College student participants (N=20; female: 13/20, 65%; age: mean 19.80, SD 1.0 years) were assessed daily across a 3-month period during the 2020 COVID-19 pandemic (June to November), resulting in 1173 unique observations. Participants were eligible if they met the following criteria: participants must be unmarried, be English speakers, be full-time undergraduate students aged between 18 and 22 years, and own a primary Android smartphone that is compatible with the ecological momentary assessment (EMA) phone-based survey apps and study wearable devices.

This study was part of a larger intensive longitudinal study for examining student mental health that included physiological assessments, sleep tracking, and daily emotional and behavioral reports. The procedures of this study were approved by the institutional review board (approval number: 2019-5153) at University of California, Irvine. All individuals provided written informed consent prior to participation.

Herein, we describe the procedures that are relevant to the purposes of our investigation. Participants were first instructed on how to wear the noninvasive device (ie, Oura ring [Ōura Health Ltd]) that continuously assessed sleep, activity, and physiology throughout the day and during sleep [26]. Participants completed daily surveys on affect by using a smartphone app. To maintain high adherence, participants received reminders via text message, email, or phone call if there were more than 2 days of inactivity.

We first examined variables for normality and heteroscedasticity. To examine the association between sleep (TST, sleep efficiency, and SOL) and affect variables (PA, NA, and COVID-19 worry), we first conducted multilevel models by using the restricted maximum likelihood approach. This approach improves estimates of variance components and fixed effect SE estimates in smaller samples by separating the estimates of the fixed effects from the variance components [31]. We predicted PA, NA, and COVID-19 worry as a function of the fixed effects of time and between- and within-subject sleep variables while controlling for previous-day affect. Next, we examined the association between sleep and affect variables by using multiple regression models. Previous studies have found that sex is linked with sleep outcomes [32-34]. Therefore, sex was included as a covariate in these regression models. Finally, we conducted hierarchical linear regressions to examine whether sleep variability variables contribute to affect dynamics above and beyond the average sleep variables.

A total of 20 college students (female: 13/20, 65%) completed this study, providing 1623 (mean 43.49, SD 25.51 days/person) nights of usable Oura ring sleep data. Completion rates for EMA studies were high (83%).

Table S1 in Multimedia Appendix 1 provides descriptive statistics and the bivariate correlations between the key variables. The averages for participants’ TST and TST variability were 6.84 and 1.8 hours, respectively. Participants also experienced an average of 65.04 and 12.01 minutes of WASO and SOL, respectively, with a sleep efficiency of 86.52% across the study. Further, on average, participants reported low levels of NA (mean 21.79), moderate levels of PA (mean 45.27), and low levels of COVID-19 worry (mean 17.41) across the study. We assessed the bivariate correlations between the COVID-19 worry mean and PA and NA means (between-person level) and found no significant correlations between these constructs. Specifically, the correlation between the COVID-19 worry mean and PA mean was −0.19 (P=.43), and the correlation between the COVID-19 worry mean and NA mean was 0.43 (P=.06).

The multilevel models of the relation between sleep and affect revealed that participants with a higher SOL (b=−1.09, SE 0.36; P=.006) and TST on the prior day (b=−0.15, SE 0.05; P=.008) had lower PA on the next day, while controlling for previous-day PA. No within-subject differences were observed in predictions for next-day PA. No associations between daily sleep and NA and between daily sleep and COVID-19 worry were found. Table S2 in Multimedia Appendix 2 shows more details.

The regression model for predicting PA from average TST accounted for 34% of the variance in average PA (adjusted R²=0.26; F_2,17=4.24; P=.03). Specifically, higher TST was associated with lower PA (b=−0.36, SE 0.12; P=.009). Other sleep variables were not associated with average PA. Sleep was not associated with average NA or COVID-19 worry. Table S3 in Multimedia Appendix 3 shows more details.

Sleep patterns across the daily and long-term scales were associated with daily and average affect and affect variability, which were assessed over a 3-month period. These findings are consistent with those of prior work suggesting that poor sleep confers a heightened risk for affective disturbances that are prevalent in mood disorders, such as depression [4]. The link between sleep variability and affect variability may provide a window into how such patterns develop over time.

Individuals with longer sleep times on the previous night experienced lower PA on the next day. Similarly, individuals with longer average TSTs over the study period reported lower PA. Previous studies however have suggested a positive association between sleep duration and greater PA (eg, a study by Galambos et al [25]). During the COVID-19 pandemic, young adults’ lives have changed dramatically [35], and it is possible that they have been sleeping for longer than the recommended hours; therefore, there may be a curvilinear association between TST and PA. Future studies may benefit from examining this curvilinear association in a larger sample. Notably, consistent with prior studies [36], sleep did not predict NA. Low PA may be a potential pathway through which sleep duration has implications for mental health. Low PA is both a significant predictor of the onset of depression and a characteristic of depressive disorders. When low PA is directly treated, symptoms of anxiety and depression, as well as other disorders with anhedonic features, are known to improve [37]. Thus, improving daily sleep may be particularly important for preventing PA decline, thereby potentially interrupting the pathogenesis of illness states, such as mood disorders.

TST variability predicted higher affect variability across all affect domains (ie, PA, NA, COVID-19 worry); thus, it may be a proximal predictor of mood disturbances. Affect variability is prevalent in various mood-related disorders, such as depression, bipolar disorder, and anxiety disorders [38]. Our findings have important implications. First, interventions that support sleep stability may indirectly reduce affect variability and therefore prevent clinically significant mood disturbances. For example, in bipolar disorder, variable sleep and affect during euthymia predict worse long-term outcomes, including episodic relapses [39]. Second, our findings suggest that the stabilization of affect may be an early marker or predictor of the efficacy of transdiagnostic sleep interventions that target mood disorders with anhedonic features [39]. Future research may benefit from experimentally investigating the effect of regulating sleep time on affect dynamics in clinical populations.

Our findings also provide specific relevance to the COVID-19 context. Individuals with a higher SOL experienced higher COVID-19 worry variability, and individuals with a higher sleep efficiency reported lower COVID-19 worry variability. The specificity of the associations between SOL and sleep efficiency, and COVID-19 worry variability (ie, not NA in general) may suggest that the association between sleep and arousal-related affect is unique. Sleep disturbances may contribute to the lower capacity to adaptively overcome stress and therefore may be associated with higher stress-related sleep reactivity and cognitive presleep hyperarousal [40]. Future research may benefit from experimentally investigating the association between sleep and specific subtypes of affect.

Our findings carry some limitations. First, whether sleep causally predicts affect remains unclear. Daily affect may also predict multiple sleep indices [36]. The bidirectional link between sleep and affect may result in a cyclical pattern of sleep disturbance impacting affect, which in turn may contribute to greater sleep disturbances. Second, the small sample size of this study was not diverse in terms of age, which limits the generalizability of our findings to other age groups. However, the peak prevalence of affect variability is among adolescents and young adults (ie, individuals aged 16-24 years) and gradually declines with age [38]. Considering the small sample size of this study, future studies may benefit from examining the association between sleep patterns and affect dynamics in a larger sample.

Fluctuating sleep patterns are associated with affect dynamics, such as average PA and affect variability, across all affect domains (ie, PA, NA, COVID-19 worry) at the daily and long-term scales. Low PA and affect variability may be potential pathways through which sleep has implications for mental health. Interventions that target sleep stability may indirectly reduce affect variability and therefore prevent mood disorders. The stabilization of affect may be an early marker or predictor of the efficacy of transdiagnostic sleep interventions that target mood disorders with anhedonic features.