Dynamic effects of psychiatric vulnerability, loneliness and isolation on distress during the first year of the COVID-19 pandemic

Abstract

The COVID-19 pandemic's impact on mental health is challenging to quantify because pre-existing risk, disease burden and public policy varied across individuals, time and regions. Longitudinal, within-person analyses can determine whether pandemic-related changes in social isolation impacted mental health. We analyzed time-varying associations between psychiatric vulnerability, loneliness, psychological distress and social distancing in a US-based study during the first year of the pandemic. We surveyed 3,655 participants about psychological health and COVID-19-related circumstances every 2 weeks for 6 months. We combined self-reports with regional social distancing estimates and a classifier that predicted probability of psychiatric diagnosis at enrollment. Loneliness and psychiatric vulnerability both impacted psychological distress. Loneliness and distress were also linked to social isolation and stress associated with distancing, and psychiatric vulnerability shaped how regional distancing affected loneliness across time. Public health policies should address loneliness when encouraging social distancing, particularly in those at risk for psychiatric conditions.

Extended Data Fig. 1 |. Associations between social isolation, distress, and loneliness.

To differentiate between subjective loneliness and objective social isolation, we measured associations between objective social isolation, distress, and loneliness. Social isolation was treated both continuously (Household size) and categorically (Living alone vs Living with others). We report results of linear mixed models using both frequentist statistics (thresholded at p < .001 two-sided, without multiplicity correction) and Bayesian statistics (practical significance defined as <2.5% of posterior estimates in region of partial equivalence [ROPE].) Boxplots present medians, first and third quartiles, and 1.5 × the interquartile range (whiskers). Gray circles denote the mean for each category. A) Histogram of mean household size. 768 participants (21.36%) of participants reported that they lived alone at baseline; 649 participants reported living alone at every timepoint throughout their participation. B) Psychological distress was positively associated with household size (B = 0.29, CI = [0.19, 0.40], p < .001), such that an increase of one additional household member was associated with an increase of 0.29 units of distress, although this effect was consistent with the null hypothesis based on Bayesian models (99.8% in ROPE; see Supplementary Table 8). C) Respondents who lived alone reported less distress than those who lived with others (B = −0.56, CI = [−0.76,− 0.39], p < .001), and this effect was of undecided significance based on Bayesian models (10.69% in ROPE; Extended Data Table 1). Loneliness still predicted distress when controlling for Household Size or Living Alone (see Extended Data Table 1 and Supplementary Table 8). D) In contrast to Psychological Distress, Loneliness was negatively associated with household size (B = −0.28, CI = [−0.33, −0.22], p < .001), such that an increase of one additional household member was associated with a reduction of 0.28 units loneliness, and this effect was practically significant (0.11% in ROPE; see Supplementary Table 9). E) Consistent with results of continuous models, individuals who lived alone reported an increase of 0.49 units loneliness compared to those living with others (B = 0.49, CI = [0.46, 0.52], p < .001), and differences were practically significant (0% in ROPE; see Extended Data Table 2).

Extended Data Fig. 2 |. Impact of social distancing-related stress.

We evaluated associations between self-reported social distancing and mental health during the pandemic. Each figure depicts a random subset of 1000 participants. Although we observed statistically significant associations with numerous measures of distancing based on linear mixed models using frequentist statistics (see Supplementary Tables 14–17), the only practically significant predictor of psychological distress and loneliness based on Bayesian statistics was individual differences in self-reported stress associated with distancing. We note that p-values are two-sided and do not include multiplicity correction. A) Individuals who reported higher average distancing-related stress also reported higher psychological distress across the pandemic (B = 0.5, CI = [0.44, 0.55], p < .001), such that an increase in one unit of average distancing-related stress was associated with 0.5 units higher psychological distress. This effect was practically significant when modeled alone (0.13 in ROPE; see Supplementary Table 14) and of undecided significance when controlling for other social distancing measures (4% in ROPE; see Extended Data Table 2). B) We also observed positive associations across individuals between mean distancing-related stress and mean loneliness (B = 0.25, CI = [0.23, 0.28], p < .001), such that an increase in one unit of average distancing-related stress was associated with 0.25 units higher loneliness. This effect was practically significant based on Bayesian models whether or not other distancing measures were included in the model (0.02% in ROPE; see Extended Data Table 3 and Supplementary Table 15).

Fig. 1 |. Study description.

a, Between 4 April 2020 and 13 November 2020, 3,655 participants enrolled in a 6 month study that consisted of a set of internet-based questionnaires to be completed every 2 weeks. Data collection proceeded from April 2020 through May 2021. b, Participants were invited to complete questionnaires for 24 weeks. At each interval, participants were asked to complete the psychosocial impact of COVID-19 survey, which included questions about social context as well as the three-item loneliness scale, as well as the Kessler-5 and DSM-XC. Additional questionnaires were administered at baseline, which were used to compute a PPS and regional estimates of social distancing based on zip code (Methods), and at the end of the study. The current paper focuses on the relationship between loneliness and psychological distress, and whether these factors vary as a function of one’s likelihood of having a psychiatric diagnosis and social distancing. c, Baseline questionnaire data from 174 participants who had previously undergone structured clinical interviews for diagnosis at NIH were used to construct a classifier to predict each participant’s likelihood of having had a psychiatric diagnosis. This classifier was applied to baseline questionnaire data from all participants to generate a PPS for each individual. For complete details, see ref. . d, Participants represented all US states and territories, as well as 16 countries outside of the United States. Zip code information for US participants (n = 3,614) was used to supplement self-report data with regional estimates of social distancing based on cell phone mobility data (Methods). PPS, patient probability score; NIEHS, National Institute of Environmental Health Sciences.

