Sleep disturbances and depression are bidirectionally associated among college student athletes across COVID-19 pandemic exposure classes

Abstract

College athletes may be vulnerable to sleep disturbances and depression during the COVID-19 pandemic as a result of large shifts in social and athletic obligations. In a national sample of college athletes, we examined the associations between sleep disturbances and depression across two timepoints, using COVID-19 exposure as a moderator. Data were collected from 2098 NCAA Division I, II, and III college athletes during two timepoints, from April 10 to May 23, and from August 4 to September 15, 2020. First, a latent class analysis was conducted with five indicators of levels of COVID-19 exposure to determine different exposure profiles. Second, to examine the directionality of associations between sleep disturbance and depression, a cross-lagged panel model was added to the latent class membership structural equation model; this allowed for testing of moderation by COVID exposure class membership. Four highly homogeneous, well-separated classes of COVID-19 exposure were enumerated: Low Exposure (57%); Quarantine Only (21%); High Other, Low Self Exposure (14%); and High Exposure (8%). COVID-19 exposure class membership did not significantly moderate associations between sleep disturbances and depression. However, student athletes significantly differed in T2 depression by their COVID-19 exposure class membership. Depression and sleep disturbances were positively correlated at both timepoints (r_T1 = 0.39; r_T2 = 0.30). Additionally, cross-lagged associations were found such that T2 depression was associated with T1 sleep disturbances (β = 0.14) and vice versa (β = 0.11). These cross-lagged associations were not significantly affected by athletes’ level of COVID-19 exposure during the beginning of the pandemic.

1. Introduction

Collegiate athletics programs have begun to recognize that sleep is an important determinant of athlete mental health, well-being, and performance (Kroshus et al., 2019). However, it is unclear how the disruptions that resulted from the coronavirus (COVID-19) pandemic and cancellation of collegiate sports affected the sleep – mental health relationship. Given the well documented bidirectional relationship between sleep and depression (Fang et al., 2019; Khurshid, 2018) and the recently reported rise in prevalence rates for both sleep disturbances (Gao & Scullin, 2020; Marelli et al., 2020) and depression (Al Omari et al., 2020; Islam et al., 2020; Z.-H. Wang, Yang, et al., 2020; Yu et al., 2021) that emerged during the COVID-19 pandemic, we were specifically interested in examining this connection among collegiate athletes.

Prior to the COVID-19 pandemic, college athletes had been identified as particularly vulnerable to sleep disturbances due to intense training schedules, frequent travel across time zones, and juggling sport, academic, and social obligations (Brauer et al., 2019). Compared to non-athlete students, college athletes generally report higher rates of sleep disturbances: 39-61% report obtaining fewer than 7 h of sleep per night, compared to 36% of non-athlete students, and 19-42% report experiencing sleep disturbances, compared to 27% of non-athlete students (Becker et al., 2018; Brauer et al., 2019; Carter et al., 2020; Mah et al., 2018). This high prevalence of sleep disturbances is of particular concern because sleep disturbances are both a cause and consequence of adverse mental health outcomes (Kroshus et al., 2019), in particular, depression (Duffield et al., 2021). For example, previous epidemiological studies have shown that 29% of young adults who experience clinical levels of sleep disturbances (i.e., mild to severe insomnia) also report comorbid clinical depressive symptoms (Baglioni et al., 2011; Gress-Smith et al., 2015; Taylor et al., 2011). Similarly, the DSM-5 (American Psychiatric Association, 2013) lists sleep disturbances (i.e., insomnia or hypersomnia) as one diagnostic criterion of major depressive disorder and other mood disorders. Although well-established as a symptom of depression (American Psychiatric Association, 2013), prior evidence suggests associations between sleep disturbances and depression are bidirectional (Bowman et al., 2021; Ettensohn et al., 2016; Fang et al., 2019; Nguyen et al., 2022).

The bidirectional nature of this relationship is complex, yet there are several potential biobehavioral mechanisms that explain the co-occurrence (for a review, see Fang et al., 2019). First, Fang et al. (2019) propose an inflammation hypothesis that suggests the relationship between sleep disturbances and depression is mediated by elevated circulating inflammatory cytokine cascades (e.g., interleukin-6 and tumor necrosis factor-alpha) resulting from previously accrued sleep deficiencies. Second, Fang et al. (2019) also propose the monoamine hypothesis of the bidirectional nature of sleep disturbances and depression. This hypothesis assumes alterations in the levels of monoamines (e.g., serotonin, norepinephrine, and dopamine) are the cause of depression (Krishnan & Nestler, 2008). Dysregulation of these same monoamine neurotransmitters are also associated with rapid eye movement (REM) sleep abnormalities (Khurshid, 2018; Pace-Schott & Hobson, 2002). Third, there is some evidence to suggest a genetic overlap between sleep disturbances and depression (Gehrman et al., 2011; Lind et al., 2017; Stein et al., 2018). In other words, genes that influence insomnia also affect depression (Fang et al., 2019). Fourth, dysregulation of circadian rhythms may influence mood and contribute to the development of depressive symptoms (Fang et al., 2019; Monteleone et al., 2011; Patke et al., 2017). Finally, behaviors influenced by symptoms of either sleep disturbances or depression may subsequently lead to the other. For example, among college athletes specifically, sleep disturbances may impair daytime concentration and athletic performance, negatively affecting athletes’ ability to participate in their sport (e.g., sustaining physical injury, being sidelined), and exacerbating depressive symptoms (Bonnar et al., 2018; Fang et al., 2019; Fullagar et al., 2015). Conversely, depressive symptoms may reduce athletes’ energy or desire to participate in their sport, increasing stress or rumination, athletic absenteeism, and/or poor sleep hygiene behaviors (e.g., excessive daytime napping) that are associated with exacerbating (or greater) sleep disturbances (Sokić et al., 2021).

