Late and Instable Sleep Phasing is Associated With Irregular Eating Patterns in Eating Disorders

Abstract

Background

Sleep problems are common in eating disorders (EDs).

Purpose

We evaluated whether sleep-phasing regularity associates with the regularity of daily eating events.

Methods

ED patients (n = 29) completed hourly charts of mood and eating occasions for 2 weeks. Locomotor activity was recorded continuously by wrist actigraphy for a minimum of 10 days, and sleep was calculated based on periods of inactivity. We computed the center of daily inactivity (CenDI) as a measure of sleep phasing and consolidation of the daily inactivity (ConDI) as a measure of daily sleep rhythm strength. We assessed interday irregularities in the temporal structure of food intake using the standard deviation (SD) of frequency (I^FRQ), timing (I^TIM), and interval (I^INT) of food intake. A self-evaluation of other characteristics included mood, anxiety, and early trauma.

Results

A later phasing of sleep associated with a lower frequency of eating (eating frequency with the CenDI rho = −0.49, p = .007). The phasing and rhythmic strength of sleep correlated with the degree of eating irregularity (CenDI with I^TIM rho = 0.48, p = .008 and with I^INT rho = 0.56, p = .002; SD of CenDI with I^TIM rho = 0.47, p = .010, and SD of ConDI with I^INT rho = 0.37, p = .048). Childhood Trauma Questionnaire showed associations with variation of sleep onset (rho = −0.51, p = .005) and with I^FRQ (rho = 0.43, p = .023).

Conclusions

Late and variable phasing of sleep associated robustly with irregular pattern of eating. Larger data sets are warranted to enable the analysis of diagnostic subgroups, current medication, and current symptomatology and to confirm the likely bidirectional association between eating pattern stability and the timing of sleep.

In eating disorders, timing of sleep was late and irregular. Later sleep associated with a lower frequency and later and variable sleep with more irregular eating.

Introduction

The effects of chronic starvation in anorexia nervosa (AN) and of fluctuating eating patterns in bulimia nervosa (BN) on the sleep regulating processes were recognized in polysomnographic studies (PSG) more than two decades ago [1]. Robust data confirm a high prevalence of sleep problems in eating disorders (EDs). For instance, in a longitudinal study among 400 female patients diagnosed with an ED, approximately half of the patients reported sleep problems [2]. The association between sleep and eating patterns is likely bidirectional: while eating rhythms affect sleep [3], patients deprived of adequate sleep experience difficulty in controlling appetite and binge impulses [4]. This is supported by experimental human and animal studies as reviewed previously [3, 5]. Furthermore, patients with poorer sleep quality in self-report show a worse outcome of ED [6, 7].

Most of the reported sleep problems in EDs have to do with the desired and actual timing of sleep [4]. The immediate consequence of such impairment is insomnia, a subjective complaint dependent on subjective needs. A mismatch between the desired and actual timing of sleep and wakefulness, as well as an inability to maintain sleep, can have various causes, including starvation, stress, or trauma [8–10]. Such insults may act on the neural systems that regulate sleep itself, or their input systems, most importantly, the circadian clock [5]. Regardless of the cause and pathways involved, aberrations in the daily timing of sleep and wake, as well as the timing of eating, can have severe consequences on health [5, 11–13].

A promising methodological advance in psychiatric research is measuring the activity and sleep parameters objectively by actigraphy [14–16]. Wrist-worn accelerometric devices are increasingly used to determine phases of rest and activity in field studies across multiple days [16]. As compared to some other psychiatric illnesses [17, 18], knowledge on the rest:activity patterns in ED is very limited. The few existing studies involved recordings limited to few days and/or very small data sets [19, 20], or the actigraphy device was used to measure activity instead of sleep [21]. Thus, the previous studies did not inform about characteristics or correlates of sleep patterns among patients with an ED.

Data on daily eating has been collected to assess frequency, spacing, regularity, skipping, and timing of eating [22]. The temporal distribution of eating events typically associates with sleep timing: we eat while we are awake. Sleep:wake timing, on the other hand, is believed to be strongly influenced by the light:dark cycle in conjunction with the circadian clock. Generally, the 24 hr rhythmic biological processes, including sleep timing are phase coupled to the circadian clock oscillator, which, in turn, is phase coupled or entrained by the day:night oscillations of the solar day. On the other hand, exogenous factors other than light, such as social cues and, most importantly, eating timing, can affect circadian phasing and rhythmicity [3, 5, 23, 24]. Notably, eating timing is particularly dysregulated in patients with an ED, exhibiting binge and/or restrictive eating behaviors [25–27]. However, while monitoring the temporal aspects of feeding is feasible at high resolution in animals and has been shown to affect rest:activity rhythmicity [11, 28], human research on eating patterns in ED has mainly focused on retrospective self-report and content of food consumed.

