An Examination of Bidirectional Associations between Physical Activity and Mood Symptoms among Individuals Diagnosed and at Risk for Bipolar Spectrum Disorders

Abstract

Objectives:

Activation, a construct including energy and activity, is a central feature of Bipolar Spectrum Disorders (BSDs). Prior research found motor activity is associated with affect, and this relationship may be stronger for individuals with BSDs. The aims of this study were to investigate bidirectional relationships between physical activity and mood and evaluate whether bipolar risk status moderated potential associations.

Methods:

Young adults at low-risk, high-risk, and diagnosed with BSD participated in a 20-day EMA study in which they wore an actiwatch to measure physical activity and sleep/wake cycles. They also reported depressive and hypo/manic symptoms three times daily. Multilevel linear models were estimated to examine how bipolar risk group moderated bidirectional relationships between physical activity and mood symptoms at within-day and between-day timescales.

Results:

Physical activity was significantly associated with subsequent mood symptoms at the within-day level. The relationship between physical activity and depressive symptoms was moderated by BSD risk group. An increase in physical activity resulted in a greater reduction of depressive symptoms for the BSD group compared to the low-risk and high-risk groups.

Conclusions:

Interventions targeting activity like behavioral activation may improve residual inter-episode mood symptoms.

Introduction

Bipolar spectrum disorders (BSDs), including Bipolar I Disorder (BD I), Bipolar II Disorder (BD II), Bipolar Disorder Not Otherwise Specified, and Cyclothymic Disorder, are characterized by periods of elevated and expansive mood (e.g., mania) and periods of low mood (e.g., depression). With a prevalence rate ranging from 4.4–6.7% in the United States (Merikangas et al., 2007), BSDs are associated with marked functional impairment, comorbidity, and reduced life expectancy (Conus et al., 2014; Kessing et al., 2015). Despite the significant public health consequences associated with BSDs, the etiology remains unclear.

With the introduction of the National Institute of Mental Health’s “Research Domain Criteria (RDoC)” initiative, there has been an emphasis on understanding psychiatric conditions by evaluating objective, quantifiable constructs across multiple levels of analysis, rather than relying on symptom-based criteria (Insel et al., 2010). Motor activity and sleep-activity patterns, components of the arousal and regulatory systems domain, may be particularly relevant to improving our understanding of BSDs. A growing body of literature suggests that activation, a construct comprised of objective physical activity levels and subjective energy levels, plays a key role in the pathogenesis of BSDs (Scott et al., 2017). This is reflected in the changes made to the diagnostic criteria for BSDs in the DSM-5: in the DSM-IV, increased activity or energy was categorized as a criterion B symptom (i.e., a secondary symptom not required for diagnosis), whereas the updated DSM-5 recognizes that increased activity or energy is necessary for diagnosis (American Psychological Association, 2013). Further, findings from a systematic review of factor analytic studies supported the notion that activation and mood are distinct constructs and not consequences of a singular underlying dimension (Scott et al., 2017).

This modification to the DSM-5 criteria has substantial empirical support. Research has found that adults being treated for bipolar disorder (BD) had lower mean physical activity levels than controls and were sedentary for more than 75% of the day (Janney et al., 2014). Similar patterns were found among euthymic individuals in sustained recovery from BD I and II (Ng et al., 2015; Salvatore et al., 2008). Meta-analytic evidence is consistent, indicating that individuals with BSDs have average levels of physical activity comparable to healthy controls, but spend substantial time (more than ten hours each day) engaging in sedentary behavior (Vancampfort et al., 2016). These findings have significant implications for the course of BSDs, given that low levels of physical activity increase the risk for heart disease and other metabolic conditions, and the prevalence rate of cardiovascular disease is elevated among individuals with BSDs (Crump et al., 2013). Indeed, among individuals with BSDs, cardiovascular disease is a leading cause of excess mortality (Correll et al., 2017).

Physical activity also is independently associated with affective states. Low physical activity levels have been linked to greater negative mood (Ingram et al., 2020) and more depressive symptoms (Poole et al., 2011). These findings have been replicated in prospective studies; a meta-analysis of exercise interventions to improve mood found that aerobic and resistance-oriented exercise programs significantly improved depressive symptoms (Rethorst et al., 2009). Further, although mania is characterized in part by increased motor activity, there is a paucity of literature examining the relationship between physical activity and manic symptoms (Vancampfort et al., 2016). A small study of individuals in a manic episode found that actigraphy-derived physical activity correlated modestly with manic symptoms (Minassian et al., 2010).

Recent research has attempted to integrate these findings and better understand the dynamic associations between physical activity, energy, mood, and sleep, particularly as they apply to BSDs. For example, in a large study of participants with and without mood disorders, an increase in motor activity levels (measured via wrist actigraphy) was significantly associated with a decrease in sad mood, whereas there was a bidirectional relationship between energy and physical activity, as well as between physical activity and sleep (Merikangas et al., 2019). Of note, the impact of physical activity on subsequent mood and energy was stronger for individuals with BD I compared to controls, providing further evidence that disturbed patterns of activation are associated with BSDs (Merikangas et al., 2019).

