Mental health after COVID-19 death-related loss in individuals with eating disorders: A multi-country nested matched case-control study

Abstract

Objective:

The COVID-19 pandemic caused millions of deaths worldwide and significantly impacted people with eating disorders, exacerbating symptoms and limiting access to care. This study examined the association between COVID-19 death-related loss—defined as the death of a family member, friend, or acquaintance due to COVID-19—and mental health among people with pre-existing eating disorders in the United States (US), the Netherlands, and Sweden.

Method:

Participants with a history of eating disorders completed a baseline survey early in the pandemic (US: N = 511; Netherlands: N = 510; Sweden: N = 982) and monthly (US, the Netherlands) or bi-annual (Sweden) follow-ups from April 2020 to May 2021. The surveys assessed pandemic impact on eating disorder-related behaviors and concerns, anxiety, depression, sleep disturbances, and COVID-19-related deaths. A matched nested case-control design was used to compare individuals experiencing a death-related loss with matched controls.

Results:

A substantial proportion experienced a COVID-19 death-related loss (US: 33%; Netherlands: 39%; Sweden: 17%). No significant differences were found between cases and controls on eating disorder, anxiety, depression, or sleep outcomes.

Discussion:

Despite the expected influence of COVID-19 death-related loss on mental health, our study found no significant association. Reactions to pandemics may be highly individual, and practitioners may want to consider broader losses—such as bereavement, missed educational experiences, relationship disruptions, financial instability, and employment challenges—when addressing patients’ needs. Future research should continue to explore how death-related loss impacts eating disorder risk and progression.

Introduction

Early cases of COVID-19 were detected in December 2019 and the virus rapidly evolved into a global pandemic. Highly infectious and capable of causing severe respiratory distress (Fu et al., 2020), COVID-19 caused a marked increase in mortality, leading to a global health crisis (Centraal Bureau voor de Statistiek, 2020; Juul et al., 2022; Woolf et al., 2020; Zhang et al., 2023). The pandemic had significant mental health repercussions, particularly for those with a history of psychiatric disorders (Devoe et al., 2023; Vindegaard & Benros, 2020; Xiong et al., 2020). Restrictions implemented to curb virus spread adversely affected general mental health and exacerbated eating disorder symptoms, while limiting access to specialized care (Cooper et al., 2022; Gao et al., 2022; Ludvigsson, 2023; Pashakhanlou, 2022; Rozanova et al., 2020; Schlegl et al., 2020; Thompson et al., in press). Pandemic-related concerns such as food access, lack of social support, and infection fears, were associated with eating disorder severity during the pandemic’s first year (Thompson et al., 2023).

The World Health Organization (WHO) estimated that COVID-19 caused 7 million deaths worldwide (World Health Organization, 2024). Understanding the impact of death-related loss on vulnerable populations, including individuals with eating disorders, is crucial. No studies have explored how COVID-19 death-related loss affects individuals with eating disorders, and limited studies have examined the role of death-related loss generally in relation to eating disorders (Degortes et al., 2014; Grogan et al., 2020; Lie et al., 2021; Pike et al., 2008; Pike et al., 2006; Reid et al., 2020; Su et al., 2016; Su et al., 2015). Research indicates that unexpected deaths of close relatives can increase the risk of developing eating disorders (Lie et al., 2021; Su et al., 2016). Some studies have linked death-related loss to disordered eating in community and college samples (Cerniglia et al., 2014; Malinauskiene & Malinauskas, 2018; Smyth et al., 2008).

The study aim was to examine the associations between COVID-19 death-related loss and mental health outcomes among individuals with lived experiences of eating disorders in the United States (US), the Netherlands, and Sweden over the first year of the pandemic. Given the number of worldwide excess deaths during the first year of the pandemic, including in these countries (Fig 1), we expected that many participants would report knowing someone who died from COVID-19. We hypothesized that individuals who experienced a COVID-19 death-related loss would have worse eating disorder, anxiety, depression, and sleep disturbances compared with controls.

Fig 1. Monthly cumulative COVID-19 deaths per 100,000 people in the United States, Netherlands, and Sweden from April–Dec 2020.

Data used to create this figure were collected by Centraal Bureau voor de Statistiek (2020); Statistiska Centralbyrån (n.d.); United States Census Bureau (n.d.); World Health Organization (2024).

Methods

We present a matched nested case-control analysis of the association between COVID-death related loss and mental health outcomes in individuals with pre-existing eating disorders. This substudy is part of a larger investigation into the impact of the COVID-19 pandemic on individuals with eating disorders (Birgegård et al., 2022; Termorshuizen et al., 2020). A statistical analysis plan was developed and agreed upon by the authors before analysis. Study reporting follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines and checklist (von Elm et al., 2007) (Supplementary Appendix).

