Is There a 24 Hour Rhythm in Alcohol Craving and Does it Vary by Sleep/Circadian Timing?

Abstract

Increasing evidence implicates sleep/circadian factors in alcohol use; however, the role of such factors in alcohol craving has received scant attention. Prior research suggests a 24-hour rhythm in related processes (e.g., reward motivation), but more research directly investigating a rhythm in craving is needed. Moreover, prior evidence is ambiguous whether such a rhythm in alcohol craving may vary by sleep/circadian timing. To examine these possibilities, 36 late adolescents (18–22 years of age; 61% female) with regular alcohol use but without a current alcohol use disorder were recruited to complete smartphone reports of alcohol craving intensity six times a day for two weeks. During these two weeks, participants wore wrist actigraphs and completed two in-lab assessments (on Thursday and Sunday) of dim light melatonin onset (DLMO). Average actigraphically-derived midpoint of sleep on weekends and average DLMO were used as indicators of sleep and circadian timing, respectively. Multilevel cosinor analysis revealed a 24-hour rhythm in alcohol craving. Findings across the sleep and circadian timing variables converged to suggest that sleep/circadian timing moderated the 24-hour rhythm in alcohol craving. Specifically, people with later sleep/circadian timing had later timing of peak alcohol craving. These findings add to the growing evidence of potential circadian influences on reward-related phenomena and suggest that greater consideration of sleep and circadian influences on alcohol craving may be useful for understanding alcohol use patterns and advancing related interventions.

INTRODUCTION

The intensity of the craving to consume alcohol predicts later alcohol use, and in adults with alcohol use disorder (AUD), predicts relapse (Schneekloth et al. 2012; Serre et al. 2018; Padovano et al. 2019). Although differing viewpoints on craving exist, it is frequently defined as the desire to use a drug, seen as a dimensional process that varies in intensity, and that exists in populations with and without substance use disorders (Shiffman 1987; Sayette et al. 2000; Naqvi et al. 2015; Serre et al. 2015). The current study adopts these viewpoints on craving and focuses on craving in late adolescent alcohol drinkers without current AUD. Alcohol use peaks in this age window, with some late adolescent drinkers developing severe and chronic AUD, and others transitioning to non-problematic alcohol use patterns. Given the importance of craving in alcohol consumption and susceptibility of late adolescent drinkers to develop AUD, understanding factors influencing craving may be especially important for understanding how to interrupt progression to AUD.

One such potential factor is daily rhythmicity, i.e., an observed cyclical fluctuation over 24 h that may be due to endogenous, social, or environmental rhythms. Preliminary evidence based on a retrospective questionnaire suggests that alcohol craving may emerge at systematic times of day in individuals with AUD (Danel et al. 2003), but no studies (to our knowledge) have examined a 24 h rhythm in craving in a prospective fashion or outside the context of AUD. The possibility that there may be a 24 h rhythmicity in alcohol craving, especially in populations at-risk for developing AUD, has received scant attention. Identification of 24 h patterns in alcohol craving intensity may characterize “windows of vulnerability” that indicate when someone is most likely to crave and potentially use alcohol. Additionally, identifying a 24 h rhythm in craving would add to the growing evidence that reward motivation phenomena are influenced by diurnal, and possibly circadian, factors.

Craving and 24-hour rhythms

There are two main reasons to suspect that cravings exhibit a 24 h rhythm. First, such a rhythm in craving may occur because of rhythms in the timing of social and contextual factors, e.g., timing of work and bar hours. Second, naturalistic and laboratory studies suggest that reward-related phenomena, such as positive affect, wanting, and risk-taking, exhibit a 24 h rhythm (Hasler et al. 2008; Murray et al. 2009; Miller et al. 2015; Byrne & Murray 2017; Itzhacki et al. 2019). On average, this rhythm rises from the morning until the late afternoon/evening, then declines until the following morning, after which the cycle beings anew. These findings suggest the presence of an underlying circadian rhythm in general reward motivation, e.g., desire/wanting, that may influence reward-related drives, such as alcohol craving.

Prior work examining 24 h rhythms in substance craving have been inconclusive. Some studies find that substance craving in diurnally active persons is characterized by low early morning craving, followed by an increase in craving later in the morning which may then remain stable or increase slightly across the day, e.g., nicotine, cocaine, alcohol (Preston et al. 2009; Epstein et al. 2010; Chandra et al. 2011; Miranda et al. 2019). In contrast, two other studies examining whether biological time predicts alcohol and nicotine craving intensity found null results (Shiffman et al. 1996; Padovano et al. 2019). Finally, three studies found more traditional 24 h rhythmic fluctuations in cigarette craving (Gilbert & Pope 1982; Teneggi et al. 2002, 2005).