However, it remains unclear whether the relationships between activation and mood are a consequence of BSDs or reflect underlying vulnerability processes that would be present in individuals at increased risk for developing a BSD. As put forth in the reward hypersensitivity model of BSDs, it is hypothesized that individuals with high reward responsiveness are at increased risk of developing a BSD compared to those with moderate reward sensitivity (Alloy et al., 2016; Alloy & Nusslock, 2019; Depue & Iacono, 1989). In brief, according to this model, symptoms of BSDs arise when individuals who are highly sensitive to rewards experience life events that excessively activate (e.g., goal-striving events) or deactivate (e.g., goal failure events) the reward system, which leads to manic or depressive symptoms, respectively (Alloy et al., 2016; Alloy & Nusslock, 2019). Consistent with this theory, prospective findings show that reward hypersensitivity measured via self-report or behavioral task is a vulnerability for BSDs and predicts first lifetime onset of BSDs among adolescents with no prior history of BSDs (Alloy et al., 2012). If dysregulated patterns between mood and physical activity reflect an underlying vulnerability or risk factor for BSDs, then we would expect to see these patterns among individuals with BSDs, as well as those at high-risk for BSDs.

To address the aforementioned questions, the present study utilized an ambulatory assessment approach, a robust method for understanding how processes unfold in real-time with fine-grained temporal resolution. Ecological momentary assessment (EMA), characterized by multiple repeated assessments in individuals’ daily lives, facilities a deeper understanding of within-person processes in participants’ naturalistic environments (Titone et al., 2020). Furthermore, by using wrist-worn actigraphic devices, motor activity can be used as a proxy of physical activity (Allison et al., 2021; Merikangas et al., 2019) and be objectively measured without relying on subjective self-report measures. This advance is particularly important for studying physical activity patterns, because research shows that individuals tend to overestimate how much physical activity they engage in and to underestimate time spent being sedentary (Dyrstad et al., 2014; Vancampfort et al., 2017).

The aims of the present study were to conduct secondary data analyses to investigate bidirectional associations between physical activity levels and mood symptoms among young adults at low-risk for developing a BSD due to moderate reward sensitivity, at high-risk for developing a BSD due to reward hypersensitivity, or diagnosed with a BSD (also with high reward sensitivity). Given that ambulatory assessment allows for a more fine-grained understanding than traditional laboratory approaches, these relationships were assessed at two timescales: within-day (e.g., is physical activity in the morning associated with mood in the afternoon? and vice versa), as well as between-day (e.g., is mood each day associated with next-day physical activity levels? and vice versa). Finally, to assess whether any observed relationships between physical activity and mood reflect a vulnerability to or consequence of BSDs, we examined whether bipolar risk group moderated any potential associations between physical activity and mood symptoms. This study is uniquely positioned to evaluate the temporal precedence of these associations, which has important implications for detecting periods of heightened risk and ultimately, preventing the onset or relapse of mood episodes. We hypothesized that there would be a bidirectional relationship between physical activity and mood symptoms such that an increase in physical activity was associated with a decrease in depressive symptoms and an increase in manic symptoms, at both the within-day and between-day levels. We also hypothesized these relationships would be stronger for the BSD group compared to the low- and high-risk groups, and stronger for the high-risk group compared to the low-risk group, which would indicate that the cross-domain reactivity was a characteristic of BSD, and also elevated among individuals at-risk.

Methods

Participants

A subset of individuals in Project TEAM (Teen Emotion and Motivation Study; Alloy et al., 2012), a prospective study of the onset and course of BSDs in adolescents and young adults, participated in this study. For Project TEAM, participants were recruited using a two-wave screening process designed to identify adolescents at high and low risk for BSDs. Almost 10,000 adolescents from high schools and colleges in Philadelphia were screened using two self-report measures of trait reward sensitivity: the Behavioral Activation Scale (BAS) of the Behavioral Inhibition System/Behavioral Activation System Scales (BIS/BAS; Carver & White, 1994) and the Sensitivity to Reward (SR) subscale of the Sensitivity to Punishment Sensitivity to Reward Questionnaire (SPSRQ; Torrubia et al., 2001). Individuals with moderate reward sensitivity (40^th-60^th percentile on both the BAS and SR) or high reward sensitivity (85^th-99^th percentile on both measures) were eligible for the next screening phase. Individuals with moderate reward sensitivity (MRew) were considered at low risk for developing a BSD, whereas those with high reward sensitivity (HRew) were considered to be at high risk.

During the second screening phase, the expanded Schedule for Affective Disorders and Schizophrenia – Lifetime (exp-SADS-L) diagnostic interview (Alloy et al., 2008; Endicott & Spitzer, 1978) was administered to rule out youth with a history of DSM-IV-TR (American Psychological Association, 2013) or Research Diagnostic Criteria (RDC; Spitzer et al., 1978) BSD, hypomanic episodes, psychotic disorder, or current psychotic symptoms. Further information about recruitment and enrollment for Project TEAM can be found in Alloy et al. (2012).

Participants are a subset of Project TEAM participants who were recruited several years (M=2.85, SD=2.48) after their initial enrollment in Project TEAM and invited to take part in a 20-day EMA study on social and circadian rhythms, mood symptoms, and reward sensitivity. Data collection began in September 2014 and concluded in May 2017. Participants were eligible for this study if they completed both the actigraphy and EMA components (n=113). The sample was 61% female and racially diverse (57% White, 25% Black, 8% Asian, 4% Other, 3% Multiracial, 1% American Indian/Alaska Native, and 3% preferred not to respond¹). Approximately 11% identified as Hispanic or Latinx, 79% were not Hispanic or Latinx, and 11% chose not to report their ethnicity. The mean age was 22.03 (SD=2.20). Approximately 79% of participants were in the low-risk group (n=32), 46% in the high-risk group (n=52), and 25% in the BSD group (n=28).

