Translation, adaptation, validation and performance of the American Weight Efficacy Lifestyle Questionnaire Short Form (WEL-SF) to a Norwegian version: a cross-sectional study

Design, respondents and setting

The present study was conducted with a cross-sectional design including 225 morbidly obese patients accepted for bariatric surgery with laparoscopic sleeve gastrectomy (LSG) in a Western Norwegian hospital. We included two subsamples in the study; 114 consecutive non-operated patients from pre-operative outpatient consultations, and 111 consecutive operated patients from outpatient consultations one year after surgery, all within the period from October 2012 to May 2013.

The outpatient consultations started with a multidisciplinary informative plenary meeting, wherein the patients were shortly introduced to the present study. Voluntary participation was emphasized. Written information about the study was distributed with the questionnaires. Informed consent was obtained, and the questionnaires were collected at the end of the day before the respondents left the hospital.

The inclusion criteria were morbidly obese patients eligible for LSG (BMI ≥ 40, or ≥ 35 with comorbidity) and age between 18 and 60 years. Patients were excluded if they were physically or mentally disabled and could not fill in the forms.

Translation and adaptation (aim 1, research question a)

According to the recommendations in the guidelines by the World Health Organization (WHO), we performed a five step, systematic approach to translation and adaptation of the questionnaire (WHO, 2007). Initially, two registered dietitians who are native speakers of Norweigian and professionally familiar with the concepts about morbidly obese patients did an independent forward translation of the WEL-SF from American-English to Norwegian. Next, a consensus panel of four people comprised of the research group compared the original version with the two translated versions. The group reconciled the forward translations into one common version by identifying inadequate concepts or expressions. Third, two blinded backward translations into English were done by a surgeon and health educator, both of whom were native speakers of Norweigian. Furthermore, the consensus panel compared the original version and the translated version with respect to conceptual- and cultural equivalence and agreed on a Norwegian version for pretesting. Finally, two nurses, a registered dietitian and a bariatric surgeon, were asked to assess the feasibility of the items in the Norwegian version for the bariatric patients. They found the questionnaire to be of clinical relevance for the population.

The Weight-Efficacy Lifestyle Questionnaire Short Form (WEL-SF)

In 2012 a short version of the original WEL (Clark et al., 1991) was developed — from 20 questions and 5 situational components to 8 questions and 1 situational component representing “confidence in ability to resist eating” (Ames et al., 2012). Three of the questions are related to emotional eating situations, two to availability, one to social pressure, one to positive activities and one to physical discomfort. The WEL-SF correlated highly significantly with the WEL, accounting for 94% of the variability in the original questionnaire, and was found to be a psychometric valid measure of eating self-efficacy (Ames et al., 2012). The instrument range scores on a Likert-scale from 0 (not at all confident) to 10 (very confident), with sum scores between 0 and 80. High scores are associated with high eating self-efficacy.

Validating instruments

The instrument selection was based on a theoretical expected association with eating self-efficacy (Fayers & Machin, 2007). Weight loss maintenance after bariatric surgery requires a balance between energy intake and energy expenditure. It has been stated that this demands self-management skills in both eating behavior and physical activity (Sallis et al., 1988; Morin, Turcotte & Perreault, 2013; Wing et al., 2001). Due to this, we obtained the Self-efficacy for Physical Activity Questionnaire (SEPA) (Fuchs & Schwarzer, 1994) as one of the validating instruments. Based on social cognitive learning theory, we also assumed that individuals with high efficacy levels toward challenging life obstacles in general would be more likely to report high confidence in adequate manners of eating (Bandura, 1977; Sherer & Maddux, 1982). Thus, the General Self-efficacy Scale (GSE) (Luszczynska, Scholz & Schwarzer, 2005) served as a second validating measure. Furthermore, as the outcome expectations and measures of success in bariatric surgery is sustained weight loss and health related quality of life, we wanted to calculate the association between eating self-efficacy (WEL-SF) and health related quality of life, both in general and weight specifically. For this purpose the Short Form 36 (SF-36) (Ware, 2000) and the Impact of Weight on Quality of Life-Lite (IWQOL-Lite) (Kolotkin et al., 2001) were chosen as the third and fourth validating instruments.

