Psychometric properties of the Eating Loss of Control scale among post-bariatric patients

Abstract

Background:

Assessing the complexities of eating behaviors in patients who undergo bariatric surgery is challenging. The Eating Loss of Control Scale (ELOCS), a measure of loss-of-control eating, has not yet been evaluated psychometrically among bariatric surgery patients.

Objective:

This study presents a psychometric examination of the ELOCS in a postoperative bariatric surgery patients.

Setting:

Academic medical center in the United States.

Methods:

171 post-bariatric treatment-seeking adults (82.5% female, 54.4% White) with LOC-eating completed the ELOCS and measures assessing eating psychopathology and mood. Confirmatory factor analysis (CFA) was used to test fit for a 1-factor solution. Exploratory factor analysis (EFA) examined alternative factor structures.

Results:

CFA revealed poor fit for a 1-factor structure, χ² = 229.375(135), p < .001, CFI = .917, TLI = .906, RMSEA = .067. EFA data suggested an alternative factor solution, χ² = 157.76(118), p = .009, CFI = .965, TLI = .955, RMSEA = .047. Factor 1 (α = .88) reflected behavioral aspects and Factor 2 (α = .92) reflected cognitive/emotional aspects of LOC-eating. Bivariate correlations with measures of eating and other psychopathology suggested good construct validity for factors.

Discussion:

Findings suggest possible differences in the construct validity of the ELOCS among post-bariatric patients. The 1-factor solution previously supported in clinical and non-clinical groups demonstrated poor fit. EFA revealed a possible alternative 2-factor solution that aligns with emerging literature, suggesting that LOC-eating presents differently in post-bariatric patients.

Conclusions:

Researchers interested in LOC-eating among bariatric patients should consider use of the ELOCS and testing the proposed alternative factor structure.

Introduction

Bariatric surgery is an effective treatment resulting in significant weight loss; however, weight and psychosocial outcomes are variable, particularly in the long-term postoperative period [1–4]. Accumulating evidence suggests that loss of control eating (LOC-eating) [5], a central feature of certain eating disorders, is common among bariatric patients, and is as an important prognostic indicator with respect to surgical outcomes [6–8]. Assessing the features and manifestations of disordered eating is challenging and there are various clinical, self-report, and interview methods [9]. The Eating Loss of Control Scale (ELOCS) was designed to comprehensively assess the severity and frequency of LOC-eating episodes [5]. While previously validated among non-clinical and clinical participant groups [10–12], this study aimed to fill an important gap by examining the psychometric properties of the measure among postoperative bariatric patients.

Defined as the subjective experience of being unable to stop or control the eating episode, LOC-eating is distressing to the individual and is a hallmark symptom of binge-eating. LOC-eating is consistently associated with greater eating and general types of psychopathology [e.g., depression; 7, 13] Moreover, among postoperative bariatric patients, LOC-eating is predictive of poorer weight loss [7]. Given the clinical importance, establishing a psychometrically valid instrument is vital to accurately characterizing the construct among this patient group. Further, the clear discrepancies in the size of LOC-eating episodes between pre and postoperative states, and along the postoperative trajectory, demand a validated measurement tool designed to accurately assess LOC-eating changes across various stages of the bariatric process [14].

The ELOCS is one of six measures that assess LOC-eating to varying degrees. Alternative measures include a single stand-alone assessment, Loss of Control over Eating Scale [LOCES; 15], a continuous measure of multiple types of eating and disordered eating behaviors, the Three Factor Eating Questionnaire [16], a diagnostic investigator-based interview, the Eating Disorder Examination [17] and its parallel self-report version, the EDE-Questionnaire [18], and a self-report screener, the Questionnaire on Eating and Weight Patterns -Revised [QEWP-R; 19]. Nuanced differences regarding assessment approach exist for each of these instruments, but importantly, none of the LOC items pertaining to these measures have previously been validated among bariatric patients. However, some studies exploring other types of eating measures have found that the factor structures vary when comparing bariatric and non-bariatric samples [20–22].

This study aimed to address this gap by examining the underlying factor structure and reliability of the ELOCS among postoperative patients. It was hypothesized that the factor structure for bariatric patients would differ from the original validated structure given the research supporting variant factor structures when comparing bariatric and non-bariatric samples [20–22].