Altogether, this evidence suggests that craving intensity changes across the day, but it is unclear whether such changes reflect a systematic 24 h pattern. Many prior studies only examined a linear effect of time-of-day effect, which models a continuous increase or decrease in craving throughout the day, rather than a sinusoidal effect, which models a rhythmic pattern in craving throughout the 24 h. Studies examining a linear time-of-day effect do not speak to whether there may be a rhythm in craving. Moreover, some of these studies used time-of-day data that were aggregated across participants; however, this aggregated approach may mask rhythmic effects if there is substantial variability in the timing or shape of the 24 h rhythm across participants or days. Given these mixed findings, many of which did not directly examine alcohol craving, there is a need to further examine the possibility of a 24 h variation in alcohol craving with more sophisticated analytic methods that directly test for the presence of a daily rhythmic pattern.

Role of individual differences in sleep/circadian timing

If alcohol craving exhibits a systematic 24 h rhythm within people, there may also be individual variations in the timing and amplitude of this rhythm that systematically varies in accordance with a person’s sleep/circadian timing. We jointly refer to individual differences in the timing of sleep and circadian processes as “sleep/circadian timing,” because the timing of sleep and circadian processes are tightly related.

While an increasing number of studies demonstrate 24 h rhythms in reward-related functioning, only a few (all examining positive affect) have also investigated whether these rhythms differ across individuals in accordance with their sleep/circadian timing. Specifically, people who self-report preferences for later sleep timing show a 24 h rhythm in positive affect that reaches its peak later in the day and has a smaller amplitude (Porto et al. 2006; Hasler, Germain, et al. 2012; Miller et al. 2015). Since both positive affect and alcohol craving may reflect activation of the reward system, individual differences in sleep/circadian timing may influence the 24 h rhythm activation of alcohol craving in a similar manner to positive affect. However, it is reported that people with later sleep/circadian timing often consume more drugs and alcohol, are more impulsive and disinhibited, and have a greater sensitivity to reward (Hasler, Smith, et al. 2012; Kang et al. 2015; Hasler et al. 2017). These characteristics, particularly the greater alcohol use, suggest that individuals with later sleep/circadian timing may have stronger craving for alcohol, and this stronger craving may be related to 24 h rhythms characterized by higher peaks in craving, i.e., greater amplitude of the rhythm in craving. Altogether, prior evidence suggests sleep/circadian timing could predict a later peak timing in alcohol craving, but it is unclear whether sleep/circadian timing may predict a shallower or steeper amplitude in the craving rhythm.

Current study

This study uses ecological momentary assessment (EMA) data from 36 participants to test two possibilities. First, this study tests whether within-person variation in alcohol craving across the day exhibits a 24 h pattern. Identifying such a pattern can provide better insight into the timing of intense cravings that might lead to alcohol use. It would converge with prior research findings of circadian rhythmicity in other reward-related processes, which would inform our mechanistic understanding of patterns of alcohol craving that future research could further elucidate. Second, this study provides the first test of whether sleep/circadian timing influences the timing and shape of 24 h alcohol craving patterns. We hypothesize that individuals with later sleep/circadian timing also have a later peak in craving timing. However, given the ambiguity of prior research regarding how individual differences in sleep/circadian timing may influence the amplitude of a 24 h rhythm in alcohol craving, we explore how sleep/circadian timing relates to the amplitude but do not make a firm hypothesis on this possibility. Note, that a dataset from this study was previously used to examine whether circadian misalignment from the weekday to the weekend is related to alcohol use (Hasler et al. 2019), but associations between alcohol craving and individual differences in sleep/circadian timing have not been previously examined.

METHODS

Thirty-six healthy late adolescents (18–22 y old, M_age = 21.26 y; 22 females; 69% White) reporting regular alcohol use were recruited as part of a larger study examining the relations among alcohol use, circadian rhythms, and reward related functioning. Regular alcohol use was defined as a minimum of one standard drink per week over the past 30 d (assessed via a web- or phone-based timeline follow-back). This study sought to recruit adolescents across a range of alcohol use levels, from light to heavy. Based on a timeline follow-back interview over the past 30 d conducted at the beginning of the study, the median number of drinks reported was 21; reports of drinking varied substantially across participants as the standard deviation was 29 drinks (mean = 31.68 drinks, minimum = 7 drinks, maximum = 136 drinks).

Participants were excluded for: having a current AUD (past AUD if remitted >1 y ago was not exclusionary), use of substances other than alcohol, cannabis, and nicotine, significant current medical, e.g., cardiovascular disorder, head injury with loss of consciousness, or psychiatric, e.g., major depression, bipolar disorder, any current sleep disorders other than insomnia, or medication use that might interfere with sleep and/or reward function. Individuals with insomnia were included given that insomnia was expected to be prevalent among regular alcohol users and/or people with late sleep timing. Finally, participants with extreme habitual sleep times (bedtimes later than 02:00h, rise times later than 10:00h) or habitual sleep durations >9 h or <6 h were also excluded due to the practical challenges of studying participants with extreme sleep times and concern over having extreme outliers in a relatively small sample.