Procedure

Participants completed a diagnostic interview at all study visits (i.e., approximately every six months after study enrollment) for Project TEAM to assess the onset and course of mood episodes, psychiatric disorders, and psychiatric symptoms. The baseline interview assessed lifetime symptoms and disorders and the follow-up interviews probed the interval since the last study visit. Participants completed an in-person study visit to learn how to wear the actigraphy device and respond to the EMA prompts on their cell phones. For 20 days, participants answered questions three times daily about hypomanic and depressive symptoms. EMA survey delivery was randomized within four-hour blocks such that participants were prompted to respond to surveys once in the morning (M=10.56, SD=1.24 hours), afternoon (M=15.11, SD=1.46 hours), and evening (M=20.09, SD=1.40 hours). Mood ratings were collected via FluidSurveys, a web-based survey platform with text alerts. EMA responses were considered valid if the participant started the survey within one hour of receiving it. Data from EMA surveys initiated over an hour after receipt were excluded. Participants also continuously wore an actigraph to measure sleep/wake cycles and physical activity levels. The dataset generated and analyzed during the current study is available from Dr. Alloy on a reasonable request.

Measures

Structured Clinical Diagnostic Interview

Psychiatric diagnoses (DSM-IV-TR and RDC) were determined by clinical interviewers at Project TEAM baseline using the expanded Schedule for Affective Disorders and Schizophrenia – Lifetime (exp-SADS-L) and at biannual follow-ups using the expanded Schedule for Affective Disorders and Schizophrenia – Change (exp-SADS-C; Alloy et al., 2008; Endicott & Spitzer, 1978). See Alloy et al. (2012) for details. Prior research has shown the exp-SADS has high inter-rater reliability (κ>.96; Alloy et al., 2008). Of note, participants were ineligible for Project TEAM if they had previously been diagnosed with a BSD, hypomanic episode, or cyclothymia. Thus, participants in this study with BSD experienced their index hypo/manic episode while enrolled in Project TEAM and were diagnosed during one of the follow-up visits for Project TEAM.

Reward Sensitivity Screening Measures

The Behavioral Inhibition System/Behavioral Approach System (BIS/BAS) Scale.

The BIS/BAS is a self-report questionnaire that measures sensitivity to rewards (behavioral activation) and punishments (behavioral inhibition). It is comprised of 20 statements about approach and avoidance tendencies that are rated on a four-point Likert scale. Original psychometric evaluations showed it had good internal consistency and retest reliability (Carver & White, 1994), and it had strong internal consistency (α=.80) in the Project TEAM sample as well (Alloy et al., 2012). The measure includes one BIS scale and three BAS subscales (reward responsiveness, fun-seeking, and drive), which are summed to create an overall BAS score.

Sensitivity to Punishment Sensitivity to Reward Questionnaire (SPSRQ).

Reward sensitivity also was assessed using the SPSRQ, a self-report measure comprised of 48 yes/no items that ask about specific types of rewards and punishments (Torrubia et al., 2001). The SR subscale (24 items) was used alongside the BAS score to select Project TEAM participants. Internal consistency of the SR scale in the screening sample was α=.76, and the SR and BAS scores were correlated (r=.40). Notably, the high reward sensitivity (HRew) group that was selected based on this screening procedure was significantly more likely to develop a BSD than participants in the moderate reward sensitivity (MRew) group (Alloy et al., 2012).

Mood Symptom Measures

Altman Self-Rating Mania Scale (ASRM).

Momentary hypo/manic symptoms were assessed three times each day using four items from the Altman Self-Rating Mania Scale (Altman et al., 1997). The ASRM is a well-validated self-report questionnaire that measures key symptoms of hypo/mania. Items assessed via EMA were inflated self-confidence, happiness, talkativeness, and reduced need for sleep. Participants rated how they felt at that moment on a five-point Likert scale from 0–4 with anchors at 0 (not at all), 2 (moderately), and 4 (extremely) and responses were summed.

Beck Depression Inventory-II (BDI-II).

Momentary depressive symptoms also were measured three times each day using selected items from the BDI-II, which is a widely used, psychometrically robust self-report questionnaire (Beck et al., 1996). Four key items central to depressive phenomenology were selected for the EMA survey: sadness, low self-esteem, low energy, and hopelessness. Participants rated the extent to which they felt each symptom on a 1–4 Likert scale with four indicating greater endorsement. Responses from the four items were summed to create a composite score.

Sleep Time and Activity Parameters

Sleep/wake cycles and physical activity were measured using the Actiwatch Spectrum (Philips Healthcare, Bend, OR). Participants wore this device on their non-dominant wrist throughout the 20-day EMA period, only removing it in situations where it could get wet. The watch allowed for continuous, non-invasive measurement with minimal participant burden. Data were collected in one-minute epochs, and the Actiware software was used to compute total sleep time (Titone et al., 2020). Actigraphy has been validated as a reliable metric of sleep; it is correlated with both polysomnography and sleep diaries among healthy sleepers and clinical samples (Cole et al., 1992; Kaplan et al., 2012; Ng et al., 2015).

Average hourly physical activity was computed from the one-minute epochs. For analyses examining between-day associations, the average daily activity score was calculated from the average hourly activity counts. For the within-day analyses, we computed average activity within four daily intervals that were based on the timing of sleep/wake cycles and the EMA prompts. For each day, interval one was the mean physical activity between wake time and prompt one, interval two was the mean physical activity between prompt one and two, interval three was the mean physical activity between prompt two and three, and interval four was the mean physical activity between prompt three and sleep. On average, participants received EMA prompts every 4.84 hours (SD=1.85). Interval length differed depending on the timing of the prompt as well as variability in sleep and wake time, so average physical activity within each interval was calculated to account for the different number of hours in each interval.

