Socio-demographic and clinical characteristics are presented in Table 1. The sample comprised 877 patients with a mean age of 36 years (SD = 13.4, range 25-84); 59.7% the participants were female. Most were married or partnered (75.8%) and had a primary level of education (66.4%). As for place of residence, 40.5% of the participants live in a large city (>100,000 inhabitants); 26.6%, in a medium-sized city (10,000-100,000), and 32.8%, in a small-sized city (<10,000). As for clinical characteristics, the most frequent cancer types were colon (47.4%) and breast (31.4%), stage I-II (55.4%).

Thirty-one physicians participated in this study; 80.6% (n = 25) were female; mean age was 36.8 years (SD = 8.3, range 28-62 years), with 12.7 years of experience (SD = 8.1, range 4-37 years). No significant differences were detected between male and female oncologists with respect to age (t = 0.32, p=.749) or years of experience (t = 0.31, p = .756). Most were super-specialists (71%) and worked at a public center.

Brief Symptom Inventory (BSI-18; Derogatis, 2001) includes 18 items divides into three dimensions (Somatization, Depression, and Anxiety) and the total score, the Global Severity Index (GSI) to assess general psychological distress in clinical and community populations (Appendix 1). Respondents were asked to answer in relation to how they had felt over the last 7 days and each item was rated on a 5-point Likert scale from 0 (not at all) to 4 (extremely). Raw scores are converted to standardized T scores which are characterized by a distribution with a mean of 50 and standard deviation (SD) of 10. According to the cutoff recommended by Derogatis (2001), a T-score ≥63 was used as indicative of caseness. In this study, Cronbach’s alpha was .75-.88.

Multidimensional Scale of Perceived Social Support (MSPSS; Zimet, Powell, Farley, Werkman, & Berkoff, 1990) is a 12-item, self-report scale that measures perceptions and adequacy of social support from three sources: family, friends, and significant others. The total score reflected the total degree of social support that individuals received. Items are scored on a 7-point Likert scale and total scores range from 7 to 84, with higher scores indicating greater perceived support. In this study, Cronbach’s alpha for the scale was .89.

Demographic survey. The following data were obtained: age, sex, marital status, educational level, occupational field, tumor site, stage, and treatment. The oncologist-related variables were sex, age, years of experience, type of medical oncologist (generalist vs. specialized), and type of hospital (academic vs. non-academic).

NEOcoping is a national, multicenter, cross-sectional, prospective study of the Continuous Care Group of the Spanish Society of Medical Oncology (SEOM) conducted between July 2015 and February 2019 in 15 Medical Oncology departments in Spain, from 10 autonomous communities. The protocol was approved by the Ethics Committee of each hospital and by the Spanish Agency for Medicines and Medical Devices (AEMPS) and all participants signed informed consent forms prior to inclusion. The study consists of self-report scales that the patients completed at the beginning and end of adjuvant treatment. Each questionnaire contained written instructions and specified that completion was voluntary and anonymous. Patients completed all pre-treatment questionnaires during the week following their first visit to the Medical Oncology Department to decide on adjuvant treatment.

The population consisted of consecutive patients with histologically confirmed, non-advanced, surgically-treated breast and colorectal cancer for which international clinical guidelines considered adjuvant treatment to be an option. Patients were excluded if they were under 18 years of age, had been treated with preoperative radio-or chemotherapy, only hormone therapy, or adjuvant radiotherapy without chemotherapy, and if they had any serious mental illness that prevented them from understanding the study. We screened 1,003 patients; 877 were eligible for this analysis and 126 were excluded.

Data analyses were conducted in five stages. In the first stage, basic sample and item descriptive statistics were obtained. In the second stage, the dimensionality of the BSI-18 item scores was assessed using exploratory factor analyses (EFAs) and a multifaceted approach (detailed below). Given that a clear, interpretable solution was attained at this stage, multiple-group confirmatory factor analyses (CFAs) were then performed and factorial invariance was assessed in groups determined by sex, age, and tumor site. In the fourth stage, validity relations were appraised via (a) product-moment correlations between BSI-18 raw scale scores and MPSSS scores (practical validity) and (b) a structural equation model, in which the CFA was extended to include MPSS scores (theoretical validity). Finally, mean pre-/post-treatment changes were assessed by means of a two-wave structural model.