Methods

Participants

Participants included 171 adults seeking treatment for eating/weight concerns following bariatric surgery. Patients were six months (M = 6.32 months, SD = 1.52; Range 4–9 months) post-surgery. The majority of patients (n = 148) underwent sleeve-gastrectomy surgery, while the remaining participants underwent Roux-en-Y gastric bypass. Participants were recruited from a single clinical site, and recruitment continued until a priori minimum sample size of trial-eligible patients (n = 140) was achieved. The current analyses included all patients who presented for clinical intake. No compensation was provided at the time of the intake. Patients were considered eligible for the trial if they were between the ages of 18 and 65, and they endorsed LOC-eating at least once weekly. LOC-eating was based on participants’ subjective experiences, considered independent of the amount consumed (i.e., an “unusually large amount of food” was not required), and assessed by trained doctoral-level interviewers using a semi-structured investigator-based interview (Eating Disorder Examination—Bariatric surgery version). Exclusion criteria for the clinical trial included: current use of medications known to influence eating or weight, current substance dependence, or severe psychiatric condition that required immediate treatment (e.g., current active suicidal intent or current mania). All participants provided informed consent, and this study received approval from the MASKED FOR REVIEW.

Participants were primarily female (n = 141, 82.5%), and racial-ethnic distribution was as follows: White non-Hispanic (52.4%), Black non-Hispanic (32.4%), Latina/o (10.0%) and other (5.3%). The mean age and body mass index (BMI) were 45.5 (SD = 11.0) years and 37.2 (SD = 7.1) kg/m², respectively. Pre-surgical BMI was 47.2 kg/m² (SD = 8.8). Mean percent excess weight loss (%EWL) was 45.5% (SD = 17.4%) and percent total weight loss (%TWL) was 26.12 (SD = 12.47). About half of participants (52.9%) had a lifetime history of binge-eating disorder prior to surgery.

Measures

Body Mass Index (BMI).

Pre and post-surgical BMI were calculated based on measured height and weight. Pre-surgical BMI was obtained from participants’ medical records and post-surgical BMI was obtained approximately six months (M = 6.32 months, SD = 1.52; Range 4–9 months) post-surgery at the initial study evaluation.

Eating Disorder Examination—Bariatric surgery version [EDE-BSV; 2] is a semi-structured interview assessing different forms of overeating and eating disorder psychopathology during the past 28 days, modified for bariatric surgery patients. This study used the alternative three-scale structure (dietary restraint, weight/shape dissatisfaction, and weight/shape overvaluation) which demonstrated superior psychometric properties among bariatric surgery patients [23]. The dietary restraint subscale pertains to conscious efforts to limit food intake for shape and weight reasons, while the two other subscales reflect body image concerns and overvalued ideas regarding weight and shape. Responses are coded on a 0 to 6 scale, with scores of 4 or higher indicating a clinical level of impairment. Average scores on the three subscales are summed and averaged to compute a global eating disorder psychopathology score. The EDE-BSV also characterizes LOC-episodes, including eating episodes that feel out of control and/or eating in which a person has difficulty stopping, regardless of the quantity of food consumed.

Yale Food Addiction Scale [YFAS; 24] is a 25-item self-report measure of food addiction based on the symptoms outlined for substance dependence in DSM-IV-TR. Psychometrically, the YFAS demonstrates adequate internal consistency, convergent validity, and incremental validity in predicting binge eating [24], including binge eating among bariatric-surgery patients [25]. Reliability was excellent in this sample (Cronbach α = .91).

Beck Depression Inventory Second Edition [BDI-II; 26] is a 21-item self-report measure of current symptoms of depression. It has been widely validated in clinical populations, including bariatric surgery patients [27]. Reliability was excellent in this sample (Cronbach α = .92).

Eating Loss of Control Scale [ELOCS; 5] is a 20-item scale assessing subjective experience of loss of control over eating in the past 28 days, with each item containing two parts: part one assesses the frequency of specific LOC-eating events; and part two assesses severity or the degree to which the person experienced feeling out of control, where 0—Not at all hard to stop to 10—Extremely hard to stop). Severity scores are calculated by taking the average of the severity items. Note that severity scores are sometimes calculated using all 20-items, and sometimes calculated using 18-items, consistent with the validated factor structure for this study group. The method of computing the mean score is indicated where appropriate. Factor structure is investigated using the severity rating only. Higher scores are indicative of greater LOC-eating severity.

Statistical Analyses

Descriptive information related to the study group and variable distributions were examined. Next, a univariate model consistent with the available published literature was fit using confirmatory factor analysis (CFA) and maximum likelihood estimation. For the current work, χ², comparative fit index (CFI) Tucker-Lewis index (TLI) and Root Mean Square Error of Approximation (RMSEA) were used [28]. A value of ≥.95 was considered indicative of good fit for CFI and TLI, whereas a value of ≤ .06 was considered good fit for RMSEA [29].

