Diet and lifestyle behaviour disruption related to the pandemic was varied and bidirectional among US and UK adults participating in the ZOE COVID Study

Mandatory public health initiatives to control COVID-19 led to dramatic changes to daily routines, social isolation, economic insecurity, and altered food environments. Prior reports suggested potential exacerbation of unfavourable behaviours (more sedentary time, increased snacking, reduced fresh food intake) and weight gain, but findings were inconclusive. Major life events are known to influence alcohol intake, sleep, diet, and physical activity, and these behaviours are interrelated with adult weight gain and associated morbidities. The study’s purpose was to quantify how diet and lifestyle behaviours changed during the pandemic, identify which populations experienced the greatest disruption, and examine how these changes related to body weight. Using the ZOE COVID Study app, the authors conducted a retrospective longitudinal observational cohort analysis of large UK and US samples, developed a composite disruption index, and investigated associations between changes in behaviours and weight, as well as how pre-pandemic behaviour patterns predicted peri-pandemic changes.

Small European studies reported worsened diet and lifestyle behaviours during lockdowns, including increased sedentary time, more snacking, less fresh food intake, and weight gain, though evidence was mixed. Literature indicates that major life events can shift alcohol consumption, sleep, diet, and physical activity; these behaviours influence weight trajectories and chronic disease risk. Excess body weight is tied to greater COVID-19 severity and intersects with social determinants of health and food security. Prior cohort and meta-analytic evidence support links between sleep duration, fruit and vegetable intake, alcohol consumption, physical activity, snacking frequency, and cardiometabolic outcomes. Collectively, existing studies motivated investigation of pandemic-related behaviour changes at scale and their health implications, while recognizing gaps in general population evidence.

Design: Retrospective longitudinal observational cohort study embedded in the ZOE COVID Symptom Study mobile application (free on iOS/Android). Recruitment and timing: UK data collection 31 July–25 September 2020; US data collection 25 September–30 November 2020. Participants: After exclusions (replicates, age <18, pregnancy, non-UK/US, implausible anthropometrics), n=896,286 provided peri-pandemic data and n=291,871 provided both pre- and peri-pandemic data. UK participants were assigned an Index of Multiple Deprivation (IMD) by geographic area. Cohorts: UK discovery (n=380,847), UK validation (n=448,321; modified app flow), US replication (n=67,118). Measures: Retrospective questionnaires assessed two time points: (1) pre-pandemic (February 2020) and (2) peri-pandemic (the prior month). Instruments included the validated Leeds Short Form Food Frequency Questionnaire (LSF-FFQ) and additional items (fast food; eggs; refined carbs; fermented foods). Variables included diet quality score (DQS), food group frequencies, number of main meals and snacks, sleep (weekday/weekend duration), physical activity, alcohol (frequency and units), food access, and self-reported height and weight (for BMI). Disruption Index (DI): Composite of five domains—DQS, alcohol frequency, physical activity, snacking frequency, and weekday sleep duration. One point was assigned for any change (increase or decrease) per domain; range 0–5 (bidirectional, direction-agnostic). Analytic approach: Descriptive statistics summarized pre- and peri-pandemic values, changes, and proportions increasing/decreasing each variable. Structural equation modelling (SEM) examined associations between changes in behaviours (sleep, physical activity, DQS, snacking, alcohol frequency/quantity) and body weight change (overall, increase-only, decrease-only), adjusting for age and stratifying by deprivation (low IMD 8–10 vs high IMD 1–3) and by baseline BMI category (normal vs overweight/obese). Factor analysis (varimax rotation) identified pre-pandemic diet/lifestyle patterns (“healthier” vs “less healthy”); quartiles were compared for demographic characteristics and peri-pandemic changes. Geospatial visualization mapped DI and weight change across UK regions. Sensitivity analyses compared discovery vs validation cohorts and handled app-flow differences. Ethics: Approved by Partners Human Research Committee (2020P000909) and King’s College London (REMAS ID 18210; LRS-19/20-18210). App registered at ClinicalTrials.gov (NCT04331509).

