Rates of depression have increased worldwide during the COVID-19 pandemic1, highlighting the importance of identifying modifiable factors for targeted approaches to prevention and intervention during a global crisis. Even before the pandemic, substantial literature had established social support as a robust protective factor for depression, both in general2 and during times of stress3, demonstrating promise for reducing the population-level burden of depression. Accordingly, a growing number of studies on the overall relationship between social support and mental health during the COVID-19 pandemic have linked greater social support to reduced risk for depression and other negative mental health outcomes4,5,6. However, further work is needed for several reasons.

First, many studies of social support and depression during the COVID-19 pandemic have relied on relatively modest cross-sectional samples in specific populations (for example, elderly, health-care workers, students, pregnant individuals)4,5,7,8,9,10. Second, such studies have tended to focus on overall social support rather than systematically examining subtypes of social support5,10,11, which include tangible supports characterized by instrumental help for everyday or crisis needs, emotional supports characterized by a listening or confiding ear, informational supports characterized by advice or knowledge from others and positive social interactions12. Although pre-pandemic research has shown that emotional support tends to be the strongest predictor of mental health compared with other forms of support2,13, during a pandemic, tangible support may become more emotionally salient than other forms of support due to the risk of illness, contagion and other daily disruptions, while the relevance of positive social interactions may be diminished in the context of pervasive social distancing. Understanding which subtype(s) of social support most influence depression risk during a global crisis14 could highlight which aspects should be addressed specifically, or together, in both individual- and population-level interventions.

Third, different groups of people may benefit from increased social support. For example, the protective role of social support may vary by intrinsic factors such as age and sex, which may influence both depression risk and how social support impacts this risk, or by pre-pandemic factors such as previous mental health histories, which may increase vulnerability to subsequent distress. In addition, risk factors may emerge during the pandemic such as financial stressors, which either limit one’s ability to benefit from social support or indicate where social supports are particularly needed. A better understanding of potential effect modifiers, across a range of intrinsic, pre-pandemic and during-pandemic characteristics, could provide insight into where interventions aimed at enhancing social support could have the largest impact (for example, targeting pre-existing or concurrent risk factors).

Given these gaps and unique challenges to both mental health and social support access during the COVID-19 pandemic, more detailed work in large prospective cohorts is needed to quantify the extent to which social support may protect against elevated depressive symptoms in the context of a major societal stressor—specifically, to examine which types of social support (combined and alone) may prove most helpful and to identify who might benefit most. One opportunity to rigorously investigate these questions is within the All of Us Research Program (AoU), an ongoing, diverse US nationwide research cohort where a longitudinal survey focused on mental health, coping and other experiences, including social support, was administered in the first year of the COVID-19 pandemic (Fig. 1). Using longitudinal data across three survey waves completed by AoU participants (N = 69,066), we tested associations between perceived social support and its subtypes with depression risk and assessed potential effect modification by intrinsic (for example, sex, age), pre-pandemic (for example, previous mood disorder) and pandemic-related (for example, financial stress) risk factors.

Fig. 1: Timeline of survey and EHR-based measures of social support, depressive symptoms and key covariates.
figure 1

Sociodemographic factors and pre-pandemic mood disorder diagnoses were assessed before the pandemic, while social support and depressive symptoms were measured in three survey waves starting from May 2020. COPE, COVID-19 Participant Experience survey. RAND-MOS, RAND Medical Outcomes Study. PHQ-9, Patient Health Questionnaire-9.


Sample characteristics are summarized in Extended Data Table 1. Participants were predominantly female (66%) and self-reported white (82%), with an average age of 59 years (s.d. = 16; range = 18.1–87.8) and with 8% reporting any pandemic-related financial stressors. The median income of participants ($87,500) was above the median US household income in 2020 ($71,186) (US Census Bureau). On average, 16% of study participants met criteria for moderate to severe symptoms of depression (hereafter, referred to as ‘depression’ to be concise) across the survey months of May, June and July, and 14% of the sample were identified to have a pre-pandemic mood disorder diagnosis according to linked electronic health record (EHR) data.

