Simulation data for novel weight suppression score (NWSS)

A random set of 1,566 observations was created using Stata 18 [29] (Stata code is in italics below for easy replication). This number was adjusted post-hoc to create a sample size matching our observational data. Given that the average weight of adults (ages 20+) in the US is approximately 185 pounds [30], three separate Poisson distributions (N = 1566 each) with means approximately equal to 185 were created. A brief discussion accompanies the methods for the simulation data, whereas the results and discussion section below are reserved for the patient dataset. The simulation was conducted prior to the analysis of the observational data to conceptualize the difference between the two approaches. The objectives were to (1) propose a novel weight suppression score (based on clinical experience) rather than a crude difference measure, (2) visualize the distribution of the two approaches, and (3) create a variable that could be used in future research either as a moderator or outcome (rather than just a predictor) in regression models.

set seed 5249

set obs 1566

gen e1 = rpoisson(185)

gen e2 = rpoisson(185)

gen e3 = rpoisson(185)

The first distribution was determined as the lowest adult weight (LW), the second distribution as the highest adult weight (HW), and the third as the current weight (CW).

gen LW = e1

gen HW = e2

gen CW = e3

A dataset was then created by removing all implausible values: simulated observations were kept only if the lowest adult weight was less than the highest adult weight and if the current weight was less than or equal to the highest weight and greater than or equal to the lowest weight. Because this simulation data is random, the final values are unlikely to reflect what is reported in the observational data; they are offered to show how the NWSS reflects a normally distributed variable that can be dichotomized for analyses.

The final dataset for analysis contained 287 observations.

Traditional weight suppression (TWS) method

The traditional weight suppression (TWS) approach subtracts the highest adult weight from the current weight [9], shown in Fig. 1.

Fig. 1

Distribution of traditional weight suppression using simulated data (N = 287)

As expected, this is not normally distributed (median < mean). This variable may be used as a linear predictor, but the approach poses methodological challenges for dichotomizing the variable for indicator analysis.

Importantly, this method is also flawed because it would assign the same value to someone with a lifetime high of 200 lbs. who is currently 100 lbs. (reduced weight by 50%) as it would to someone with a lifetime high of 400 lbs. who is currently 300 lbs. (reduced weight by 25%) (see persons C and H in Table 1, color-coded to indicate the same values). In addition, this approach lacks a nuanced context of weight history.

Novel weight suppression score (NWSS) method

Next, a lifetime (adult) midpoint variable (LM) was created by averaging the lowest adult weight (LW) and the highest adult weight (HW), as shown in Fig. 2.

Fig. 2

Distribution of lifetime midpoint using simulated data (N = 287)

As expected, this is normally distributed (the median is approximately equal to the mean). Next, we generate the NWSS by dividing the lifetime (adult) midpoint (LM) by the current weight (CW), creating more stringent criteria for suppression than the TWS approach (Fig. 3).

gen NWSS = LM/CW

hist NWSS

Fig. 3

Distribution of novel weight suppression score using simulated data (N = 287)

As expected, this is again normally distributed (the median is approximately equal to the mean), which may lend itself to use as a linear outcome or a dichotomized variable for indicator analysis. Conceptually, this captures a more notable phenomenon of weight suppression, which might be associated with a different clinical picture (which we put to the test below). Using this novel method, values above the mean could represent a status of weight suppressed. Technically, values above one could represent the weight-suppressed group using this definition, but we use the mean for consistency in comparing the two approaches. Many values below the mean would be considered weight suppressed using the TWS approach, warranting clarification on the most clinically relevant classification, potentially explaining inconsistent findings in previous research. Table 1 illustrates the differences in the WS values that would be generated by the TWS versus NWSS methods for eight hypothetical case examples (chosen to compare the two measurement approaches, which are on different scales). Matching colors indicate matching scores for numerous case examples using each WS calculation.

Importantly, the values assigned by TWS do not discern between the person who lost 50% of their body weight (person C) and the person who lost 25% of their body weight (person H). Such differences might be registered by the brain/body differently, where relatively higher percentages may be associated with more ED symptoms, but this needs to be empirically tested by holding the highest weight constant. In contrast, NWSS more appropriately considers the larger context of one's weight history (adult weight range), which makes it a more sensitive indicator likely to capture different clinical characteristics (e.g., assigning a much higher WS value to person C versus person H). The following analysis compares the associations of these two WS approaches in a clinical sample.

Intake data from private nutrition counseling practice

Data collected come from a HIPAA-compliant online intake process (single time point) at a private (cash-based) group nutrition counseling practice in Los Angeles, CA. Clinicians are registered dietitian nutritionists specializing in the nutritional management of eating and substance use disorders, but patients sought counseling for a wide range of reasons. The study was approved by the University of California Los Angeles Institutional Review Board (IRB# 20-008829). Subjects were eligible for enrollment if they consented to participate in the study (20.7% of the potential sample opted out). Because the current study asked about one’s highest and lowest adult weights, analysis was restricted to those ages 21 and above. Study enrollment began in September 2020 and ended in April 2024. Final analysis occurred in April 2024.

