The sample consisted of 568 cancer patients; 59.8% (n=338) were women and the average age was 59.1 years (SD=12.1, range 26-84). Most patients were married or lived with a partner (77.1%) and had completed primary education (58.3%). The most common employment status was retired (60.0%). The sample's clinical characteristics revealed that the most common types of cancer were colon (40.5%, n=230), and breast (33.5%, n=190). All relevant socio-demographic and medical characteristics are included in Table 1.

Patients were recruited by 30 medical oncologists from 14 Spanish hospitals; 78.1% (n=25) of these specialists were female; mean age was 35 years (SD=7.4, range 27-62 years), and 11.9 years of experience in caring for cancer patients (SD=8.8, range 3-37 years). Most were super-specialists (68.8%) working at a public, teaching hospital (53.1%).

SDM-Q-9 is a brief, valid, and reliable questionnaire that evaluates the SDM process from the patient's perspective (Kriston et al., 2010), adapted to Spanish (De las Cuevas et al., 2015). The questionnaire contains nine items, each describing one step of the SDM process (Simon et al., 2006), it was developed to assess the degree to which patients feel involved in the decision-making process. The items are scored from 0 to 5 on a six-point Likert scale ranging from ″completely disagree″ (0) to ″completely agree″ (5). Standard scoring is a simple sum score with values between 0 and 45. Internal-consistency (alpha) reliability estimates are generally high in patients with chronic diseases: .98 (Germany), .94 (U.S.), and .88 (Spain).

SDM Questionnaire-Physician's version (SDM-Q-Doc) is a questionnaire that evaluates the physician's perspective and how well they follow SDM with their patients (Scholl et al., 2012). It was adapted and validated to Spanish (Calderon et al., 2017). The questionnaire consists of nine items, each of which describes one step of the process. The items are scored from 0 to 5 on a six-point Likert scale as “completely disagree” (0) to “completely agree” (5). A simple sum-score with possible values between 0 and 45 is obtained. In this study, Cronbach's alpha for the scale was .90.

SIS is a 4-item scale that was created to ascertain patients’ degree of satisfaction with the information provided by their physician about the disease, risk of recurrence, side effects of treatment, and time dedicated to informing them. The scale provides two subscales: satisfaction with the information provided and satisfaction with the time dedicated. Items are scored from 0 to 4 on a five-point Likert scale ranging from ″completely disagree″ (0) to ″completely agree″ (4); the higher the score, the greater the satisfaction with the information provided. The scale revealed a Cronbach's alpha value of .82 in our study.

The patients’ medical and demographic variables included were: age, gender, marital status, educational level, occupational sector, tumor site, stage, and time since diagnosis. The oncologist-related variables included age, gender, years of experience, area of specialization (general (treating all kinds of tumors) vs. super-specialized (treating one specific subtype of tumor) and type of hospital (academic vs. non-academic).

This is a multi-institutional, prospective, transversal, observational study that pooled consecutive patients recruited at 14 Spanish teaching hospitals from June 2015 to May 2017. The study is part of a research program investigating patients with cancer; it is funded by the Continuous Care Group of the Spanish Society of Medical Oncology (SEOM). The study was approved by the Ethics Review Board at each institution and by the Spanish Agency of Medicines and Medical Devices (AEMPS). Inclusion criteria were being ≥18 years of age, having a histologically confirmed, non-advanced, solid tumor surgically treated for which international clinical guidelines consider adjuvant treatment to be an option. Patients with metastatic disease, treated with preoperative radio- or chemotherapy, or with adjuvant hormonal or radiotherapy without chemotherapy were excluded. Similarly, physical ailments, comorbidity, and/or age precluding chemotherapy, and personal, psychological, family, sociological, geographical, and/or underlying medical condition that, in the investigator's opinion, could hinder the individual's ability to participate in the study were also cause for exclusion, since these patients did not have to decide on adjuvant therapy. The evaluation was performed in all cases approximately one month following surgical resection, in the context of patients’ first visit with the oncologist to decide on adjuvant chemotherapy. Data collection procedures were similar at all hospitals. Participation was voluntary, anonymous, and would not affect their care in the slightest. The participants completed the questionnaires individually, with no limit on time. Data were collected and updated by medical oncologists, specifically trained to comply with the study requirements, via a web-based platform (www.neocoping.es). Of the 627 patients screened, 59 were not eligible (17 did not meet inclusion criteria; 23 met exclusion criteria and 19 had incomplete data).