After adjusting for sociodemographic and clinical factors, overall social support was inversely associated with depression (adjusted odds ratio (AOR) [95% confidence interval (CI)] = 0.45 [0.44–0.46], P < 2.0 × 10−16) (Table 1). As shown in Fig. 2a, the predicted probabilities of depression were elevated at low levels of reported social support and declined towards zero at the highest levels of social support. This was also consistent when depression was modelled as a continuous symptom outcome (Extended Data Table 2) and in lagged sensitivity analyses, where overall social support at the first survey wave was prospectively associated with depression in the subsequent 1- or 2-month windows (AORs 0.66–0.68; Extended Data Table 3) even after excluding individuals with elevated symptoms at the first wave.

Table 1 Inverse-probability weighted mixed-effects logistic regression analyses examining the association between overall social support or its subtypes with depression
Fig. 2: Model-predicted probabilities of having moderate–severe depressive symptoms against different levels of social support.
figure 2

a, Overall social support. b, Tangible support. c, Emotional/informational support. d, Positive social interaction. The bands indicate 95% CIs for the predicted probabilities.

Among subtypes, emotional/informational support (Fig. 2c) showed the largest inverse association with depression (AOR [95%CI] = 0.42 [0.41–0.43], P < 2.0 × 10−16), followed by positive social interaction (Fig. 2d) (AOR [95% CI] = 0.43 [0.41–0.44], P < 2.0 × 10−16) (Table 1). Tangible support (Fig. 2b) was also associated with a smaller but nonetheless significant reduction in depression odds (AOR [95% CI] = 0.63 [0.61–0.65], P < 2.0 × 10−16).

When examining combinations of social support subtypes, a marked dose–response gradient (Fig. 3 and Table 2) emerged. Compared with those reporting lower support on all subtypes, those with higher tangible support alone showed a modest reduction in the odds of depression (AOR [95% CI] = 0.87 [0.79–0.96]) compared with those with higher positive social interaction (AOR [95% CI] = 0.45 [0.38–0.54]) or emotional/informational support (AOR [95% CI] = 0.39 [0.34–0.44]) alone. However, endorsing higher levels of at least two subtypes of support was linked to greater reductions in the odds of depression, with a particularly strong protective association among those endorsing both emotional/informational support and positive social interaction (AOR [95% CI] = 0.22 [0.19–0.25]). Ultimately, endorsing higher levels of all three subtypes of support appeared most protective (AOR [95% CI] = 0.15 [0.14–0.16]).

Fig. 3: Association between specific combinations of social support subtypes and depression.
figure 3

Having a particular subtype refers to endorsing above-average levels of that subtype. Reference group (not shown) is those with lower support on all three subtypes. The error bars indicate 95% confidence intervals for a given odds ratio estimate.

Table 2 Inverse-probability weighted mixed-effects logistic regression analysis examining the association of specific combinations of social support types with depression

Last, we identified overall effect modification by sex (β = −0.033, interaction P = 1.4 × 10−2), age (β = 0.48, interaction P = 1.6 × 10−3), pre-pandemic mood disorder (β = −0.056, interaction P = 1.4 × 10−2) and COVID-related financial stress (β = 0.050, interaction P = 8.1 × 10−4). As shown in Fig. 4 and supported by stratified results (Extended Data Table 4), female participants and younger individuals had higher (two- to fourfold) predicted probabilities of depression than male participants and older individuals, respectively. In both cases, these effects appeared to attenuate at higher levels of social support. A similar pattern was observed for those reporting any pandemic-related financial stressors compared with those without any such stressors.

Fig. 4: Model-predicted probabilities of having moderate–severe depressive symptoms against different levels of overall social support, stratified by potential effect modifier.
figure 4

a, Sex assigned at birth. b, Current age. c, Pre-existing mood disorder diagnosis (lifetime). d, COVID-related financial stressors. The bands indicate 95% confidence intervals for the predicted probabilities.


In a large prospective cohort of 69,066 adults participating in the nationwide AoU, higher levels of social support were associated with a 55% reduction in the odds of elevated depressive symptoms (depression) during the early months of the COVID-19 pandemic. Greater perceived support across multiple domains appeared most protective, with individuals reporting higher levels of tangible, emotional/informational and positive social interaction supports showing a more than sixfold reduction in the odds of depression compared with those without. Moreover, those at higher risk of depression during the pandemic—female participants, younger individuals below age 60, those with a previous mood disorder diagnosis and those reporting pandemic-related financial stressors—showed attenuated odds of depression with higher levels of social support.