Study Population

Participants were recruited from the US (N = 511), the Netherlands (N = 510), and Sweden (N = 982) early in the COVID-19 pandemic (April-May 2020 in the US and Netherlands; May-July 2020 in Sweden). The main prospective cohort study has generated several substudies, including the present study (Birgegård et al., 2022; Goode et al., 2023; Termorshuizen et al., 2020; Thompson et al., 2023). Eligible participants self-reported a current or past eating disorder, were aged 18+ years in the US and 16+ in the Netherlands and Sweden, and were recruited through various sources: prior research studies (with verified diagnoses and re-contact consent) in Sweden and some US participants (Bulik et al., 2020; Thornton et al., 2018), social media in the US and the Netherlands (e.g., Twitter, Facebook, Instagram), the Dutch Eating Disorder Register (NER), and Proud2BeMe, a Dutch online community that provides support for people with body image or eating disorder issues. Participants self-reported their eating disorder history from a list that included anorexia nervosa (AN), bulimia nervosa, binge-eating disorder, avoidant/restrictive food intake disorder, purging disorder, night-eating syndrome, other specified feeding or eating disorder, and other eating disorders. Individuals selecting ‘don’t know/prefer not to answer’ or ‘I never had an eating disorder’ were ineligible.

Procedures

Online surveys assessed the challenges faced during the COVID-19 pandemic, focusing on eating disorders and mental health. The surveys were developed in English and translated to Dutch and Swedish. US participants were invited to complete a baseline and 12 monthly follow-up surveys (13 waves). The Netherlands participants were invited to complete a baseline and 13 monthly follow-up surveys (14 waves), and Sweden participants were invited to complete baseline, 6-month, and 12-month follow-up surveys (3 waves). The survey schedules reflected the research capacities of each site and their ethical permissions.

Selection of Cases and Controls

Participants self-reported monthly, starting at the second month in the Netherlands and third month in the US, whether they knew anyone personally who had died from COVID-19. Sweden participants were given this question only at 12-month follow-up. Participants could choose multiple responses from: “Yes, family member,” “Yes, friend,” “Yes, acquaintance, neighbor, someone I’m not close to,” and “No, no one in my immediate circle has died.” Participants reporting a COVID-19 death-related loss during the study were defined as cases, while those reporting no loss were controls. This created the unmatched nested case-control samples. In the US, there were 146 cases and 229 unmatched controls, in the Netherlands, 114 cases and 233 unmatched controls, and in Sweden 108 cases and 520 unmatched controls. A participant flow chart is shown in Fig 2.

Fig 2.

Flow Chart of Participants Included in the Study Analyses.

Matching Procedure

Matching was performed to balance potential confounding factors between cases and controls based on self-reported eating disorder status (either current or past at the study baseline), gender identity, and completion of a wave at the case’s corresponding index date. For the US and the Netherlands, where the monthly surveys included the question about COVID-19 death-related losses, the index date for the case was determined as the survey wave in which the first death-related loss was reported. Controls were matched 1:1 without replacement. The matched nested case-control samples included 142 cases and 142 controls in the US, 113 cases and 113 controls in the Netherlands, and 107 cases and 107 controls in Sweden (Fig 2). Matching was successful for over 97% of cases in each cohort.

Variables and Definitions

Exposure

The exposure was self-reported loss to death of a family member, friend, acquaintance, neighbor, or someone personally known to the respondent either due to or possibly due to COVID-19. This was measured monthly from the second wave in the Netherlands and the third wave in the US, and at 12-month follow-up in Sweden.

Covariates/Confounds

Covariates included the baseline measure of each outcome, gender identity, age, COVID-19 lockdown (i.e., shelter-in-place/stay-at-home order, quarantine, or voluntary or mandatory self-isolation), COVID-19 exposure (individually and among family members), lost employment due to the pandemic, current eating disorder treatment, and worry about COVID-19 infection. Four survey questions were averaged to measure worry about COVID-19 infection (Cronbach’s alphas at baseline were .81, .78, and .72 for the US, Sweden, and the Netherlands respectively). Gender and age were fixed covariates measured at baseline. In the US and the Netherlands, other covariates were time-varying and collected monthly.

Outcomes

Pandemic-Related Impact on Eating Disorder Illness Behaviors.