Self-efficacy for physical activity (SEPA)

Self-efficacy for physical activity refers to the belief of being capable to stick to an exercise program even under unfavorable circumstances. The questionnaire was first developed in German by Fuchs & Schwarzer (1994) and assesses self-efficacy for physical activity using a 12-item measure on a Likert-scale ranging from 1 (very confident) to 7 (not confident at all). The sum score ranges from 12 to 84. High scores indicate high levels of perceived physical self-efficacy. The instrument was positively correlated with general self-efficacy and with specific self-efficacy expectations toward cancer screening and healthful eating behavior (Fuchs & Schwarzer, 1994), and has been translated and adapted to Norwegian conditions (Jenum et al., 2003).

General self-efficacy scale (GSE)

The General self-efficacy scale (GSE) contains general questions measuring an individual’s confidence in his or her personal competence to fulfill difficult tasks (Luszczynska, Scholz & Schwarzer, 2005). The instrument measures a person’s ability to cope with a broad range of demanding unspecific situations in life, and thereby assess his or her optimistic self-belief toward difficulties in general. The questionnaire has been translated, psychometrical tested and adopted for studies worldwide (Schwarzer et al., 1997; Scholz et al., 2002; Røysamb, Schwarzer & Jerusalem, 1998). The GSE contains 10 items on a Likert-scale ranging from 1 (completely wrong) to 5 (completely right). The sum scores ranges from 10 to 40. High scores indicate high levels of general self-efficacy.

Short form 36 (SF-36)

SF-36 is the most widely used generic self-report health questionnaire, which is based on a multidimensional model of health (Ware, 2000). The scale assesses health related quality of life outcomes, known to be most directly affected by unspecific disease and treatment and was first translated and adapted to Norwegian in 1998 (Loge et al., 1998). The 36 items are measuring 8 different aspects (subscales) of health related quality of life. The 8 subscale scores can be summed into two domains: physical component sum score (PCS) and mental component sumscore (MCS). The sub scores are transformed into a scale where high scores indicate high health-related quality of life. A score = 50 represents the average PCS and MCS scores in the US population. The psychometric properties of the SF-36 are well documented (Ware, 2000) and are validated for use in a Norwegian morbidly obese population (Karlsen et al., 2011).

Impact of weight on quality life-lite (IWQOL-lite)

Impact of Weight on Quality of life-lite is a validated, 31-item self-report measure of obesity-specific quality of life (Kolotkin et al., 2001). The questionnaire consists of a total score and scores on each of five scales; physical function, self-esteem, sexual life, public distress and work—exhibiting strong psychometric properties (Kolotkin et al., 2001). The subscores are transformed into a scale from 0–100 where high scores indicate high obesity specific quality of life. The version in use is linguistically-, but not yet psychometrically, validated in a Norwegian morbidly obese population.

Socio-demographic and clinical data

Socio-demographic variables of age, gender, marital status, level of education and work participation were recorded. The clinical variables include initial weight, weight loss, BMI, changes in BMI, height, diabetes, hypertension, psychiatric disorder, muscular- and skeletal pain and weather the respondents had undergone surgery or not. Changes in BMI were collected retrospectively. Data were coded and registered as categorical or continuous variables.

Statistical analysis

Data are presented as mean and standard deviation (SD) or number (%) unless otherwise stated. Between-group comparisons at baseline were analyzed using independent samples t-test for continuous variables and Pearson’s chi-square test for categorical variables. We employed two-tailed tests and considered P values <.05 statistically significant. The statistical analysis was conducted using SPSS version 21.0.