Exploratory Factor Analysis (EFA) was used to examine if under-retaining factors was a contributor to observed issues of misfit. Rotations were used to aid in interpretations of potential factors. Items with factor loadings greater than .40 were retained; factor loadings greater than .71, .63, .55., .45, and .32, were interpreted as indicative of excellent, very good, good, fair, and poor fit, respectively [30]. CFA and EFA analyses were completed using Mplus Version 7.0 [31], and the remaining analyses were completed using IBM SPSS. Finally, psychometric properties for the alternative factor structure were explored, including internal consistency and construct validity based on correlations with measures of depression and eating-disorder psychopathology.

Results

On the EDE-BSV, patients reported a median number of six LOC-eating episodes in the last 28 days. Utilizing the ELOCS, the frequency of episodes varied by items (See supplemental Table 1 for median and IQR for number of episodes and mean and SD for severity items). The mean severity rating for the 20-item scale was 5.37 (SD = 2.90) and did not vary as a function of race, ethnicity or gender.

In the first step, a univariate model including all items was fit using maximum-likelihood estimation and delta parameterization. The data suggested issues with a one-factor fit (Table 1). Similarly, a univariate model consistent with the validation paper [18-items; 5] as well as the published revised version without the proposed correlated error terms [16-items; 12] were specified, and fit indices suggested poor fit and possible misspecification. Exploratory factor analysis (EFA) was undertaken to determine if under-retaining factors could be contributing to the observed problems with fit.

Table 1.

Fit Statics for Alternative Factor Structures in Wave I Sample

Model	Overall fit indices
	χ2 (df)	CFI	TLI	RMSEA
Model 1	273.212(170), p < .001	0.912	0.902	0.063
1 factor 20 items
Model 2
1 factor 18 items (Blomquist et al., 2014)	220.375(135), p < .001	0.917	0.906	0.067
Model 3
1 factor 16 items (Hopwood et al., 2018)	177.673(104), p < .001	0.921	0.909	0.068
Model 4†
2 factor 18-item version of the scale	187.413(134), p = .002	0.953	0.946	0.051

Note. CFA = Confirmatory Factor Analysis, CFI = Comparative Fit Index, TLI = Tucker-Lewis Index, RMSEA = Root Mean Square Error of Approximation.

^†these data represent a CFA of the model suggested by EFA within the same sample

As multiple samples [5, 12] failed to find support for two items “Ate unhealthy food choices” and “Feel out of control when you have not eaten an unusually large amount of food,” the subsequent EFA utilized the 18-item scale. The results of EFA suggested several, possible alternatives to the univariate model, all of which demonstrated improved fit indices. In addition to examining fit indices, which supported a possible 2, 3, or 4-factor solution, rotated factor loadings were examined to determine the smallest number of interpretable factors. Across several rotations, 2 and 4-factors emerged as interpretable and a good fit to the data. As parsimonious solutions are generally considered more favorable [28], a 2-factor solution was subjected to further psychometric analyses.

Table 2 includes the factor loadings for the proposed 2-factor solution. Broadly, the items appear to capture the behavioral (factor 1) and cognitive/emotional (factor 2) aspects of LOC-eating. Four items failed to load at the fair level onto either Factor 1 or Factor 2 [30]. When less conservative thresholds (i.e., < 0.40) are applied, only two items which refer to eating until uncomfortably full and eating rapidly failed to load onto either factor. It was hypothesized that time from surgery could significantly influence the strength of the relationship between the items and the latent factors, and therefore the items were retained at this stage. The items did not significantly cross load, suggesting a degree of independence between factors. The internal consistency reliability for both Factors 1 and 2 was good to excellent: Cronbach α = .88 and .92, respectively.

Table 2.

EFA 18 item 2 factor solution: Geomin Rotation Factor loadings

	Factor 1 “Behavioral”
ELOCS 1. Go out of your way to get food you were craving	0.708	−0.123
ELOCS 3. Give up control over what you ate BEFORE started to eat	0.851	−0.008
ELOCS 4. Give in to an impulse to eat even though not hungry	0.652	0.162
ELOCS 5. Ignore an interruption to keep eating	0.500	0.055
ELOCS 7. Keep eating even though you thought you should stop	0.610	0.246
ELOCS 8. Eat much more rapidly than normal	0.299†	0.259
ELOCS 9. Eat until you feel uncomfortably full	0.368†	0.350
ELOCS 10. Ate large amount of food when not physically hungry	0.408	0.089
ELOCS 16. Give up even trying to control eating	0.589	0.239
		Factor 2 “Cognitive/Emotional”
ELOCS 2. Feel helpless to control eating urges	0.372	0.453
ELOCS 11. Feel embarrassed about how much you were eating	−0.087	0.841
ELOCS 12. Feel disgusted, depressed, or very guilty while eating	0.123	0.758
ELOCS 13. Afraid of losing control over eating	−0.176	0.924
ELOCS 14. Feel driven or compelled to eat	0.276	0.532
ELOCS 15. Hard to stop eating once started	0.341	0.489
ELOCS 17. Feel upset by the feeling that you couldn’t stop eating	0.049	0.859
ELOCS 18. Hard to stop thinking about food you were craving	0.319	0.441
ELOCS 19. Feel out of control when you have eaten an unusually large amount of food	0.004	0.811