The complex bidirectional relationship between sleep disturbances and depression may be further complicated by the presence of external stressors. Indeed, since the start of the COVID-19 pandemic, epidemiological studies have documented a 27% increase in the prevalence of sleep disturbances (Gao & Scullin, 2020; Marelli et al., 2020). Further, prevalence rates of experienced depression symptoms in college student samples were reported to be as high as 57% during the pandemic, which represents a stark increase from pre-pandemic prevalence rates in college students which were closer to 34% (Al Omari et al., 2020; Gress-Smith et al., 2015; Islam et al., 2020; Z.-H. Wang, Yang, et al., 2020; Yu et al., 2021). In their study, Marelli et al. (2020) attributed prevalence rate increases to contextual factors related to the COVID-19 pandemic, such as social isolation, alterations to daytime routines and activities, exposure to the COVID-19 virus and/or fear of oneself or a loved one becoming ill with the life-threatening virus. College athletes, like nonathletes, have experienced these general stressors, but also have faced unique ones that resulted from the cancellation of collegiate sports, including a reduction in sport/physical activity, uncertainty about when sport competitions would be resumed, and ongoing questions about athletic eligibility and scholarships (Andreato et al., 2020; Di Fronso et al., 2020; Rubio et al., 2020). At the beginning of the COVID-19 pandemic, student-athletes’ worry about COVID-19 significantly predicted their psychological distress; when combined with having been quarantined, having experienced changes in class delivery, and having had their sport seasons cancelled, these variables accounted for over 50% of their general psychological distress (Moore, et al, 2022). What is not known, however, is how these contextual factors may relate directly, or indirectly, to how athletes experience sleep disturbances and depressive symptoms.

A particularly salient contextual factor is exposure to the COVID-19 virus. Indeed, exposure to the COVID-19 virus has been shown to be a risk factor for both sleep disturbances (Deng et al., 2021) and depression (Deng et al., 2021; Li et al., 2021; X. Wang, Yang, et al., 2020) in college student samples. Given the bidirectional relationship between sleep disturbances and depression, exposure to the COVID-19 virus may alter how depression and sleep disturbances influence each other. For example, exposure to the COVID-19 virus may exacerbate individuals’ stress or deplete their ability to cope with stress (Bridgland et al., 2021) that, in turn, may moderate the relationship between sleep disturbances and depression (Burke et al., 2005; Vargas et al., 2015). Specifically, higher levels of stress would be expected to strengthen the association between sleep disturbances and depression (Vargas et al., 2015). Although examined in parts within samples of nonathletes, to date, such relationships have not been examined among college athletes.

In regard to the sleep-depression relationship, and the potential influences of COVID-19 related factors, there are several critical limitations within the extant literature. First, most COVID-19 research on sleep disturbances and depression has focused on general samples of college students. Because college athletes have experienced unique stressors in relation to COVID-19, they may be particularly vulnerable to sleep disturbances and depression and thus should be studied separately (Uroh & Adewunmi, 2021). Second, of the few COVID-19 studies conducted with student athletes, the majority have been cross-sectional; thus, the directional effects of the COVID-19 pandemic on the sleep disturbances - depression relationship have not been tested. Third, there is considerable ambiguity in what constitutes direct and indirect exposure to COVID-19 (e.g., knowing others quarantined or diagnosed with COVID-19), and whether different types of exposure exert differential influence on associations between sleep disturbances and depression. Thus, to address these limitations, within the context of COVID-19 exposure, we examined the bidirectional relationship of sleep disturbances and depression in a large, nationally-based sample of collegiate athletes. To do so, we first determined distinct classes of COVID-19 exposure based on student-athletes reported direct and indirect experiences with the virus. Drawing from existing research and theory (Baglioni et al., 2011; Kroshus et al., 2019), we hypothesized that (1) greater sleep disturbances in April/May 2020 (T1) would be associated with greater depression symptom frequency in August/September 2020 (T2); (2) greater depression symptom frequency at T1 would be associated with greater sleep disturbances at T2; and (3) the bidirectional relationships between sleep disturbances and depression symptom severity would be moderated by COVID-19 exposure, such that the magnitude of the bidirectional relationships would be greater in college athletes who had the most direct and indirect exposure to COVID-19.

2. Methods

2.1. Procedure

The study was part of a larger investigation on the longitudinal effects of COVID-19 on college athlete well-being. Data were initially collected at two time periods. T1 was collected from April 10 through May 1, 2020 as an appendage to the end of the NCAA’s Student-Athlete Advisory Committees (SAAC) study (NCAA, 2020), as well as from April 17 through May 23, 2020 through solicitation of NCAA DI, DII, and DIII athletic departments. These data collection phases started 31 and 38 days, respectively, after the NCAA cancelled winter championships and spring sport seasons. Following consent, participating college athletes completed questionnaires online via Qualtrics. Participants who completed the first survey were invited by email to complete a three to four-month follow-up survey (T2). The T2 surveys were sent out August 4, 2020 and closed September 15, 2020. All procedures were approved by the University of North Texas Institutional Review Board. A total of 5913 athletes responded to the first invitation, of which, 2098 (35.5%) responded at T2.

