Changes in Population Health-Related Behaviors During a COVID-19 Surge: A Natural Experiment

Abstract

Background

The study of impact of lockdowns on individual health-related behaviors has produced divergent results.

Purpose

To identify patterns of change in multiple health-related behaviors analyzed as a whole, and their individual determinants.

Methods

Between March and August 2020, we collected data on smoking, alcohol, physical activity, weight, and sleep in a population-based cohort from Catalonia who had available pre-pandemic data. We performed multiple correspondence and cluster analyses to identify patterns of change in health-related behaviors and built multivariable multinomial logistic regressions to identify determinants of behavioral change.

Results

In 10,032 participants (59% female, mean (SD) age 55 (8) years), 8,606 individuals (86%) modified their behavior during the lockdown. We identified five patterns of behavioral change that were heterogeneous and directed both towards worsening and improvement in diverse combinations. Patterns ranged from “global worsening” (2,063 participants, 21%) characterized by increases in smoking, alcohol consumption, and weight, and decreases in physical activity levels and sleep time, to “improvement” (2,548 participants, 25%) characterized by increases in physical activity levels, decreases in weight and alcohol consumption, and both increases and decreases in sleep time. Being female, of older age, teleworking, having a higher education level, assuming caregiving responsibilities, and being more exposed to pandemic news were associated with changing behavior (all p < .05), but did not discriminate between favorable or unfavorable changes.

Conclusions

Most of the population experienced changes in health-related behavior during lockdowns. Determinants of behavior modification were not explicitly associated with the direction of changes but allowed the identification of older, teleworking, and highly educated women who assumed caregiving responsibilities at home as susceptible population groups more vulnerable to lockdowns.

During COVID-19 related lockdowns, most individuals changed their health-related behaviors. Older, teleworking, highly educated women who assumed caregiving responsibilities at home seemed to be particularly vulnerable to both favorable and unfavorable behavioral changes.

Introduction

Law-enforced stay-at-home measures implemented during the COVID-19 pandemic, popularly known as lockdowns, have created highly disruptive scenarios impacting many aspects of life, including health-related behaviors (i.e., tobacco smoking, alcohol consumption, physical activity, diet, weight, and sleep). Given their role as preventable risk factors for non-communicable diseases (NCDs) [1], understanding the impact of lockdowns on health-related behaviors is of utmost importance to provide more effective mitigation strategies in the context of future extreme events.

Previous studies have investigated the impact of lockdowns on health-related behaviors with divergent results. Some studies have reported that people decreased their physical activity levels during lockdown [2–4]. Indeed, a study carried out among 13,754 university students in Spain reported decreases in moderate and vigorous physical activity (29.5% and 18.3%, respectively), as well as a large reduction in walking time (84.3%) [2]; a study in France showed that 53% out of 37,252 adults participating in the NutriNet-Santé cohort decreased their physical activity [3]; and another study looking at objectively measured physical activity in 97 younger adults (18–35 years) in the United States, reported decreases in daily step count, light and moderate-to-vigorous physical activity [4]. In comparison, other studies have observed that individuals became more active during lockdowns [5–7]. The German Motorik-Modul (MoMo) cohort study, including 1,711 children and adolescents (4–17 years), showed that, despite the decline in organized activities (e.g., sports), participants reported an overall increase in non-organized physical activity and in their adherence to WHO physical activity guidelines during lockdown [5]. Another study investigating physical activity behavior in 1,098 Canadian adults presented that individuals that were considered both inactive and active before the pandemic reported to be more active during lockdown (33% and 40% increase, respectively) [6]; and similarly, a study carried out in Belgium described increases in physical activity among low and high active adults (58% and 36% increase, respectively) [7].

Similar contradictory findings have been observed for other health-related behaviors such as smoking, alcohol, diet, body weight, or sleep [3, 8–16]. In regards to smoking, heterogeneous changes have been described among adult smokers, as has been reported by different online cross-sectional surveys in the United States (41% of smokers increased and 20% decreased consumption) [12], France (27% increase and 19% decrease) [14], and Australia (7% increase and 3% decrease) [9]. The same online surveys also reported changes in alcohol consumption during lockdown among drinkers (40% increase and 16% decrease in the United States; 11% increase and 24% decrease in France; 27% increase and 18% decrease in Australia) [9, 12, 14]. Changes in sleep quality, sleep time, and the presence of sleep disturbances during lockdown have been described among adults in Italy and Australia [9, 17], and among respiratory patients in Portugal [18]. In regards to changes in diet and body weight, a recent scoping review focused on the impact of lockdowns in various population groups across the globe and identified 23 studies reporting changes both towards favorable and unfavorable diets [8]. Favorable changes—including increased consumption of fresh products, home cooking, and reduced comfort food—were reported among adolescents and adults in Italy, Spain, Brazil, Colombia, and Chile [13, 19, 20]. Likewise, unfavorable changes, mostly related to reductions in fresh product consumption and increases in comfort foods, were reported among adolescents and adults in Italy, Spain, and France [3, 19–21]. Of note, this review also considered other health-related behavior associated with dietary changes (i.e., alcohol consumption, physical activity), and identified similar differing results in various settings across the globe [3, 8, 13, 21–23].