In this study, we wanted to correlate eating patterns with sleep based on objective, actigraphy-derived data in a representative patient sample from a tertiary care ED program. We quantitatively assessed eating patterns employing a method we introduced recently [29]. As the accelerometry data from the device we used has been validated against PSG [30, 31], the data obtained could be translated into sleep scores. Furthermore, we used self-evaluation of other characteristics that have an impact on eating and sleep, including mood, anxiety, and early trauma. Our hypothesis was that sleep-phasing instability would associate with eating pattern irregularity in ED patients. This would be consistent with the view that the daily shaping of sleep and eating patterns relies at least in part on a common regulatory system or that the control systems of eating and sleep strongly interact.

Methods

Inclusion and Exclusion

Patients were eligible for participation if they had been diagnosed with an ED according to DSM-5 criteria [32], treated as outpatients, capable of consenting, and between the ages of 18 and 65 inclusively. Patients were excluded if their current level of suicidality or symptoms required hospitalization. The protocol for this study was carried out in accordance with The Code of Ethics of the World Medical Association (Declaration of Helsinki) and was approved by the local Ethics Committee. All participants provided written informed consent at the first in-person visit.

Procedure

Study data were collected and managed using REDCap electronic data capture tools [33]. The patients were instructed to provide simultaneous recordings of eating and sleep. Patients documented hourly their mood and eating events, as well as onsets and offsets of rest in journals for 2 weeks (Supplementary Fig. 1) beginning at their inscription (V1). At the same time, participants were outfitted with accelerometric wrist-worn devices (GENEActiv, Activinsights) for measurement of (in)activity and were instructed to wear the devices continuously for 2 weeks. At V2, participants completed a survey of scales selected for the measurement of current mood and eating. Online questionnaires were completed in the clinic or online at the patient’s residence. Participants also participated in a semistructured clinical interview for mood by research assistants. Diagnoses and current medication were obtained by medical chart review.

Participation in the study was offered to 55 patients and 36 (65.5%) of those gave informed consent. For the current analysis, we consider the final data set as the 29 patients for whom continuous accelerometry data were available for a minimum of 10 days with simultaneous journal data [34, 35]. Data were excluded due to too short valid recording of sleep (n = 4) or missing report for eating (n = 3).

Questionnaire Measures

We used the Eating Disorder Examination Questionnaire (EDEQ) [36] to measure the current severity of EDs. Additionally, the scales to measure state-like symptoms included the Clinically Useful Anxiety Outcome Scale (CUXOS) [37], and the revised Reinforcement Sensitivity Theory Questionnaire (rRSTQ) [38] to describe personality features. The rRSTQ was recently used in ED patients and showed good psychometric properties [39]. We also asked about traumatic experiences using the first six items for experiences before the age of 17 from the Childhood Trauma Questionnaire (CTQ), as well as Item 7 “Have you had any other major upheavals, including as an adult, that you think may have shaped your life or personality significantly?” [40]. We used the total sum score of the seven items.

Interview

Mood symptoms were scored during the past week to measure the intensity of the symptoms using the Montgomery–Asberg Depression Scale (MADRS) [41]. While true manic symptoms were not present, the Young Mania Rating Scale (YMRS) [42] was considered as a measure for peak severity of arousal, including agitation, or irritability. We used the Social and Occupational Functioning Scale (SOFAS) [32] to measure functionality.

Descriptors of Eating

To assess the daily pattern of eating and mood, participants charted the time of any food intake (hereafter, the expression meal is used as a synonym for any eating occasion during waking hours over the 2 week period between two in-person visits (V1 and V2). Eating frequency refers, thus, to the daily number of meals or any eating occasions. To visualize daily eating patterns, we generated eatograms by entering eating occasions from the hourly charts into MATLAB (The MathWorks) based custom scripts. Eatograms were displayed in a double plot mode (x-axis = 48 hr, y-axis = days) in order to better discern patterns as is customary for actogram displays in the study of circadian rhythms.

To assess the temporal aspects of eating behavior (eating pattern) quantitatively, we computed three indices based on documented eating events across the 2 week recording period as previously described [29]. Briefly, I^FRQ indicates the standard deviation (SD) of the daily number of meals (SD of meal frequency) and I^TIM represents the SD of the daily timing of meals. We determined I^TIM by calculating the average of the SDs of the timing of each daily meal (Meal 1 across days, Meal 2 across days, etc.). I^INT represents the SD of the intermeal interval. For this index, we first calculated the SDs of the intervals between each consecutive daily meal and, then, calculated their average. These indices served as dimensional measures to assess regularity in the temporal pattern of food intake. A larger index value reflects a more irregular eating pattern, whereas 0 indicates no variation in meal frequency, times, or intermeal interval.

