Cross‐cultural validation of the patient‐practitioner orientation scale among primary care professionals in Spain

Abstract Background In recent decades, many self‐report instruments have been developed to assess the extent to which patients want to be informed and involved in decisions about their health as part of the concept of person‐centred care (PCC). The main objective of this research was to translate, adapt and validate the Patient‐Practitioner Orientation Scale (PPOS) using a sample of primary care health‐care professionals in Spain. Methods Baseline analysis of PPOS scores for 321 primary care professionals (general practitioners and nurses) from 63 centres and 3 Spanish regions participating in a randomized controlled trial. We analysed missing values, distributions and descriptive statistics, item‐to‐scale correlations and internal consistency. Performed were confirmatory factor analysis (CFA) of the 2‐factor model (sharing and caring dimensions), scale depuration and principal component analysis (PCA). Results Low inter‐item correlations were observed, and the CFA 2‐factor model only obtained a good fit to the data after excluding 8 items. Internal consistency of the 10‐item PPOS was acceptable (0.77), but low for individual subscales (0.70 and 0.55). PCA results suggest a possible 3‐factor structure. Participants showed a patient‐oriented style (mean = 4.46, SD = 0.73), with higher scores for caring than sharing. Conclusion Although the 2‐factor model obtained empirical support, measurement indicators of the PPOS (caring dimension) could be improved. Spanish primary care health‐care professionals overall show a patient‐oriented attitude, although less marked in issues such as patients’ need for and management of medical information.

in improved health outcomes and greater resource use efficiency. [3][4][5] In the last decades, many self-report instruments have been developed to assess the extent to which patients want to be informed and involved in decisions about their care, the most widely used of which are the Control Preference Scale 6 and the Autonomy Preference Index. 7 Research has shown that most patients desire a collaborative or autonomous role in their medical decisions and that many of them do not feel as involved as they would want. 3,8 Furthermore, less perceived involvement or a mismatch between preferred and experienced involvement has been shown to be related to poorer satisfaction, adherence and quality of life. [9][10][11][12][13] Although not all variations in patients' perceptions of being involved are caused by an actual deficit in health-care professionals' behaviour, the above results clearly reinforce the need to enhance PCC and improve patient participation. To achieve this aim, health-care providers must develop certain communication and social skills-and an obvious prerequisite is to hold favourable beliefs and attitudes towards this model of care. Consequently, assessing these attitudes becomes a relevant issue in the research and implementation of PCC in routine practice and in academic curricula. While a number of instruments have been developed for that purpose, this is lesser extent than for patients. [14][15][16] One such instrument is the Patient-Practitioner Orientation Scale shown to be associated with more patient-centred behaviours in consultations, 18 while congruence between patients and physicians' orientations has been demonstrated to be associated with greater patient satisfaction, [19][20][21] trust and endorsement of physicians, 19 and fewer referrals. 22 The PPOS, which has been translated from English to several other languages, [23][24][25] has been widely used to assess preferences for patient orientation among health-care professionals, medical students and even patients, since it is applicable to the general population. [26][27][28][29][30] However, few studies have reported the psychometric properties of the instrument or, more specifically, its factorial validation. [31][32][33] Since (to our knowledge) no studies have been published on use of the PPOS in Spain, our aim was to translate, adapt and carry out a psychometric validation of this scale using a sample of health-care professionals.

| ME THODS
This study analyses PPOS baseline data for participants, who were recruited during 2016, for a cluster randomized controlled trial that aimed to assess the impact of a virtual community practice intervention on health-care professionals (general practitioners and nurses) at the primary care level. 34 Primary care centres from Results: Low inter-item correlations were observed, and the CFA 2-factor model only obtained a good fit to the data after excluding 8 items. Internal consistency of the 10-item PPOS was acceptable (0.77), but low for individual subscales (0.70 and 0.55).
PCA results suggest a possible 3-factor structure. Participants showed a patient-oriented style (mean = 4.46, SD = 0.73), with higher scores for caring than sharing.

Conclusion:
Although the 2-factor model obtained empirical support, measurement indicators of the PPOS (caring dimension) could be improved. Spanish primary care health-care professionals overall show a patient-oriented attitude, although less marked in issues such as patients' need for and management of medical information.

K E Y W O R D S
health-care professionals, person-centred care, PPOS, primary care, validation 3 Spanish autonomous communities (Canary Islands, Catalonia and Madrid) were contacted via their managers and invited to participate. Centres were randomly selected, while a balanced north/ south geographical representation was maintained within each region. In-person meetings were held in each centre to explain the study in detail to interested professionals. Those who agreed to participate signed the informed consent and received a password to access a web interface where they could fill out the PPOS questionnaire. Participants' allocation to the intervention group or control group was only disclosed after the questionnaire was completed. The scale is scored on a 1-6 Likert scale (totally disagree-totally agree). Items, except for 9, 13 and 17, are written in a physician-oriented style; scoring is therefore reversed in such a way that a higher score indicates a patient-oriented style. Scores for the overall scale (18 items) and sharing and caring subscales (9 items each) are divided by their corresponding number of items and thus range between 1 and 6.

| Sociodemographic and professional data
Data were collected on age, sex, profession (general practitioner or nurse), years' experience, tutorship of medical residents/medical or nursing students (yes/no), and patients attended per day. 5. In a final equivalence testing step, the back-translation was compared with the original instrument by the study directors in Spain (LPP, AIGG, CJBG and CO) and, after some minor revisions, the Spanish version was considered ready to use.