In sum, studies on the impact of lockdowns on health-related behaviors report divergent results that could be attributed to different types of restrictions and severities of lockdowns [24], differences in populations under study in terms of demographic, clinical, and socio-environmental conditions [7, 12, 16], or the possibility that humans behave heterogeneously to same life-threatening events.

While these early findings help in assessing the effect of lockdowns, their ability to understand population behavior is limited because most studies have studied health-related behavior in isolation [3–11, 13]. However, much data suggest that health-related behaviors and their changes often co-occur and, therefore, should be studied as a whole [25–28]. Further, health-related behaviors are influenced by both downstream and upstream social determinants of health [29–31], implying that considering these factors is essential to studying behavioral changes. This is in agreement with the main frameworks of social determinants of health [29, 30] showing the importance of focusing on the individual aspects associated with health-related behaviors, and on the social, structural, and organizational ones [31]; and reinforcing the idea that health-related behaviors are shaped by the social, economic and political environment in which individuals live [32].

Therefore, we aimed to identify patterns of change in multiple health-related behaviors during the first COVID-19 law-enforced lockdown in a population-based study in Catalonia, Spain, and to examine the individual and social determinants of these patterns.

Methods

Study Design and Participants

COVID-19 cohort in Catalonia (COVICAT) is a population-based study aimed at describing the health impact of the COVID-19 pandemic on the adult population of Catalonia, Spain [33–35]. COVICAT was built on five pre-existing cohort studies established before the outbreak [33, 34]. Participants were asked to complete a questionnaire and provide blood samples between May and December 2020. Further details on the recruitment for our cohort study have been published elsewhere [34, 35]. Briefly, participants (50% of the eligible sample) were younger (mean (SD) 55 (8) vs. 57 (9), p < .01) and had higher education level (45% vs. 29% university studies, p < .01) and lower BMI (27 (5) vs. 28 (5), p < .01) than non-participants [34]. For the present analysis, we used a pre-post study design and considered the 1st national COVID-19-related lockdown as a “natural experiment” (i.e., intervention) in which all Catalan residents were subjected to the same combination of restrictions imposed by public health authorities [36, 37]. Data collected before this intervention was defined as “pre”, and data collected in 2020 as “post”.

We excluded participants who responded between October and December 2020 because a different set of restrictions were put in place (n = 34), and participants who did not respond about changes in their health-related behavior (n = 21), resulting in a sample of 10,032 participants (99.4% out of the total 10,087 in COVICAT). The study was approved by the Parc de Salut Mar Ethics Committee (CEIm-PS MAR, number 2020/9307/I) and informed consent was obtained from all participants.

1st National COVID-19-Related Lockdown in Catalonia, Spain

In response to the global pandemic of COVID-19, Spanish national authorities declared a state of alarm on March 14, 2020, and the country entered the first period of lockdown lasting until June 21, when a “new normality” was established [38]. The 1st national COVID-19-related lockdown lasted 14 weeks and was divided into five consecutive phases (viz., Phases 0, 0.5, 1, 2, and 3). The length and transition between these phases were uneven throughout the Spanish territory and depended on health, epidemiological, and social indicators. In the Catalan provinces, Phase 0 lasted at least eight weeks and was characterized by a very severe lockdown, in which individuals were only allowed to leave their residences to buy food or medicines, seek urgent healthcare or go to work if considered essential [36]. A progressive relief of restrictions characterized the following phases until only recommendations on mandatory face-mask wearing, social distancing, and avoidance of agglomeration in public spaces were put in place [38]. Notably, restrictions in Catalonia were reinforced by sanctions (i.e., fines) imposed on those who failed to comply with or resist the established rules.

Procedures

Participants answered questionnaires primarily on a study website (94.5% of the sample), or telephone-administered by an interviewer for participants unfamiliar with web-based approaches (5.5% of the total sample).