For establishing temporal precedence, physical activity during intervals one (prior to mood prompt one), two (prior to mood prompt two), and three (prior to mood prompt three) and mood symptoms assessed at prompts one, two, and three were used for analyses examining the impact of physical activity on subsequent mood. Analyses examining the impact of mood on physical activity used mood symptoms, as assessed at prompts one, two, and three during the three EMA surveys, and physical activity intervals two, three, and four.

Data Analysis

The analytic approach was designed and planned as a specific secondary data project prior to model-building. Data were analyzed using R version 3.6.2 (R Core Team, 2013). Multilevel modeling was utilized because the data were nested within three levels (person-level, day-level, and observation-level) and multilevel models provide more accurate estimates for data clustered within participants than traditional regression techniques. Models were estimated using the lme4 (Bates et al., 2015), lmerTest packages (Kuznetsova et al., 2017), and restricted maximum likelihood. Each model included two random intercepts (participant and day) and fixed slopes.

To examine bidirectional associations between physical activity and mood both within-day and between-day, four sets of multilevel models were estimated, with physical activity levels and mood symptoms each serving as focal predictors across each time scale. Two sets of models examined the impact of mood symptoms on same-day and next-day physical activity levels, and the other two sets of models evaluated the impact of physical activity levels on same-day and next-day mood symptoms. To ensure temporal precedence in the within-day analyses, the predictor variables (e.g., mood or physical activity) were lagged by one observation. For the between-day analyses, the composite variables representing average daily mood symptoms and average daily physical activity (which were calculated based on observation-level data) were lagged by one day.

The predictor variables were separated into two components: a between-persons component reflecting between-person differences in participants’ average levels across all observations, as well as a within-persons component that reflected within-person variability (i.e., the extent to which participants’ responses across observations deviated from their individual mean). In all analyses, the within-persons component was the focal predictor, and the between-person effects were included as a covariate. The predictors and outcome variables were z-standardized. All models also controlled for sex, age, BSD risk group (centered on low-risk for BSD), and previous-night total sleep time (z-standardized). Body mass index (BMI) also was included as a covariate in models where the dependent variable was physical activity level.

To investigate whether BSD risk group moderated within-person associations between physical activity and mood, cross-level interactions between BSD risk group (a person-level variable) and the focal predictors were added to the models in addition to their component variables. These models also controlled for the interaction between BSD risk group and the person-level version of the predictor (e.g., between-person component), as well as the component variables. To obtain all possible pairwise comparisons among the three BSD risk groups, all models were re-estimated using a different reference category (i.e., diagnosed with BSD) for BSD risk group. For these models, only the unique pairwise comparisons were interpreted. Significant interactions were probed by calculating simple slopes using the reghelper (v1.0.2; Hughes, 2021) package.

Results

First, we examined whether there were any within-day associations between physical activity levels and subsequent mood symptoms. As shown in Table 1, at the within-person level, there was a main effect of physical activity levels on depressive and hypomanic symptoms such that an increase in physical activity was prospectively associated with a decrease in depressive symptoms (B=−.06, SE=.01, p<.001) and an increase in hypomanic symptoms (B=.11, SE=.02, p<.001). Findings from interaction models (i.e., Step 2) demonstrated that BSD risk group moderated the within-person relationship between physical activity and depressive symptoms, but not hypomanic symptoms. As shown in Figure 1, the association of physical activity with subsequent depressive symptoms was stronger and more negative for the BSD group compared to the low-risk group (B=−.10, SE=.04, p<.01) and the high-risk group (B=−.07, SE=.03, p=.04). The simple slopes for the BSD and high-risk groups were significantly different than zero (B=−.12, SE=.02, p<.001 and B=−.05, SE=.02, p=.01, respectively). The association of physical activity with later depressive symptoms for the low-risk group was small and not statistically different from zero.

Table 1.

Within-day associations between physical activity and subsequent depressive and hypomanic symptoms

	Depressive Symptoms						Hypomanic Symptoms
	Step 1			Step 2			Step 1			Step 2
Level 1	B	SE	p	B	SE	p	B	SE	p	B	SE	p
PA (within-person)	−.06	.01	<.01	−.02	.03	.44	.11	.02	<.001	.10	.02	<.001
PA X High-risk (ref: low-risk)				−.03	.03	.34				<.01	.03	.95
PA X BSD (ref: low-risk)				−.10	.04	<.01				.02	.03	.46
PA X BSD (ref: high-risk)†				−.07	.03	.04				.03	.03	.36
Level 2
Group
High-risk (ref: low-risk)	.18	.14	.18	.19	.14	.17	−.14	.14	.32	−.14	.14	.34
BSD (ref: low-risk)	.30	.15	.05	.31	.16	.05	.13	.16	.43	.13	.11	.51
BSD (ref: high-risk)†	.12	.14	.40	.12	.14	.40	.27	.15	.07	.03	.03	.10
Observations	3666			3666			3666			3666
Days	19			19			19			19
Participants	112			112			112			112
ICC (person level vs all)	.32			.33			.40			.40
Individual (person-level) Variance	.32			.33			.37			.37
Day-level Variance	<.001			<.001			.001			.002
Residual (observation-level) Variance	.67			.67			.55			.55

Notes: PA = physical activity; ref: = reference; low-risk = moderate reward sensitivity; high-risk = high reward sensitivity; BSD = bipolar spectrum disorders; PA (within person) is centered on each participant’s mean; All analyses control for past-night’s sleep, sex, age, and between-person PA; Step 1 models only contain main effects; Interaction components were added in Step 2; Step 2 models also control for the interaction between group and between-person PA; PA, sleep, depressive, and hypomanic symptoms were z-standardized

^†These estimates were ascertained from separate models, identical to the primary models, except the reference group for BSD risk group was changed from low-risk to high-risk to ascertain all pairwise comparisons.