| Questionnaire translation
The final Spanish version of the PPOS was pre-tested on first twelve adult patients attended at the two primary care centre participants in this study and their responses were analysed to identify whether any modifications were necessary, which resulted not to be the case.

| Statistical analyses
The distribution and descriptive statistics for the items (missing values, frequencies, means, standard deviations, asymmetry and kurtosis) were calculated. Floor and ceiling effects for each item were defined as more than 85% of participants scoring 1 (totally disagree) and 6 (totally agree), respectively. Also calculated were the mean inter-item correlations, corrected item-to-scale correlations and Cronbach's α after excluding each item. In order to assess whether the data fit the 2-factor model proposed for the scale (sharing and sharing), a confirmatory factor analysis (CFA) was performed. Missing values were handled by using full information maximum likelihood estimation, which does not require the imputation of missing values, but uses all the available data to estimate population parameters. 36 However, in the presence of non-normal data this technique can produce negatively biased standard errors, leading to an erroneous rejection of the null hypothesis. For this reason, we first assessed non-normality by means of Yuan, Lambert and Fouladi's normalized estimate of multivariate kurtosis, applicable to data with missing values 37 (a value outside the range −3,3 is indicative of multivariate kurtosis).
In the case of non-normality, standard errors were corrected using the robust method proposed by Yuan and Bentler. 38 The model was refined by repeating the analysis after excluding items with non-significant coefficients or low R 2 values. Its fit was then assessed by means of the chi-square test (or the Yuan-Bentler's scaled chi-square, 38 in the case of performing robust estimation); since this statistic is very sensitive to sample size, we calculated several other recommended goodness-of-fit indices [39][40][41] : Finally, we assessed associations of the obtained scales with participants' sociodemographic and professional characteristics. Due to the clustered nature of the study design (ie professionals clustered into centres), we used a 2-level mixed multiple regression model, with fixed effects for professionals (level 1) and random effects for centre (level 2), to adjust for correlated observations within the clusters.
Analyses were performed with SPSS 21.0 and EQS 6.2 software.  Table 1 shows sociodemographic and professional characteristics of the professionals. Mean age was 47.7 years (SD 8.8), and 76% were female. Over half (59%) were general practitioners, and the remaining 41% were nurses. Mean years' experience was 22.0 (SD 8.84), mostly in primary care (mean 17.7; SD 8.92). The professionals attended a mean of 29 patients per day (SD 11.3), and 25.5% had tutored residents/medical and nursing students.
Item #8 had 15 missing values (4.7%), whereas 9 more items had between 1 and 4 missing values ( Table 2). There were no ceiling or floor effects for any item. Distributions were asymmetric, with most participants stating some level of disagreement (slightly, moderately or totally disagreed) with the physician-oriented style. Item #9, written in patient-oriented terms (Patients should be treated as if they were partners with the doctor, equal in power and status), disagreement was high (80%). Favourable or less critical responses with the physician-oriented style were obtained for 4 items, specifically, item #5 (Patients should rely on their doctors' knowledge and not try to find out their conditions on their own), item #8 (Many patients continue asking questions even though they are not learning anything new), item #10 (Patients generally want reassurance rather than information about their health) and item #18 (When patients look up medical information on their own, this usually confuses more than it helps).

| Dimensionality and internal consistency
Mean inter-item correlation was 0.15 (median 0.14). The 3 items written in a patient-oriented style (#9, #13 and #17) obtained the lowest mean correlations (0.08, 0.07 and 0.09, respectively). Table 3  We carried out a CFA for the 2-(correlated) factor model. We used the maximum likelihood (with missing values) estimation method, with robust standard errors due to the non-normal distribution of the data (multivariate kurtosis normalized estimate 25.5).
All items obtained significant coefficients and, as shown in Table 4,  Table 5. The first, second and third components include 5 items on information, 6 items on the patient-physician relationship and 4 items favouring technical aspects of medicine and an asymmetric relationship between patient and professional, respectively. Items #4, #11 and #14 saturated above 0.30 in 2 or 3 components. When 2 components were extracted, the above-mentioned first and third components collapsed into a single dimension (not shown in the table).

| Associations with sociodemographic and professional variables
Multilevel mixed regression models did not point to any significant association between the 10-item PPOS overall or its subscales and age, sex, profession, years' experience, tutorship, or patients attended per day.

Mean b (SD)
Asym. Kurt. For our sample of Spanish health-care professionals in the primary care sector, the proposed structure of the PPOS with 2 correlated factors (representing the sharing and caring dimensions) only obtained an acceptable fit to the data after excluding 8 items (almost half of the total scale with 18 items). Inter-item correlations were observed to be low overall; the 3 items phrased in a patient-oriented style were the most uncorrelated, pointing out to a possible methodological effect that has also been observed in other studies for 2 of these items. 31,32 The PCA also pointed to poor functioning of several of the remaining items, which loaded similarly on more than a single component. Of the extracted components, 2 (C1 and C2) can be partially assimilated,

| LI M ITATI O N S
A limitation of this study is that, although the sample was not small, it was not large enough to be randomly split into 2 subsamples: to explore the latent structure and to act as a validation sample