Changes in health-related behaviors during lockdown were covered by specific questions on changes since the beginning of lockdown on (i) tobacco smoking (quit smoking, decrease smoking, no changes in never smokers, no changes in former smokers, no changes in current smokers, increase smoking, resume smoking after quitting, start smoking); (ii) alcohol consumption (quit alcohol, decrease drinking alcohol, no changes in abstainers, no changes in regular drinkers, increase drinking alcohol, resume drinking alcohol after quitting, and start drinking alcohol); (iii) physical activity (much less physical activity, less physical activity, a bit less physical activity, no changes in physical activity, a bit more physical activity, more physical activity, and much more physical activity); (iv) sleep time (decrease, no change, and increase); and (v) weight (decrease, no change, and increase). We also obtained information on pre- and post-lockdown smoking status (never, former, or current), alcohol consumption (yes, no), physical activity (low, moderate, or high, based on the International Physical Activity Questionnaire (IPAQ) classification [39]), average hours of sleep (including daytime sleep), and height and weight, that allowed the calculation of body mass index (BMI) and classification as underweight (<18.5 kg/m²), normal weight (18.5-24.9 kg/m²), overweight (25–29.9 kg/m²), and obese (≥30 kg/m²) [40].

Demographic data included sex, age, country of origin, region of residence, education level, household income, and work status during lockdown (working in the usual place, partially teleworking, teleworking, or not working). The presence of chronic conditions was obtained from a combination of self-report and a clinical diagnosis obtained from electronic health records and included hypertension, chronic obstructive pulmonary disease (COPD), diabetes, anxiety and depression, among other diseases. COVID-19 status (confirmed, probable or no case) was based on the combination of SARS-CoV-2 self-reported infection, presence of COVID-19 symptoms, COVID-19 hospitalization and presence and levels of immunoglobulins in serum [33].

Socio-environmental data included daily time spent outside during the lockdown, use of face mask outdoors and regular hand washing, frequent exposure to media news about the pandemic (>1 or ≤1 a week), number of people living at home, caregiving activities (taking full responsibility of children at home, splitting responsibility with a partner, or not having children or assuming no responsibility), access to outdoor spaces at home, having a dog and changing residence during lockdown.

Statistical Analysis

We conducted power calculations based on individual proportion of changes in health-related behaviors reported in previous studies [3, 6, 9, 11, 12, 14], and estimated that 10,000 participants allowed us to assess these proportions with a precision of 1% unit with a power >96% [41].

Descriptive characteristics of the sample are presented as absolute (n) and relative (%) frequencies for categorical variables, and as mean (SD) or median (P₂₅–P₇₅) for continuous variables according to their distribution. Changes in health-related behaviors are described using absolute (n) and relative (%) frequencies. Because of the small sample size in some of the categories (Electronic Supplementary Material 1), for the main analyses we re-categorized all self-reported changes in health-related behaviors into decrease, no change, or increase. All analyses were performed following a complete case approach, given that there was virtually no missing data (<0.2%) due to the mostly web-based administration of the questionnaire.

To extrapolate individual results about health-related behavior changes to the general population, we calibrated study data to generate estimates representing the adult population of Catalonia aged ≥20 years, using health population data from the 2019 National Statistics Institute and the Catalan Health Survey. We calculated sampling weights using the iterative proportional fitting and used their trimmed (at 99%) values to estimate population changes in health-related behaviors [33].

To identify patterns of change in health-related behavior during the lockdown, we performed a multiple correspondence analysis (MCA) followed by k-means cluster analysis. The MCA, multivariate exploratory analysis for categorical variables, provided us with a visual and structural organization for the variables and categories and allowed us to identify underlying structures (i.e., dimensions) using all variables of self-reported change in health-related behavior simultaneously (Fig. 1) [42, 43]. Then, we applied a k-means clustering algorithm to the dimensions obtained from the MCA and selected the optimal number of clusters using the Calinski–Harbarasz criterion [44, 45]. We reported the absolute (n) and relative (%) frequencies of each cluster, (i.e., patterns of changes in health-related behavior) and described their characteristics according to pre- and post-lockdown health-related behaviors.

Fig. 1.

Multiple correspondence analysis, changes in health-related behaviors during lockdown.

To assess the characteristics associated with patterns of change in health-related behaviors, we built multivariable multinomial logistic regression models using the resulting clusters as the outcome variable. All demographic, clinical, socio-environmental, and behavioral characteristics (i.e., age, sex, country of origin, education level, household income, work status during the lockdown, presence of chronic conditions (COPD, hypertension, diabetes, anxiety/depression), daily time spent outside during the lockdown, frequent exposure to media news about the pandemic, number of people living at home, caregiving activities, access to outdoor spaces at home, having a pet, changing residence during the lockdown, smoking status, alcohol consumption, physical activity levels, sleep time, and BMI before the pandemic) were considered as potential determinants of behavioral change. The final model included determinants if they were related to at least one of the clusters with p < .05, or if they modified (>10% change in regression coefficient) the estimates of the remaining variables in the model. Questionnaire administration (website, telephone), time of questionnaire (before, after “new normality”), and sanitary region were included as additional covariates.

In a secondary analysis, we investigated additional characteristics associated with a change in each health-related behavior. We conducted sensitivity analyses using the non-aggregated categories of change on health-related behaviors to investigate the extent to which the re-categorization of change affected our results. Analyses were performed using Stata (version 14.0; Stata Corp LP; College Station, TX, USA). Statistical significance was set at a p < .05.