Descriptors of Sleep

Sleep was determined in accordance with an actigraphy-based method that has been previously validated against polysomnography using the accelerometry devices employed here [30, 31]. Raw accelerometry data were extracted from the GENEActiv devices as binary files containing x, y, and z acceleration values sampled at 10 Hz resolution. These values were median filtered using a rolling 5 s window, and an estimate of the arm angle was calculated as follows: angle = $ta{n}^{-1}\left(\frac{a_z}{\sqrt{a_x^2+a_y^2}}\right)$ (where, a_x, a_y, and a_z are the median-filtered acceleration values). Next, the arm angles were averaged across 5 s epochs. Sleep was then defined as a 5 min period with arm angle change between 5 s epochs of less than 5° [30], which provided >90% sensitivity and >80% accuracy for sleep detection in reference to polysomnographic scoring [30]. Accordingly, each 5 s bin was assigned a score of 1 for sleep or 0 for no sleep, and the data were averaged to 1 min bins.

The actigraphy-based sleep score data were converted into double-plotted inactograms, which illustrate the sleep scores across an entire recording period (Fig. 1) using custom MATLAB (The MathWorks) scripts. Self-reported bedtime data from hourly charts were cross-checked for quality with inactograms. As the minimum continuous recording period was 10 days among patients, we only used the first 10 days of actigraphy data for analysis; days with more than 1 hr of nonwear were excluded from the calculation of values.

Fig. 1.

Examples of sleep and eating patterns in patients with an eating disorder. We present two cases. The first shows a relatively regular eating frequency and timing. The sleep is well consolidated around a midpoint of sleep as compared to a midpoint reported in the general population. The second shows very irregular eating, dispersed across the daily cycle, including nighttime, with fragmented sleep and a late midpoint of sleep. Left: daily sleep based on actigraphy-derived sleep scores is displayed as double-plotted inactograms with eating events indicated (filled circles). Tick marks indicate continuous sleep at 5 min resolution. Right: Circular plots representing the average distribution of sleep across the daily cycle. Vector (red) direction indicates the center of inactivity and, thus, the average phasing of the sleep:wake cycle, while vector length indicates the consolidation of daily inactivity as a measure of sleep fragmentation.

Center of daily inactivity (CenDI) is the mean direction of inactivity timing and describes the daily distribution of sleep in radian based on the sleep scores, providing information about the phasing of the sleep:wake cycle, which can be seen as a reflection of chronotype. Consolidation of daily inactivity (ConDI) represents the mean resultant vector length corrected for total duration of sleep, which describes the degree of daily sleep consolidation and, thus, informs on sleep continuity. Both values were calculated using the Circular Statistics Toolbox for MATLAB [43]. “Sleep onset” and “Sleep Offset” were calculated in hours and mark the beginning and end of the “Nighttime Sleep” period. “Nighttime Sleep” was determined by applying a rolling window with a length that is equal to 20% of daily total sleep time divided by ConDI, with total sleep time defined as the sum of all 1 min bins scored as sleep per each daily cycle. The window length had a 120 min cutoff, that is, a 120 min window length was applied to all those days where the calculated window length would exceed this value. To calculate “Nighttime Sleep,” the 24 hr time span defining one daily cycle was centered at mean CenDI, that is, a single cycle stretched ±12 hr from mean CenDI. The rolling window as defined above was then moved by 1 min increments across the 24 hr cycle as defined above and periods of sleep ended when the moving window started to encompass zero sleep scores. The longest sleep period per 24 hr cycle calculated in this manner was then defined as the “Nighttime Sleep” period. “Sleep onset” and “Sleep Offset” were defined as the beginning and end of this period and superimposed on inactograms for visual inspection of accuracy (data not shown). The calculated timing of sleep showed good correlation with the patient-reported timing (objective vs. subjective sleep midpoint (rho = 0.85, p < .001; Supplementary Fig. 2); this confirms the appropriateness of the actigraphy-derived variables. Notably, a similar approach has been recently reported to correlate well with PSG-determined sleep onsets and offsets [31]. Correlations between the dimensional sleep variables are presented in Supplementary Table 1 and correlations between the dimensional eating variables are presented in Supplementary Table 2.