Among social support subtypes, emotional/informational support showed the largest protective association with depression, followed by positive social interactions. During the acute time frame of the early pandemic, social support may reduce feelings of loneliness and isolation amid social distancing15,16 and provide outlets to process heightened uncertainty and stress4 related to the evolving pandemic. Findings are consistent with our pre-pandemic observations from the UK Biobank13 showing that, among more than 100 potentially modifiable factors, confiding in others—which relates to the use and availability of emotional support—had the strongest protective association with incident depression in adults. This underscores the importance of trusted interpersonal connections for mitigating depression risk, possibly via enhanced affect and cognitive regulation17. Recent work during the pandemic in a longitudinal cohort found that perceived quality of one’s relationships was protectively associated with psychiatric risk18, reinforcing the importance of social connection quality, not simply quantity—which may be especially relevant for modes of social media and online interactions that have emerged since the pandemic.

In addition to examining subtypes separately, we probed the consequences of different combinations of support. In a dose–response fashion (Fig. 4), individuals with higher support on all three subtypes showed the largest decrease in odds of depression, followed by those with higher support on at least two subtypes then those with higher support on one subtype alone. Most individuals endorsed support across either all subtypes or none, while a subset of individuals (8.6%) endorsed higher tangible support alone. Tangible support was less strongly associated with depression compared with other subtypes. We had hypothesized this association would be stronger during the pandemic, given potential stressors of infection, illness and quarantine that could incur practical challenges requiring assistance. However, we also adjusted for demographic factors such as marital status that may overlap with availability of tangible support at home, which could have attenuated observed associations.

Overall, results suggest that boosting social support on multiple fronts could have the most beneficial influences on mental health risk. Interestingly, higher emotional support and positive social interaction together seemed to show greater protective relevance for depression versus either subtype alone—even in the absence of substantial tangible support. Endorsing higher positive social interaction alone was relatively rare (<2%), perhaps because time spent in positive experiences with others may naturally facilitate asking for tangible and/or emotional support; consistent with this, positive social interactions showed the highest correlations with both subtypes. Identifying which social support domains are perceived as high quality versus lacking for a given individual may help in tailoring interventions. Social support-enhancing interventions might focus on the creation/use of support groups that provide a safe space for individuals to share concerns or interests19,20 and include individual skill-building and network identification21,22. Certain self-care practices (for example, joining an online yoga group) may also increase positive social interactions.

We found that intrinsic, pre-pandemic and during-pandemic factors may modify the relationship between social support and depression. Women generally had higher elevated depressive symptoms compared with men but showed larger reductions in depression odds in the presence of higher social support. This is consistent with a recent longitudinal study during the pandemic that observed stronger protective associations between social participation/trust and any depressive symptoms in women versus men14. We expand previous work by identifying additional effect modifiers, including age and financial stressors related to the pandemic. Probing these results revealed that social support was consistently linked to larger reductions in depression odds among participants at higher risk (younger individuals, those with previous mood disorder diagnosis, those endorsing pandemic-related financial stressors). Thus, adults more vulnerable to depression—for intrinsic or environmental reasons—may benefit more from the protective influences of social support.

Strengths and limitations

Our study offers the following contributions. First, it makes use of the uniquely large and diverse data from the AoU to study the effects of social support during the early COVID-19 pandemic (an exemplar crisis impacting all facets of society), which presented unique challenges to both mental health and social support access. Second, conceptually, our study aimed to add the existing literature by examining not only different subtypes of social support but also the implications of their combinations and for whom the effects of social support on depression may be most pronounced. Methodological strengths of this study included its longitudinal, multi-wave design in a large, diverse cohort with linked previous data and ongoing research participation, and nuanced operationalization of social support using an established survey instrument12. While some studies may have focused on quantity of supportive individuals14, focusing on perceived quality of support is an important contribution of our research as social support may be concentrated within a few individuals providing high-quality support. In addition, our study assessed the relationship of social support to clinically elevated symptoms of depression, which incur the greatest impairment and disability requiring prevention at the population level.