Results

Out of the 10,032 participants, 59% were female, the mean (SD) age was 55 (8) years, and 9% had been diagnosed with COVID-19 (Table 1). Prior to lockdown, 16% of participants reported being current smokers, 82% reported a usual intake of alcohol, and 17% reported low levels of physical activity. A total of 39% of participants were overweight and 17% were obese. Mean (SD) daily sleep was 7 (1) hours (Table 2, Supplementary Material 1).

Table 1.

Descriptive Demographic, Clinical and Socio-Environmental Characteristics of 10,032 COVICAT Participants

Descriptive characteristics of COVICAT participants	N = 10,032
Sex, n (%)
Female	5,918 (59)
Male	4,114 (41)
Age (years), m (sd)	55.3 (8)
Country of origin (Spain), n (%)	9,669 (96)
Working during confinement, n (%)
No	3,834 (38)
Yes, teleworking	3,050 (30)
Yes, partially teleworking	911 (9)
Yes, in my usual place	2,237 (22)
Education level, n (%)
Less than primary	140 (1)
Primary	1,065 (11)
Secondary	4,200 (42)
University	4,604 (46)
Other	23 (0.2)
Equivalized household income (€), med (P25–P75)	1,549 (1,014–2,313)
Pre-lockdown BMI categories, n (%)
Underweight	52 (0.5)
Normal weight	4,303 (43)
Overweight	3,931 (39)
Obese	1,726 (17)
Presence of any chronic conditiona (yes), n (%)	7,815 (78)
COPD (yes), n (%)	710 (7)
Hypertension (yes), n (%)	1,823 (18)
Diabetes (yes), n (%)	370 (4)
Anxiety/depression (yes), n (%)	1,976 (20)
COVID-19 case, n (%)
Confirmed (with and without serology)	887 (9)
Probable	222 (2)
No case	8,923 (89)
HADS anxiety (0–21), med (P25–P75)	6 (3–8)
HADS depression (0–21), med (P25–P75)	3 (1–6)
Self-reported overall health status, n (%)
Excellent	639 (6)
Very good	3,497 (35)
Good	4,678 (47)
Regular	1,097 (11)
Bad	121 (1)
Frequency of exposure to COVID-19 news, n (%)
Daily or frequently (≥1 a week)	8,073 (85.8)
Rarely (<1 a week)	1,340 (14.2)
Change of residence during lockdown (yes), n (%)	365 (4)
Access to outdoor spaces in the residence (yes), n (%)
Never or rarely	1,201 (14)
Occasionally	3,828 (45)
Frequently	3,470 (41)
Living alone (yes), n (%)	1,371 (14)
Number of people at home, med (P25–P75)	3 (2–4)
Taking care of children at home, n (%)
No children/no care	5,718 (57)
Split responsibility with partner	2,512 (25)
Assuming full responsibility	1,802 (18)
Pet ownership (yes), n (%)	3,431 (34)
Dog, n (%)	1,859 (18)
Cat, n (%)	1,407 (14)
Other pets, n (%)	617 (6)
Wearing a face-mask outside (yes), n (%)	9,832 (98)
Washing your hands (yes), n (%)	9,998 (99.7)
Exposure to greenness at home (yes), n (%)	8,748 (87)
Daily time outside during lockdown (yes), n (%)	2,403 (25)
Health region of residence, n (%)
Alt Pirineu i Aran	11 (0.1)
Barcelonès	5,671 (57)
Camp de Tarragona	697 (7)
Catalunya Central	176 (2)
Girona	542 (5)
Lleida	722 (7)
Metropolità Nord	1,506 (15)
Metropolità Sud	604 (6)
Terres de l’Ebre	64 (0.6)
Questionnaire administration, n (%)
Study portal website	9,485 (95)
Telephone-administered	547 (5)
Time of questionnaire
Before “new normality” (before June 21)	7,886 (79)
After “new normality” (June 21 onwards)	2,146 (21)

Some variables had missing values: 5 in country of origin, 1,461 in Equivalized household income, 39 in region of residence, 619 in frequency of exposure to news about the pandemic, 1,533 in access to outdoor spaces in the residence, 619 in daily time outside during lockdown.

^aPresence of any chronic conditions was obtained from a combination of self-report and a clinical diagnosis obtained from electronic health records. This included diseases from all body systems categories included in the International Classification of Diseases, ninth revision, Clinical Modification (ICD-9-CM) that ever had a diagnosis of 12 months or longer to be considered chronic.

BMI body mass index; COPD chronic obstructive pulmonary disease; HADS Hospital Anxiety and Depression scale; m mean; med median; sd standard deviation; y years.

Table 2.