Descriptive analyses were conducted for every SMD-Q-9 item and explored means standard deviations and distributions of the item scores. To assess the factorial structure of the scale, the sample was randomly split into two groups and different Exploratory Factor Analysis (EFA) solutions based on previous reported results were performed on the first split-half sample. The adequacy of the inter-item correlation matrix to be factor analyzed was first assessed using the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy. The different hypothesized solutions were next fitted by using robust, unweighted least squares estimation with mean-and-variance corrected fit statistics as implemented in the FACTOR program (Lorenzo-Seva & Ferrando, 2013). They were: (a) unidimensional (as the scale was initially designed to be single-trait), (b) unidimensional with item 1 omitted (De las Cuevas et al., 2015) and (c) bidimensional with correlated factors. From the EFA results, a simple and clearly interpretable bifactor structure (Lorenzo-Seva & Ferrando, 2013) could be specified. This structure was next fitted to the entire sample with FACTOR by using the same estimation procedure described above.

In all the tested solutions above, the goodness-of-fit indices used to assess model-data fit were: (a) RMSEA, with its 95% confidence interval (as a measure of approximate fit); (b) Goodness-of fit-index (GFI), and (c) the root mean square of the standardized residuals (z-RMSR), (as absolute measures of fit), and (d) the comparative fit index (CFI), (as a relative measure of fit with respect to the null independence model). We followed the usual rules in deciding model appropriateness (Schermelleh-Engel, Moosbrugger, & Müller, 2003). In addition to model-data fit measures, additional indices of appropriateness for assessing the strength and replicability of the solution (H index) as well as closeness to unidimensionality (ECV index) were also obtained (Ferrando & Lorenzo-Seva, 2017).

Once the proposed structure had been fitted and found appropriate, scores based on this structure were obtained and their reliability was assessed by using the omega coefficient (McDonald, 1999). Finally, construct and external validity were assessed on the basis of these scores by using product-moment correlations and univariate ANOVA-based mean-group comparisons using Bonferroni corrections. For all the tests conducted, bilateral statistical significance was set at p≤.05.

Item means ranged from 2.59 (item 8) to 3.81 (item 1) and the mean sum of SDM-Q-9 was 3.15 (SD=0.9). In general, the item scores were negatively skewed and with high kurtosis values. So, we decided to use the underlying-variables approach, and fit the FA models to the inter-item polychoric correlation matrix (more details in (Ferrando & Lorenzo-Seva, 2013). This approach is quite feasible here given that the test is short and the sample reasonably large. Finally, regarding model adequacy, the KMO index (.87) suggested that the inter-item correlations were substantial and appropriate for being factor analyzed. Table 2 presents the descriptive statistics corresponding to the SDM-Q-9 items.

Because a clear final structure was attained in the entire sample, only a summary of the previous exploratory results will be provided here. The unidimensional model with the original 9 items was untenable by all the standards. Omitting item 1 considerably improved the fit bringing it to the lower limits of acceptability and providing an ECV estimate of .80, which means that 80% of the common variance of the item scores can be explained by a general factor.

The solution in two factors had an excellent fit and was interpretable: factor 1 clustered the items 1, 2, 3, 4, and 5 that assess the information and explanations provided to the patient by the physician about treatment and the advantages and disadvantages of the different options. Factor 2 clustered the items 6, 7, 8, and 9 that appraise the choice of the best treatment option for the patient. However, items 1 and 9 were factorially complex, with substantial loadings on both factors. Factor 1 was stronger and better defined, with a replicability H index of .82 (Ferrando & Lorenzo-Seva, 2017) whereas that of factor 2 was only .75. Finally, the estimated inter-factor correlation was rather high: r=.61.

To sum up, the initial analyses suggest that a unidimensional solution omitting item 1 is almost acceptable, whereas the bidimensional solution fits very well and has a clear interpretation, although it consists of two short and highly correlated factors, one of which is relatively weak, with low replicability. Consequently, no highly reliable scores can be expected to be derived from this factor. In view of these results we considered that the most appropriate and parsimonious solution for the SDM-Q-9 items was a bifactor solution (Rodriguez, Reise, & Haviland, 2016) based on all 9 items, with a general factor that describes the entire SDM process, and a group factor defined by items 1 to 5 related to the information and treatment options provided to the patient. The bifactor solution is justifiable, given the data's essential unidimensionality; it also maintains the parsimony and strength advantages of the unidimensional solution (clear interpretation and high reliability of the general factor). At the same time, the additional inter-item covariance between items 1 to 5 that cannot be explained by the general factor is modeled as a group factor, thereby avoiding potential bias on the general factor due to unmodeled inter-item covariance.