Several limitations should be noted. First, as with many survey instruments, our measure of social support captures self-reported perceptions of support, which may be influenced by concurrent depressed mood, thereby inflating the association. Notwithstanding, our primary analyses were consistent with lagged models in which social support at the first wave was associated with subsequent depression even after removing individuals with elevated depressive symptoms at that wave, suggesting the association was not driven purely by contemporaneous mood states; however, further causal inference analyses are warranted23. Measuring depression with the PHQ-9 also has its limitations24,25, although it is one of the most-used measures of depressive symptoms. Different time frames of reference for the different measures (for example, past month versus past 2 weeks) may also complicate recall. Second, we selected available variables from the COVID-19 Participant Experience (COPE) survey to examine their potential role as effect modifiers, but there may be other unmeasured intrinsic or pandemic-related factors (for example, lockdown exposure) that may also influence the association between social support and depression. Our pre-pandemic mood disorder variable was also conservatively defined using linked EHR data, which were available for most, but not all, participants. Third, given that most COPE participants were research volunteers with Internet access who were relatively older on average, were well educated and endorsed minimal financial stressors related to the pandemic, and generally showed higher socioeconomic status (for example, educational attainment, home ownership, household income) compared with overall AoU participants26, this warrants caution as our sample may not fully generalize to more disadvantaged populations, to younger adults or to the broader US population, although we used inverse-probability weighting as an attempt to match COPE survey participants more closely with the broader and more diverse AoU study cohort.


Social connection is increasingly recognized as a key public health priority27. While the links between social support and depression are well established2, a more nuanced understanding of this relationship—including which sets of support subtypes are most relevant for depression risk, and who may benefit most, during a highly stressful global crisis—could inform targeted interventions to enhance resilience and reduce the population-level burden of depression. Individuals reporting higher levels of social support were at substantially reduced risk of elevated depressive symptoms during the COVID-19 pandemic. The perceived availability of emotional support and positive social interactions—and their combination—more so than tangible assistance, was key. Individuals more vulnerable to depression (for example, women, younger individuals, those with previous mood disorder diagnosis or those experiencing financial stressors) may particularly benefit from enhanced social support during a major stressor, supporting a precision prevention approach.



The Institutional Review Board of the AoU approved all study procedures, and participants with a demonstrated capacity to consent provided (signed) informed consent to share EHR, survey and other study data with qualified investigators for broad-based research. The Mass General Brigham has an institutional data use agreement with Vanderbilt University Medical Center, which operates the All of Us Data and Research Center. In addition, Mass General Brigham researchers have an Institutional Review Board approval that covers their access and analysis of the Registered and Controlled Tier data through the Researcher Workbench.

Cohort description

The AoU28 has enrolled more than 482,000 participants as of April 2022. AoU participants consist of individuals over age 18 (inclusive) who currently reside in the US or a US territory. More than 80% of participants are from communities that have been underrepresented in biomedical research based on the following characteristics: race and ethnicity, age, sexual orientation and gender identity, low income and educational attainment, rural residence and disability. By the first pandemic survey, 309,832 AoU participants had been enrolled.

Study sample

The COPE survey was administered electronically to AoU participants to assess the longitudinal impact of the pandemic and included questions on COVID-19 symptoms, physical and mental health, social distancing, economic impacts and coping strategies. The first three waves of the COPE survey (administered in May, June and July of 2020) included assessments of social support and depressive symptoms (Fig. 1; the May 2020 survey was open between 7 May 2020 and 29 May 2020; the June survey between 2 June 2020 and 26 June 2020; the July survey between 7 July 2020 and 25 September 2020). For context, official lockdowns and social distancing mandates in the United States were first implemented in March 2020. While the pandemic response varied between regions, overall, restrictions in the United States began to ease in June 2020. Vaccines did not become available to the general public until after December 2020 (ref. 29). A total of 69,066 respondents completed the COPE survey at least once across these three timepoints, with 22% (n = 14,856) completing all three timepoints.

It has been previously reported that research volunteers tend to be healthier and have higher socioeconomic status compared with the underlying source population30. In recent previous work, we found some evidence of ‘healthy volunteer bias’ among the COPE survey participants and demonstrated the utility of inverse-probability weighting in offsetting potential bias31. We thus calculated inverse-probability weights for COPE survey completion at each timepoint using sex assigned at birth, self-reported race and ethnicity, birthplace, educational attainment, marital/partnership status, health insurance status, employment status, home ownership and current age as predictors to use as weights in the primary analyses.