2.2. Participants

Participants (N = 2098; 20.9% male; M _age = 20.10 years, SD _age = 1.27 years) who completed both the initial survey and three to four-month follow-up survey represented over 80 college athletic departments across NCAA DI (72%), DII (12%), and DIII (16%) and over 25 sports. See Table 1 for demographic characteristics of the sample.

Table 1.

Sample demographics.

Demographic Characteristic	Proportion
Gender Identity
Female	78.8%
Male	21.2%
Ethnic Identity
Hispanic	8.5%
Non-Hispanic	91.5%
Racial Identity
White/Caucasian	84.1%
Black/African American	7.3%
Native American/Alaska Native	0.6%
Asian/Asian American/Pacific Islander	3.3%
Middle Eastern/Arab	0.1%
Mixed	2.9%
Academic Year
Year 1	23.3%
Year 2	26.4%
Year 3	28.8%
Year 4	18.0%
Year 5 +	3.4%
Age
18 years old	8.6%
19 years old	26.3%
20 years old	28.9%
21 years old	22.6%
22 years old	11.1%
23+ years old	2.5%
NCAA Division
Division I	71.7%
Division II	12.3%
Division III	15.9%
Sport Type
Team	50.2%
Individual	49.8%
Sports (Most Frequent, >4%)
Track & Field	13.0%
Soccer	12.2%
Swimming & Diving	11.3%
Cross Country	8.5%
Softball	8.1%
Volleyball	6.1%
Rowing	6.0%
Basketball	4.3%

Note. Participants reported their age (18; 19; 20; 21; 22; 23+), year of study (1st Year; 2nd Year; 3rd Year; 4th Year; 5th or 6th Year), international student status (International Student; Not International Student [US student]), NCAA Division level (DI; DII; DIII). The sports were then grouped by type (Individual; Team).

2.3. Measures

Demographics and sports-related questions. Participants reported their age (18; 19; 20; 21; 22; 23+ years), gender identity (e.g., man; woman, nonbinary), race identity (White/Caucasian; Black/African American; Asian/Asian American/Pacific Islander; Mixed/Biracial), ethnicity (Hispanic/Latinx; Not Hispanic/Latinx), relationship status (Single; In A Relationship/Married), year of study (1st Year; 2nd Year; 3rd Year; 4th Year; 5th or 6th Year), international student status (International Student; Not International Student [US student]), NCAA Division level (DI; DII; DIII), and sport. The sports were then grouped by type (Individual; Team). Sport type was categorized this way to capture potential differences in social support provided by playing an individual vs. team sport.

PROMIS™ Sleep-Related Disturbances short form (PROMIS-SD-SF). The PROMIS-SD-SF is a short-form questionnaire of the PROMIS™ Sleep-Related Disturbances form, and is a widely used and well-validated instrument for measuring sleep disturbances (Yu et al., 2011). This scale contains 8 items on sleep-related disturbances over the past two weeks and was assessed at both the initial survey and the two-month follow-up survey (Buysse et al., 2010; Yu et al., 2011). The scale items assess qualitative aspects of sleep disturbances, such as: “My sleep was restless [not at all; a little bit; somewhat; quite a bit; very much]”; “I had trouble staying asleep [never; rarely; sometimes; often; always]”; and “My sleep quality was … [very poor; poor; fair; good; very good]” (Buysse et al., 2010; Yu et al., 2011). All items are rated on a 1 to 5 scale, with possible raw scores ranging from 8 to 40, and higher scores indicating more sleep-related disturbances. The short form scale correlates strongly with the longer form (r = .96) and with other similar measures, such as the Pittsburgh Sleep Quality Index (PSQI; r = 0.83). It also has greater measurement precision than the PSQI and the Epworth Sleepiness Scale (i.e., higher reliability with fewer total items) (Buysse et al., 2010; Yu et al., 2011). In the current study, the scale demonstrated good internal consistency at both T1 and T2 (both α′s = 0.92).

COVID-19 questions. COVID-19 items were developed based on exposure events that had been occurring within the U.S. since the beginning of the pandemic. Items included: “Did someone close to you, such as a family member, have to be quarantined due to COVID-19? (Yes/No)”, “Did you have to be quarantined due to COVID-19? (Yes/No)”, “Was someone close to you, such as a family member, diagnosed with COVID-19 (Yes/No)”, “Were you diagnosed with COVID-19 (Yes/No)”, and “Did someone close to you, such as in your family, become severely ill with, or die from, COVID-19 (Yes/No)”.