Figure 1. Within-Day Association of Physical Activity with Subsequent Depressive Symtoms.

Note: BSD=Bipolar Spectrum Disorder

Next, we investigated whether these effects were observable across a larger timescale (i.e., whether physical activity predicted next-day mood). As seen in Table 2, there were no main effects of within-person physical activity on next-day mood symptoms. There also were no significant interactions, suggesting BSD risk group did not moderate the association between physical activity and next-day depressive or hypomanic symptoms (all ps>.05).

Table 2.

Between-day associations between physical activity and next-day depressive and hypomanic symptoms

	Depressive Symptoms						Hypomanic Symptoms
	Step 1			Step 2			Step 1			Step 2
Level 1	B	SE	P	B	SE	P	B	SE	P	B	SE	P
PA (within-person)	−.01	.02	.58	.01	.04	.88	<.01	.02	.92	−.04	.04	.21
PA X High-risk (ref: low-risk)				<.01	.05	.98				.05	.04	.28
PA X BSD (ref: low-risk)				−.07	.06	.20				.09	.05	.07
PA X BSD (ref: high-risk) †				−.07	.05	.13				.04	.04	.32
Level 2
Group
High-risk (ref: low-risk)	.19	.16	.23	.19	.16	.24	−.14	.16	.41	−.12	.16	.46
BSD (ref: low-risk)	.31	.18	.09	.32	.18	.08	.19	.19	.30	.20	.19	.29
BSD (ref: high-risk) †	.12	.16	.45	.13	.17	.43	.33	.17	.06	.32	.17	.06
Observations	1736			1736			1736			1736
Participants	113			113			113			113
ICC (person level vs all)	.45			.46			.47			.53
Individual (person-level) Variance	.44			.45			.42			.48
Residual (observation-level) Variance	.53			.53			.48			.42

Notes: PA = physical activity; ref: = reference; BSD = bipolar spectrum disorders; low-risk = moderate reward sensitivity; high-risk = high reward sensitivity; PA (within person) is centered on each participant’s mean; All analyses control for past-night’s sleep, sex, age, and between-person PA; Step 1 models only contain main effects; Interaction components were added in Step 2; Step 2 models also control for the interaction between group and between-person PA; PA, sleep, depressive, and hypomanic symptoms were z-standardized

^†These estimates were ascertained from separate models, identical to the primary models, except the reference group for BSD risk group was changed from low-risk to high-risk to ascertain all pairwise comparisons.

Results from models examining the opposite direction, specifically whether mood symptoms were prospectively associated with within-day or next-day physical activity levels, are presented in Tables 3 and 4. As seen in Table 3, an increase in within-person depressive symptoms predicted a decrease in subsequent same-day physical activity (B=−.09, SE=.01, p<.001) and an increase in within-person hypomanic symptoms significantly predicted an increase in same-day physical activity (B=.13, SE=.01, p<.001). Additionally, there was a main effect of BSD risk group such that participants with BSDs had significantly lower physical activity levels on average than individuals in the high-risk group (B=−.26, SE=.13, p=.04; B=.30, SE=.12, p=.01). None of the interactions included in Step 2 reached significance, indicating BSD risk group was not a moderator of within-day relationships between mood and later physical activity.

Table 3.

Within-day associations between depressive and hypomanic symptoms and subsequent physical activity levels

	Physical Activity
	Step 1			Step 2
Depressive Symptoms	B	SE	P	B	SE	P
Level 1
Depressive Sx (within-person)	−.09	.01	<.001	−.09	.02	<.01
Depressive Sx X High-risk (ref: low-risk)				.01	.04	.80
Depressive Sx X BSD (ref: low-risk)				.01	.04	.83
Depressive Sx X BSD (ref: high-risk)†				<.01	.03	.97
Level 2
Group
High-risk (ref: low-risk)	.16	.12	.18	.18	.12	.15
BSD (ref: low-risk)	−.09	.14	.50	−.08	.14	.59
BSD (ref: high-risk) †	−.26	.13	.04	−.25	.13	<.05
ICC (person level vs all)	.25			.25
Individual (person-level) Variance	.25			.25
Day-level Variance	<.001			<.001
Residual (observation-level) Variance	.75			.75
Hypomanic Symptoms
Level 1
Hypomanic Sx (within-person)	.13	.01	<.001	.10	.03	<.001
Hypomanic Sx X High-risk (ref: low-risk)				.04	.04	.21
Hypomanic Sx X BSD (ref: low-risk)				.03	.04	.42
Hypomanic Sx X BSD (ref: high-risk) †				−.01	.03	.72
Level 2
Group
High-risk (ref: low-risk)	.19	.12	.10	.17	.12	.13
BSD (ref: low-risk)	.17	.12	.40	−.11	.13	.41
BSD (ref: high-risk) †	−.30	.12	.01	−.28	.12	.02
Observations	3980			3980
Days	19			19
Participants	112			112
ICC (person level vs all)	.24			.24
Individual (person-level) Variance	.23			.23
Day-level Variance	<.001			<.001
Residual (observation-level) Variance	.74			.74