Pre-Lockdown Health-Related Behaviors and Changes During Lockdown in 10,032 COVICAT Participants

Pre-lockdown health-related behaviors and changes during lockdown	COVICAT All	Patterns of change in health-related behaviors
		Global worsening	Worsening physical activity and weight	No changes	Mixed changes	Improvement
	N = 10,032	n = 2,083	n = 3,595	n = 1,426	n = 380	n = 2,548
Pre-lockdown smoking, n (%)
Never	4,157 (41.4)	679 (32.6)	1,724 (47.9)	612 (42.9)	0 (0.0)	1,142 (44.8)
Former	4,236 (42.2)	827 (39.7)	1,582 (44.0)	610 (42.8)	0 (0.0)	1,217 (47.8)
Current	1,639 (16.3)	577 (27.7)	289 (8.0)	204 (14.3)	380 (100.0)	189 (7.4)
Pre-lockdown alcohol, n (%)
Abstainer	1,797 (17.9)	311 (14.9)	686 (19.1)	307 (21.5)	51 (13.4)	442 (17.4)
Drinker	8,235 (82.1)	1,772 (85.1)	2,909 (80.9)	1,119 (78.5)	329 (86.6)	2,106 (82.6)
Pre-lockdown physical activity, n (%)
Low	1,689 (16.8)	264 (12.7)	33 (0.9)	245 (17.2)	85 (22.4)	1,062 (41.7)
Moderate	2,878 (28.7)	471 (22.6)	562 (15.6)	677 (47.5)	105 (27.6)	1,063 (41.7)
High	5,465 (54.5)	1,348 (64.7)	3,000 (83.5)	504 (35.3)	190 (50.0)	423 (16.6)
Pre-lockdown sleep time (h), m (sd)	6.97 (1.0)	7.2 (0.9)	6.85 (0.9)	6.92 (0.9)	7.10 (1.2)	6.93 (0.9)
Pre-lockdown weight, n (%)
Underweight	52 (0.5)	11 (0.5)	13 (0.4)	7 (0.5)	5 (1.3)	16 (0.6)
Normal weight	4,303 (43.0)	925 (44.6)	1,586 (44.2)	589 (41.4)	179 (47.2)	1,024 (40.2)
Overweight	3,931 (39.3)	774 (37.3)	1,386 (38.6)	573 (40.3)	137 (36.1)	1,061 (41.7)
Obese	1,726 (17.2)	366 (17.6)	603 (16.8)	254 (17.8)	58 (15.3)	445 (17.5)
Self-reported changes in smoking during lockdown, n (%)
Decrease	380 (3.8)	0 (0.0)	0 (0.0)	0 (0.0)	380 (100.0)	0 (0.0)
No change	9,010 (89.8)	1,487 (71.4)	3,595 (100.0)	1,397 (98.0)	0 (0.0)	2,531 (99.3)
Increase	642 (6.4)	596 (28.6)	0 (0)	29 (2)	0 (0.0)	17 (0.7)
Self-reported changes in alcohol during
lockdown, n (%)
Decrease	1,054 (10.5)	137 (6.6)	375 (10.4)	78 (5.5)	85 (22.4)	379 (14.9)
No change	7,812 (77.8)	1,169 (56.1)	3,054 (85.0)	1,318 (92.4)	244 (64.2)	2,027 (79.5)
Increase	1,166 (11.6)	777 (37.3)	166 (4.6)	30 (2.1)	51 (13.4)	142 (5.6)
Self-reported changes in physical activity during lockdown, n (%)
Decrease	5,546 (55.3)	1,484 (71.2)	3,439 (95.7)	0 (0.0)	194 (51.0)	429 (16.8)
No change	2,073 (20.7)	223 (10.7)	156 (4.3)	1,426 (100.0)	80 (21.1)	188 (7.4)
Increase	2,413 (24.0)	376 (18.1)	0 (0.0)	0 (0.0)	106 (27.9)	1,931 (75.8)
Self-reported changes in sleep time during lockdown, n (%)
Decrease	1,738 (17.3)	1,287 (61.8)	0 (0.0)	81 (5.7)	74 (19.5)	296 (11.6)
No change	6.490 (64.7)	648 (31.1)	2,499 (69.5)	1,345 (94.3)	243 (63.9)	1,755 (68.9)
Increase	1,804 (18.0)	148 (7.1)	1,096 (30.5)	0 (0.0)	63 (16.6)	497 (19.5)
Self-reported changes in weight during lockdown, n (%)
Decrease	1,134 (11.3)	53 (2.5)	0 (0.0)	0 (0.0)	41 (10.8)	1,040 (40.8)
No change	4,422 (44.1)	612 (29.4)	1,605 (44.6)	1,044 (73.1)	177 (46.6)	984 (38.6)
Increase	4,476 (44.6)	1,418 (68.1)	1,990 (55.3)	382 (26.8)	162 (42.6)	524 (20.6)

Some variables had missing values: 49 in sleep time pre-lockdown and 20 in weight categories pre-lockdown

Changes in Health-Related Behaviors During Lockdown

During lockdown most participants reported changes in physical activity levels (55% decrease, 24% increase) and weight (11% decrease, 45% increase), while levels of smoking, alcohol, or sleep remained stable (Table 2, Fig. 2). Extrapolation to the general population showed that most people in Catalonia were likely to change their physical activity levels (49% decrease, 22% increase) and weight (14% decrease, 35% increase), but not smoking, alcohol, or sleep (Supplementary Material 2).