Patient Health Questionnaire (PHQ-2). The PHQ-2 is a short-form questionnaire of the PHQ-9, and is a widely used and well-validated instrument for measuring the severity of depressive symptoms (Kroenke et al., 2003). This scale contains 2 items that assess the frequency (from “Not at all” (0) to “Nearly all Days” (3)) with which individuals have experienced both affective and somatic symptoms related to depression and depressive disorder over the past two weeks. Compared with the PHQ-9, the PHQ-2 cut-off scores generally have statistically lower sensitivity or specificity to detect probable clinical depression. However, the PHQ-2 sensitivity to detect probable clinical depression is highest when compared to a semi-structured interview, which most closely replicates clinical interviews by trained professionals, rather than a full structured interview (Levis et al., 2020). Additionally, the PHQ-2 continuous scores correlate similarly to the PHQ-9 scores when associated with other similar measurements, such as the BDI-II (r _PHQ-2 = 0.74 vs r _PHQ-9 = 0.75) (Seo & Park, 2015). In the current study, the scale demonstrated good internal consistency at both T1 (α = 0.76) and T2 (α = 0.83).

2.4. Statistical analysis plan

Data Preparation. First, data were examined for normality based upon skew and kurtosis values, as well as examination of histograms for outliers. For the variables of interest, the first time point had missing data ranging from 0% (race) to 10.4% (PROMIS item 7); the second time point had missing data ranging from 2.6% (COVID item 3) to 9.8% (PROMIS item 4). However, Little’s Missing Completely at Random (MCAR) test was nonsignificant for T1 data after accounting for gender identity and race/ethnicity identities and for T2 data. Therefore, missing data were likely MCAR at T2 and missing at random (MAR) at T1 based upon gender and racial identities, which meets the full information maximum likelihood (FIML) assumption that data is missing due to a MAR process (Enders & Baraldi, 2018; Moore et al., 2020). Missingness is handled by FIML for latent class analysis (LCA), including the validity steps when other variables are added to the LCA model conducted within a structural equation model (SEM) framework (Masyn, 2013; Moore & Little, in press). The following analyses were conducted in Mplus version 8.4 (Muthén & Muthén, 2019).

Latent Class Analysis Enumeration. A k-fold or split sample approach was taken to build further support for the LCA enumeration solution by randomly splitting the data into two subsamples (2-fold; n = 1014; n = 1084) and independently conducting the LCA enumeration process on both halves (Grimm et al., 2017; Masyn, 2013; Moore & Little, in press). The LCA enumeration was done for 1 through 5 classes, since 5 classes is the largest number that can be accurately enumerated with five, binary indicators (Moore & Little, in press). Each class model was run using random starts beginning with 100 and 50 (starts and to completion) and maximized for each class solution at 10,000 and 5000. Class solution quality was assessed by the model convergence, log likelihood (LL) replication, condition quality number, and smallest class size. The best fitting solution for the 1 class through 5 class models were compared across the following fit metrics: lowest information criterion values (AIC, BIC, SABIC, CAIC, AWE), nonsignificant loglikelihood ratio tests (LRTs), higher Bayes Factor and correct model probability relative to most likely solution (cmP(K)), and relative improvement (improvement relative to the LL improvement of going from the 1- to 2-class solutions) (Masyn, 2013; Moore & Little, in press). These fit metrics must be examined together to make a gestalt decision about the best fitting model, knowing that the BIC and SABIC tend to identify the best fitting solution and AWE tends to reach its lowest value at a conservative point (i.e., 1- to 2-classes before the optimal solution) (i.e., 1- to 2-classes before the optimal solution; Nylund et al., 2007). The best fitting model characteristics were then examined to assess classification quality (entropy, model estimated proportion, modal class assignment proportion, Average Posterior Probabilities, and Odds of Correct Classification), class homogeneity (item endorsements less than 0.30 or greater than 0.70) and separation (odds ratios less than 0.20 or greater than 5). These metrics were examined to obtain evidence of the classification quality and within class homogeneity and distinction between the classes. The best fitting solution for the two random samples were then compared based upon their classification qualities, homogeneity, and separation. This best fitting model was then enumerated with the whole sample. The best fitting latent class solution was then used with the whole sample to examine relationships with the covariates.

Latent Class Analysis of Covariate Relationships. The SEM was expanded to include demographics, plus depression and sleep disturbances from T1 and T2 to examine the bidirectionality of the relationship between depression and sleep disturbances as well as moderation by latent class COVID-19 exposure membership. The demographic covariates (i.e., male, nonwhite, DI, and team) were included in the SEM to assess class membership differences, and control for their effects on the sleep and depression scores at both timepoints, as differences by these demographics have been seen previously (Duffield et al., 2021; Kroshus et al., 2019). Depression scores and sleep disturbances scores were included from T1 (April/May 2020) and T2 (Aug/Sept 2020) with their means, variances, and covariances varying by class membership to test for moderation by COVID exposure class.

We used the BCH method of adding covariates (Bolck et al., 2004) due to the sample size and the relatively high entropy (Moore & Little, in press); this method resulted in stable classes when the above variables were added. First, an omnibus, nested chi-square test of the covariates predicting class membership was conducted to determine if the covariates (gender, race, NCAA Division, and sport type) should be individually tested in follow-up comparisons. Second, the means and variances for the outcomes of class membership (T2 sleep disturbances and depression) were tested for differences by class membership with omnibus nested chi-square tests for each parameter. Finally, within time correlations and the cross-lag regressions between the T1 and T2 depression and sleep disturbances were tested for significant moderation by class membership with omnibus chi-square tests. Finally, 95% bias-corrected accelerated confidence intervals (95% BCa CI) were determined with 5000 bootstrapped samples.

