1 BACKGROUND

Following bariatric surgery, the majority of patients experience sustainable weight loss and significant improvements to physical impairment, obesity-related comorbidities and health-related quality of life (HRQOL).1-5 Furthermore, up to 80% of patients have reported being satisfied with their overall treatment satisfaction 5 years after surgery.6 Although bariatric surgery is followed by positive outcomes for most patients, a cohort study of almost 25 000 patients who had undergone bariatric surgery in Australia found an increase in psychiatric illness presentations after surgery, as well as increased occurrences of deliberate self-harm and suicidal deaths.7 These findings are supported in a recent systematic review exploring the risk of suicide and self-harm after bariatric surgery.8

The awareness of impaired mental health has led to an increased focus on evaluating such concerns after bariatric surgery. Mental illness preoperatively has been associated with lower total weight loss and the risk of increased weight regain in patients following Roux-en-Y gastric bypass.9 Furthermore, mental illness can contribute to a higher risk of early readmission after bariatric surgery.10, 11 A review article found that psychological factors like depression and anxiety may increase the risk of drop-out from follow-up consultations,12 which is disturbing, as lifelong follow-up after bariatric surgery is recommended.13, 14 This knowledge has led to changes to the guidelines for post-bariatric surgery follow-up, most recently to the British guidelines.15 In the British guidelines, a preoperative and postoperative mental health evaluation to identify those patients in need of psychological follow-up is recommended. For patients in need of mental health treatment, the follow-up should be with trained personnel, preferably a psychologist.15

As a way to organize a person-centred follow-up after bariatric surgery, patient-reported outcome monitoring with a clinical feedback system (PRO/CFS) appears promising,16-18 in particular regarding psychological health. A PRO/CFS is a system in which the patient completes various digital questionnaires prior to the consultation with a healthcare professional and their responses form the basis for the content of the communication between the patient and healthcare professional during the consultation.16 This enables the healthcare professional to start the consultation by addressing what the patient has reported as most bothersome. The use of a digital PRO/CFS to provide an instant analysis of the patient's responses and present them in a visual report has been found to be advantageous for the clinical consultation16 and the possibility to adapt the consultation to the patient's needs has been reported to be important to patients.19

The Norse Feedback (NF) is a multidimensional questionnaire developed for mental health treatment in Norway.20 The system was developed for a PRO/CFS and is computer-adaptive in that it automatically selects the items relevant to the individual patient based on what he/she answered to a set of trigger questions following an initial administration of the full item set.20 Traditional questionnaires developed with a nomothetic approach have been criticized for omitting important concerns experienced by many patients and for lacking relevance to particular patients because they are aimed at capturing problematic issues across the entire population.21 An ideographic adaptation in which the questionnaire is adapted to the patient's situation has been recommended.22 The NF contains a broad measure of symptoms and processes, many of which may be relevant for patients who have undergone bariatric surgery, such as suicidal ideation and substance use. However, which of these scales, if any, is most appropriate for use in bariatric surgery patients is not yet clear because the NF has thus far only been used and evaluated in mental and behavioural health settings.20

The aim of this study was to test the internal consistency and construct validity of the scales in the NF in a population of patients who had been accepted for or already undergone bariatric surgery. More specifically, we hypothesized that (a) the individual scales in the NF would demonstrate satisfactory internal consistency; (b) the items in each scale would demonstrate sufficient unidimensionality. To explore the relationship between the Obesity-related Problems scale (OP) and the NF we hypothesized that the pattern of correlations with the NF scales would be moderate and positive for scales that represent common mental health issues (such as Sad Affect and Somatic Anxiety), and food and socially relevant scales. This is consistent with previous research demonstrating that individuals with OP tend to experience a wide range of psychological issues. We anticipated lower correlations between the OP and less typically relevant scales of the NF.

2 METHODS 2.1 Ethical considerations

The study was approved by the Norwegian Centre for Research Data, Department of Data Protection Services (reference number 282738). All participants signed an informed consent form and were recruited to the study by an outpatient nurse. The PRO/CFS is integrated as a standard part of the consultation, so there was no additional burden on those patients who were willing to participate. The design of the implemented PRO/CFS is described elsewhere.23

2.2 Population

Patients who were accepted for or had already undergone bariatric surgery were eligible for inclusion in the study. The criteria for undergoing bariatric surgery were a body mass index (BMI) of over 40 kg/m2 or 35–39.9 kg/m2 with obesity-related comorbidities. When accepted for surgery, the patient attends a preoperative consultation, and follow-up at 3, 12, 24, 60 and 120 months after surgery. Each time the patient completes the PRO/CFS prior to the consultation at a bariatric outpatient clinic.