Notes.

^†items loaded below .40 threshold but were conservatively retained

Table 3 reports the descriptive information and correlations between the two factors, BMI, depression, and eating-disorder psychopathology. Dietary restraint was associated significantly with the cognitive/emotional aspects of LOC-eating only, and generally, eating-disorder variables were more strongly related to the cognitive/emotional factor as compared with the behavioral LOC-eating factor.

Table 3.

Psychometric Properties and Evidence of Construct Validity for the Two-Factor ELOCS Model

		M (SD)	1	2	3	4	5	6	7	8	9	10
1	ELOCS Factor 1	5.03 (1.78)
2	ELOCS Factor 2	5.81 (2.24)	.73**
3	Pre-Surgical BMI	47.24 (8.75)	−.02	.02
4	Current BMI†	37.20 (7.13)	.08	.01	.64**
5	% EWL	45.51 (17.44)	−.16*	−.04	−.31**	−.56**
6	Depression	11.80 (9.90)	.37**	.39**	0.14	.18**	−.28**
7	Food Addiction	2.87 (1.82)	.54**	.54**	.00	.06	−.14	.55**
8	Weight/Shape Overvaluation	2.74 (1.90)	.34**	.31**	.05	.00	−.04	.37**	.40**
9	Weight/Shape Dissatisfaction	2.91 (1.56)	.35**	.47**	.27**	.25**	−.35**	.47**	.40**	.39**
10	Dietary Restraint	3.09 (1.84)	.08	.18*	.13	.01	.14	.08	.06	.18*	.18*
11	Global Eating Psychopathology	2.91 (1.24)	.36**	.45**	.20**	.11*	−.10	.43**	.41**	.76**	.70**	.66**

Notes. ELOCS = eating loss of control scale; % EWL = percent excess weight loss

^†due to the skew in the distribution of current BMI, Kendall’s tau correlation test were used, all other variables used Pearson’s Correlation

^**Correlation is significant at the 0.01 level (2-tailed).

Discussion

The current study explored the psychometric validity of the ELOCS among post-bariatric surgery patients with regular LOC-eating. Results indicated that the existing, validated 1-factor structure for the ELOCS was a poor fit among these patients. Subsequent analyses supported an 18-item, alternative factor structure in which LOC-eating was defined by two distinct factors—a behavioral factor and a cognitive/emotional factor. Bivariate correlations with measures of eating and other psychopathology suggested good construct validity across both factors. A valid factor structure is essential for assessing differences between groups of patients (e.g., possible demographic differences or differences among clinical groups) as well as differences within patients over time (e.g., before and after bariatric surgery).

In the current study group of postoperative bariatric surgery patients, the frequency of LOC-eating episodes varied substantially; the EDE-BSV indicated a median of six episodes in the past 28 days, while the ELOCS, which assesses the number of days of various types of LOC episodes occur, indicated a median range of 0 to 10 of the last 28 days. Although such findings might simply reflect well-known variability between self-report and interview-based assessments of the same behaviors [9, 32, 33], the findings also suggest the possibility that LOC-eating experiences are more variable in post-bariatric patients than what is captured by the EDE-BSV interview. Importantly, responses may be influenced by both eating pathology as well as the proposed mechanism of surgical intervention. For example, one of the most commonly endorsed items describes eating large amounts when not physically hungry. Lack of hunger is frequently endorsed in post-bariatric patients. Generally lower levels of hunger, and the associated decrease in energy intake, are thought to be one contributing factor to weight-loss observed in the first year [34].