Notes: Sx = symptoms; ref: = reference; low-risk = moderate reward sensitivity; high-risk = high reward sensitivity; BSD = bipolar spectrum disorders; All analyses control for past-night’s sleep, sex, body mass index, age, and between-person depressive/hypomanic Sx; Depressive Sx and Hypomanic Sx (within person) are centered on each participant’s mean; PA, body mass index, sleep, depressive, and hypomanic symptoms were z-standardized; Step 1 models only contain main effects; Interaction components were added in Step 2; Step 2 models also control for the interaction between group and between-person Sx

^†These estimates were ascertained from separate models, identical to the primary models, except the reference group for BSD risk group was changed from low-risk to high-risk to ascertain all pairwise comparisons.

Table 4.

Between-day associations between depressive and hypomanic symptoms and next-day physical activity

	Physical Activity
	Step 1			Step 2
Depressive Symptoms	B	SE	P	B	SE	P
Level 1
Depressive Sx (within-person)	<.01	.02	.84	−.04	.04	.32
Depressive Sx X High-risk (ref: low-risk)				.14	.05	.42
Depressive Sx X BSD (ref: low-risk)				.05	.05	.29
Depressive Sx X BSD (ref: high-risk) †				.01	.03	.71
Level 2
Group
High-risk (ref: low-risk)	.16	.15	.28	.17	.15	.27
BSD (ref: low-risk)	−.12	.17	.50	−.11	.17	.53
BSD (ref: high-risk) †	−.28	.15	.03	−.28	.16	.08
ICC	.49			.49
Individual (person-level) Variance	.39			.39
Residual (observation-level) Variance	.41			.41
Hypomanic Symptoms
Level 1
Hypomanic Sx (within-person)	−.01	.02	.50	.07	.04	.07
Hypomanic Sx X High-risk (ref: low-risk)				−.09	.04	.04
Hypomanic Sx X BSD (ref: low-risk)				−.10	.05	.02
Hypomanic Sx X BSD (ref: high-risk) †				−.02	.04	.65
Level 2
Group
High-risk (ref: low-risk)	.20	.14	.40	.18	.14	.21
BSD (ref: low-risk)	−.14	.16	.31	−.14	.16	.41
BSD (ref: high-risk) †	−.33	.15	.03	−.31	.15	.04
Observations	172			172
	0			0
Participants	112			112
ICC	.46			.46
Individual (person-level) Variance	.35			.35
Residual (observation-level) Variance	.41			.41

Notes: Sx = symptoms; ref: = reference; low-risk = moderate reward sensitivity; high-risk = high reward sensitivity; BSD=bipolar spectrum disorders; All analyses control for past-night’s sleep, sex, body mass index, age, and between-person depressive/hypomanic Sx; Depressive Sx and hypomanic Sx (within person) are centered on each participant’s mean; PA, body mass index, sleep, depressive, and hypomanic symptoms were z-standardized; Step 1 models only contain main effects; Interaction components were added in Step 2; Step 2 models also control for the interaction between group and between-person Sx

^†These estimates were ascertained from separate models, identical to the primary models, except the reference group for BSD risk group was changed from low-risk to high-risk to ascertain all pairwise comparisons.

Last, we investigated whether there were associations between depressive and hypomanic symptoms and subsequent physical activity at the between-day timescale. As shown in Table 4 (Step 1), neither depressive symptoms nor hypomanic symptoms had significant unconditional within-person associations with next-day physical activity (B=−.003, SE = .02, p = .84, B=−.01, SE=.016, p=.50, respectively). Individuals in the BSD group had lower average levels of physical activity compared to the high-risk group in both the model with depressive symptoms as the focal predictor (B=−.28, SE=.15, p=.03) and the model with hypomanic symptoms as the focal predictor (B=−.33, SE=.15, p=.02). There was a significant effect of between-person hypomanic symptoms on next-day physical activity (B=.19, SE=.06, p<.01). As seen in Figure 2, results from the models estimated with interaction terms revealed that BSD risk group moderated the association between within-person hypomanic symptoms, but not depressive symptoms, and average next-day physical activity. The relationship between hypomanic symptoms and next-day average physical activity levels was more negative for the high-risk group (B=−.09, SE=.04, p=.04) and the BSD group (B=−.10, SE=.05, p=.02) compared to the low-risk group. Simple slope analysis indicated that participants in the low-risk group had a trending within-person association between hypomanic symptoms and next-day physical activity (B=.07, SE = .04, p=.07), whereas hypomanic symptoms did not have a significant within-person association with next-day physical activity for participants in the high-risk group (B=−.02, SE = .03, p=.33) or the BSD group (B=−.04, SE=.03, p=.16).

Figure 2. Association of Hypomanic Symptoms with Next-Day Physical Activity.

Note: BSD=Bipolar Spectrum Disorder

Discussion

This study used ambulatory assessment methods to examine bidirectional relationships between physical activity levels and mood symptoms (i.e., hypomanic and depressive symptoms) at two different time scales among a diverse sample of young adults at low-risk, high-risk, and diagnosed with a BSD. Physical activity was significantly associated with subsequent depressive and hypomanic symptoms, and this relationship only was observed at the within-day (not between-day) level. Further, the within-day association between physical activity and subsequent depressive symptoms was moderated by BSD risk group such that an increase in physical activity resulted in a greater reduction of depressive symptoms for the BSD group compared to the other two groups. There also was a main effect of mood symptoms on subsequent physical activity, but only when examining within-day associations. However, moderation analyses revealed that there was an interaction between hypomanic symptoms and BSD risk group on physical activity, although none of the simple slopes analyses reached significance. Several of these findings warrant elaboration.