This study was approved by the university Institutional Review Board and complies with ethical standards for biological rhythms research (Portaluppi et al. 2010). Written informed consent was obtained for all participants.

Procedures

Participants completed 14 d of ecological momentary assessments in which they reported on their alcohol craving six times a day. Participants also wore an actigraph to continuously and behaviorally assess sleep characteristics during this time. EMA assessments were scheduled such that participants each received the first assessment at their self-reported habitual wake time, which was further tailored to their habitual weekday and weekend wake times). Participants also reported on their prior night sleep and prior day alcohol consumption during this morning assessment. Afterwards, the next four assessments were sent every ~3 h, and the last assessment was sent during the participant’s self-reported habitual bedtime (which was also further tailored to their habitual weekday and weekend bedtimes). Participants were sent reminders after 30 and 60 min if they did not complete the assessment, and each assessment remained opened to be completed until the next assessment was sent. All assessments were administered via smartphones, through which a text message was sent that included a link to an online assessment system.

During this 14 d period, participants came into the sleep laboratory for two overnight visits in which dim light melatonin onset (DLMO) was assessed. The order of these visits was counter-balanced to address potential task habituation effects during each visit’s fMRI scan (not discussed here). Specifically, these overnight assessments were counter-balanced across participants, such that they occurred on either a subsequent Thursday and Sunday or a subsequent Sunday and Thursday.

Measures

Ecological momentary assessments.

During each assessment, participants reported how much they were currently craving alcohol as well as about other variables. Intensity of alcohol craving was assessed by participants responses to a single question asking, “How much are you craving alcohol right now?” on a Likert scale from “very slightly or not at all” (score equal to 1) to “extremely” (score equal to 5). Similar ecological momentary assessments of craving have been used in a wide range of substance use studies and predict substance use in populations with and without diagnosed substance use disorders (Serre et al. 2015). During each morning assessment, participants were also asked about their prior night sleep timing. These self-reported sleep timing characteristics were used to aid in setting bed and wake times in actigraphic sleep data (see below). In addition to these sleep questions, participants also reported how many alcoholic drinks they consumed the prior day, i.e., “How many alcoholic drinks did you have yesterday?”. Reports of prior day drinking were shifted back by one day, e.g., responses made on Thursday were shifted to Wednesday, so that reports of drinking indicated number of drinks for the same day that alcohol craving was assessed. This variable was then dichotomized to create a variable indicating whether the participant consumed alcohol that day (0 = did not consume alcohol, 1 = consumed alcohol).

On average, participants completed approximately five out of the six daily alcohol craving assessments for a total of 2,513 assessments out of the total of 3,060 potential EMA craving assessments (82.1% completion rate). The time-of-completion of these assessments almost spanned the entire 24 h period, and assessments were completed across all hours of the day except for 04:00 and 05:00h (Figure 1). It is also notable that there were few assessments (>30) completed at 02:00, 03:00, 06:00, and 07:00h. Overall, the EMA assessments in this study approximately covered the 24 h timespan in which this study seeks to estimate a rhythm in craving, though estimated craving between 02:00 and 07:00h may be less reliable than estimates for other parts of the day.

Figure 1.

Completion times of alcohol craving EMA assessments throughout the 24-hour day.

Sleep/circadian timing

Actigraphic average midpoint of sleep on weekends.

Participants wore an Actiwatch Spectrum Classic (Philips Respironics, Bend, OR) during the 14 d study period. On average, actigraphy data was available for 11.70 out of the 14 d (SD = 3.26; 82.7% completion rate). All actigraphs were set to record activity on medium sensitivity at 1 min epochs. Participants were instructed to press an event recording marker on the watch to indicate when they started to try to fall asleep and when they woke up for the day. These bed and wake time event markers were used to adjust the automatic actigraphic estimates of bed and wake time. If event markers were not available, participant self-reported bed and wake times from morning EMA assessments were used, and if these self-reports were not available, then consensus among study personnel was used to determine actigraphic bed and wake times.

After making these adjustments, midpoint of sleep, defined as the average of the bed and wake times on all weekend days recorded during the study period, were averaged to estimate a person’s typical midpoint of sleep on weekend days. Weekend days were used as they should provide the closest approximation of the midpoint of sleep on “free days,” i.e., days in which people have no social constraints on their sleep schedule. People with later midpoints of sleep on free days typically have a delayed sleep timing in which they go to bed and wake up later than people with earlier midpoints of sleep on free days (Roenneberg et al. 2003). Average actigraphic midpoint of sleep from all weekends was used to obtain a more reliable estimate of individual differences in sleep timing (Rushton et al. 1983). Weekend actigraphy was available for an average of 3.31 out of 4 d (SD = 1.03). Actigraphy data was not available for two participants; therefore, these participants were excluded from analyses involving midpoint of sleep.

Dim Light Melatonin Onset.