The surgical population had undergone laparoscopic sleeve gastrectomy (LSG) or biliopancreatic diversion with duodenal switch (BPD/DS). The patients had their consultation in the bariatric outpatient clinic between February 2018 and January 2020. For patients with several consultations during the course of the study, we included only their first consultation in our analyses (the first consultation could be at any follow-up consultation). All patients met with an outpatient nurse.

2.3 Outcome measures

The NF and the OP were a part of the PRO/CFS that the patients responded to prior to the consultations and were thereby administered at the same time.

2.3.1 Mental health

The NF is a multidimensional questionnaire developed for use with a digital PRO/CFS. The NF version 2.0 consists of 102 potential items that load on 23 different scales in four domains, three treatment process scales and four single items that do not load on any scales (see Table 1).20 Twelve of the NF scales consist of four or more items, whereas seven scales consist of three items. Higher scores indicate a higher symptom load. The scale composition of the NF is the result of an iterative process involving patient and provider feedback, as well as quantitative scale development methods. Since the NF scales are designed to be used somewhat independently and the full NF is rarely administered after the initial completion, we elected to maintain the scale structure of the existing NF, treating each scale of the NF as an independent measure. In other words, we do not conduct analyses of all items in the NF as a whole but treat each existing scale as an independent measurement. This is consistent with the development process of the NF in mental health settings.20 After an initial complete assessment in mental-health settings, some NF scales are automatically skipped for certain patients when they are algorithmically determined to be likely not important for care. The specifics of that algorithm are not relevant in this case since patients only completed the measure one time and so could reasonably complete all scales. In this project, we applied 20 scales, four single items and one two-item scale. For the Substance Recovery scale, we did not perform a CFA analysis, as only 20 patients had responded to this scale. The Substance Recovery scale only opens when the Substance Use scale is elevated above an empirical threshold. After considerations in a user-panel, including two patients and a clinician, we excluded the scales Alliance and Needs in treatment, as the items in these scales are specific to mental health treatment and not directly relevant for patients undergoing obesity treatment.23 Because the NF scales are designed to be used independently and in one score per scale, exclusion of these scales for research purposes should not affect the validity of the questionnaire.20

TABLE 1. Structure of domains, scales and number of items included in each scale in the Norse Feedback questionnaire Domains and scales included in the Norse Feedback Number of items Symptom Expression Eating Problems 5 Sad Affect 3 Somatic Anxiety 5 Substance Use 4 Suicidality 4 Trauma Reactions 4 Resources Readiness for Recovery 3 Recovery Environment 5 Social Safety 6 Problem-maintaining Processes Need for Control 4 Hopelessness 5 Internal Avoidance 5 Irritability 3 Self-Criticism 8 Situational Avoidance 3 Social Avoidance 3 Worry 3 Personal Consequences Cognitive Problems 6 General Functioning 3 Substance Recovery 4 Treatment Process Scales Alliancea 4 Needs in treatmenta 5 Medication 2 Individual Items I take care of my physical health I am sleeping very badly at the moment My sexuality and/or sex life is difficult for me I am impulsive in a way that troubles me a Scales not included in the clinical feedback system, as the items were not considered relevant for follow-up to bariatric surgery.

The items have seven-point response categories, ranging from This is completely true for me to This is not at all true for me, as well as the response choice Not relevant/Do not know/Refuse to answer.

When a patient completes the NF for the first time, the patient responds to all items. If the patient's scale score falls under the defined threshold for that scale, the patient will only answer one trigger item from that scale the next time he or she is administered the NF. This ensures an individualization of the questionnaire, as the patients are not asked items that are likely to be irrelevant to them. However, if the patient's response later switches to an empirically predefined threshold on the trigger item, the scale will re-open to its full length. This is considered a key feature in idiographic adaptation.22 The assessment of the adaptive part of the NF does not fall within the scope of this study and will be performed at a later stage.

2.3.2 Obesity-specific quality of life

The OP scale is a questionnaire developed to measure the psychosocial consequences of obesity.24 The questionnaire has been validated for use in Norway,25 as well as validated for both a surgical and non-surgical population of people with obesity.24 The questionnaire consists of eight items that measure the impact of obesity on social activities, public activities and intimate relations on a four-point scale (definitely bothered, mostly bothered, not so bothered and definitely not bothered). Lower scores indicate fewer psychosocial consequences of obesity. In research, the score is transformed to a 0–100 scale, where a lower score indicates better psychosocial functioning. In clinical consultations, the individual items provide the most useful information. The OP has been shown to correlate positively with anxiety, depression and overall mood measured using the Hospital Anxiety and Depression Scale (HADS).24

2.4 Analyses

Demographic data are presented in terms of frequency and percentages (%) for dichotomous data and mean with standardized deviation (SD) and median for continuous data.