The COVID-19 pandemic’s impact on eating disorder behaviors in the previous two weeks was assessed using four Likert items that inquired whether participants had (1) “binged on food that I (or my family) have stockpiled” (binge eating) (2) “restricted my intake more because of COVID-19 related factors” (restriction) (3) “engaged in more compensatory behaviors (e.g., self-induced vomiting, excessive exercise, misuse of laxatives and/or water pills) because of COVID-19 related factors” (compensatory behaviors) and (4) “felt anxious about not being able to exercise” (anxiety about exercise). Participants completed the measures at each wave, except anxiety about exercise at baseline in the Netherlands, due to an oversight in translating this item into the Dutch questionnaire during the rapid assembly of the study at the pandemic’s onset. The response options ranged from 1 = not at all to 4 = daily or more. Each variable was analyzed separately due to low Cronbach’s α’s for the combined items at baseline: US = .60, the Netherlands = .69, and Sweden = .61.

Pandemic-Related Eating Disorder Concerns.

Seven items measured pandemic-related eating disorder concerns over the past two weeks on a scale of 1 = not at all concerned to 4 = very concerned. Participants rated (1) “having access to enough food” (2) “having access to food consistent with current meal plan/style of eating” (3) “worsening of one’s eating disorder due to lack of structure” (4) “worsening of one’s eating disorder due to lack of social support” (5) “worsening of one’s eating disorder due to increased time living in a triggering environment” (6) “not being able to afford food needed for recovery due to loss of income related to COVID-19” and (7) “not being able to afford eating disorder treatment due to loss of income related to COVID-19.” Participants completed these at each wave.

Anxiety Symptoms.

Anxiety symptoms over the past two weeks were measured using the Generalized Anxiety Disorder-7 Screener (GAD-7) (Spitzer et al., 2006). Total scores range from 0 to 21 and higher scores indicate higher severity. Participants completed the measure at each wave. Cronbach’s α at baseline was good to excellent (US = .91, Netherlands = .88, Sweden = .91).

Depression Symptoms.

Depression symptoms during the pandemic were assessed at one-year follow-up in all countries using the question, “Do you think you have felt more down, depressed, or hopeless since the pandemic began in early spring of 2020?”. Response options included “No - my mood has stayed about the same”, “No - I have felt less down, depressed, or hopeless since the pandemic began,” and “Yes - I have felt more down, depressed, or hopeless since the pandemic began.” Responses were recoded into binary categories: “0 = no worsened depression” and “1 = worsened depression.”

Sleep Disturbances.

Pandemic-related sleep disturbances were assessed at the final follow-up in all countries with the question, “Over the past year, have you had problems with sleep? Please select all that apply.” Response options included: “My sleep patterns have not changed much”, “I have had a period of insomnia, either not being able to fall asleep or not being able to stay asleep”, “I have periods when I was sleeping too much”, and “I have had anxiety dreams related to the pandemic.” Responses were recoded to a binary variable: “0 = no sleep disturbances” for those who selected “My sleep patterns have not changed much”, and “1 = sleep disturbances” for all other non-missing responses.

Statistical Analysis

Sample Size

The main study needed to become operational rapidly to investigate the early impact of the pandemic and monitor changes over time. Recruitment was restricted to 1–2 months, determining the sample sizes, which were not based on a priori hypotheses or power calculations.

Primary Analysis

The prevalence of death-related loss was calculated, and associations between COVID-19 death-related loss and study outcomes were tested in the matched nested case-control sample. Longitudinal outcomes were analyzed with multilevel modeling (MLM) to accommodate missing data (Kwok et al., 2008). A sequential model-building process was followed beginning with an unconditional means model (null model). Each outcome was initially modeled with participant and matched pair random effects (three-level) and compared to a model with only the participant random effect (two-level). Model selection was based on intraclass correlation coefficients (ICC) ≥ 0.10. Linear and quadratic unconditional growth models assessed changes over time. Time was defined as months since the index and was entered as a fixed and random effect. A quadratic model was retained if it reduced the Akaike information criterion (AIC) by two units or more; otherwise a linear model was retained (Burnham & Anderson, 2004). Next, a minimally adjusted conditional growth model was tested, adding case/control status as a predictor and the baseline level of the outcome as a covariate. Finally, a fully adjusted model incorporating additional covariates selected through a forward stepwise inclusion and backwards elimination approach (α < .20 for inclusion, < .05 for elimination) was tested. This method helps prevent overloading the model with predictors, which is advised for smaller sample sizes (Nezlek, 2008).

Cross-sectional outcomes were analyzed with analyses of covariance (ANCOVAs) and logistic regression. For Sweden, where COVID-19 death-related loss was only measured at the last wave, ANCOVA was used. For the depression and sleep outcomes measured at the last wave in each country, logistic regression was employed. Minimally and fully adjusted models were generated using the same covariate entry procedures as for MLM.

