms. The mean age was 50 years (SD=10.4 years). More than one-half of the patients were married and had finished secondary school. A total of 44% of these patients reported not having an occupation. The majority were Catholic (84%) (see Table 1).

Nearly 70% of the sample was found in clinical stages II-III; 51% were under active treatment, and 28% under surveillance. A total of 24% of the sample had undergone multidisciplinary cancer treatment, mainly (40%) surgery and chemotherapy (Table 2).

Psychometric analysis

Before initiating the analyses, item 1 was excluded from all the validation tests because its format and objective are not found in the remainder of the items. The remainder of the items demonstrated adequate distribution among the response options (asymmetry <2). All the items exhibited the capacity to discriminate between the extreme scoring groups; they were retained for subsequent analyses.

A principal axis factoring EFA with oblimin rotation was performed with the 19 items (Item 1 was excluded). This analysis showed a KMO=.910 with significant Bartlett’s sphericity test (p=.000) that indicated the non-identity of the correlation’s matrix and an adequate sample size. The factorial structure explained 56.07% of the total explained variance that was distributed in three factors. High internal consistency showed a Cronbach’s alpha coefficient = 0.938.

The CFA showed non-normal multivariate distribution with a c.r. kurtosis > 2.0. The CFA of the 19 items version showed good adjustment, as indicated by the following values: CMIN/df = 2.964 p = 0.000; NFI = 0.884; RFI = 0.869; CFI = 0.919, and RMSEA = 0.081, with a 95% confidence interval (95% CI) between 0.072 and 0.090 (Table 3). The convergent validity was acceptable, the AVE values for each factor were 0.57, 0.43, and 0.54 respectively; the McDonald’s Omega Coefficients were 0.82 for the factor 1, 0.80 for the factor 2, and 0.74 for the factor 3.

To evaluate discriminative validity, correlation between factors were estimated. Analyses showed F1 - F2 (r=0.846, p<0.01), F1 - F3 (r=0.498, p<0.01), and F2 - F3 (r=0.561, p>0.01). Correlations between factors were lower EVA square root, therefore factors 2 and 3 showed adequate discriminative validity.

A factorial structure with 13 final items, distributed in three factors, was obtained from the evaluation of the short version. This structure explained 57% of the total explained variance, which is distributed uniformly among the factors. Additionally, high internal consistency is exhibited through Cronbach’s alpha coefficient = 0.912. Compared to the original structure, the items 3,8,14 and 19 were eliminated from the EIS. From the AF, we discarded the Item 2; and, from the IS, we dispensed Item 11. Item 7 and 8, deriving from the EIS, were incorporated into the AF (Table 4).

The CFA of the short version, the model obtained showed slightly better adjustment, as indicated by the following values: CMIN/df = 2.259, p = 0.000; NFI = 0.934; RFI = 0.911; CFI = 0.961, and RMSEA = 0.065, with a 95% confidence interval (95% CI) between 0.051 and 0.079 (Table 3 and Figure 1). The convergent validity was adequate because the AVE values for each factor were 0.69, 0.55, and 0.55 respectively; and the McDonald’s Omega Coefficients were 0.82 for the factor 1, 0.77 for the factor 2, and 0.63 for the factor 3.

Figure 1 Original Model of three correlated factor model and final model of three correlated-factor model. Both were estimated by Maximum Likelihood in the sample of 300 participants. 

In order to evaluate discriminative validity, correlation between factors were estimated. Analyses showed F1 - F2 (r=0.468, p<0.01), F1 - F3 (r=0.468, p<0.01), and F2 - F3 (r=0.718, p>0.01). Correlations between factors were lower EVA square root, therefore factors showed adequate discriminative validity.

The item 1 (which evaluates the size of the social network) demonstrated a mean of 11.13 (SD = 10.21) with a range of 1-60. This item demonstrated a positive correlation in terms of the obtained scores in the EIS --Factor 1-- (r = 0.277; p <0.001); AS --Factor 2-- (r = 0.227; p <0.000); IS --Factor 3-- (r = 0.182; p = 0.002), and the total of the score (r = 0.280; p <0.000). In addition, it discriminated among the extremes of the total scores obtained (t = -13.147; gl = 89; p <0.001). For these reasons, we decided to retain the item in the structure of the survey as an independent component that contributes to the understanding of SS.

In sum, the Mexican version of the MOS SSS for BC obtained two factorial structures with good adjustment: a large (19 item) and a short version (13 items), both distributed among three factors: (a) EIS; (b) IS, and (c) AS, plus one item that evaluates the size of the patient’s SS network (Figure 1).

Additionally, we found (in standardized score) that patients with BC reported that AS was the highest perceived support (M=87.23; SD=19.19), followed by IS (M=80; SD=24.62); EIS was the lowest factor (M=79.75; SD=25.19).