The NF scales with four or more items were examined for structural validity (unidimensionality) using robust confirmatory factor analysis (CFA) for ordinal data, with weighted least squares and mean and variance adjustment (WLSMV) estimation, using the R package lavaan (version 0.6–7).26 Patients who did not complete all items on the given scale were excluded from the analyses because the WLSMV estimator in lavaan does not handle missing values. Factor loadings ≥0.4 were considered satisfactory. We evaluated goodness of fit using the root mean square error of approximation (RMSEA), standardized root mean square residual (SRMR), the comparative fit index (CFI), and the Tucker-Lewis index (TLI). Model fit was indicated by RMSEA close to or below 0.06, SRMR close to or below 0.08, and CFI as well as TLI close to or above 0.95.27, 28 To evaluate the internal consistency of scale items, factor loadings from the CFA were used to calculate composite reliability (CR) values for the NF scales, with scores ≥0.7 being considered satisfactory.29 CR is a CFA-based estimate of reliability and is not biassed as the more common Cronbach's alpha. Nevertheless, we also calculated Cronbach's alpha, as this is the most reported analysis for internal consistency with a threshold of 0.7 or higher considered as satisfactory,30 and report mean inter-item correlation, where ideal range is considered between 0.2 and 0.4.31 Thus, a total of at least 11 indices were compared for each scale with four or more items: three internal consistency metrics (CR, alpha, mean inter-item correlation), four CFA model fit indices, and factor loadings for each item. For scales with three items, we do not report model fit indices as they are uninformative, but the other seven indices were computed.

SPSS version 26 was used for other analyses. Hypothesis (c) was used to study the relationship between the OP and the NF, and we performed correlation analyses of the 19 different scales, as well as the four single-item indices and the one two-item index in the NF and the correlation to the total score of the OP, as the OP is designed to be used as a unidimensional measure. Given the large number of pairwise comparisons, we interpret effect sizes rather than statistical significance. Pearson's correlation coefficients ˂ 0.1were considered trivial, 0.1–0.29 as small, 0.30–0.49 as moderate and ≥0.5 as strong.32 SPSS version 26 was used for statistical analyses33 and exact two-sided p-values reported.

3 RESULTS

Out of 259 eligible patients, 213 (82%) are included in the analyses. The included patients had a mean age of 41.1 years, a mean BMI before surgery of 43.5 kg/m2, and 93% of the patients had undergone LSG. The majority of the patients (76.1%) were women and 27% of the patients reported receiving treatment for depression at the time of data collection. See Table 2 for demographic details on the patients included in the analyses. Two patients were excluded due to not giving consent to participate, whereas 44 patients were excluded because they had not completed the PRO/CFS when they met for their follow-up consultation. The patients included did not differ from those excluded from the study regarding age, biological sex, BMI before surgery or type of surgery. In the population not included in the analyses, more patients were divorced (11.4% vs. 3.8%), had undergone treatment for depression (20.5% vs. 12.7%), and fewer had education at university college level or higher (19.5% vs. 33.0%). In terms of the follow-up consultation, more patients in the population not included met for their 120-month follow-up (31.8% vs. 20.7%) and fewer for the preoperative consultation (15.9% vs. 20.7%).

TABLE 2. Demographic information on the 213 patients included in the analyses Demographic Included in the analyses (n = 213) Age Mean (SD) 41.1 (10.7) Median 41.0 Biological sex (female), n (%) 162 (76.1) BMI before surgery Mean (SD) 43.5 (5.5) Median 42.3 Type of surgery, n (%) LSG 198 (93.0) BPD/DS 14 (6.6) Depressiona, n (%) 27 (12.7) Marital status, n (%) Married/cohabitant 134 (64.1) Single 66 (31.6) Divorced 8 (3.8) Widow/widower 1 (0.5) Education, n (%) University college or higher 67 (33.0) Highschool or primary school 136 (67.0) Time point for follow-up, n (%) Preoperative 44 (20.7) 3 months 27 (12.7) 12 months 22 (10.3) 24 months 23 (10.8) 60 months 53 (24.9) 120 months 44 (20.7) Abbreviations: BDP/DS, biliopancreatic diversion with duodenal switch; LSG, laparoscopic sleeve gastrectomy; n, number. SD, standard deviation. a The patients are asked whether they are undergoing treatment for depression.

For descriptive information on the scales and individual items of the NF in bariatric surgery follow-up, see Table 3. The analyses demonstrated that the floor effect ranged from 5.7% to 88.2% (with floor effects detected in the scales Sad Affect, Somatic Anxiety, Substance Use, Suicidality, Trauma Reactions, Need for Control, Hopelessness, Irritability, Situational Avoidance, Social Avoidance, Worry, Cognitive Problems, General Functioning and Medication, and the single items Physical Health, Sleep, Sexuality, and Impulsivity), whereas the ceiling effect ranged from 0% to 13.1%.