Sensitivity Analyses

Sensitivity analyses explored the robustness of the findings. First, the primary analysis was repeated on participants who reported a current eating disorder at baseline. Secondly, ANCOVAs compared cases and controls on outcomes at the index time point only, to provide insight into potential differences during the period closest to death-related loss, which could reflect the most severe responses. Thirdly, we performed the MLM analysis in the unmatched case-control sample.

General Analysis Procedures

We interpreted estimates of the association between COVID-19 death-related loss and outcomes using an alpha level of 0.05, due to the study’s exploratory nature. Interpretation prioritized the fully adjusted model for its relevance for testing the study hypotheses. Data from each country were analyzed separately, owing to data sharing restrictions. Information on software and missing data are provided in the Supplement.

Results

Prevalence of Death-Related Loss

The prevalence of COVID-19 death-related loss over the course of the study was 38.9% (n = 146/375) in the US, 32.8% (n = 114/347) in the Netherlands, and 17.2% (n = 108/628) in Sweden (Fig 3). In both the US and the Netherlands, death-related losses were most frequently reported early in the study (see Figure 3 note for further detail).

Fig 3. Sample Prevalence of COVID-19 Death-Related Loss in the US, the Netherlands, and Sweden Cohorts.

Percentages are based on available data and exclude participants who dropped out of the study before the initial measurement of death-related loss. The ‘other person’ category includes acquaintances, neighbors, or other individuals personally known. The ‘any’ category percentage may be less than the combined total of the other categories, as participants may have experienced death-related losses in multiple categories. The following information describes the pattern of death-related loss reporting over time, specifically among participants who reported experiencing a death-related loss (i.e., 100% reflects the total percentage of death-related losses reported by those affected). In the Netherlands, deaths were most frequently reported in the first survey that included this question (36%) and the following month (10%). Following this initial peak, the percentage was lower in subsequent surveys (ranging from 1% to 8%) until the final survey, where 23% were reported. This rise is likely due to targeted efforts encouraging participants to complete the final survey, which re-engaged individuals who had not been actively participating and included those who had experienced death-related losses between their last completed survey and the final one. In the US, the percentage of reported losses was highest in the first survey that included this question (37%) and in the following month (13%), with lower percentages in later surveys (3% to 10%), including the final survey (8%).

Participant Characteristics

Table 1 presents the sample characteristics for the matched nested case-control participants in the primary analysis. Participants were mainly women in their twenties to thirties and many had a history of AN. In the US, most participants identified as White (93%) and non-Hispanic (94%) (see Supplement). In the US and the Netherlands, the majority were acutely ill and receiving treatment. These characteristics reflect those of the overall cohorts (Birgegård et al., 2022; Termorshuizen et al., 2020).

Table 1.