Fig. 2 |. Associations between distress, loneliness and PPS over time during the COVID-19 pandemic.

Analyses focused on psychological distress, a mental health outcome measure operationalized through biweekly responses on the Kessler-5 scale, as a function of time, loneliness and PPS. Each figure depicts a random subset of participants, with locally estimated scatterplot smoothing regression to capture the overall trend (purple line). We depict only findings from linear mixed models that were practically significant based on Bayesian models (<2.5% of posterior estimates in region of partial equivalence (ROPE)) and statistically significant at P < 0.001 (two-sided) in frequentist models to account for the large sample size. Multiplicity correction was not applied. For complete results, see Table 2. a, PPS was positively associated with average psychological distress (B = 0.77(0.02), confidence interval (CI): [0.72, 0.82], P < 0.001; 0% in ROPE), such that mean psychological distress across time was 0.77 units higher in individuals with a likely diagnosis (purple) relative to those likely to have no diagnosis (yellow) based on PPS. b, We observed positive associations between average loneliness and average distress (B = 0.95(0.03), CI: [0.89, 1.00], P < 0.001; 0% in ROPE), such that individuals with 1 unit higher loneliness reported 0.95 units higher distress across time. c, Changes in loneliness over time within individuals were also positively associated with changes in psychological distress (B = 0.57(0.02), CI: [0.53, 0.60], P < 0.001; 0% in ROPE), such that an increase of 1 unit loneliness was associated with an increase of 0.57 units distress at that time. d, Although psychological distress was positively associated with both loneliness and PPS, associations between PPS and average loneliness were only moderately correlated (r = 0.44, P < 0.001), indicating that psychiatric vulnerability and loneliness capture separate constructs.

Fig. 3 |. Relationships between psychiatric vulnerability, regional distancing and loneliness vary over time.

We used linear mixed models to evaluate associations between regional estimates of social distancing, time and loneliness as a function of PPS (Extended Data Table 5). a–c, Model predictions (a) and observed data (b,c) as a function of average regional distancing based on quartiles. Models revealed interactions between all factors (B = 0.64, CI: [0.36, 0.91], P < 0.001, 0.08% in ROPE), such that individuals in communities of low regional distancing (top rows) exhibited stable relationships over time, reporting higher loneliness at times of less distancing, whereas individuals in regions with higher rates of distancing (bottom row) showed changes in associations between distancing and loneliness over time. This effect also interacted with PPS (B = 0.39, CI: [0.27, 0.51], P < 0.001, 0.06% in ROPE): individuals with low likelihood of having a psychiatric diagnosis based on PPS (yellow) showed changes in the relationship between distancing and loneliness over time regardless of regional distancing rates, such that more regional distancing was associated with higher loneliness in late intervals, whereas those with a high probability of having a psychiatric diagnosis based on PPS (purple) showed no change over time if they lived in areas with low rates of social distancing (top row), and generally reported more loneliness at times of less distancing in the community. Panel a depicts marginal effects, with error bands representing the 95% confidence intervals. Spaghetti plots in panels b and c reflect linear regressions between regional distancing and loneliness for each of 700 randomly selected participants as a function of regional distancing. Panel b depicts relationships during early enrollment (first quartile of intervals) while panel c depicts late enrollment (fourth quartile of intervals). All P values are two-sided and do not include multiplicity correction.

Fig. 4 |. Loneliness mediates associations between self-reported distancing and psychological distress.

We used mediation analyses to evaluate whether the relationship between distancing-related stress and psychological distress was mediated by changes in loneliness. PPS and living alone were treated as moderators, and we report results with and without moderators in Extended Data Table 6. Coefficients in Fig. 4 depict results from models including moderators. a, Single-level mediation was used to examine relationships across participants. There was a significant association between distancing-related stress and loneliness (path a), such that individuals who reported higher distancing-related stress on average also reported higher loneliness. This effect was moderated by living alone, such that associations were stronger in those living alone (dark red) than in those living with others (tan); error bands denote standard error of the means. Loneliness was positively associated with distress, while controlling for distancing-related stress (path b), such that lonelier individuals reported higher psychological distress regardless of distancing-related stress. This effect was moderated by PPS, such that associations were stronger in individuals with who were more vulnerable. We observed significant mediation by loneliness (path a*b), such that individual differences in loneliness explained 43.5% of the variance between distancing-related stress and psychological distress. b, We used multilevel mediation to examine whether changes in loneliness also mediated dynamic associations between distancing-related stress and psychological distress over time within individuals. Results were highly consistent with findings across individuals. Participants reported higher loneliness at time points when they reported higher distancing-related stress (path a), and this effect was stronger in those living alone. Fluctuations in loneliness were positively associated with psychological distress when controlling for distancing-related stress (path b), and these within-person associations were stronger for individuals with high PPS scores. Finally, the dynamic association between distancing-related stress and psychological distress was reduced when controlling for fluctuations in loneliness (path a*b), such that loneliness explained 25.5% of the total effect. All P values are two-sided and do not include multiplicity correction. ***, P < .001; *, P < .05.