3. Results

3.1. COVID-19 LCA class enumeration

First, we conducted the LCA class enumeration separately on the two random halves of the data to determine the best fitting solution in each random half; we then compared the best fitting solution across those two random subsamples. A gestalt review of the fit indices, the smallest class size, and interpretability of the class solutions supported the four-class solution for both subsample 1 and subsample 2 (Table 2 ). The classification quality of the four-class solution was supported for both samples, including having high overall entropy, Average Posterior Probabilities, and Odds of Correct Classification (Table 3 ). Next, the item homogeneity pattern for the classes was similar across the two subsamples and the majority of the item probabilities supported within class homogeneity of responses (Figure 1 ). The odds ratios of the item probabilities comparing classes pairwise supported that the classes were distinct (i.e., separated) from each other (Table 4 ).

Table 2.

LCA model fit information for COVID exposure in subsamples 1 and 2.

									RI	H0: K classes; H1: K+1 classes
	Model (K-class)	LL	npar	AIC	BIC	SABIC	CAIC	AWE	(K,K+1)	LRTS	Adj LMR	Boot-strapped	BF(K, K+1)	cmP(K)
Sample 1 (n = 1014)	1-class	−2224.345	5	4458.69	4483.30	4467.42	4488.30	4532.91	na	622.90	<.001	<.001	0.000	0.000
	2-class	−1912.897	11	3847.79	3901.93	3867.00	3912.93	4011.07	na	74.99	0.0015	<.001	0.000	0.000
	3-class	−1875.400	17	3784.80	3868.47	3814.47	3885.47	4037.14	0.12	10.58	0.0562	0.0789	36.524	0.973
	4-class	−1858.233	23	3762.47	3875.66	3802.61	3898.66	4103.86	0.06	12.07	0.017	0.030	2.50E+06	0.027
	5-class	−1852.200	29	3762.40	3905.13	3813.02	3934.13	4192.86	0.02	na	na	na	na	0.000
Sample 2 (n = 1084)	1-class	−2345.655	5	4701.31	4726.25	4710.37	4731.25	4776.19	na	674.12	<.0001	<.0001	0.000	0.000
	2-class	−2008.596	11	4039.19	4094.06	4059.13	4105.06	4203.94	na	88.43	<.0001	<.0001	0.000	0.000
	3-class	−1964.382	17	3962.76	4047.57	3993.57	4064.57	4217.37	0.14	10.58	0.0562	0.0789	0.119	0.106
	4-class	−1941.285	23	3928.57	4043.30	3970.25	4066.30	4273.04	0.07	4.30	0.335	0.375	1.48E+08	0.894
	5-class	−1939.134	29	3936.27	4080.93	3988.82	4109.93	4370.60	0.01	na	na	na	na	0.000

Note. SABIC (Sample-Adjusted BIC); CAIC (Consistent AIC); Approximate Weight of Evidence (AWE); Relative Improvement (RI); LL Ratio Test Statistic (LRTS); Bayes Factor (BF); Approximate Correct Model Probability (cmP).

Table 3.

Classification quality for the 4-class model solutions for subsamples 1 and 2, and the whole sample.

Subsample 1 (Entropy = .906)	Model Estimated
	Proportion	95% C.I.	mcaP	AvePP	OCC
High COVID Exposure	7.8%	[.028, .106]	10.1%	1.000	11.76
Low Self-, High Other-Exposure	16.1%	[.127, .209]	13.6%	0.841	27.52
Quarantined Only	22.2%	[.197, .311]	20.6%	0.922	41.32
Low Exposure	53.8%	[.444, .565]	55.7%	0.999	858.12
Subsample 2 (Entropy = .889)
High COVID Exposure	11.2%	[.089, .140]	9.6%	0.855	6.77
Low Self-, High Other-Exposure	11.2%	[.089, .131]	12.1%	1.000	7.97
Quarantined Only	19.7%	[.156, .290]	18.6%	0.886	31.66
Low Exposure	57.9%	[.472, .621]	59.7%	0.991	79.97
Whole Sample (Entropy .880)
High COVID Exposure	8.4%	[.089, .140]	9.4%	0.989	10.85
Low Self-, High Other-Exposure	13.9%	[.089, .131]	13.0%	0.915	5.67
Quarantined Only	21.0%	[.156, .290]	19.9%	0.879	27.28
Low Exposure	56.7%	[.472, .621]	57.7%	0.973	27.49

Note. mcaP (modal class assignment Proportion); AvePP (Average Posterior Probabilities) > .80 represent good posterior probability assignment to modal class; OCC (Odds of Correct Classification) which is odds of model estimated class assignment relative to random assignment by class proportion; OCC >5 supporting adequate class separation and precision.

Figure 1.

Item Probability Profiles for COVID Exposure for Subsample 1, 2, and the full sample

Note. Class 1 (High COVID Exposure); Class 2 (Low Self-, High Other-Exposure); Class 3 (Quarantined Only); Class 4 (Low Exposure).

Table 4.

Class homogeneity and separation for the 4-class solution with the whole sample.