Fig. 2.

Changes in individual health-related behaviors during lockdown in Catalonia and their patterns identified using hypothesis-free methods. Individual distributions of changes in health-related behaviors as absolute (n) and relative (%) frequencies, and characterization of patterns of change in health-related behaviors, according to the distribution of changes in individual health-related behaviors.

Following multiple correspondence and cluster analyses, we identified five patterns of behavioral change, ranging from general worsening to improvement (Fig. 2). Overall, 86% of participants were included in patterns characterized by behavioral changes. We labeled the identified clusters as (i) “global worsening”, including a group of individuals characterized by an increase in smoking, alcohol consumption, and weight, and a decrease in physical activity levels and sleep time (n = 2,083, 21%); (ii) “worsening physical activity and weight”, including participants characterized by a decrease in physical activity levels and an increase in weight and sleep time (n = 3,595, 36%); (iii) “mixed changes”, including a group characterized by a decrease in smoking, alcohol consumption, and physical activity levels (n = 380, 4%); (iv) “improvement”, including participants characterized by an increase in physical activity levels, weight reduction, a decrease in alcohol consumption, and both increases and decreases in sleep time (n = 2,548, 25%); and (v) “no changes”, including participants without substantial changes in any of the health-related behaviors (n = 1,426, 14%) (Table 2, Fig. 2).

Determinants of Change in Health-Related Behavior and Weight During Lockdown

Using “no change” as the reference category, being female, older age, changing work status towards teleworking or partially teleworking, or not working during the lockdown, having a university degree, and being frequently exposed to media about the pandemic news were associated with a higher likelihood of belonging to both worsening and improvement patterns of change (Fig. 3, Supplementary Material 3). Assuming caregiving responsibilities was associated with a higher likelihood of belonging to the “global worsening” and “improvement” clusters. Further, having a dog was associated with a lower likelihood of belonging to the “global worsening” and “worsening physical activity and weight” clusters (all p < .05). Of note, the country of origin, household income, presence of chronic conditions, daily time spent outside during lockdown, number of people living at home, access to outdoor spaces at home, and changing residence during lockdown were not significantly associated with the patterns of change.

Fig. 3.

Adjusted association among demographic, clinical, socio-environmental and pre-lockdown health-related behavior characteristics and patterns of health-related behaviors changes (multivariable multinomial logistic regression). Adjusted associations resulting from a multivariable multinomial logistic regression model using cluster of “no change” as reference category. Reference categories were: working in the usual place for changes in work during lockdown; no children/no responsibility for caregiving activities; never smoker for pre-lockdown smoking; moderate physical activity for pre-lockdown physical activity; 7–9 hours per day sleep; underweight or normal weight for pre-lockdown BMI. Abbreviations: BMI, body mass index; h, hours; PA, physical activity; RRR, relative risk ratio.

Similar findings were observed in post-hoc analyses in which the reference categories were “worsening physical activity and weight” or “improvement” (Supplementary Material 4a, b). Secondary analyses investigating determinants of change in individual health-related behaviors (instead of behavioral patterns) showed similar associations, with some exceptions: change in work status was no longer associated with a higher risk of changing tobacco consumption; assuming caregiving activities were related to increases in smoking, alcohol consumption, physical activity levels, and weight, as well as with both directions of change in sleep time; and anxiety/depression was significantly associated with increases in smoking, alcohol consumption, weight, and with both increases and decreases in sleep time (Supplementary Material 5a-e). Sensitivity analyses using non-aggregated categories of self-reported change in health-related behaviors identified nine clusters that were similar to that of the main analysis (Supplementary Material 6) and similar predictors (Supplementary Material 7).

Discussion

In this population-based study, we provided evidence that up to 86% of participants modified their behavior in the context of an extreme event (i.e., COVID-19 law-enforced lockdown), and that these changes were heterogeneous and directed both towards worsening or improvement in diverse combinations. Further, we showed that the main determinants of behavior modification were common to both favorable and unfavorable changes, and were broad in nature, as they included demographic, socio-environmental, and behavioral characteristics. Taken together, these results provide evidence that lockdowns disproportionally changed the population’s health-related behavior and, if persistent in time, may become triggers of a rising NCDs epidemic [46].