The average severity rating, using the 20-items version of the total scale, and responses of 0—Not at all hard to stop to 10—Extremely hard to stop was 5.37, and the average was highly similar for the 18-item version of the scale (M = 5.40). This severity rating appears most similar to other clinical study groups with BED 6.38 [5] and 6.30 [12], compared with a non-clinical study groups 1.57 [12]. These results suggest that individuals presenting with LOC-eating after bariatric surgery are reporting difficulties that appear comparable to levels reported for treatment-seeking clinical groups of patients diagnosed with BED. Clinically, such findings for patients with LOC-eating following bariatric surgery suggest that postoperative consideration of existing evidenced based treatments that address LOC-eating in the context of binge eating, such as cognitive behavior therapy [35] might warrant consideration.

Given the degree of misfit across published factor structures and the problems associated with relying too heavily on modification indices [36], EFA was conducted to determine if under retaining factors was contributing to the observed misfit. A 2-factor model was thought to be the most appropriate solution for several reasons. First, the fit statistics appeared similar to other alternative solutions in which a greater number of factors were retained. Parsimony is an important principle in factor analysis, as a higher number of factors is thought to reduce the likelihood of generalizability [28]. Second, a review of the items comprising the 2-factor solution revealed face-valid (i.e., easily interpretable) factors, including a factor comprised of the behavioral aspects of LOC-eating and a factor comprised of the cognitive and emotional aspects of LOC-eating. These two factors were internally consistent and related to other types of eating symptomology as well as depression. These results contrast those found previously with non-bariatric patients, which supported a unidimensional structure [12]. The present findings suggest, that among post-bariatric patients, LOC-eating may best be understood as including distinct elements related to the actual eating behavior as well as the cognitive and emotional experience of this eating. Patients show variability (e.g., higher or lower levels) in each of these factors; thus, these findings seem to highlight the potential importance of understanding patient’s subjective experience of LOC-eating in addition to previous measures’ narrower focus merely on frequency of specific behaviors. It also indicates that a single total ELOCS score may inaccurately characterize a post-bariatric patient’s level of eating difficulties. Hypothetically an individual may score very high on all the psychological symptoms (Factor 2) of LOC-eating, but using a total score, their overall level of eating concerns would be characterized as minimal. Measuring these difficulties more accurately will allow for understanding of how the symptoms change over time as well as the potential for targeted interventions to address the aspects of LOC-eating that may be most challenging for the patient. These data also align with emerging literature suggesting that the original factor structure(s) of eating-disorder measures appear to have poor fit among bariatric patients [22]. While the body of literature is limited at this time, and most often explored at the pre-surgical stage, it is evident that more studies are needed to explore psychometrics of eating disorder measures beyond LOC-eating after bariatric surgery.

While the 2-factor solution demonstrated good fit overall, a single item – eating more rapidly than usual-appeared to not be significantly related to either factor. This item might be unrelated to LOC-eating among post-bariatric patients given the recommendations to engage in slower eating post-surgery. However, a conservative decision was made to retain the item at this stage, as the association between eating rapidly and LOC-eating may differ as a function of time (months post-surgery) or surgery type [38]. Additional independent data are needed to replicate this factor structure as well as determine the degree to which eating rapidly should be considered relevant to the behavioral expression of LOC-eating among post-bariatric patients.

Study findings should be considered within the context of several limitations. First, the conclusive validity of the 2-factor structure is unknown, and it will remain as such, until it is independently replicated. Second, the current data are limited to understanding LOC-eating among post-bariatric patients who are presenting with concerns about this exact phenomenon. However, the range in frequency of LOC-eating justifies psychometric analysis in this study group. Third, a relatively small sample size was utilized. Small samples increase the probability of sampling error as small samples may be less likely to approximate the population of interest. Independent replication of these findings would increase confidence in model fit and would also be necessary before interpreting any parameter estimates. A final limitation is related to the restricted assessment of construct validity. For example, information related to convergent validity with another measure of LOC-eating and the predictive validity of the two discrete LOC-eating factors remain unknown. Ultimately, the current study represents an important step towards improving the psychometric utility of the measure among this unique patient group, and the conclusions serve to clarify future directions.

Conclusions

LOC-eating is a central feature of some specific eating disorders and an important prognostic indicator with respect to surgical outcomes. The ELOCS is one of few validated LOC-eating instruments, and this study was the first to explore its psychometric validity among postoperative bariatric patients. Results demonstrated that the existing, validated 1-factor structure was a poor fit and instead, supported two distinct factors—a behavioral factor and a cognitive/emotional factor. Researchers interested in LOC-eating among bariatric patients should consider use of the ELOCS as well as testing the proposed alternative factor structure.

Highlights.

• LOC-eating among post-bariatric patients is prognostic of surgical outcomes

• The ELOCS assesses LOC-eating

• These data suggest differences for using the ELOCS among postoperative patients

• Researchers should consider the ELOCS and testing the proposed factor structure