Participant Characteristics at Baseline and at the Index Time Point

	US			Netherlands			Sweden
	Overall	Cases	Controls	Overall	Cases	Controls	Overall	Cases	Controls
	N = 284	N = 142	N = 142	N = 226	N = 113	N = 113	N = 214	N = 107	N = 107
Baseline
Eating disorder diagnosisa
Anorexia nervosa	186 (65)	92 (65)	94 (66)	151 (67)	78 (69)	73 (65)	157 (73)	77 (72)	80 (75)
Bulimia nervosa	99 (35)	53 (37)	46 (32)	55 (24)	21 (19)	34 (30)	75 (35)	40 (37)	35 (33)
Binge-eating disorder	80 (28)	39 (27)	41 (29)	25 (11)	13 (12)	12 (11)	39 (18)	22 (21)	17 (16)
Other	134 (47)	69 (49)	65 (46)	92 (41)	66 (58)	68 (60)	103 (48)	52 (49)	51 (48)
Eating disorder statusb, c
Current eating disorder	136 (48)	68 (48)	68 (48)	136 (60)	68 (60)	68 (60)	48 (22)	24 (22)	24 (22)
Past eating disorder	148 (52)	74 (52)	68 (48)	90 (40)	45 (40)	45 (40)	166 (78)	83 (78)	83 (78)
In eating disorder treatment	172 (61)	82 (58)	90 (63)	121 (54)	68 (60)	68 (60)	32 (15)	14 (13)	18 (17)
Gender identityd
Woman	282 (99)	141 (99)	141 (99)	220 (97)	110 (97)	110 (97)	212 (99)	106 (99)	106 (99)
Man	<5 (<5)	<5 (<5)	<5 (<5)	6 (3)	<5 (<5)	<5 (<5)	<5 (<5)	<5 (<5)	<5 (<5)
Gender diverse	0	0	0	0	0	0	0	0	0
Sex assigned at birth
Female	280 (99)	140 (99)	140 (99)	223 (99)	111 (98)	112 (99)	212 (99)	106 (99)	106 (99)
Male	<5 (5)	<5 (<5)	<5 (<5)	<5 (<5)	<5 (<5)	<5 (<5)	<5 (<5)	<5 (<5)	<5 (<5)
Age (years)c,e	32.12 (10.03)	33.67 (10.33)	30.59 (9.50)	-	-	-	-	-	-
16–21	24 (8)	9 (6)	15 (11)	46 (20)	23 (20)	23 (20)	<5 (<5)	<5 (<5)	<5 (<5)
22–29	118 (42)	50 (35)	68 (48)	104 (46)	52 (46)	52 (46)	45 (21)	35 (33)	10 (9)
30–39	81 (29)	46 (33)	35 (25)	49 (22)	27 (24)	22 (19)	121 (57)	34 (32)	87 (81)
40–49	39 (14)	24 (17)	15 (11)	12 (5)	5 (4)	7 (6)	39 (18)	30 (28)	9 (8)
≥ 50	21 (7)	12 (8)	7 (5)	15 (7)	6 (5)	9 (9)	6 (3)	6 (6)	0
Index time point c
In lockdown	92 (36)	51 (39)	41 (33)	90 (49)	47 (42)	43 (38)	107 (50)	52 (49)	55 (51)
COVID-19 exposure	71 (25)	45 (32)	26 (19)	52 (23)	12 (11)	40 (35)	85 (40)	48 (45)	37 (35)
Lost job due to pandemic	63 (22)	27 (19)	36 (25)	16 (9)	9 (10)	7 (8)	14 (7)	7 (7)	7 (7)
In current eating disorder treatment	174 (62)	82 (58)	92 (66)	117 (52)	54 (48)	63 (56)	29 (14)	15 (14)	14 (13)
Worry about COVID-19f	4.24 (1.29)	4.37 (1.25)	4.12 (1.32)	3.95 (1.16)	4.01 (1.15)	3.89 (1.17)	3.60 (1.23)	3.76 (1.35)	3.44 (1.11)

Note. The index time point signifies the initial report of a COVID-19 death-related loss in cases throughout the 12 months of the study, and it also corresponds to the same survey wave among control participants within matched case-control pairs. In each matched pair, the index time point can correspond to wave 3 to wave 12 for US participants and wave 2 to wave 13 for Dutch participants. The variables described at the index time point represent potential confounding variables, which were measured monthly. The table summarizes their values at the index time point, denoted as Time = 0 in the longitudinal analyses.

^aPercentages could sum to more than 100 as some participants reported having histories of more than one type of eating disorder. The ‘other’ category included diagnoses of avoidant/restrictive food intake disorder, atypical anorexia nervosa, purging disorder, night-eating syndrome, other specified feeding or eating disorder, eating disorder not otherwise specified, or other eating disorder.

^bThe distinction between current and past eating disorders was based on a participant’s response to the question, ‘Which of the following best describes your experience?’. Endorsing ‘I currently have an eating disorder’ represented a current eating disorder, and endorsing ‘I had an eating disorder in the past and have no current symptoms’ or ‘I had an eating disorder in the past and still have some lingering symptoms’ corresponded to a past eating disorder.

^cBased on available data.

^dThe ‘gender diverse’ category included non-binary, gender diverse, and two-spirit.

^eAge was measured by age bands in the Netherlands and Sweden datasets.

^fLikert scale score range of 1 to 7 where 1 = Not at all worried, 4 = Somewhat worried, and 7 = Very worried.

At the index time point, defined as the wave when a case first reported a COVID-19 death-related loss or the matched person-moment among controls, many participants reported being in eating disorder treatment, being in lockdown, recent COVID-19 exposure, job loss due to the pandemic, and being “somewhat worried” about contracting COVID-19. After the index point, US participants completed an average of 2.6 (SD = 2.8) follow-ups, whereas participants in the Netherlands completed an average of 3.7 (SD = 3.9). At the one-year follow-up, 47% and 66% of the US and Netherlands samples completed the final survey.