There were bidirectional associations between physical activity and hypo/manic symptoms, and physical activity and depressive symptoms, reflecting a reciprocal relationship between these phenomena. In contrast, prior work found a unidirectional relationship such that physical activity influenced mood, but mood did not influence physical activity (Merikangas et al., 2019). This may reflect differences in how mood was operationalized in the two studies: because the aims of this study were to examine processes associated with BSDs, we specifically measured hypomanic and depressive symptoms (key features of BSD) using a composite score based on several items, whereas Merikangas and colleagues (2019) used a single-item question asking to what extent participants felt very happy or sad. This discrepancy highlights the nuances that result from how affective states are measured; physical activity may be differentially related to certain facets of depressive or manic mood states. Indeed, although sadness is a component of depression, depression is a multifaceted construct. We employed a more comprehensive approach by assessing depressive and hypomanic symptoms, capturing both poles of the mood spectrum, to examine subthreshold mood symptoms that may be particularly relevant to the phenomenology of BSD.

Notably, BSD risk group moderated the association of physical activity with subsequent within-day depressive symptoms. As depicted in Figure 1, an increase in physical activity was associated with a greater reduction in depressive symptoms for the BSD group compared to the high-risk and low-risk groups, suggesting individuals with BSDs are more sensitive to changes in physical activity. Indeed, whereas physical activity appears to be beneficial for the overall sample in that it is associated with lowered depressive symptoms, this relationship was even stronger for the BSD group. The mechanisms driving these within-day effects are likely complex and involve a myriad of interconnected systems, both biological and psychosocial. One possible explanation asserts that increased physical activity enhances an individual’s ability to recover from stressors and return to their resting physiological state, increasing overall emotional resilience (Bernstein & McNally, 2018; Calvo et al., 1996). This theory accounts for within-day, but not between-day effects; however, it is possible that sleep serves as a “reset,” preventing carry-over effects from the prior day. Indeed, deep sleep has been shown to restore the area of the brain that regulates emotions, and thus, the effects of sleep on mood may override the effects of physical activity (Ben Simon et al., 2020). This hypothesis needs to be explicitly tested.

It is important to note that the physical activity construct measured in this study is not synonymous with exercise, and although our findings are relevant to the ample literature examining the relationships between planned exercise and affect, we would be remiss to assert that we are measuring the same phenomenon. However, a growing body of literature has found that exercise and aerobic activity interventions are effective in treating internalizing symptoms (Hearing et al., 2016); for example, a meta-analysis of eleven exercise intervention studies yielded a large effect size (d=1.42; Stathopoulou et al., 2006). Specifically examining individuals with BSD, a cross-sectional study found infrequent exercise was associated with greater depressive symptoms and time spent depressed, whereas increased exercise was associated with more time spent manic (Sylvia et al., 2013); however, causality cannot be inferred and it may be that individuals engaged in more exercise because they were manic and had increased energy.

Research examining everyday physical activity, outside the context of planned exercise, is limited. However, our results are consistent with other work that found that walking interventions also had a large effect on depressive symptoms (Robertson et al., 2012), and prevalence rates of depression among women with moderate levels of ambulatory activity were half that of primarily sedentary women (McKercher et al., 2009).

The moderating role of BSD diagnosis on the relationship between physical activity and depressive symptoms has important clinical relevance. Although BSDs are considered episodic conditions, residual inter-episode mood symptoms are common. For example, in a study of adults with mood disorders, inter-episode mood lability (defined as mood fluctuations in the absence of a major depressive episode or hypo/manic episode) was present in almost 50% of the sample (62.9%; Benazzi, 2004). Moreover, inter-episode mood instability is associated with poor prognosis, increased risk of relapse, and more frequent hospitalizations (Patel et al., 2015). In turn, it has been suggested that the treatment of residual mood symptoms may be a worthwhile endeavor to improve overall functioning among those with BSDs (Saunders et al., 2016). Moreover, it has been noted that depressive symptoms, and not mania, drive the burden of disease due to their pervasiveness (Miller et al., 2014). For example, in a longitudinal study of BD I and BD II, participants endorsed depressive symptoms 30–50% of the study duration, but manic symptoms only 1–10% of their time in the study (Judd et al., 2002, 2003). Thus, focusing on strategies that regulate depressive symptoms may be more beneficial than targeting hypo/manic symptoms.

A secondary goal was to evaluate whether physical activity-mood symptom relationships reflected a vulnerability for BSDs. If that was the case, we would expect there to be a significant difference between the high-risk and low-risk groups. We did not see a main effect of the high-risk group on any of the outcomes, nor did we observe any significant interactions between the high and low-risk groups. Thus, there is not sufficient evidence to suggest that the physical activity-mood symptom association is characteristic of those at-risk for BSDs due to heightened reward sensitivity. One explanation is that the physical activity-mood relationship is a consequence of BSD that only is observable in those already diagnosed with a BSD.