Our study grouping health-related behavior changes in distinct patterns show that up to 86% of the population modified their behavior, a higher proportion than observed previously when studying behaviors in isolation [3, 6, 9, 11, 12, 14]. We found that physical activity was the main distinctive behavior in all patterns of change and was usually reported in combination with opposite changes in weight. Further, changes in tobacco smoking and alcohol consumption were also captured in the same clusters, in agreement with studies reporting the co-occurrence of smoking and alcohol consumption [25, 26]. In line with previous evidence reporting sleep disturbances and alterations in daily routines during lockdown [16, 47], we observed changes in sleep time in the four patterns of change, but they had no clear direction and were not considered as differentiating between clusters. Overall our results support the need to consider all health-related behaviors (and their inter-relations) as a whole and not individually.

Our study provides evidence that the same set of demographic, socio-environmental, and behavioral characteristics are predictive of changes in health-related behaviors. Consistent with previous research [3, 12], being a female, younger, teleworking, and highly educated were indistinctly associated with changes in health-related behaviors. A novel finding was that having a dog was associated with a lower likelihood of changing health-related behavior towards any identified patterns. It might be possible that pet owners had a higher likelihood of “not changing” their health-related behaviors, given that dog walkers were among the few allowed to spend time outside during strict law-enforced lockdown. Of novelty and relevance, we identified that assuming caregiving responsibilities at home was associated with a higher likelihood of belonging to the “global worsening” and “improvement” patterns, reflecting on the extra burden imposed by household tasks and family responsibilities at home (on top of teleworking). In this case, it might be possible that the increasing number of tasks assumed at home gave individuals fewer opportunities to be healthier (e.g., less time to practice exercise, less time to cook balanced meals, more stressors triggering tobacco, and alcohol consumption) or that changes in household dynamics given by the extra time spent at home could motivate individuals to assume healthier behaviors (e.g., cooking home meals, practicing physical activity at home with children, reducing smoking and alcohol, and sleeping longer). Finally, being frequently exposed to media news about the pandemic was associated with a higher likelihood of belonging to the “global worsening,” “worsening physical activity and weight” and “improvement” patterns, with the idea that elevated exposure to a highly disruptive event such as a pandemic, may become an additional cause of anxiety and stress symptoms [35, 48] and may trigger extreme changes in human behavior.

Interestingly, and contrary to expected, we did not find a relation between access to outdoor home spaces and changes in health-related behaviors. This finding might reflect the heterogeneity of what constitutes outdoor spaces at home (i.e., balcony, terrace, and garden) and may require further investigation. We also did not find an association between country of origin and changing residence during lockdown with changes in health-related behaviors, most likely because 96% of participants were Spanish and did not change their residence. Similarly, we did not find an association between chronic respiratory, cardiovascular, metabolic, or mental health conditions and patterns of change in health-related behaviors. However, in secondary analysis using each behavior individually, we did find an association between the presence of anxiety/depression and a higher likelihood of increasing smoking and alcohol consumption, increasing weight and changing sleep time (both increase and decrease), in agreement with previous literature suggesting that lockdowns were incredibly disruptive in subjects with underlying health conditions [9, 12]. It is plausible that mental health affected smoking, alcohol, weight (i.e., diet), and sleep behaviors, but not physical activity could explain why the association was only observed when these factors were examined individually but not when assessed in combination.

While previous behavioral theories suggest that patterns of change in behavior could be driven by context or individual choices [49–51], our findings illustrate the dynamics and complexity of human behavior in which context and individual characteristics usually interact in shaping behavior. Specifically, the “worsening physical activity and weight” pattern could be interpreted as contextually driven and could reflect on a group of highly active participants who changed their behavior by reducing physical activity and consequently increasing weight due to the unavoidable situation of being forced to stay at home. Similarly, the “mixed changes” pattern might capture a group of participants that, given the impossibility to go out and socialize, were contextually driven to reduce their usual smoking and alcohol consumption, but also reduced their physical activity. In contrast, individually driven patterns of change could capture personal decisions within the lockdown context. The “global worseners” pattern is likely to include individuals whose situation changed in such a way that led them to engage in overall unhealthy behaviors including reduced physical activity and sleep time, and increased smoking, alcohol, and weight, thus reflecting more individually-driven changes in behavior than only the context. Another possibility is that for those individuals the context was too adverse compelling them to adopt these overall unhealthy behaviors. In the same way, the “improvers” pattern could be seen as a group that took benefit of having to stay at home to change their health-related behaviors towards a healthier lifestyle as they increased their physical activity levels, lost weight, and reduced their alcohol consumption. These findings confirm that response to lockdown was heterogeneous, as facing the same disruptive scenario and type of restrictions, individuals’ reactions vary depending on their personal decisions and context [49].