Items	Class Homogeneity (Probability of Item Endorsement)				Class Separation (Odds Ratios)
	Class 1 (8%)	Class 2 (14%)	Class 3 (21%)	Class 4 (57%)	Class 1 v 2	Class 1 v 3	Class 1 v 4	Class 2 v 3	Class 2 v 4	Class 3 v 4
Quarantined: OTHER	0.983	0.859	0.630	0.105	9.58	34.30	500.00	3.58	52.20	14.60
Quarantined: SELF	1.000	0.099	0.868	0.024	2.99E+07	4.96E+05	1.31E+08	0.017	4.37	263.00
Diagnosed with COVID-19: OTHER	1.000	1.000	0.000	0.003	1.00	1.07E+13	1.18E+09	1.07E+13	1.18E+09	1.10E-04
Diagnosed with COVID-19: SELF	0.300	0.011	0.068	0.000	37.20	5.89	1.40E+06	0.158	3.78E+04	2.38E+05
Severe COVID-19	0.189	0.209	0.023	0.002	0.882	9.89	96.80	11.20	110.00	9.79

Note. The bold class homogeneity values support the class members being homogeneous in their endorsement (>0.70) or nonendorsement (<0.30) of the item. The bold class separation values support the two classes are separated in their responses on the given item (<0.20 or >5).

Class 1 (8%) was named High COVID-19 Exposure (self and others), because there were very high probabilities for members having been quarantined, someone close to them quarantined, and being diagnosed with COVID-19 or knowing someone diagnosed with COVID-19; further, although there was a low probability overall across classes, members of this class also had the highest probability of being diagnosed with COVID-19 themselves and second highest probability of having or knowing someone severely ill with COVID-19. Class 2 (14%) was named High Other, Low Self Exposure, because there was high probability for members knowing someone quarantined and diagnosed with COVID-19 combined with low probability of themselves being quarantined and/or diagnosed with COVID-19; further, although an overall low probability, members of this class had the highest probability of knowing someone who had a severe case of COVID-19. Class 3 (21%) was labeled Quarantine Only, because this was the only exposure item the members consistently endorsed, followed by knowing someone else quarantined; they or someone they knew being diagnosed with minor or severe COVID-19 had a very low probability. Class 4 (57%) was labeled Low Exposure, because the probability of these members experiencing any type of COVID-19 exposure was very low.

3.2. COVID-19 LCA covariate results

First, an omnibus test of the covariates predicting class membership was conducted; the change in chi-square supported further exploration of these covariates ($\Delta {\chi }_4^2$ = 28.276, p < .001). Gender and team/individual sport status did not significantly predict class membership. However, NCAA Division status and race/ethnicity did significantly distinguish class membership. Specifically, relative to membership in the Low Exposure Class, DI student-athletes were more likely than DII or DIII student-athletes to be members of the High COVID-19 Exposure Class (p = .006). Furthermore, compared to members in the Low Exposure Class, athletes of color were more likely than white athletes to be members of the High Other, Low Self Exposure (p = .001) and Quarantine Only (p = .03) classes.

The outcomes were tested for significant mean differences by class membership. Class members did not significantly differ in their T2 depression variance ($\Delta {\chi }_3^2$ = 1.725, p = .63); and T2 sleep disturbances mean ($\Delta {\chi }_3^2$ = 9.294, p = .03) and variance ($\Delta {\chi }_3^2$ = 1.595, p = .66). Class members did significantly differ in their T2 depression ($\Delta {\chi }_3^2$ = 14.539, p = .002). Specifically, Only-quarantine Exposure Class members’ depression (M = 1.76, 95% BCa CI [1.497, 1.935]) was significantly higher than Low Exposure Class members’ depression (M = 1.41, 95% BCa CI [1.272, 1.601]).

Finally, within time correlations and across time regression coefficients of the cross-lagged portion of this SEM analysis were tested for moderation by class membership. No significant moderation by class membership was found. The following relationships held across class after controlling for participants’ gender, race/ethnicity, NCAA division, and sport type: Depression and sleep disturbances were significantly (p < .001), positively correlated at both T1 (r = 0.39) and T2 (r = 0.30). T2 depression was significantly predicted by T1 depression (β = 0.36, p < .001) and sleep disturbance (β = 0.14, p < .001). T2 sleep disturbances was significantly predicted by T1 sleep disturbances (β = 0.46, p < .001) and depression (β = 0.11, p < .001; See Figure 2 ).

Figure 2.

Structural Equation Model Results including the Mean Differences in T2 Depression by COVID Exposure Class Membership

Note. Dotted paths were kept in the model regardless of significance to control for covariate effects. The solid lines represent significant relationships in the model. The differential prediction of class membership and depression means by membership are shown; however, the significant pairwise comparisons are presented in the manuscript text.

4. Discussion

The current study examined the relationship between sleep disturbances and depression in college athletes as moderated by their COVID-19 exposure class in a large, nationally-based sample of college athletes. As hypothesized, we found variability in athletes' COVID-19 exposure as well as significant bidirectional relationships between sleep disturbances and depression across the three-to four-month period that represented the beginning of the cancellation of collegiate sports and athletes return to campuses in fall 2020. Contrary to our hypothesis, however, the athletes’ COVID-19 exposure class membership did not moderate these relationships. We detail our findings, and their implications, in the context of existing literature.