Statistical Analyses

Data were analyzed employing SPSS version 25.0. Due to mostly nonnormally distributed values, we used Mann–Whitney U-Test in between-group comparisons and bivariate Spearman correlations with two-tailed significance. For descriptive purposes, significant findings were defined by p-values <.05.

Results

Description of the Cohort

All participants were female, and the median age at study onset was 26.0 years (Table 1). Seven patients had a principal diagnosis of BN, 11 AN (3 restricting and 8 bingeing/purging), 9 other specified feeding or ED (OSFED) (5 restricting, 3 purging, and 1 mixed), and 1 avoidant/restrictive food intake disorder (ARFID). The range of body mass index (BMI) was from 15.20 to 39.8 kg/m², with three patients below the limit of mild AN (<17.5 kg/m²) and three above the obesity limit ≥30 kg/m². One patient had sleep apnea and was using a continuous positive airway pressure machine, but no other patients had a primary sleep disorder. Most patients (21, 72.4%) were using psychoactive medication, including antidepressants (15, 51.7%), antipsychotics (6, 20.1%), hypnotics (8, 27.6%), and stimulants (2, 5.9%).

Table 1.

Sociodemographic and clinical descriptors of a sample of 29 eating disorder patients

Characteristic
	Median	IQR
Age	26.0	9.75
BMI, kg/m2	20.9	6.25
Depression (MADRS)	20.0	12.0
Manic symptoms (YMRS)	4.0	4.0
Functionality (SOFAS)	58.0	12.5
Anxiety (CUXOS)	35.0	23.3
EDEQ, global score	4.0	1.60
Trauma (CTQ)	18.0	9.0
Reward sensitivity (rRSTQ)	55.0	8.75

Depression (MADRS) = Montgomery-Åsberg Depression Scale total score; Manic symptoms (YMRS) = Young Mania Rating Scale total score; SOFAS = Social and Occupational Functional Assessment Scale; Anxiety (CUXOS) = Clinically Useful Anxiety Outcome Scale self-rating scale total score; EDEQ = Eating disorder examination questionnaire mean global severity score; Trauma (CTQ) = Childhood Trauma Questionnaire total sum score of Items 1–7; Reward sensitivity = Revised Reinforcement Sensitivity Theory Questionnaire (rRSTQ) total score;

BMI, body mass index; IQR, interquartile range.

The description of mood, anxiety, and personality based on the diagnostic groups are presented in Table 1. Most patients had depressive symptoms to some extent, with 15 (51.7%) reporting moderate (MADRS 18–34), and 2 (6.9%) severe depressive symptoms (≥35) [44]. While none of them had a hypomanic episode, four patients were scored to the range of mild manic symptoms (YMRS scores ≥12 but ≤20 [45]). On average, the patients also scored high on CUXOS, with 11 (37.9%) scoring above severe anxiety (≥41). The median eating frequency in the sample was 3.75 (interquartile range [IQR] 1.84), median CenDI 3:40AM (IQR 1.57), median sleep onset 24:18 AM (IQR 3:03), and median sleep offset 7:46 AM (IQR 1:51).

Correlation Between the Descriptors of Sleep Pattern and Eating Pattern

A later timing of sleep associated with a lower frequency (FRQ) of eating as seen in a correlation of FRQ with CenDI and sleep onset, respectively (Table 2). Furthermore, I^TIM and I^INT, both describing irregular eating, were moderately/strongly correlated with all descriptors of the timing of sleep (CenDI, sleep onset, and sleep offset). I^TIM also correlated with SD of CenDI, indicating that variability in the timing of sleep associated with the variability in meal timing.

Table 2.

Correlations between sleep and eating pattern in a sample of 29 eating disorder patients

	Frequency		IFRQ		ITIM		IINT
	Rho	p	Rho	p	Rho	p	Rho	p
CenDI	−0.49	0.007	−0.30	0.12	0.48	0.008	0.56	0.002
SD	−0.32	0.091	−0.96	0.62	0.47	0.010	0.37	0.048
ConDI	−0.24	0.21	−0.074	0.70	0.077	0.69	0.21	0.28
SD	0.010	0.96	−0.037	0.85	0.025	0.90	−0.17	0.39
Onset	−0.42	0.024	−0.22	0.26	0.41	0.028	0.53	0.003
SD	−0.23	0.24	0.009	0.96	0.075	0.70	0.15	0.44
Offset	−0.28	0.14	−0.15	0.44	0.42	0.022	0.46	0.012
SD	0.22	0.26	0.15	0.43	0.00	1.0	−0.070	0.72
Nighttime sleep	0.13	0.49	0.13	0.52	−0.13	0.52	−0.15	0.43
SD	−0.34	0.075	−0.20	0.30	0.13	0.52	0.12	0.54