Course of Mental Health and Wellbeing Over Time

Some outcomes showed average declines from the index point, as indicated by statistically significant, negative linear time fixed effects (US: 5/12 outcomes, Netherlands: 6/12 outcomes). For example, restriction symptoms in the US and Netherlands decreased slightly after the index point (Supplementary Table 1). Some outcomes showed stable levels from the index point (US: 7/12; Netherlands: 4/12). Others showed average U-shaped trajectories, characterized by an initial decline followed by an increase, indicated by quadratic time effects (US: 0/12; Netherlands: 1/12). For most outcomes, random intercepts and slopes were significant, indicating significant variability among participants’ outcome scores at the index time point and rate of change to the one-year follow-up.

At the final wave, many participants reported worsened depression symptoms since the pandemic began: 74% in the US (99/133), 62% in the Netherlands (92/149), and 44% in Sweden (94/214). Additionally, 82% in the US (109/133), 70% in the Netherlands (105/149), and 63% in Sweden (134/214) reported pandemic-related sleep disturbances, including insomnia (US: 67%, 89/133; Netherlands: 65%, 99/149; Sweden: 57%, 123/214), excessive sleep (US: 34%, 45/133; Netherlands: 20%, 30/149; Sweden: 13%, 27/214), and anxious dreams about the pandemic (US: 23%, 30/133; Netherlands: 3%, 5/149; Sweden: 37%, 80/214).

Case-Control Differences in Mental Health and Wellbeing Over Time

The comparisons between cases and controls on study outcomes are shown in Fig 4 and Table 2, with complete results available in Supplementary Table 1. In the US, the Netherlands, and Sweden, cases did not have significantly worse outcomes on any outcome measured.

Fig 4.

Case-Control Differences in Mental Health and Wellbeing Over Time After Cases Experienced a COVID-19 Death-Related Loss.

Table 2.

Associations between COVID-19 death-related loss and mental health and wellbeing outcomes in individuals with eating disorders: Primary analyses in the US (N = 284), the Netherlands (N = 226), and Sweden (N = 214) nested matched case-control samples.

Outcomes	US				Netherlands				Sweden
	Baseline-adjusted		Fully adjusted		Baseline-adjusted		Fully adjusted		Baseline-adjusted				Fully adjusted
	Est. (s.e.)	p	Est. (s.e.)	p	Est. (s.e.)	p	Est. (s.e.)	p	F	ηp2	p	MD	F	ηp2	p	MD
Longitudinal in US and the Netherlands (cross-sectional in Sweden a )
Binge eating	<.01 (0.06)	.95	<.01 (0.06)	.99	−.12 (0.08)	.13	−.13 (0.08)	.09	0.39	<.01	.53	0.05	0.11	<.01	.74	0.03
Restriction	.08 (0.08)	.37	.04 (0.08)	.63	.09 (0.09)	.30	.05 (0.08)	.52	0.08	<.01	.78	−0.03	0.41	<.01	.53	−0.07
Compensatory behaviors	.09 (0.09)	.30	.06 (0.09)	.50	.08 (0.09)	.41	.02 (0.09)	.83	0.16	<.01	.69	0.04	0.03	<.01	.86	0.02
Exercise anxiety	.10 (0.10)	.28	.07 (0.10)	.75	.06 (0.12)	.64	.03 (0.12)	.81	0.29	<.01	.59	0.06	0.30	<.01	.86	0.02
Food access	.11 (0.07)	.12	.06 (0.06)	.19	0.07 (0.07)	.34	.03 (0.07)	.64	0.001	<.01	.98	−0.002	0.40	<.01	.53	−0.04
Meal plan	.08 (0.08)	.27	.03 (0.08)	.54	−.08 (0.09)	.35	−.10 (0.08)	.24	0.95	<.01	.33	0.09	0.21	<.01	.65	0.04
Lack of structure	.11 (0.10)	.25	.06 (0.09)	.56	.02 (0.11)	.86	<.01 (0.10)	.96	0.61	<.01	.44	−0.09	1.43	<.01	.23	−0.13
Lack of support	.01 (0.09)	.91	−.04 (0.09)	.67	.01 (0.10)	.94	−.03 (0.09)	.77	4.72	.02	.03	−0.23	8.42	.04	<.01	−0.29
Triggering environment	.12 (0.11)	.26	.05 (0.11)	.39	<.01 (0.10)	.98	−.02 (0.10)	.81	0.04	<.01	.84	0.02	0.004	<.01	.95	−0.01
Food costs	−.04 (0.08)	.59	−.07 (0.07)	.63	−.08 (0.06)	.22	−.08 (0.07)	.20	0.02	<.01	.88	0.01	0.01	<.01	.92	−0.01
Treatment costs	−.06 (0.07)	.37	−.08 (0.07)	.26	−.09 (0.06)	.11	−.10 (0.06)	.07	0.11	<.01	.74	0.04	0.03	<.01	.85	0.01
General anxiety symptoms	.90 (0.51)	.07	.58 (0.49)	.22	−.74 (0.49)	.14	−.90 (0.47)	.06	2.25	.01	.14	−0.87	2.36	.01	.13	−0.80
	Unadjusted		Fully adjusted		Unadjusted		Fully adjusted		Unadjusted				Fully adjusted
	OR (95% CI)	P	OR (95% CI)	p	OR (95% CI)	p	OR (95% CI)	p	OR (95% CI)			p	OR (95% CI)			p
Cross-sectional, final wave
Depression symptoms	0.91 (0.41, 2.03)	.82	0.83 (0.37, 1.88)	.66	1.02 (0.52, 1.98)	.96	0.84 (0.42, 1.69)	.63	1.26 (0.73, 2.16)			.41	1.16 (0.63, 2.14)			.62
Sleep disturbances	0.89 (0.36, 2.20)	.80	0.89 (0.36, 2.20)	.80	1.11 (0.55, 2.26)	.77	1.11 (0.55, 2.26)	.77	1.49 (0.86, 2.61)			.16	1.29 (0.71, 2.34)			.41