Both race/ethnicity and NCAA Division status significantly predicted class membership. Division I student athletes were more likely than Division II or III student athletes to be a part of the High COVID-19 exposure class, which may be due to the expectations and opportunities that were in place for these student-athletes. For example, Division I athletes, particularly those within revenue sports (e.g., football), were expected to play in competitions upon returning to campuses during Fall 2020, whereas athletes in the other NCAA Divisions did not have such expectations because many had their sports championships cancelled or shifted to Spring 2021 (Bengel, 2020). Athletes of color were more likely to be a part of the High Other, Low Self and Quarantine Only exposure classes. This difference is consistent with what was happening within the broader U.S. society where historically marginalized individuals (e.g., racial/ethnic minorities) were more likely to be employed in essential work positions, and thus more likely to be exposed to (and become infected with) COVID-19 (Tai et al., 2021).

Consistent with the study hypotheses, we found a bidirectional relationship across time between sleep disturbances and depression in our sample of college athletes. Greater sleep disturbances at T1 were associated with higher levels of depression approximately three months later. Specifically, every-one standard unit increase in T1 sleep disturbances was associated with a 0.14 increase in T2 depression severity scores even after accounting for T1 depression, gender, race, NCAA division, and sports type, and COVID-19 exposure class. Additionally, greater depression symptoms at T1 was associated with elevations in sleep disturbances approximately three months later. Specifically, every-one standard unit increase in T1 depression symptoms was associated with a 0.11 increase in T2 sleep disturbances, accounting for T1 sleep disturbances, gender, race, NCAA division, sport type, and COVID-19 exposure class. Our results corroborate recently reported estimates in college athlete samples taken during the COVID-19 pandemic (Duffield et al., 2021; Grandner et al., 2021). Additionally, findings are consistent with pre-COVID research (Baglioni et al., 2011) where the effect of sleep on subsequent depression has been shown to be slightly larger than the effect of depression on subsequent sleep. Our findings improve upon prior cross-sectional research that has documented the co-occurrence of sleep disturbances and depression in college athletes prior to COVID-19 (Duffield et al., 2021). Furthermore, our study extends research from non-athlete student samples (Fang et al., 2019; Khurshid, 2018), establishing the prospective, bidirectional relationship of sleep disturbance and depression within collegiate athletes and during a global pandemic. A next step would be to examine potential group differences in bidirectional sleep-depression associations in college-athlete and non-athlete student samples simultaneously to identify potential protective and risk factors across groups.

In contrast to our study hypothesis, COVID-19 exposure class membership did not moderate the bidirectional relationships between sleep disturbances and depression across time. In other words, the strength and direction of the associations between sleep disturbances and depression were consistent across the sample, regardless of the athletes differential direct, and indirect, exposure to COVID-19. One potential explanation for our results is that college athletes were quick to adapt to the COVID-19 circumstances and experienced lower negative mental health outcomes over time through routine engagement in sport or physical activity (Di Fronso et al., 2020; Rubio et al., 2020). Indeed, one study comparing the impact of the COVID-19 pandemic on mental health and well-being in elite athletes, recreational athletes, and non-athletes found that both recreational and elite athletes had less depression and greater well-being than non-athletes or athletes who had become physically inactive (Sokić et al., 2021). These findings suggest that routine participation in sport and physical activity may be a protective factor against negative mental health outcomes due to COVID-19 exposure and disturbances to social, academic, and sport schedules. Based on our findings, a next step for further investigation would be to directly compare sleep-depression associations, as moderated by COVID-19 exposure, in college athlete and non-college athlete samples to identify if significant group differences exist. Such comparison studies may be helpful to identify those at most risk during global health crises.

Despite the nonsignificant moderator findings, exploratory mean comparisons in sleep disturbances and depression by COVID-19 exposure class membership revealed that class members significantly differed in average T2 depression. Specifically, Quarantine Only class members had higher reported levels of T2 depression than Low Exposure class members. This result mirrors a multitude of studies showing that quarantined respondents have a higher likelihood of exhibiting symptoms of depression than those not quarantined (Kaparounaki et al., 2020; Sun et al., 2021; Tang et al., 2021). At T2 in our study, the athletes had returned to their campuses and were participating in both school and their sports under strict medical protocols. Consistent with college/university athletic departments’ and NCAA’s policies, quarantined athletes were removed from their teams, placed by themselves in rooms within designated quarantine buildings, and not allowed to exercise or engage in any physical training activities. In addition, contact with others (e.g., teammates, friends, family) was highly restricted. Thus, it is not surprising that athletes who were in quarantine, and experiencing such social isolation and loss of physical activity, would be experiencing more depressive symptoms than those with low exposure (and thus not in quarantine). Prior studies have shown that social support (Hagiwara et al., 2017) and physical activity (Wyshak, 2001) during COVID-19 are negatively associated with depression; generally, athletes who are more physically active or who have higher levels of support have reported lower depression (Rubio et al., 2020; Sokić et al., 2021; Şenışık et al., 2021).