Melatonin concentrations in saliva samples collected in dim light conditions were used to determine DLMO. During dim light conditions (<15 lux at any angle of gaze, confirmed for each visit via a light meter) in overnight laboratory visits, saliva samples were collected in Salivettes (Sarstedt, Newton, NC) every 0.5 h starting 6 h before, and ending 1 h after each participant’s habitual bedtime. Dim light conditions started 1 h before the first saliva collection. During dim light procedures, participants remained seated (other than trips to the bathroom) to control for posture effects, refrained from eating or drinking within 10 min of sample collection, and were not allowed to consume any caffeine, bananas, or chocolate throughout the in-lab visits (Burgess 2010). Otherwise, after eating or drinking, participants were asked to rinse their mouths with water 10 min prior to each sample collection. After collection, saliva samples were frozen at −80 C and shipped overnight on dry ice for radioimmunoassaying by Solid Phase, Inc. (Portland, ME) using commercially available kits (Buhlmann, Amherst, NH). The DLMO was calculated as the clock time when levels exceeded the mean of three consecutive baseline samples plus twice the standard deviation of those samples (Voultsios et al. 1997; Molina and Burgess 2011). DLMO from both Thursday and Sunday assessments were averaged to create a more reliable indicator of individual differences in circadian timing (Rushton et al. 1983). No clear time of DLMO could be determined for four participants, and, therefore, these participants were excluded from analyses involving DLMO.

Note, that while midpoint of sleep on weekend days and DLMO are conceptually and methodologically distinct (sleep timing vs. circadian timing), they are both indices that have been used to approximate the timing of a person’s endogenous circadian rhythm. Accordingly, the midpoint of sleep and DLMO are correlated in this study (r = .35), though to a lesser degree than prior research (Crowley et al. 2006; Kantermann and Burgess 2017). By using two related, yet distinct, markers of sleep/circadian timing, converging findings across these indicators would provide stronger evidence for a role of sleep/circadian timing in alcohol craving and diminish the chance that findings are driven by idiosyncrasies or measurement error related to one assessment type.

Analytic strategy

SAS Enterprise Guide 6.1 (SAS Institute Inc., Cary, NC, USA) was used to estimate a three-level multilevel 24 h cosinor model that tested whether there is a 24 h rhythm in alcohol craving (Zhang et al. 2018; Itzhacki et al. 2019). We focused on a single 24 h harmonic (versus more complicated models with multiple harmonics) to establish preliminary evidence of a potential circadian rhythm in craving. Cosinor models seek to detect whether sinusoidal curve significantly describes observed time-series data. The simplest form of this equation is f(t) = c₀ + c₁sin(2πt ⁄ P), with time (t) and period (P). The periodic function has its baseline value (c₀) at the origin (t= 0), rises to c₀ + c₁ at t = P⁄4, drops to c₀ - c₁ at t = 3P⁄4, and returns to baseline at t = P. This captures the periodic effect if the phase begins at t = 0. To allow a phase shift, a cosine term is included as regressor (i.e., c₂cos(2πt ⁄ P)). Given that this study seeks to model a rhythm which repeats every 24 h, P = 24 and a significant fit of the model indicates the presence of a 24 h sinusoid pattern of alcohol craving.

The three-level cosinor approach used in this study was extended from prior studies that utilized a two-level cosinor model (Hasler, Germain, et al. 2012; Miller et al. 2015). A three-level multilevel model was used because alcohol craving assessments were nested within days, and days were in turn nested within people. In support of using a three-level approach, there was significant variability in random intercepts at both the person-level (ICC = .17, σ² = .09, p < .001) and day-level (ICC = .15, σ² = .08, p < .001). In addition to these random intercepts, random slopes for sine and cosine terms were added because there was significant variability across days and across people in the slopes of the sine (σ² = .06, p < .001; σ² = .03, p = .001) and cosine terms (σ² = .02, p = .002; σ² = .06, p < .001). To isolate a 24 h pattern from a linear increase throughout the day, time-of-assessment was included as a fixed-effect covariate. Following the equation syntax of Raudenbush and Bryk (2002), the equation for this basic three-level model is (note that the subscript p = person-level, d = day level, a = assessment-level):

$$\begin{gathered} \text{Alcohol}\hspace{0.17em}{\text{craving}}_{pda}={\pi }_{0pd}+{\pi }_{1pd}\text{sine}\left(2{\pi \text{t}}_{pda}/24\right)+{\pi }_{2pd}\text{cosine}\left(2{\text{πt}}_{pda}/24\right)+{\pi }_{3pd}(\text{time-} \\ {\text{of-assessment})}_{pda}+e_{pda} \end{gathered}$$ (1)