Internal validity (aim 2, research question a and c)

Data quality was examined comparing mean values for each item with standard deviation, median, percentage of missing values and extent of ceiling and floor effects. Optimal floor- and ceiling effects were defined to stay between 1% and 15% (Mchorney & Tarlov, 1995). Internal consistency was assessed by calculating Cronbach’s alpha coefficients. According to Clark & Watson (1995) the alpha coefficient should be benchmarked at .80 to raise reliability to an acceptable level. To eliminate the risk of a potentially false high reliability coefficient, we also calculated alpha if single items were deleted (Polit & Beck, 2008). Further, we measured the internal item convergence in terms of each items’ correlation with the rest of the scales’ total score. A minimum item-total correlation was benchmarked at the level of .3 (Fayers & Machin, 2007). In order to examine face validity, nine bariatric patients, included from the outpatient consultations one year after surgery, evaluated the questionnaire. The scales feasibility was assessed by four professional health workers.

Construct validity and factor analysis (aim 2, research question c)

To examine the structural validity of the WEL-SF we applied principal component analysis (PCA) with a varimax rotation (Tabachnick & Fidell, 2006). The Kaiser–Meyer–Olkin measure and Bartlett’s test of spherity were computed to determine whether the data in this sample were suitable for PCA. Following Kaiser’s criterion, eigenvalues of 1.0 were chosen to ensure that the extracted components accounted for a reasonably large proportion of the total variance (Tabachnick & Fidell, 2006). The PCA was first applied on the total sample (n = 225) and then on each subsample to compare the component structure between samples.

Convergent validity (aim 2, research question b)

Convergent validity was tested by comparing Pearson correlation coefficients between the WEL-SF and SF-36, the Impact of Weight on Quality of Life (IWQoL-Lite) questionnaire, the Self-Efficacy for Physical Activity Questionnaire (SEPA) and the General Self-Efficacy Scale (GSE).

Sensitivity (aim 2, research question a and d)

Multiple linear regression analyses were conducted to evaluate whether the WEL-SF discriminated between non-operated and operated patients, adjusted for age, gender, work participation, marital status and education.

Ethical approval

The study was approved by the Regional Committee for Medical Research Ethics in Western Norway and performed in accordance with the Helsinki Declaration (Saksnr 2012/1481).

Characteristics of the participants

We included 225 Caucasian morbidly obese patients (69.3% women) accepted for bariatric surgery; 114 patients prior to surgery and 111 patients one year post-surgery. All patients that were asked to take part in the study agreed to participate, giving a response rate of 100%. A very high item response was obtained with missing values of only 0.4%. The missing items were not substituted. The distribution of answers was right-skewed with no floor effect and a ceiling effect of 8.9% for the entire sample; respectively 0.9% and 17.1% for the non-operated and operated subsample. A further characteristic of the respondents and description of data is shown in Tables 1–2 and Fig. 1.

Face validity of the WEL-SF

To be able to compare the results from the original WEL-SF with the psychometric properties obtained in the translated version, it is of major concern that the item construction in the two versions is semantically equivalent. Banduras’ test-theoretical approach to the development of self-efficacy scales worked as a guide during the item evaluation. We aimed to take the reader’s perspective using everyday vocabulary jargon. Furthermore, we aimed to avoid ambiguous or multi-barreled items that include different types of attainments within the same item, where the respondents may have different levels of perceived efficacy. Item 4 in the American WEL-SF (I can resist overeating when I am watching TV (or use the computer) may, in a Norwegian context, represent a double-wording problem in which it refers to disparate situations challenging eating self-efficacy. To assess our assumptions toward this potentially double-wording problem, we extracted the PC- item into a new item 9: “I can resist eating too much when I am using my PC/Ipad” and placed it elsewhere in the questionnaire. The mean score for the TV-item in the non-operated group was 6.07 and in comparison 8.90 for the PC-item. The difference was respectively the same in the operated group. The respondents seemed to experience significantly less eating efficacy while watching TV than when using the computer. As most respondents were checking the same, high response point on the PC-item, followed by a ceiling effect and low variability, it did not add relevant clinical information. We therefore decided to eliminate the PC-item from the questionnaire and maintained the original item. We worded the item closer to the original global WEL: “I can resist eating too much when I am watching TV”.