Sample characteristics

Demographic characteristics of the study sample are described in Table 2 (separated by gender), with additional columns for those screening positive for EDs, as operationalized by the short version of the Eating Disorder Examination Questionnaire (EDE-QS), and those meeting the criteria for UPFA, as operationalized by the modified Yale Food Addiction Scale version 2.0 (mYFAS 2.0). In the study sample, 177/287 (61.7%) screened positive for an ED (mean EDE-QS score = 16.2, SD = 8.5), 116/287 (40.4%) met the criteria for either moderate or severe UPFA (mean UPFA score = 3.6; SD = 3.5), and 104/272 (36.2%) met the established thresholds for both simultaneously. Of the 116 individuals who met the criteria for UPFA, 10.3% (n = 12) did not screen positive for an ED. Further, of the 171 persons who screened positive for an ED, 42.7% (n = 73) did not meet the criteria for UPFA. 98/287 (34.1%) did not meet thresholds for either ED or UPFA. Thus, the overlap between UPFA and EDs observed in this sample was comparable to prior studies in clinical settings [31]. The only gender difference that emerged was the association between TWS and screening positive for ED and UPFA, which was observed among women but not non-women. However, the NWSS did detect these associations among men. Further, given that 55.7% of the sample reported a BMI of \(\ge\) 25 kg/m2, differences in the associations of NWSS versus TWS with symptoms of disordered eating and UPFA were explored for individuals with BMI < versus \(\ge\) 25 kg/m2. Overall, similar conclusions were derived from each BMI subgroup as with the entire sample, and dichotomizing by BMI resulted in small cell sizes that prevented the ability to conduct all analyses using that approach. Thus, for simplicity and appropriate power, results are reported for the whole sample.

Table 2 Demographic characteristics of study sample by eating disorder and ultra-processed food addiction positive screens (N = 287)Measures

Standard demographic data on age, gender (non-binary collapsed into non-woman due to small cell size), race/ethnicity (non-Hispanic Black, Hispanic/Latin, Asian, Other/Mixed, Prefer Not to Say collapsed into non-White due to small cell sizes), education, parent’s level of education, self-reported current or previous alcohol/substance use disorder (yes/no), and self-reported height and weight, including lowest and highest (excluding pregnancy) adult weights was collected. Two participants reported lifetime highest weights that were less than lifetime lowest weights, and those observations were dropped. There were no missing data.

Initial symptom survey

This screening tool is used by Oxford Biomedical Technologies to assess eligibility for the Lifestyle Eating and Performance (LEAP) Mediator Release Test (https://www.nowleap.com/) by certified LEAP therapists. This screening tool has yet to be formally validated but is used by clinicians to assess the frequency of various somatic symptoms. Eighty symptoms are classified as never/rarely; mild/occasional; mild/frequent; severe/occasional; and severe/frequent. Indicator variables were classified as mild/occasional or less and compared to mild/frequent or more. Three variables: (1) food cravings; (2) binge eating or drinking; and (3) purging (all methods) were selected for analysis because they were hypothesized to correlate with the two WS measures compared in the current study (Table 3). Data generated from this instrument can be considered preliminary, as follow-up studies are needed. Notwithstanding, research using this symptom survey has been published [32].

Table 3 Chi-Squared analysis of associations between indicator variables with each weight suppression calculation, dichotomized at the mean (N = 287)Everyday discrimination scale

The Everyday Discrimination Scale was originally developed to detect experiences of racism encountered by African Americans [33] and has been validated for population health research on racism and health [34]. Questions ask about the frequency of perceived discrimination (from never to almost every day). Expanded versions of the measure ask about the reasons for these experiences. Given the explosion of research on weight stigma in recent years [35], listing weight as a reason for perceived discrimination has become increasingly common. Given that 24.4% of our study sample reported weight as a reason for experiencing discrimination, this variable was hypothesized to correlate with the two WS measures compared to the current study (Table 3).

Ultra-processed food consumption

A food frequency questionnaire (FFQ) developed by the private practice asked about the frequency of various foods consumed (0 = never, 1 = 1–3 times/month, 2 = 1–2 times/week, 3 = 3–4 times/week, 4 = 5–6 times/week, 5 = once/day, 6 = twice/day, 7 = three times/day, and 8 = four or more times/day). The NOVA classification [36] was used to classify ultra-processed foods (UPFs) based on the following: sugar-sweetened beverages, diet beverages, fried food, sweets/desserts, refined grains, alternative dairy products, and protein powders. An index was created combining the frequency of these foods, which created a normally distributed variable that was dichotomized at the mean to compare those with above-average UPF consumption to those below average (see Table 3). While this instrument is not validated, it was included as one minor exploratory variable to detect potential differences between WS measurement methods.

Eating disorder examination question short form (EDE-QS)

The widely used EDE-Q has a validated shortened version, which is a 12-item version with a 4-point response scale that asks about the frequency of behaviors over the last 7 days: (1) 0 days; (2) 1–2 days; (3) 3–5 days; (4) 6–7 days [37]. The total score ranges from 0 to 36, where scores at or above 15 are determined to be likely to have an ED (sensitivity = 0.83, specificity = 0.85) [38], which was used as an indicator variable (Table 3). For the individual item analysis, indicator variables were created for experiencing the symptom 3–7 days and compared to 0–2 days (Table 4).

Table 4 Chi-squared analysis of associations between EDE-QS items (< 2 Vs. 3–7 Days) with each weight suppression calculation, dichotomized at the mean (N = 287)Modified Yale food addiction scale 2.0 (mYFAS 2.0)

The widely used Yale Food Addiction Scale has a validated shorted version, which is a 13-item questionnaire based on DSM-5 criteria for substance use disorder [39]. Two questions indicate clinical significance, and one must be present for a positive screen. Questions have different thresholds for meeting criteria and are then classified as no (0–1 symptoms or does not meet criteria for clinical significance), mild (2–3 symptoms), moderate (4–5 symptoms), or severe food addiction (6+ symptoms). An indicator variable was created for those with moderate/severe food addiction and compared to individuals with none/mild since this is a clinical population with high rates of eating and substance use disorders. For the individual item analysis, indicator variables were created based on the established thresholds for that item (Table 5).

Table 5 Chi-Squared analysis of associations between ultra-processed food addiction symptoms with each weight suppression calculation, dichotomized at the mean (N = 287)