$$\hspace{1em}{\pi }_{0pd}={\beta }_{00p}+r_{0pd}$$

$$\hspace{1em}\hspace{1em}\hspace{1em}{\beta }_{00p}={\gamma }_{000}+u_{0p}$$

$$\hspace{1em}{\pi }_{1pd}={\beta }_{10p}+r_{1pd}$$

$$\hspace{1em}\hspace{1em}\hspace{1em}{\beta }_{10p}={\gamma }_{100}+u_{1p}$$

$$\hspace{1em}{\pi }_{2pd}={\beta }_{20p}+r_{2pd}$$

$$\hspace{1em}\hspace{1em}\hspace{1em}{\beta }_{20p}={\gamma }_{200}+u_{2p}$$

$$\hspace{1em}{\pi }_{3pd}={\beta }_{30p}$$

$$\hspace{1em}\hspace{1em}\hspace{1em}{\beta }_{30p}={\gamma }_{300}$$

To test whether the 24 h rhythm was influenced by sleep/circadian timing, the average actigraphic midpoint of sleep on weekends and average DLMO was used in separate models as cross-level moderators of the sine and cosine terms. The equation for this moderation model is:

$$\begin{gathered} \text{Alcohol}\hspace{0.17em}{\text{craving}}_{pda}={\pi }_{0pd}+{\text{π}}_{1pd}\text{sine}\left(2{\pi \text{t}}_{pda}/24\right)+{\pi }_{2pd}\text{cosine}\left(2{\pi \text{t}}_{pda}/24\right)+{\pi }_{3pd}(\text{time-} \\ {\text{of-assessment})}_{pda}+e_{pda} \end{gathered}$$ (2)

$$\hspace{1em}{\pi }_{0pd}={\beta }_{00p}+r_{0pd}$$

$$\hspace{1em}\hspace{1em}\hspace{1em}{\beta }_{00p}={\gamma }_{000}+{\gamma }_{001}{\left(\text{Sleep}/\text{Circadian}\hspace{0.17em}\text{timing}\right)}_p+u_{0p}$$

$$\hspace{1em}{\pi }_{1pd}={\beta }_{10p}+r_{1pd}$$

$$\hspace{1em}\hspace{1em}\hspace{1em}{\beta }_{10p}={\gamma }_{100}+{\gamma }_{101}{\left(\text{Sleep}/\text{Circadian}\hspace{0.17em}\text{timing}\right)}_p+u_{1p}$$

$$\hspace{1em}{\pi }_{2pd}={\beta }_{20p}+r_{2pd}$$

$$\hspace{1em}\hspace{1em}\hspace{1em}{\beta }_{20p}={\gamma }_{200}+{\gamma }_{201}{\left(\text{Sleep}/\text{Circadian}\hspace{0.17em}\text{timing}\right)}_p+u_{2p}$$

$$\hspace{1em}{\pi }_{3pd}={\beta }_{30p}$$

$$\hspace{1em}\hspace{1em}\hspace{1em}{\beta }_{30p}={\gamma }_{300}$$

Finally, to provide a more direct statistical test of what 24 h rhythm parameters, i.e., amplitude and acrophase, were related to individual differences in sleep/circadian timing, π_1pd and π_2pd estimates from Equation 1 were used to calculate person-specific estimates of alcohol craving acrophase and amplitude (Hasler, Germain, et al. 2012; Miller et al. 2015). Acrophase reflects the time when the peak level of the rhythm occurred (increases in acrophase indicate a later timing of peak craving); whereas, amplitude reflects the degree of variation of the rhythm (increases in amplitude indicates a larger fluctuation from the mean throughout the day). These person-level indices of acrophase and amplitude were then correlated with sleep/circadian timing variables to provide a statistical test of whether individual differences in sleep/circadian timing were related to individual differences in the timing and amplitude in the 24 h rhythm of alcohol craving.

Missing data

To examine if missingness may bias observed associations among key study variables, i.e., alcohol craving, drinking day, midpoint of sleep, and DLMO, a series of logistic regressions were conducted in which missingness on each key study variable was predicted by the other key study variables. Missingness on key study variables was not related to a person’s level of alcohol craving, number of drinking days, midpoint of sleep, nor DLMO (all p’s > .24).

RESULTS

Descriptive characteristics of and correlations among key participant characteristics are presented in Table 1. On average, alcohol craving was low (~1 on a 5-point scale) and participants reported drinking on ~10% of the study days. Despite alcohol craving being low overall¹, person-average alcohol craving was strongly related to the person-average of percentage of drinking days (r = 0.68, p < 0.001), highlighting the relevance of alcohol craving for the consumption of alcohol. Interestingly, midpoint of sleep on weekends and DLMO had negligible correlations with person-averages of alcohol craving and percentage of drinking days (all r’s ≤ |0.13|). However, such negligible associations do not preclude whether sleep-circadian timing influences intra-daily processes of alcohol craving.

Table 1.

Descriptive statistics and correlations among person-level study variables (N = 32 to 36).