There were no main effects of physical activity on next-day mood, and vice versa. Interpreted in the context of the within-day findings, we observed that there is a bidirectional relationship of physical activity and mood symptoms within a given day, but that this relationship does not carry over to the subsequent day. Potentially, this could reflect the moderating and stabilizing role of sleep (Borb & Achermann, 1999). Sleep is known to be associated with physical activity as well as emotion regulation (Kredlow et al., 2015). Sleep researchers assert that one function of sleep is to help maintain homeostasis, effectively hitting the “reset” button (Griffith & Rosbash, 2008). The lack of between-day carryover seen in our results may reflect a return to setpoint that occurs during sleep, but future studies are needed to explore this hypothesis.

The inconsistency in the within-day and between-day findings also highlights the importance of studying processes at the time scales in which they unfold, which is part of a broader theme of effective measurement in psychiatry (Moriarity & Alloy, 2021). The need for temporal specificity is a consideration for many research areas and reflects the notion that results are contingent upon the time interval in which they are studied. For example, the impact of a stressful life event may be observable one month later, but not ten years later.

However, at the between-day level, we obtained a significant interaction between hypomanic symptoms and subsequent physical activity for the high-risk and BSD groups compared to the low-risk group. As depicted in Figure 2, the relationship between hypomanic symptoms and physical activity for the high-risk and BSD groups were negative, whereas the relationship for the low-risk group was positive, although none of the slopes were significantly different than zero.

Finally, all four models examining the association between mood and subsequent physical activity revealed a significant main effect of group on physical activity, such that physical activity levels consistently were lower for the BSD group compared to the high-risk group. This is in line with previous work that found adults with BSDs have lower mean levels of activity (Janney et al., 2014). At first, this finding may appear counterintuitive because mania is characterized by increased energy, activity, and goal-striving behaviors. However, this may be indicative of a state and trait distinction: the increased energy observed during mania is specific to the episode and does not carry over to euthymic periods.

This study has several strengths and limitations worth noting. Mood symptoms were assessed comprehensively using a multi-item scale several times each day, which facilitated analysis of within-person variation. Actigraphy allowed for an objective measurement of physical activity, rather than relying on self-report, although actigraphy-derived activity counts are a proxy for physical activity and not synonymous with exercise. The sample was relatively diverse, increasing generalizability. Participants were young adults, which is the developmental period in which risk for mood disorders is elevated (Gustavson et al., 2018), which increases the validity of the at-risk group. However, the restricted age range also may reduce generalizability. Limitations include the sample size of the BSD group, which precluded an examination of BSD subtypes, and the inability to measure depressive and manic episodes, which have more clinical relevance yet are harder to study on this timescale. Additionally, participants completed the reward sensitivity measures that were used to determine risk group several years before the EMA protocol. Although the BIS/BAS and SPSRQ are designed to capture trait-based reward sensitivity that is relatively stable, there could have been changes in reward hypersensitivity status in the interim. However, considering participants completed a diagnostic interview at every project TEAM follow-up visit, we are confident the BSD grouping is accurate. Finally, we did not evaluate the possibility that the observed findings were driven by a reduction in sedentary behavior, as opposed to an increase in physical activity. Indeed, sedentary behavior could be an operative factor in the current study; a prior cross-sectional study found physical inactivity (operationalized as a lack of participation in organized or unorganized physical activities) was associated with a greater odds of depression (OR=1.43), but sedentary behaviors (e.g., time spent watching TV, playing video games, etc.) also was comparably associated with an increased risk for depression (OR=1.38). It is likely both constructs are important for predicting the course of depressive symptoms, although it is difficult to disentangle their relative contributions (Bélair et al., 2018). Clinically, our findings support the use of treatments that target not only mood states, but also activity patterns, like behavioral activation (Dimidjian et al., 2011). Behavioral activation encourages individuals to engage in activities that promote pleasure or mastery rather than follow their state-dependent action urge. Although initially developed for unipolar depression, preliminary evidence suggests behavioral activation also is beneficial for BSDs (Weinstock et al., 2016). It is interesting to note that findings from a randomized control study of behavioral activation for depression found that all participants, not just those randomized to the supplemental exercise arm, engaged in greater exercise as depressive symptoms improved over the course of the intervention. In turn, the increase in physical activity that appears to be an unintended benefit of behavioral activation may be one such mechanism of change contributing to its effectiveness (Szuhany & Otto, 2020). Examining target mechanisms that increase an individual’s likelihood of engaging in physical activity in the long-term (e.g., enjoyment, pleasure) could help promote adherence and sustained results over time. A recent study found evidence of specificity in that outdoor running/walking, cycling, and team sports were the only physical activity types associated with a reduced risk for depression (Matias et al., 2022). These findings raise the possibility that exposure to light, inherent to outdoor activities, or the socialization aspect germane to team sports, also may be other mechanistic explanations for the improved mood associated with physical activity (Matias et al., 2022). Pharmacologic interventions that stimulate activity levels also may warrant further consideration.

In conclusion, we observed a bidirectional association between physical activity and depressive and hypomanic symptoms at the within-day level, but not between-day level. The beneficial impact of physical activity on depressive symptoms was greater for individuals with BSDs, suggesting this relationship may be a feature of BSD that could be targeted to ameliorate inter-episode mood symptoms.

Highlights:

• Activation is a central feature of bipolar spectrum disorders.

• People with bipolar disorder may experience heightened cross-domain reactivity.

• For people with bipolar disorder, increased physical activity was associated with a greater decrease in depressive symptoms.

• This relationship was observed at the within-day, but not between-day timescale.

• Findings suggest interventions that promote physical activity may be important for affective stability between episodes.