Descriptive statistics at the first stage were obtained with the SPSS v23. Second stage EFAs were performed with the FACTOR program (Ferrando & Lorenzo-Seva, 2017; Lorenzo-Seva & Ferrando, 2013) and dimensionality was assessed with (a) conventional goodness-of-fit indices; (b) measures of appropriateness and essential unidimensionality, and (c) added-value analyses. The estimation procedure for all EFAs was robust, unweighted least squares with second-order (mean and variance) corrections (ULS-MV). The indices selected in were (a) the Standardized root-mean-square residual (SRMS) and the root-mean-square error of approximation (RMSEA), as measures of relative fit, and the CFI index as a measure of comparative fit with respect to the null independence model. Measures in (b) consisted of: the H index, which assesses the strength and replicability of the solution; the explained common variance (ECV) index, which determines closeness to unidimensionality, optimal implementation of parallel analysis (PA) (Timmerman & Lorenzo-Seva, 2011), and Schwarz’s Bayesian Information Criterion (BIC). Finally, added-value analyses (Ferrando & Lorenzo-Seva, 2019) assess the extent to which scores on a given factor are better predicted from the score estimates based on this factor than from the score estimates based on the overall single factor.

Multiple group CFAs were fitted using robust, weighted least squares estimation with second-order (mean and variance) corrections (WLS-MV), as implemented in Mplus (Muthén & Muthén, 2012). Model fit and appropriateness were explored with RMSEA and CFI. As for reference values, CFI values ≥.95 are indicative of good model fit (Schermelleh-Engel, Moosbrugger, & Müller, 2003), whereas RMSEA values ≤.06 indicate a satisfactory fit (Hair, Black, & Babin, 2010).

The property of measurement invariance (MI) appraised at this stage indicates that the BSI items measure the same dimensions with the same structure in all the groups to be compared. This property is a prerequisite if BSI scores are to be validly interpreted and compared. Various MI levels can be obtained; likewise, there is considerable debate regarding the appropriate level for a clinical instrument such as the BSI. Our position is that strong invariance is an attainable goal that, if achieved, sufficed to establish valid comparisons both at the individual and at the mean-group level (Millsap & Meredith, 2007). Therefore, provided that strong invariance is obtained, the mean differences sex-, age-, and tumor site-defined groups were assessed at this third, CFA stage.

The CFA model of stage 3, extended so as to include the MPSS scores as external variables, was the structural model used to evaluated validity relations at the fourth stage. The model was fitted and model-data fit was gauged using the same procedures detailed above for the third-stage analyses. Finally, the two-wave model to appraise sensitivity to post-treatment change was strongly-invariant over time; likewise, it was fitted and assessed by means of the same procedures described in the third-stage analyses.

Descriptive statistics of the eighteen BSI items can be found in Table 2. Mean item scores ranged from 0.09 (item 17) to 1.54 (item 3). BSI-18 item score distributions were unimodal and asymmetrical (positively skewed), thereby indicating that most of the values were concentrated at the lowest end of the response scale. Cronbach’s alpha estimates for raw test scores ranged from .75 to .88, which are similar to those reported in a community cancer sample (Galdón et al., 2008; Recklitis et al., 2017) that varied from .75 to .88 in adult survivors of childhood cancer (Recklitis et al., 2017) and .62-.70 in a Spanish breast cancer sample (Galdón et al., 2008).

Taking into account the descriptive results summarized above (skewed item scores), as well as the fact that the test is not very long and the sample is reasonably large, the best model choice to fit the data is nonlinear underlying-variables-approach FA (UVA-FA e.g. [Muthén & Muthén,2004]). In this modeling, the item scores are treated as ordered-categorical variables and the EFA is fitted to the inter-item polychoric correlation matrix (Ferrando & Lorenzo-Seva, 2014). As preliminary analyses, sampling adequacy was assessed by the KMO and Bartlett’s sphericity test and was considered to be very good.

In accordance with previous studies, models containing between 1 and 3 factors were fitted to the data. The 4-factor solution was not attempted because the fit of the 3-factor model was too good. Double cross-validations based on random, fully independent sub-samples were made for the EFA solutions, and the across-samples results were virtually identical in all cases (coefficients of congruence were always above .96). So, only the results based on the entire sample are reported. Results for the unidimensional and the 3-factor solution are located in the upper panel of Table 3.

As a summary, the unidimensional model has only a marginally acceptable fit in pure goodness-of-fit terms, whereas the fit of the 3-factor model is excellent by all the standards. However, the ECV value suggests that there is a strong dominant factor running through all the 18 items, and the PA-based procedure indicates the unidimensional solution as the most replicable. In contrast, the BIC results point to the 3-factor solution as being preferable.