The translation process revealed divergence in translation of the concept “overeat”. The American “overeat” can qualify as a medical diagnosis (F50.4. ICD-10) within the broader framework of eating disorders (F50. ICD-10). Culturally and semantically “overeat” was interpreted as closer to the Norwegian “eating too much”. The Norwegian “overeating” seems as such conceptually more related to the American “binge-eating” which involves a pathological pattern of compulsive food intake. As we do not assume that all bariatric patients suffer from an eating disorder, we chose to reformulate “overeating” into “eating too much”. By this reformulation we also aimed to avoid potential stigmatizing and biases.

A pretest was performed to assess face validity and feasibility. Nine patients were recruited from outpatient consultations for this purpose one year after undergoing bariatric surgery. They were asked to complete the questionnaire and thereafter express whether the questionnaire was clear and easy to understand, covering topics of interest and if any items had been difficult to answer. In addition they were asked whether the questions were relevant for their situation. The pretest presented no corrections to the items and confirmed their clearness and relevance. Some of the respondents considered the introduction-text inappropriately long and redundant. We shortened and simplified the introduction accordingly. The participants in the pretest were not included in the psychometrical test performance of the translated version of the WEL-SF.

Reliability, internal consistency and sensitivity of WEL-SF

Cronbach’s alpha coefficients were 0.92 for the whole sample, 0.89 for the non-operated sample and 0.92 for the operated sample (Table 2). The Alpha value remained high (0.86–0.92) if single items were deleted (Table 3).

Construct validity and factor analysis

The data met the Kaiser–Meyer–Olkin measure (0.89) and the Bartlett’s test of spherity criterion (p < 0.001) for performing PCA. Following the Kaiser’s criterion, components with an eigenvalue > 1.0 were contained. The PCA was performed on the entire sample (n = 225) and the eight items of the WEL-SF loaded on one component only (Table 4) with an eigenvalue of 5.04 explaining 63% of the total variance. When performing the PCA on each of the subsamples this picture did not change. The 1 component solution had an eigenvalue (explained variance) of respectively 4.5 (56.4%) and 5.2 (65%) for the non-operated and operated sample. In comparison, Ames’ one-component solution accounted for 49% of the variance.

Convergent validity

The correlation matrix for the sum score of the WEL-SF and the validating instruments covering our sample is illustrated in Table 5. The correlations ranged from .34 to .45 for all patients which represents a medium strong correlation (Cohen, 1988). Separating the groups, the correlations ranged from .12 to .37 in the non-operated group and .08 to .30 in the operated group, i.e., non to moderate strong correlations (Table 5).

Sensitivity

WEL-SF sum score was lower in the non-operated than in the operated group in unadjusted analysis (Table 1). This difference remained using multiple regression as the WEL-SF sum score was 12.55 (95% CI: −16.59, 8.51) points lower in the non-operated than in the operated group (p < 0.001) (not shown).

Strength and limitations

A cross-sectional design represents potential limitations due to its lack of time measurement (Polit & Beck, 2008). Nevertheless, we find this methodological approach appropriate for the present study due to our intention of inferring WEL-SF’s present psychometrical properties for future predictive purposes.

Data from both non-operated and operated patients strengthened the study in terms of a larger sample-size, and by bringing the opportunity to assess the WEL-SF’s sensitivity for the overall different eating pattern between the two groups. The subgroups were similar regarding socio-demographic variables, but had different health profiles, as expected. This was, nevertheless, taken into account by conducting the statistical analysis for the two subsamples in addition to the entire sample to visualize the outcome differences and similarities.

We noted some possible problems with ceiling effect in the operated group. This may be a problem if the WEL-SF is to be used for measuring change over time because of potentially low responsiveness beyond one year after surgery. Studies with longer follow-up should be performed in order to explore this issue, and caution must be taken in future studies if ceiling effects are common in Norwegian bariatric patients.

The response bias (Polit & Beck, 2008) was reduced due to the consecutively and convenient sampling procedure, contributing to a very high response rate and only one missing value. The referral of patients to the hospital from general physicians throughout the country strengthens the representativeness and generalizability of the results. Nevertheless, as most of the respondents were Caucasian, all admitted for surgery, the results may not be valid for obese patients from other ethnic groups, or for those seeking non-surgical treatment.