4.1. Implications and study limitations

Study results inform both theory and our current understanding regarding the impact of the COVID-19 pandemic on college athletes’ mental health. First, even during the ongoing pandemic, relationships across time for both sleep disturbances and depression were bidirectional and remained stable. Although the college athletes had been differentially exposed to COVID-19, their exposure class membership did not moderate these relationships. Second, college athletes classified as Quarantine Only class members had greater average T2 depression scores than their Low Exposure peers. Our study’s findings are particularly relevant in the lives of college athletes because sleep disturbances (Bonnar et al., 2018; Fullagar et al., 2015) and depression (Moore et al., 2018) have been shown to have a negative effect on sport performances and injury risk (Bonnar et al., 2018; Fullagar et al., 2015; Milewski et al., 2014; Moore et al., 2018).

Our findings suggest that approaches (e.g., therapies, psychological skills training, sleep education) employed by athletic departments and sport clinicians to treat clinical sleep disturbances and depression pre-COVID-19 pandemic may remain effective under highly stressful situations, such as the COVID-19 pandemic. Targeting athletes’ sleep problems may be one way to improve their mood and targeting mood may have downstream beneficial effects on their sleep. Additionally, athletic departments and sport clinicians should prioritize maintenance of routine participation in sport and physical activity to potentially reduce negative mental health impacts of COVID-19 exposure and quarantine (Rubio et al., 2020; Sokić et al., 2021; Şenışık et al., 2021). Our results also highlight the need for targeted mental health interventions for college athletes who are required to quarantine.

Although this study had several methodological strengths (e.g., large nationally-based sample, two time-point design), some limitations need to be acknowledged. First, pre-COVID-19 data were not collected from participants for either sleep disturbances or depression; therefore, we cannot conclude if the COVID-19 pandemic increased the prevalence of either within our sample at T1. However, generally identifying the COVID-19 pandemic as a risk factor for the prevalence of sleep disturbances or depression was not the specific aim of the study. Instead, we were interested in examining if group differences in COVID-19 virus exposure classes influenced relations between sleep disturbances and depression across time among student-athletes in the context of the spread of the first wave of the COVID-19 virus in the United States between April/May 2020 to August/September 2020. Generally, results provide insights in how rapidly progressing global stressors (e.g., pandemics) may influence well-established relationships between sleep disturbances and depression in college athletes. Second, we did not examine other potential contextual factors regarding COVID-19 pandemic related stressors and challenges (e.g., social isolation, reduction of on-campus activities) or other life events that may affect both sleeping patterns and mood (Andreato et al., 2020; Di Fronso et al., 2020; Rubio et al., 2020). Given the timing of our initial data collection, which preceded publications that identified contextual factors regarding the COVID-19 pandemic (Andreato et al., 2020; Di Fronso et al., 2020; Rubio et al., 2020), we were unaware of the subsequent findings from these studies and focused on what was known at the time. With the knowledge that has accumulated from such COVID-19 studies, future research may consider a broader range of contextual factors of the COVID-19 pandemic and life events that may moderate associations between sleep disturbances and depression to better understand underlying risk factors that contribute to these associations among college athletes.

Third, because the focus of the parent study was to examine the within-population psychological well-being of collegiate athletes, we did not collect a comparison sample of non-athlete college students. Thus, the generalizability of our results is limited to athletes and suggests that future studies could examine COVID-19 exposure as a moderator of sleep-depression relationships in non-athlete college students. Fourth, although we used a person-centered approach (i.e., LCA) for these analyses, the data were collected around two changes consistent across college athletes’ lives (i.e., closing of campuses in April/May 2020 and initial re-starting of training and classes in August/September 2020), which may not have been the optimal intervals for capturing fluctuations in how individuals’ depression and sleep disturbances related during the beginning of the COVID-19 pandemic. Although we were able to make inferences about temporal changes and directionality from the results of our cross-lagged panel model, we acknowledge that a two time-point design is a limited longitudinal approach. Future studies should consider employing more intensive longitudinal methods to track dynamic changes between sleep disturbances and depression in the context of the COVID-19 pandemic or any subsequent pandemic that may unfold globally. Finally, athletes’ specific location data were not collected; therefore, we were unable to examine regional differences in infection rates and public health response strategies. Future studies should attempt to replicate our results with all the aforementioned factors to better account for any unexplained variance within our models.

As the ongoing COVID-19 pandemic continues to effect college athletes, researchers must continue to examine the specific consequences of the COVID-19 pandemic on athletes’ mental health and well-being. Although several studies have documented effects of the COVID-19 pandemic on both sleep disturbances and depression (Gao & Scullin, 2020; Marelli et al., 2020; Yu et al., 2021), only a few studies have investigated the impacts of the COVID-19 pandemic on sleep disturbances and depression among college athletes. Our results extend these findings by examining the impact of COVID-19 exposure class membership on the bidirectional relationships between sleep disturbances and depression in a college athlete sample, a group for whom these issues may be particularly salient (Duffield et al., 2021; Kroshus et al., 2019). The current study provides a methodological and statistical framework to investigate other potential mental health relationships impacted by COVID-19 exposure (e.g., mood, anxiety, PTSD) that may interact with sleep disturbances. Our findings also provide initial support for the idea that simultaneously targeting college athletes’ sleep and mood may result in greater improvements than targeting either alone.

Funding

This study was funded by the University of North Texas through a College of Liberal Arts & Social Sciences (CLASS) grant for the UNT Center for Sport Psychology.

Declaration of competing interest

We, the authors, have no financial interests or benefits in this research to disclose.

Data availability

Data will be made available on request.