TABLE 3. Mean, standard deviations, median, minimum-maximum scores and floor and ceiling effects of the Norse Feedback Scales Mean (SD) Median Minimum-maximum score Floor effect/ceiling effect % Symptom Expression Eating Problems 2.38 (1.09) 2.2 1.00–6.00 10.5/0 Sad Affect 1.86 (1.29) 1.33 1.00–6.67 48.1/0 Somatic Anxiety 2.13 (1.27) 1.70 1.00–6.20 23.8/0 Substance Use 1.14 (0.60) 1.00 1.00–6.25 88.2/0 Suicidality 1.24 (0.68) 1.00 1.00–5.75 81.1/0 Trauma Reactions 1.81 (1.06) 1.25 1.00–6.00 41.0/0 Resources Readiness for Recovery 2.76(1.29) 2.67 1.00–7.00 13.0/0.5 Recovery Environment 2.31(1.09) 2.20 1.00–7.00 15.2/0.5 Social Safety 3.03 (1.29) 3.00 1.00–6.83 5.7/0 Problem-maintaining Processes Need for Control 2.42 (1.29) 2.25 1.00–6.50 17.2/0 Hopelessness 2.05(1.13) 1.75 1.00–7.00 25.4/0.5 Internal Avoidance 2.43(1.30) 2.20 1.00–6.40 13.8/0 Irritability 1.91 (0.92) 1.67 1.00–4.50 30.0/0 Self-Criticism 1.89 (1.03) 1.63 1.00–5.88 13.7/0 Situational Avoidance 1.67 (1.16) 1.00 1.00–7.00 51.7/0.5 Social Avoidance 2.22 (1.53) 1.67 1.00–7.00 35.5/1.4 Worry 2.54 (1.68) 2.00 1.00–7.00 28.4/1.4 Personal Consequences Cognitive Problems 2.16 (1.31) 1.67 1.00–6.50 22.7/0 General Functioning 2.50 (1.29) 2.33 1.00–6.33 20.0/0 Treatment Process Scale Medication 3.33(1.40) 4.00 1.00–7.00 18.7/1.6 Individual Items I take care of my physical health 2.98 (1.61) 3.00 1.00–7.00 25.5/0.5 I am sleeping very badly at the moment 3.15 (2.11) 3.00 1.00–7.00 36.0/10.3 My sexuality and/or sex life is difficult for me 2.89 (2.14) 2.00 1.00–7.00 40.7/13.1 I am impulsive in a way that troubles me 1.78 (1.45) 1.00 1.00–7.00 68.1/2.4 Abbreviation: SD, standard deviation.

A summary of the psychometric properties of the NF is presented in Table 4.

TABLE 4. Psychometric properties of the 19 scales of the Norse Feedback Norse Feedback scales, n = 213 Composite reliability Cronbach's alpha Mean inter-item correlation Model fit indices Factor loadings above 0.4 Eating Problems, n = 201 0.76 0.63 0.27

CFI = 0.99

TLI = 0.97

RMSEA = 0.07 (90% CI 0.00–0.13)

SRMR = 0.05

Q46: 0.55

Q104: 0.76

Q57: 0.69

Q18: 0.69

Q63: 0.39

Sad Affect, n = 205 0.91 0.83 0.64 NA In all items Somatic Anxiety, n = 202 0.87 0.79 0.43

CFI = 0.99

TLI = 0.97

RMSEA = 0.11 (90% CI 0.05–0.17)

SRMR = 0.04

In all items Substance Use, n = 180 0.98 0.90 0.74

CFI = 1.00

TLI = 0.99

RMSEA = 0.13 (90% CI 0.04–0.23)

SRMR = 0.02

In all items Suicidality, n = 197 0.94 0.80 0.54

CFI = 1.0

TLI = 1.0

RMSEA = 0 (90% CI 0.00–0.13)

SRMR = 0.02

In all items Trauma Reactions, n = 186 0.83 0.67 0.35

CFI = 1.0

TLI = 1.0

RMSEA = 0 (90% CI 0.00–0.14)

SRMR = 0.02

In all items Readiness for Recovery, n = 181 0.64 0.40 0.20 NA

Q135: 0.19

Q134: 0.53

Q133: 1.00

Recovery Environment, n = 186 0.68 0.54 0.20

CFI = 0.98

TLI = 0.96

RMSEA = 0 (90% CI 0.00–0.12)

SRMR = 0.04

In all items Social Safety, n = 199 0.82 0.74 0.33

CFI = 0.95

TLI = 0.92

RMSEA = 0.13

(90% CI 0.09–0.18)

SRMR = 0.06

In all items Need for Control, n = 201 0.76 0.70 0.38

CFI = 0.99

TLI = 0.97

RMSEA = 0.08 (90% CI 0.00–0.18)

SRMR = 0.03

In all items Hopelessness, n = 182 0.83 0.66 0.32

CFI = 0.98

TLI = 0.96

RMSEA = 0.13