On the basis of the EFA results summarized above, a bifactor SCFA solution was fitted to the entire sample data with the following specifications: factor 1 (the general factor) was defined by all 9 items, and factor 2 (the group factor) was defined by items 1 to 5. As in the previous EFAs, the bifactor model was fitted by using robust ULS estimation as implemented in FACTOR. Goodness-of-fit results are in Table 3 and indicate an excellent fit.

The general factor in Table 3 is well defined by the 9 items with all loadings >.30 except for item 1 (as expected from previous results). These loadings can be interpreted as item discriminations and are reasonably acceptable for a personality measure. The H-index is therefore rather high: 0.88, meaning that the factor is strong, well defined, and likely to replicate across different samples. The second group factor is mainly defined by items 1 and 5, which have high loadings on it and only moderate loadings on the general. In this case, the H-index is only 0.73, which means that the group factor is far weaker than the general factor, the most common result in bifactor solutions (Rodriguez et al., 2016).

To assess the invariance of the solution described thus far, a series of analyses were performed by splitting the entire sample into subsamples according to gender and pathology type. In all cases, the results were found to be essentially invariant, both in terms of item locations and item discriminations. Thus, there appears to be no differential item functioning for any of the items and the scale is expected to function with the same properties in the general population for which it is intended. Given the space limitations, the invariance results are not provided here, but are available from the authors.

The clear bifactor solution discussed above allows two summed scores to be obtained. First is the sum of the scores for all 9 items, which represents the general factor, and so, aims to measure a general dimension of perspectives regarding the decision-making processes. The second score is the sum of items 1 to 5, represents the group factor above, and measures a more specific sub-dimension of information and treatment options provided to the patient. From the H results above and also from basic psychometric principles, the total sum score is expected to be more reliable than the group sum score, and this was indeed the case. The omega reliability estimates were .90 for the total scores and .85 for the group scores. So, both scores achieve a quite acceptable degree of accuracy, and the total scores in particular would be considered as accurate enough for clinical (individual) assessment. Overall, the total scores are more representative of the entire SDM process and will be the ones used in the validity assessments below.

To ascertain whether the reliability of the total scores reflects accuracy at all trait levels, conditional reliabilities were also estimated (see Ferrando & Lorenzo-Seva, 2017). Results revealed that conditional reliabilities were >.85 for a range of trait values between two standard deviations below the mean and two standard deviations above the mean. Hence, not only do the SDM scores possess good overall reliability, but this reliability is also high for almost the entire effective trait range, and the .90 estimate reported above is thus representative of the overall precision of these scores. This result is a positive feature of the instrument and suggests that SDM would enable most respondents to be accurately assessed.

Finally, given the results summarized in this section, and for the benefit of practitioners, a normative table based on the total sum score was constructed based on the entire sample data. The table is provided as supplementary material.

Construct validity was explored by analyzing the product-moment correlations between the total SMD-Q-9 scores (as proxies for the general SDM dimension) and scores from other questionnaires aimed at measuring theoretically-linked dimensions. Results indicated that the total SMD-Q-9 score relates positively with satisfaction regarding the patient-physician relationship (r=.29, p<.001), specifically with the time dedicated (r=.40, p<.001), but not with the information provided (r= -.02, p=.313). Similar results were found with the group factor SDM-Q-9, which related positively with satisfaction regarding the patient-physician relationship (r=.36, p<.001) and time dedicated (r=.32, p<.001), but not with information provided (r=.07, p=.313).

Significant correlations were found between SDM-Q-9, patient version, and SDM-Q-Doc, physician version (r=.14, p<.001), between women (r=.21, p<.001), but not in men (r= -.04, p=.464) and estimated risk of relapse (r=.04, p=.289).

Men tended to exhibit greater satisfaction with the SDM than women (F(1,566) =10.96, p<.001) and patients over 60 years (n=288) more than younger ones (n=270) (F(1,556) =5.19, p=.023) with significant intergroup differences (F(3,552)=7.53, p<.001). Furthermore, the subsequent post-hoc Bonferroni analysis uncovered significant gender- and age-based differences (p<.005) such that women under the age of 60 years were the least satisfied group with SDM versus men of the same age, or older women (see Figure 1).

Figure 1.

Mean 9-item Shared Decision-Making Questionnaire (SDM-Q-9) scores by gender and age.

Note. Horizontally, the figure shows patients grouped by gender (man or woman) and by age (≥ or <60 years) and vertically, the score on the Shared Decision-Making Questionnaire (SDM-Q-9).

(0.05MB).