	M	SD	1	2	3	4	5	6	7
1. Age	21.26	1.15	--
2. Sex	61% females	--	.11	--
3. Person-average alcohol craving	1.32	0.32	.02	−.19	--
4. Person-average drinking days	10.30%	7.60%	.20	−.22	.68***	--
5. Actigraphic midpoint of sleep (hh:mm)	5:29	0:54	−.11	−.39*	−.13	.03	--
6. Dim Light Melatonin Onset (hh:mm)	22:06	1:23	.02	−.24	−.09	.00	.35*
7. Alcohol craving amplitude	0.33	0.29	.03	.00	.84***	.41*	−.26	−.18
8. Alcohol craving acrophase (hh:mm)	20.96	2:57	−.06	−.12	−.16	−.10	.35*	.29	−.33†

Note. Craving scale ranges from very slightly or not at all” (1) to “extremely” (5). Person-average alcohol craving and person-average drinking days are person-level averages calculated from all study days. Mean and SD of actigraphic midpoint of sleep and Dim Light Melatonin Onset are reported in 24-hour time.

^†p<.10

^*p<.05

^**p<.01

^***p<.001.

Is there a 24 hour rhythm in alcohol craving?

Mean alcohol craving averaged in 1 h time bins across the day ignoring participant nesting is displayed in Figure 2, which suggests the overall presence of a 24 h rhythm in alcohol craving. To examine whether alcohol craving exhibited a 24 h pattern throughout the day when accounting for the nesting of the data, the model from Equation 1 was estimated. This model revealed that alcohol craving did exhibit a systematic 24 h sinusoid (sine π = −0.17, p < 0.001; cosine π = 0.15, p = 0.003; Null model χ²difference (df = 4) = 1,075.11, p < .001, AIC = 4,279).² The estimated 24 h sinusoidal craving pattern is presented in Figure 3. As depicted, average estimated alcohol craving is lowest (1.02 alcohol craving units) ~08:00h, after which craving increases throughout the day and reaches its peak (1.54 alcohol craving units) ~20:00h. Note that 90% of the alcohol craving reports ranged from 1.00 to 2.00 across all days in the sample. Thus, while the 24 h rhythm in alcohol craving for the entire sample only ranged from 1.02 at its lowest point to 1.54 at its highest point, these daily fluctuations covered ~50% of the typical range of alcohol craving values reported. This suggests that while the absolute magnitude of 24 h fluctuations in alcohol craving may not be large, these fluctuations were substantial when put into the context of the range of naturally occurring alcohol craving.

Figure 2.

Overall raw mean alcohol craving throughout the 24-hour day.

Figure 3.

Estimated 24 hour rhythm in alcohol craving.

Because a 24 h rhythm in alcohol craving may be influenced by daily alcohol intake, which tends to occur later in the day, drinking day was entered as a covariate as well as a day-level moderator of the sine and cosine terms. Even after controlling for whether or not the participant drank alcohol that day, the sine and cosine terms remained significant (sine π = −0.18, p < 0.001; cosine π = 0.12, p = 0.01). Moreover, drinking day moderated the cosine term (cosine*drinking day π = 0.18, p = 0.009). Figure 4 portrays this interaction term and depicts that alcohol craving is greater throughout the entire course of a drinking day, that there is a later peak timing in craving, and a larger amplitude on days in which participants consumed alcohol. Overall, these findings demonstrate that the consumption of alcohol does not account for the 24 h rhythm in alcohol craving, and that drinking days were associated with later timing and stronger 24 h fluctuations in craving.

Figure 4.

Estimated 24 hour rhythm in alcohol craving on drinking days and non-drinking days.

Does the 24 hour rhythm in alcohol craving vary by individual differences in sleep/circadian timing?

The last set of analyses examined whether individual differences in sleep/circadian timing moderated the 24 h rhythm in alcohol craving. These analyses revealed an overall pattern that sleep/circadian timing moderated 24 h patterns in alcohol craving (actigraphic midpoint of sleep*sine γ = 0.10, p = 0.03; DLMO*sine γ = 0.05, p = 0.10; both cosine term interactions not significant, both p’s > .70). The estimated sinusoidal 24 h rhythm in alcohol craving for individuals one standard deviation above and one standard deviation below these sleep/circadian timing variables is presented in Figures 5 and 6. More specifically, individuals one standard deviation below the mean (indicating earlier timing) for both midpoint of sleep and DLMO, reached their trough in alcohol craving (1.19 and 1.15 alcohol craving units, respectively) at 08:00h and their peak (2.09 and 2.03, alcohol craving units, respectively) at ~20:00h. In contrast, individuals one standard deviation above the mean (indicating later timing) for both midpoint and DLMO reached their craving trough (1.32 and 1.26 alcohol craving units, respectively) and peak (1.96 and 1.92 alcohol craving units, respectively) 1 h later (at 09:00 and 21:00h, respectively). Overall, inspection of these figures suggests that individuals with later sleep/circadian timing have smaller amplitudes and later timing of alcohol craving rhythms.

Figure 5.

Estimated 24 hour rhythm in alcohol craving for individuals with earlier vs. later midpoint of sleep.

Figure 6.

Estimated 24 hour rhythm in alcohol craving for people with earlier vs. later DLMO.

