**3. Results**

First, a factorial analysis was performed to extract an interpretable measure of increased and more unhealthy food intake during the COVID-19 pandemic ("dysfunctional COVID-19-specific eating behavior", DCSEB). A parallel analysis, as well as Velicer's minimum average partial (Velicer, 1976), were applied to extract the optimal number of factors. Both analyses convergingly indicated the existence of one factor. Within this one factor (proportion of explained variance = 36%), four items reached standardized factor loadings of above 0.6 (Awang, 2014, Hair, 2008; see Supplemental Material). These items assess whether the individual started to eat larger portions more frequently in an unhealthier fashion, and whether they fell back into old eating patterns. Kaiser–Meyer–Olkin measures of sampling adequacy indicate values of above 0.8 for each item; sum scores were applied to subsequently summarize the scale.

Spearman correlation analyses revealed significant associations between *DCSEB* and *COVID-19-related fear* (*r* = 0.167; *p* = 0.008), *DCSEB* and *generalized anxiety* (*r* = 0.396; *p* < 0.001), and *DCSEB* and depression symptoms (*r* = 0.496; *p* < 0.001). For an overview of all correlation coefficients, see Tables S1 and S2 in the supplementary online material. To explore possible effects of obesity surgery on the psychopathological states and eating behavior, Mann–Whitney U tests were computed to identify differences between groups (*with* and *without* surgery) in each of the psychometric scales mentioned above. These Mann–Whitney U tests revealed no significant differences in the tested variables: *COVID-19-related fear* (*W* = 8288, *p* = 0.739), *dysfunctional safety behavior* (*W =* 8695.5, *p* = 0.305), and *DCSEB* (*W* = 8431.5, *p* = 0.566). However, *p*-values approached significance at α = 0.05 for the comparisons between participants with and without obesity surgery in *generalized anxiety* (*W* = 9180, *p* = 0.064) *and depression symptoms* (*W* = 9186, *p* = 0.057), and participants who underwent obesity surgery exhibited lower levels in each of these dimensions. Table 2 lists the psychometric data for the obesity patients *with* and *without* obesity-specific surgery.

**Table 2.** Psychometric data for the obesity patients with and without an obesity-specific surgery. Mean sum scores and standard deviations (in parentheses) are listed.


Note: Generalized anxiety was measured by GAD-7 (7 items, 4-point Likert scale, cut-off mild = 5, cut-off moderate = 10); depression symptoms were measured by PHQ-8 (8 items, 4-point Likert scale, cut-off ≥ 10), COVID-19-related fear, dysfunctional safety behavior, Dysfunctional COVID-19-specific eating behavior (DCSEB, see Supplementary online Material). Body mass was computed using the formula weight in kg/(height in m)2.

To assess whether obesity surgery moderates the relationship between the abovedescribed psychological dimensions and *DSCEB*, robust regression analyses were conducted for each possible predictor, using *group* (*with* vs. *without* surgery) as a moderator and *DCSEB* as the dependent variable. The strongest interest was to reveal unconditional relationships so that one regression model was computed for each predictor.

This moderator analysis revealed a significant interaction between the predictors *generalized anxiety* and group (with vs. without surgery, *b* = 0.289; *p* = 0.028, see supplemental material for illustration of the marginal effects) on *DCSEB*. The regression coefficient for generalized anxiety turned out significant (*b* = 0.227, *p* = 0.025). No differences occurred in the direct comparison between patients with and without surgery (*b* = −0.003, *p* = 0.983). The regression model accounted for 16.6% of variance. This pattern—and particularly the interaction between group and generalized anxiety—remained robust after conditioning on age, gender, and education. No other significant interaction appeared in these regression models (see supplemental online material). To further illustrate this effect, participants were divided according to common cutoffs for the GAD-7, namely participants who show no anxiety (GAD-7 score below five), people who exhibit mild anxiety (GAD-7 scores from five to nine), and participants who report moderate to severe anxiety (GAD-7 scores from 10 to 21, see [35]). The moderating effect of generalized anxiety on DCSEB before and after surgery is shown in Figure 1. Corroboratory results from a further robust regression analysis that included the categorized GAD-7 values (no anxiety vs. mild anxiety vs. moderate to severe anxiety), the group variable (with vs. without surgery), and their interaction term also indicated that while levels of DCSEB remained unchanged for individuals with surgery compared to individuals without surgery in participants with low and mild anxiety

levels, participants with high anxiety showed even more DCSEB after surgery (interaction term between surgery [reference: without surgery] and GAD-7 [dummy: mild anxiety with reference: no anxiety]: *b* = 0.049, *se* = 0.286, *t* (250) = 0.170, *p* = 0.865; interaction term between surgery [reference: without surgery] and GAD-7 [dummy: moderate and severe anxiety with reference: no anxiety]).

**Figure 1.** Generalized anxiety as a risk factor for increased levels of dysfunctional COVID-19-specific eating after obesity surgery. Group-wise box-plots indicate medians and interquartile ranges (see supplemental material for illustration of non-splitted continuous data). While for participants with no or mild manifestations of generalized anxiety (GAD-17 scores of 0 to 4, or 5 to 9, respectively), no increase in DCSEB is notable, and individuals with moderate to severe levels of anxiety (GAD-7 > 9) show increased DCSEB after obesity surgery. Whiskers extend to the most extreme data point unless there are data 1.5 inter-quartile-ranges away from the first or the third quartile, respectively. Data points beyond that are shown as a black dot.