Note. The fixed effect of the case/control predictor is shown. Positive Est.’s and MDs, and ORs > 1 indicate that exposure to COVID-19 death-related loss is associated with higher scores and increased odds for the outcome. The full model results are in Supplementary Table 1. Fully adjusted models consider an additional set of covariates including gender identity, age, lockdown restrictions, COVID-19 exposure, lost employment due to the pandemic, current eating disorder treatment, and worry about COVID-19 infection.

^aDeath-related loss was only measured at the end of the study in Sweden. CI = confidence interval, Est. = estimate, MD = mean difference, s.e. = standard error, US = United States, η_p² = partial eta squared.

Sensitivity Analyses

The various sensitivity analyses supported the primary findings (Supplementary Table 2) and showed that cases did not have significantly worse outcomes than controls in 39/40 US analyses, 40/40 Netherlands, and 27/28 Sweden analyses.

Discussion

The purpose of this study was to investigate the impact of COVID-19 death-related loss on mental health during the pandemic’s first year among individuals with a lifetime history of eating disorders. Losing someone to COVID-19 was not associated with a worse impact on eating disorder symptoms, pandemic-related eating disorder concerns, anxiety symptoms, depression symptoms, or sleep disturbances in the US, the Netherlands, or Sweden cohorts.

The limited association between COVID-19 death-related loss and mental health outcomes contrasts with other research that has reported increased symptoms associated with COVID-19 death-related losses, particularly among those with pre-existing psychiatric disorders (Joaquim et al., 2021). Several factors may explain this discrepancy. The measure of death-related loss in our study may not fully capture bereavement and grief, which consider the perceived closeness of the relationship and the emotional response, which are important for predicting clinical impairment (Lee & Neimeyer, 2022). Even if we restricted our measure to losses of family and friends, excluding acquaintances, it still might not fully reflect the magnitude of personal loss. For example, losing a coworker or a neighbor whom one interacts with daily might have a greater emotional impact than the loss of a distant relative or friend seen only occasionally or not closely involved in one’s life. In contrast, non-COVID-19 studies using similar measures, such as death-related loss based on death register data, have detected significant impacts on eating disorder course, including an increased risk for onset (Lie et al., 2021; Su et al., 2016).

Another possible interpretation of our findings is that severe pre-existing mental health symptoms and early pandemic stress may have created a ceiling effect, limiting additional worsening from further adverse experiences (Wright et al., 2021). Additionally, many participants in the US and the Netherlands were engaged in eating disorder treatment, which might have mitigated the impact of loss. Social support is essential for coping with grief (Burke & Neimeyer, 2013). Those in treatment, those who have previously received psychotherapy, or those who have managed chronic medical conditions, such as eating disorders, may display more resiliency to death-related losses because of learned positive coping skills.

Individual differences in response to loss may also explain the unexpected results. Some individuals respond to death-related losses resiliently, experiencing minimal emotional disruption, while others experience acute and prolonged grief and impairment (Burke & Neimeyer, 2013). Established risk factors for clinical impairment from loss include being female, younger, unexpectedness of the death, and low social support (Burke & Neimeyer, 2013). Coping strategies, social resources, and positive emotions enhance resilience to loss (Mancini & Bonanno, 2009). Further, bereavement may be delayed, with effects possibly emerging after the study period. Anticipatory grief due to the scale of COVID-19-related deaths or the characteristics of the deceased (e.g., older age, pre-existing health conditions) might also have reduced the impact of loss (Kumar, 2023). In our study, the timing of the first COVID-19-related death was designated as the index point for participants who faced multiple losses. However, we recognize that bereavement responses may vary depending on the nature of each loss, as individuals may react more intensely or in a more complex manner to certain losses than others, and there may be individual differences in response to multiple losses.