To more directly statistically test these visual interpretations, the individual difference estimates of amplitude and phase were correlated with individual differences in sleep/circadian timing. (see Table 1). Later alcohol craving acrophase estimates were positively correlated with later sleep timing midpoint of sleep (r₍₃₃₎ = 0.35, p = 0.04). Note that a similar direction of association emerged with DLMO, though this correlation was not statistically significant (r₍₃₀₎ = 0.29, p = 0.12). Amplitude in alcohol craving did not have statistically significant associations with sleep/circadian timing (actigraphic midpoint of sleep r₍₃₃₎ = −0.26, p = 0.14; DLMO r₍₃₀₎ = −0.18, p = 0.36). Note that individual differences in the amplitude of the 24 h rhythm of alcohol craving were strongly related to the person-level average of alcohol craving (r₍₃₄₎ = 0.84, p < 0.001) and person-level average of percentage of drinking days (r₍₃₄₎ = 0.41, p = 0.02), suggesting that people with larger amplitudes in 24 h rhythms have greater cravings for alcohol and consume more alcohol.

DISCUSSION

These findings provide novel and direct support that late adolescents with regular alcohol use, but without current AUD, exhibit a 24 h pattern in alcohol craving. Data from this study suggest that in our diurnally active sample, on average, 08:00h represents the time of the day when such individuals are experiencing lowest craving of alcohol, whereas 20:00h represents a window of vulnerability when craving may be highest. This window of vulnerability may provide a time that could be targeted in interventions seeking to reduce alcohol use and prevent at-risk young adults from developing AUD. However, findings also indicate that the timing and amplitude of daily rhythms in alcohol craving depend on individual differences in sleep/circadian timing. Specifically, late adolescents with later sleep/circadian timing have a later time of peak craving. Thus, while findings provide an overall guide to what times of the day late adolescents are least and most strongly craving alcohol, i.e., ~08:00 and ~20:00h, respectively, consideration of the sleep/circadian timing of the individual may provide an even more nuanced understanding of these 24 h patterns.

Why is there a 24 hour rhythm in alcohol craving?

The observed 24 h rhythm in alcohol craving likely generates from the influence of multiple, interrelated social, environmental, experiential, and biological factors. For instance, alcohol is typically consumed between 18:00 to 21:00h and least typically consumed between 03:00 to 06:00h (Arfken 1988). These hours coincide with the availability of alcohol and social acceptability of consuming alcohol, which is greatest in the evening when bars are typically open and when the work or school day has concluded. The observed alcohol craving patterns in the current study mirror these alcohol consumption patterns, suggesting that the 24 h pattern in craving may be (i) occurring with concurrent alcohol consumption and (ii) reinforced by prior experiences with when people typically consume alcohol. While it is likely that the observed 24 h pattern in craving may be occurring in tandem with current experiences of drinking, it is important to note that there was still a 24 h rhythm in craving even on days when participants did not report consuming alcohol.

A different reason for why alcohol craving exhibits a 24 h rhythm, even on non-drinking days, is that there may be an endogenous, circadian component to when people crave alcohol. This possibility is supported by prior work identifying similar rhythms in related reward motivation and reward-seeking behaviors, such as engagement in risk-taking and self-reports of wanting and positive affect (Hasler et al. 2008; Murray et al. 2009; Miller et al. 2015; Byrne & Murray 2017; Itzhacki et al. 2019). While many of these prior studies have used methodologies which preclude definitive identification of an endogenous circadian component, prior work has identified a unique circadian contribution to positive affect throughout the day (Murray et al. 2002; Murray et al. 2009). Given that positive affect is theorized to reflect activation of an underlying reward-approach system, such a circadian contribution may generalize to alcohol craving because craving also reflects motivation for reward (Watson et al. 1999). Theoretically, such a rhythm serves to organize appetitive behaviors during times of the day when chances of obtaining resources are greatest and the risk of obtaining those resources is lowest (Watson et al., 1999). For humans, this optimization usually means promoting appetitive behaviors during the biological day and inhibiting them during the biological night. The observed 24 h rhythm in alcohol craving may reflect this circadian organization of behavior wherein craving is promoted during the day and declines during the night.

Role of individual differences in sleep/circadian timing in alcohol craving

While findings identified an average 24 h rhythm in this sample, this rhythm depended on an individual’s sleep/circadian timing as individuals with later sleep/circadian timing tended to have later alcohol craving peaks. Because both midpoint of sleep and DLMO are distinct, yet related, indices frequently seen as approximations of the timing of an underlying circadian rhythm, the convergent results suggest that this underlying endogenous circadian rhythm may be influencing the 24 h rhythm of alcohol craving. Moreover, prior work has utilized self-reports of preferences for sleep timing when examining how sleep/circadian timing relates to 24 h patterns in reward and has called for evaluating these relations using behavioral or objective markers of timing (Miller et al. 2015). This study is the first to tie a behavioral measure of sleep timing (actigraphic midpoint of sleep) and an objective measure of circadian timing (DLMO) to a 24 h rhythm in phenomena that reflect reward processes. These markers more strongly tie biological individual differences in sleep/circadian timing to reward processes and avoid common method biases that may conflate observed associations between self-reported sleep timing and reward-related phenomena.