- Disruption Index (DI): In the UK discovery cohort with pre- and peri-data (n=201,301), most experienced moderate disruption (65% with DI≥2); 15% had high disruption (DI≥4). DI was higher among younger participants, females, and those in more deprived areas (all P<0.001), with similar patterns in the US cohort (n=14,473). - Body weight change: Mean population change was small, UK discovery mean (10th, 90th) −0.2 (−4.4, 3.6) kg. However, individual variability was large: 33% lost a mean 4.4 (−8.6, −0.9) kg; 34% gained a mean 3.7 (0.9, 6.4) kg. Similar patterns in UK validation and US replication cohorts. - DI and weight variability: Greater disruption was associated with larger, more variable bidirectional weight change. In the high DI group, mean loss and gain were −5.5 (−11.0, −1.0) kg and 4.2 (1.0, 8.0) kg, respectively; in the low DI group, −3.5 (−6.4, −0.9) kg and 3.3 (0.5, 5.4) kg (all P<0.001). Associations were stronger in less deprived areas after adjusting for age/sex, though deprivation modestly exacerbated weight gain. - Behaviour–weight associations (SEM): Among weight gainers (n=68,607), reductions in physical activity, diet quality and sleep, and increased snacking were moderately associated with weight gain. Among weight losers (n=65,327), the opposite associations were observed, with reduced alcohol frequency additionally associated with weight loss. These patterns held across deprivation strata, except sleep showed no association in more deprived areas. - Baseline BMI stratification: Reductions in physical activity and DQS and increased snacking were more strongly associated with weight gain in overweight/obese vs normal-weight individuals (e.g., β for physical activity −0.055 vs −0.014; DQS −0.044 vs −0.019; snacking 0.050 vs 0.028; all P<0.05). - Individual-level behaviour changes: Population means changed minimally, but many individuals changed substantially. In the UK discovery cohort: DQS mean change 0.2 (−2.0, 2.0); 37% increased DQS (mean +1.7) and 26% decreased (mean −1.6). Fruit/vegetable portions: 31% increased daily portions (mean +1.85) vs 22% decreased (mean −1.68). Snacks/day: minimal mean change (−0.1), but 16% increased (~+1.4/day) and 23% decreased (~−1.4/day). Weekday sleep: 15% increased vs 9% decreased (∼±1.2 h). Alcohol: frequency increases (18.2%) exceeded decreases (11.2%), but units per occasion decreased more often than increased (13.4% vs 10.6%), suggesting little change in total alcohol consumed. Food access: 92.8% no change; decreases (5.8%) exceeded increases (1.4%). Physical activity: UK had similar proportions increasing and decreasing; US saw more decreases (35.4%) than increases (24.6%). - Pre-pandemic patterns and change: Factor analysis identified “healthier” and “less healthy” pre-pandemic patterns. Those most adherent to the “less healthy” pattern showed larger improvements during the pandemic: greater DQS increases, more fruit/vegetable intake, fewer snacks, and more weight loss (UK: ~1.1 kg more loss vs “healthier” adherents). Benefits were observed irrespective of deprivation, age, sex, or country, though somewhat attenuated by deprivation. - Public health implications: Diet quality and physical activity emerged as key targets to prevent weight gain across socio-economic strata during pandemic conditions.

The pandemic functioned as a large-scale natural experiment producing heterogeneous, bidirectional changes in health behaviours and weight. While average population-level changes were minimal, stratified analyses revealed substantial inter-individual variability. Disruption disproportionately affected younger people, women, and residents of more deprived areas, aligning with known social and economic burdens, yet higher disruption also coincided with greater weight loss for some, indicating that disruption precipitated positive change in a subset. SEM highlighted physical activity and diet quality as central drivers of weight change, consistent across deprivation levels, underscoring their relevance for public health interventions. Notably, individuals with less healthy pre-pandemic behaviours were more likely to improve diet quality and lose weight regardless of deprivation, suggesting that the pandemic may have offered an opportunity for health behaviour improvements among those with the greatest scope to change. These findings temper narratives of uniformly negative behavioural impacts and emphasize the need for targeted support where vulnerabilities to weight gain persist (e.g., more deprived areas), while broadly promoting diet quality and physical activity.

This large UK–US app-based cohort study quantified bidirectional disruptions in diet and lifestyle during the COVID-19 pandemic and linked them to weight change, revealing substantial inter-individual variation despite minimal mean population change. Disruption was greater among younger individuals, women, and those in more deprived areas, and was associated with more variable weight change. Physical activity and diet quality emerged as key modifiable determinants of weight gain and loss across socio-economic strata. Individuals with less healthy pre-pandemic patterns tended to improve diet quality and lose weight irrespective of deprivation. Public policies should continue to prioritize diet quality and physical activity for all, with tailored support for younger women and economically deprived populations. Future research should identify higher-level contextual drivers (e.g., work-from-home, commuting, food environments, structural inequalities), assess sustainability of behaviour changes, and examine long-term metabolic outcomes associated with observed weight changes.

- Self-reported, retrospective data are subject to recall bias and potential misclassification. - App data collection methods changed mid-study (two UK app flows), potentially introducing measurement differences, though sensitivity analyses suggested minimal impact on population characteristics. - The Disruption Index is a crude, direction-agnostic composite of five behaviours; it does not capture the direction or magnitude nuances within each domain. - Limited covariates: no detailed measures of isolation level, mental health, comorbidities beyond baseline, job role, or furlough status; these could confound behaviour changes. - Generalizability may be limited: participants were older, more likely to reside in less deprived areas, had lower BMI, and were less ethnically diverse than national populations in the UK and US. - Regression to the mean may partly explain improvements among those with the least healthy baseline behaviours, though asymmetry in changes suggests real effects. - Weight and behavioural measures rely on self-report rather than objective assessments.