Clinical Implications and Future Directions

The lack of an association between COVID death-related loss and mental health and wellbeing highlights the need for individualized assessments of how such losses affect individuals with eating disorders. Clinicians should consider various factors, including the nature and timing of the loss, the individual’s baseline severity, the quality of the relationship with the deceased, and the individual’s social network when assessing the potential impact on mental health (Burke & Neimeyer, 2013).

The high proportion of people with eating disorders reporting a COVID-19 death-related loss underscores the importance of exploring the pandemic’s multifaceted impacts. Practitioners might consider broadly inquiring about other pandemic-related experiences and general losses, such as missed life experiences, relationship disruptions, and financial and career impacts (Kumar, 2023; Sirrine et al., 2023).

Researching bereavement’s role in eating disorders presents challenges (Reid et al., 2020). Periods of heightened mortality during epidemics and pandemics provide opportunities to investigate how bereavement impacts eating disorder symptoms. Employing more detailed measures to capture bereavement, grief, and risk factors for clinical impairment from death-related losses (Burke & Neimeyer, 2013) would facilitate a systematic investigation of how these losses influence the mental health trajectories of individuals with or at risk of eating disorders.

Limitations and Strengths

This study has several limitations. We utilized self-reported eating disorder histories instead of validated measures or diagnostic interviews. However, our recruitment approach included contacting participants with verified DSM-IV or DSM-5 eating disorder diagnoses from prior studies and a clinical eating disorder register, partially addressing this limitation. Given the urgency of data collection, we judged that expecting gold standard clinical interviews would compromise willingness to participate and feasibility, as research staff were seconded to COVID-related duties. Additionally, our sample size was relatively small given the nature of our research questions. Furthermore, our model selection process used stepwise procedures for covariate selection, which may introduce post-selection bias, but was necessary to avoid overfitting and non-convergence from including excessive predictors. The concordance of results from the minimally adjusted models supports the validity of our conclusions.

Another limitation is that the cohorts and/or countries were homogenous concerning gender, race and ethnicity, limiting generalizability. Furthermore, the countries varied in their pandemic response strategies (Ludvigsson, 2023). We included covariates such as lockdown and job loss to address potential confounds. Additionally, dropout occurred due to the longitudinal design.

As with any large-scale crisis, our intentions were to capture data as close to its onset as possible by quickly assembling and submitting materials to ethics committees. Given the unprecedented nature of the global pandemic, we developed a consensus questionnaire to broadly assess various pandemic experiences instead of using extensive validated measures for each domain. This focused approach aimed to limit participant burden and increase study retention. Additionally, without pre-pandemic data on validated measures, interpreting the impact of the pandemic on symptoms could be challenging in the absence of a baseline. To ensure that our questions captured pandemic-specific responses rather than general symptom levels, we included wording about pandemic impact; however, this choice may have limited our understanding of the outcomes.

A key strength of the study is its exploration of the previously unexplored impact of death-related loss on individuals with pre-existing eating disorders (Grogan et al., 2020). The prospective design avoids retrospective recall bias, such as the tendency for distressed individuals to emphasize past negative experiences (Reid et al., 2020). We acknowledge that differences in the number of survey waves across participating countries might complicate the interpretation of results, particularly when comparing outcomes across sites. However, we did not perform statistical comparisons across sites, and more importantly, we believe the multi-country design enhances the richness and generalizability of our findings by capturing a broader range of pandemic experiences.

Conclusion

This study provides valuable insights into the impact of COVID-19 death-related losses on individuals with pre-existing eating disorders. Contrary to expectations, these losses did not significantly worsen eating disorder symptoms or mental health and wellbeing outcomes. This finding emphasizes the need for individualized assessment of how death-related losses impact mental health and underscores the value of measures that capture bereavement and grief to better understand their complex effects.

Summary.

• Pandemics bring widespread losses across several domains. While many individuals with eating disorders faced COVID-19 death-related losses, these results showed minimal impact on mental health outcomes.

• Bereavement may be complex and extend beyond death-related loss. Responses to loss may vary individually, and clinicians should consider a range of factors, such as the nature of the loss and potential risks for bereavement-related clinical impairment, when assessing patient’s needs.

Data Availability Statement

The datasets analyzed are available from the corresponding author on reasonable request.