Results from this study could serve as evidence to help inform health promotion interventions in response to extreme events in the future [52]. For example, the role of physical activity in defining patterns of change could help modify some restrictive measures implemented by authorities during a pandemic. In addition, our findings showing that younger, teleworking, and highly educated women were more susceptible to change behavior could help design public health policies targeted to protect this population group against adding extra burden and engaging in unhealthy behaviors. Finally, future research should focus on exploring the interplay between different health-related behaviors and additional factors that could be related to the likelihood of changing behavior (e.g., having sports material at home, patterns of active transportation, motivation, or stress management), assessing the relationship between changes in health-related behaviors during the lockdown and important health outcomes such as life satisfaction, adherence to international recommendations or health-related quality of life [53, 54], and understanding the long-term effects of the observed changes, as they might become underlying risk factors for developing NCDs.

This study had some limitations. First, a large proportion of our study participants comes from a pre-existing blood donors cohort [55] and are likely to represent health conscientious individuals who might be prone to healthier behaviors. In addition, only half of the previous study participants contributed to the current research and these were younger and higher educated than the non-participants. To partially address the potential selection bias from these sources, we extrapolated changes in individual health-related behaviors to the general Catalan population and confirmed most of our reported levels of change. Second, we did not have information on diet and included weight in our analysis, but we acknowledge this is not a good proxy for diet. Third, we lacked information on factors that could relate to the likelihood of changing behavior such as having sports material at home, patterns of active transportation, motivation, or the size of households. Finally, our results might not be generalizable to other populations, as the restrictions imposed differed between countries.

Strengths of this study include the large, population-based sample of participants from all sanitary regions in Catalonia, which allowed the estimation of population-based behavioral changes and increased the external validity of our results. The study is of particular interest to understanding behavioral changes in response to extreme events because of the high number of cases and the severity of restrictions put in place during the first COVID-19 surge [36]. In addition, we had an extensive characterization of the sample before the pandemic. Finally, the availability of five health-related behaviors and the analytical approach addressing all health-related behaviors at the same time, allowed us to interpret results at the population level and prevented (positive or negative) confounding and/or interaction with other behaviors.

In conclusion, most of the population changed their health-related behaviors during lockdowns. These changes are heterogeneous, could be grouped into five patterns ranging from general worsening to improvement, and are difficult to predict. However, we identified susceptible population groups that may be more vulnerable to future lockdowns.

Compliance with Ethical Standards

Authors’ Statement of Conflict of Interest and Adherence to Ethical Standards The authors declare that they have no conflict of interest.

Authors’ Contributions L.D.-O. contributed to study conceptualization and design, analysis and interpretation of data, preparation and review of the final manuscript. A.-E.C. contributed to study conceptualization and design, preparation and interpretation of data, validation of results and review of the final manuscript. J.M. contributed to study conceptualization and design, interpretation of data and review of the final manuscript. I.C. contributed to the visualization and interpretation of data, and review of the final manuscript. S.K. contributed to the interpretation of data and review of the final manuscript. X.G. contributed to the interpretation of data and review of the final manuscript. G.C. contributed to the interpretation of data and review of the final manuscript. M.B.deB. contributed to the interpretation of data and review of the final manuscript. G.C.-V. contributed to the collection and preparation of data, project management and review of the final manuscript. A.E. contributed to the preparation of data and review of the final manuscript. A.C. contributed to the collection and preparation of data, and review of the final manuscript. B.C.M. contributed to the collection and preparation of data, and review of the final manuscript. K.S. contributed to the interpretation of data and review of the final manuscript. R.de.C. contributed to the collection and preparation of data, project management and review of the final manuscript. M. K. contributed to the collection and preparation of data, interpretation of data, project management and review of the final manuscript. J.G.-A. was the research lead contributing to study conceptualization and design, coordination and data collection, analysis and interpretation of data, validation of results, as well as preparation and review of the manuscript. All authors read and commented on previous versions of the manuscript, approved the final manuscript and agreed to be accountable for all aspects of the work.

Ethical Approval All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.

Informed Consent Informed consent was obtained from all individual participants included in the study.

Transparency Statements Study registration. The COVID-19 cohort in Catalonia (COVICAT) study was built upon five pre-existing cohort studies, established and registered before the COVID-19 outbreak (see Methods). The study was approved by the Parc de Salut Mar Ethics Committee (CEIm-PS MAR, number 2020/9307/I) and informed consent was obtained from all participants. Analytic plan pre-registration. The statistical analysis plan was not formally pre-registered.

Data availability De-identified data and materials from this study are not available in a public archive. De-identified data and materials are available upon request following institutional procedures. This manuscript adheres to the recommendations on transparency in statistical practice. Analytic codes are available upon request following institutional procedures.