Social support

Perceived social support in the past month was measured using ten items from the RAND Medical Outcomes Study Social Support Survey Instrument12, using a five-point scale ranging from ‘1 = none of the time’ to ‘5 = all the time’. Per the original instrument12, these items can be classified into subtypes of tangible support (for example, ‘someone to take you to the doctor if you needed it’), emotional/informational support (for example, ‘someone to confide in or talk to about yourself or your problems’, ‘Someone to turn to for suggestions about how to deal with a personal problem’) and positive social interaction support (for example, ‘someone to have a good time with’)12 (for a list of all items and their corresponding subtypes, see Extended Data Table 5). After excluding participants who were missing responses for all ten social support items at each wave (N = 570 in May, N = 270 in June and N = 480 in July), an overall social support score was calculated per scoring recommendations as the mean rating across all completed items, and scores for each support subtype were calculated as the mean rating across all completed items for a given subtype. Scores ranged between 0 and 5 such that one unit increase (for example, moving from ‘some of the time’ to ‘most of the time’) reflects increased perceived availability of support. Correlations between subtypes ranged from r = 0.65 to r = 0.83 (Extended Data Table 6), with the strongest correlation between emotional/informational support and positive social interaction.

To further characterize the potential impact of social support subtypes and their combinations, we derived dichotomous indicators for individuals who reported above-average versus below-average (including average) levels of each support subtype. We then created an eight-level categorical exposure variable to indicate individuals who endorsed higher levels on all three support subtypes (all three types of social support), only two subtypes (both tangible and emotional/informational support; both tangible and positive social interaction; both emotional/informational support and positive social interaction), only one subtype (tangible support alone; positive social interaction alone; emotional/informational support alone) or no subtypes.


Depressive symptoms were assessed using the PHQ-9 (ref. 32), a nine-item checklist that assesses symptoms of depression in the past 2 weeks using a four-point scale response format (ranging from ‘0 = not at all’ to ‘3 = nearly every day’). The scale is scored by summing responses to all nine items, with total scores ranging from 0 to 27. Total scores of 5, 10 and 20 represent validated cut points for mild, moderate and severe depression32,33; scores in the moderate to severe range of 10 or above thus reflect clinically elevated symptoms of depression.


We used linked EHR data to establish whether participants had a pre-pandemic history of mood disorder diagnosis at any time before 21 January 2020, the date of the first reported COVID-19 case in the United States34, defined by two or more qualifying diagnostic codes mapped to ‘Mood disorder’ code 46206005 in the Systematized Nomenclature of Medicine (a common coding scheme for harmonizing different coding vocabularies or versions across health systems). Those without qualifying codes or, to be conservative, who did not have linked EHR data (available for 45,477 participants; 66%) were considered not to meet criteria for pre-pandemic history of mood disorder. Participants also responded to the COPE survey about past-month potential financial stressors, including not having enough money to pay for housing, gas/fuel, food or medications or not having a regular place to sleep or stay. Given that COPE survey respondents generally did not endorse numerous financial stressors, we binarized this variable to indicate the presence of at least one stressor. We also extracted information from the AoU baseline survey on sex assigned at birth, age at the working index date of the COVID-19 pandemic (20 January 2020), home ownership, employment status, educational attainment, health insurance status and self-reported race and ethnicity.

Statistical analysis

Although the COPE survey was administered multiple times over the course of the pandemic, not all participants completed each assessment. To accommodate both missingness and within-subject correlations across survey measurements35, we first fitted inverse-probability-weight-adjusted, mixed-effects logistic regression models using the lme4 R package to determine the time-varying relationships between social support and depression with subject-specific random intercepts and fixed effects for the three survey timepoints (May, June and July of 2020). In models testing main effects of overall social support and specific subtypes of social support, we adjusted for potential confounding factors, including sex, age, home ownership, employment status, educational attainment, health insurance status, and self-reported race and ethnicity. We performed two-tailed tests of statistical significance and did not adjust the models for multiple comparisons. As sensitivity analyses, we assessed these relationships where depression was modelled as continuous symptoms rather than meeting the moderate to severe level. Second, we similarly fitted a mixed-effects model to test the association between the categorical variable of support subtype combinations and depression, using the category with lower support on all three subtypes as the reference group. Third, we assessed binary factors—sex assigned at birth (female versus male), current age (below versus above 65), pre-pandemic mood disorder diagnosis (any versus none) and COVID-related financial stress (any versus none)—as potential modifiers of the association between social support and depression during the pandemic. As a sensitivity analysis, we also fitted logistic regression models testing lagged associations between overall social support at the first COPE survey wave and subsequent depression at the second wave only (one-month window) or at the second or third survey wave (two-month window), excluding individuals with elevated depressive symptoms at the first wave. To query data, perform statistical analyses and generate tables and figures, we used R version 4.1.0 in a Jupyter Notebook on the AoU Workbench Controlled Tier Version R2020Q4R2.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.