To obtain further information in order to decide which dimensionality was most appropriate, the solution in three factors was next rotated to achieve maximum factor simplicity by using the Promin criterion (Lorenzo-Seva, 1999). The rotated pattern closely approached a simple structure (Bentler’s simplicity index was .98) and allocated all the items in the ‘a priori’ expected structure. Only two items: 8 (feeling blue) and 14 (hopelessness about the future) were found to be factorially complex and loaded on both the depression and anxiety factors.

The lower panel of Table 3 displays further measures aimed at comparing the unidimensional solution with the Promin rotated solution in three factors. Clearly, the rotated solution univocally demonstrates added value, because for all three factors, the true factor scores are better predicted from the corresponding estimates than from the score estimates in the single general factor (results in brackets). The panel also displays the reliability estimates for the factor scores derived from the solutions, as well as the alpha estimates for the raw total scale scores. The ordinal alpha estimate (Zumbo, Gadermann, & Zeisser, 2007) has been also included as theoretical upper limit for the reliability of the total scores.

The results summarized so far, suggest that the best solution for the BSI-18 in this population is a second-order solution, with 3 strongly related primary factors that are clear, meaningful, replicable, and lead to reliable scores, as well as a general second-order factor that runs throughout all the items and that can be understood as “psychological distress.” This general factor is also strong and replicable, and provides reliable scores: both factor score estimates and the simpler sum scores can be considered to be reliable enough to be used in clinical assessment. This general schema leads to the CFA solutions used in the next stage.

For the same reasons detailed above, and also for consistency in the procedures, all the structural models that follow (CFAs, extended validity model, and two-wave model) used the same UVA strategy as the EFA models in the previous section: variables were treated as ordered-categorical, and robust statistics based on mean-and-variance corrections were used for assessing model-data fit.

Based on the EFA results in the previous section, a second-order CFA model with three primary factors was fitted by using again a cross-validation schema based on independent random samples. Across-sample results of the proposed model were virtually the same in terms of both, parameter estimates, and goodness-of-fit measures. So, it is justified to consider a common solution, that, in principle, holds for the entire sample. The second-order loadings estimated in the full sample were: Depression=.98; Anxiety=.84; Somatic=.69. Based on this common solution, strong invariance assessments were next carried out. Table 4 shows the strong invariance results for sex, age, and tumor location. In all cases, the strong-invariance model based on the second-order solution has a remarkably good fit based on all the indices considered.

As discussed above, if measurement invariance is achieved, it can be assumed that the same dimensions are measured in the different groups and that the items function in the same way in these groups. Therefore, differences in group mean scores can be validly interpreted as reflecting ‘true’ group differences in the dimensions being measured. So as to interpret the mean differences in Table 4, we note that, for identification purposes, the means are always fixed to zero in the first group and are freely estimated in the remaining. Results can be summarized as follows: no mean differences in any of the primary factors were obtained for tumor location. Finally, and as regards age groups, the results clearly suggest that the levels in all three factors tend to decrease with age.

Validity results in Table 5 can be summarized as follows. First, the profile of raw validity coefficients in panel (a) and standardized coefficients in panel (b) is virtually the same, which is expected. However, the disattenuated standardized coefficients are higher in all cases, as they should be, and the differences are more pronounced, thereby making the pattern in (b) more easily interpreted. Second, the structural model has quite an acceptable fit.

As for the substantive results, the Depression factor is, clearly, the one that most strongly and negatively correlated with the MSPSS scales. Somatization and Anxiety correlate significantly in all cases, albeit effect sizes are small. Finally, the relations revealed with respect to the general factor might well be due to the fact that it mostly reflects depression. Overall, then, less perceived social support appears to be associated with all BSI factors, but especially with Depression.

Table 6 presents the results for the two-wave longitudinal model. The constraints here were similar to those in the previous multiple-group analyses but were imposed over time instead of over groups. Thus, strong invariance over time was assumed and the mean of each dimension at Time 1 was set to zero. Results can be summarized as follows. First, the models exhibited fairly acceptable. Second, significant pre-test post-test changes were observed for the three dimensions: a mean increase in Depression and Somatization and a mean decrease in anxiety. In terms of effect sizes (Cohen’s d), the Depression and Anxiety effects would be qualified as small, whereas the somatization effect is more substantial and would qualify as medium.