These findings mirror those in the positive affect literature wherein individuals with later self-reported sleep/circadian timing had later timing in positive affect throughout the day (Porto et al. 2006; Hasler, Germain, et al. 2012; Miller et al. 2015). To the extent that the sleep and circadian timing variables used in this study reflect individual differences in the endogenous timing of the circadian rhythm, the associations with the timing of peak alcohol craving are quite intuitive. If an individual’s endogenous circadian rhythm is delayed in time, then any observed 24 h pattern that is influenced by an endogenous circadian rhythm should also be delayed in time. It is also possible that individual differences in sleep/circadian timing are associated with timing of other activities, e.g., timing of work, that may influence the timing of the 24 h rhythm in alcohol craving.

Limitations and future directions

The current study has limitations that are important to consider when interpreting the findings. First, the study sample consisted of late adolescent alcohol drinkers without an AUD, and findings may not generalize to other populations, particularly those with AUD. While this study recruited late adolescents reporting a range of alcohol use, alcohol craving may not be as high or operate in similar ways as in individuals with AUD, in which craving may be more persistent and difficult to resist. Future questions include whether 24 h rhythms in alcohol craving manifest in individuals with AUD and whether such fluctuations affect alcohol consumption and ability to abstain from alcohol. It is also unclear whether the rhythm in alcohol craving reflects rhythms in actual alcohol consumption and whether findings regarding individual differences in sleep/circadian timing extend to alcohol consumption. The temporal precedence and causal directions among craving, alcohol consumption, and sleep/circadian factors are unclear and future research utilizing more sophisticated assessments of these variables, e.g., event-contingent measurement of alcohol use, will be needed to better understand the directionality.

Additionally, this study utilized a sample of late adolescents, many of whom were undergraduate college students. Given the unique demands on the sleep schedules of undergraduate population, e.g., varying class start times throughout the week, work and study hours, findings regarding sleep/circadian timing may not directly translate to populations with a more consistent sleep schedule. Examining 24 h patterns in craving and how they relate to sleep/circadian timing in other populations, particularly clinical samples with alcohol use disorder or other addictions, represents an important area for future research.

Furthermore, this study sought to examine whether the 24 h rhythm in alcohol craving was associated with individual differences in sleep/circadian timing, but had a relatively small person-level sample size to do so. Given the small person-level sample size, findings at this level of analysis may be unreliable and future research should use larger sample sizes to replicate these findings. For instance, neither sleep/circadian timing variables had associations with craving amplitude that were statistically significant. A true null relationship seems unlikely given that prior work suggests that sleep/circadian timing should influence the amplitude, although it is not entirely clear in what direction (Hasler, Germain, et al., 2012; Hasler et al., 2017; Miller et al., 2015). Given that the craving amplitude was strongly associated with frequency of drinking (r = .41), understanding factors that influence this amplitude may be especially important to understanding and intervening on alcohol consumptions. Future research should utilize a larger sample size to further examine this association in order to provide a more reliable and precise estimate that would allow for more substantive interpretation; effect sizes observed in this study could be used to inform power analyses of such future studies (Schönbrodt & Perugini 2013).

Finally, this study observed a 24 h rhythm in alcohol craving, but it is unable to parse the origin(s) of this rhythm. While prior research suggests that such a rhythm would be partly driven by an underlying circadian rhythm, the current study did not utilize the laboratory-based protocols, e.g., forced desynchrony or constant routine, capable of making such a determination. It possible that this observed rhythm originates entirely from co-occurring social rhythms, though it seems more likely to be a dynamic process resulting from both endogenous circadian and exogenous social influences. Moreover, reports of alcohol craving during early morning hours when people tend to be asleep (e.g., 02:00 to 07:00h) were more sparse than other times of day, and estimation of the rhythm in alcohol craving during that time may be less reliable, though mean levels of alcohol craving during this time in Figure 2 all suggest minimal craving. Future research using forced desynchrony and constant routine protocols will be necessary to definitively determine whether an endogenous circadian rhythm in craving exists and to better capture alcohol craving at times when people traditionally tend to be asleep. Such protocols are costly and intensive to conduct, and findings from this study provide initial evidence justifying allocation of resources to conduct them.

CONCLUSION

These findings demonstrate a 24 h rhythm in the intensity of alcohol craving in a late adolescent population with regular alcohol use, with greatest craving during the evening and least craving in the morning. The exact timing of this rhythm depended on a person’s sleep/circadian timing, with a later peak in craving occurring with later sleep/circadian timing. Although findings should be examined in other populations, particularly those with AUD, the results implicate that rhythmic factors, such as circadian rhythms, influence alcohol craving, suggesting that such factors may provide novel targets for research and interventions seeking to understand and reduce problematic alcohol use.