CenDI = center of daily inactivity. This measure serves as approximate of the midpoint of sleep, which is known to reflect chronotype (ref needed); SD = standard deviation as a descriptor of variance in each measure; ConDI = vector length of the mean midpoint of sleep during 24 hr, normalized for the length of sleep; Onset = mean sleep onset in hours; Offset = mean sleep offset in hours; Nighttime sleep = mean total nighttime sleep; Frequency = the daily number of meals (any eating occasions during the day). I^FRQ = the SD of the daily number of meals (SD of meal frequency). I^TIM = the SD of the daily timing of meals. We determined I^TIM by calculating the SDs of each daily meal and, then, by averaging these SDs. I^INT is the SD of the intermeal interval. For this index, we first calculated the SDs of the intervals between each consecutive daily meal and, then, averaged these SDs. These indices served as dimensional measures to assess regularity in the temporal pattern of food intake. A larger index value reflects a more irregular eating pattern, whereas 0 indicates no variation in meal frequency, times, or intermeal interval. Rho = Spearman’s rank correlation coefficient.

We ordered the individual patient data by increasing variability of the intervals between eating occasions (index I^INT; Fig. 2), CenID (Supplementary Fig. 3), and I^TIM (Supplementary Fig. 4). Observation of these panels provides further evidence for the association between patterns of sleep and eating.

Fig. 2.

Inactograms of 29 eating disorder patients with eating events indicated. Plots are ordered by increasing variability of the intervals between eating occasions (index I^INT). The vector in the adjacent circular plots indicates the center of daily inactivity and consolidation of the daily inactivity, that is, the phasing (vector direction) and consolidation (vector length) of the sleep–wake cycle, respectively.

Correlations of Sleep and Eating Patterns With Other Clinical Characteristics

Associations between sleep and eating patterns were not better explained by other clinical features (Table 3). However, a higher eating frequency correlated with less severe depressive symptoms (MADRS) and better functionality. A higher EDEQ score correlated with later offset of sleep, higher SD of nighttime sleep, and lower eating frequency. A higher CTQ total score correlated with later sleep onset and with I^FRQ (Table 3).

Table 3.

Correlations between sleep and eating patterns with mood in a sample of 29 eating disorder patients

	MADRS	YMRS	SOFAS	CUXOS	EDEQ total	rRSTQ	CTQ total
CenDI	−0.076	0.11	0.028	−0.007	0.14	−0.013	−0.314
SD	−0.17	0.24	−0.29	0.092	0.096	−0.11	0.019
ConDI	0.062	−0.19	−0.081	−0.23	0.004	−0.31	−0.31
SD	−0.10	0.21	0.02	0.079	0.13	0.14	0.091
Onset	−0.13	0.004	0.035	−0.063	−0.12	0.025	−0.51*
SD	0.34	0.16	−0.23	0.19	0.27	0.016	0.030
Offset	0.023	0.016	−0.093	0.19	0.39*	0.20	0.023
SD	−0.015	0.27	0.082	−0.12	−0.077	−0.006	0.26
Nighttime sleep	0.093	−0.14	−0.13	0.25	0.26	0.077	0.32
SD	0.30	0.25	−0.19	0.22	0.53**	0.18	0.13
Frequency	−0.36*	−0.20	0.39*	−0.25	−0.42*	−0.077	0.15
IFRQ	−0.27	0.045	0.15	−0.040	−0.33	0.13	0.43*
ITIM	−0.18	0.15	0.012	0.13	0.25	0.18	0.098
IINT	0.006	0.002	−0.16	0.034	0.20	0.14	−0.036

CenDI = angle of the midpoint of sleep as reflecting chronotype; SD = standard deviation as a descriptor of variance in each measure; ConDI = vector length of the mean midpoint of sleep during 24 hr, normalized for the length of sleep; Onset = mean sleep onset in hours; Offset = mean sleep offset in hours; Nighttime sleep = mean total nighttime sleep in minutes; Frequency = the daily number of meals (any eating occasions during the day). I^FRQ = the SD of the daily number of meals (SD of meal frequency). I^TIM = the SD of the daily timing of meals. We determined I^TIM by calculating the SDs of each daily meal and, then, by averaging these SDs. I^INT is the SD of the intermeal interval. For this index, we first calculated the SDs of the intervals between each consecutive daily meal and, then, averaged these SDs.

MADRS, Montgomery–Åsberg Depression Scale total score; YMRS, Young Mania Rating Scale total score; SOFAS, Social and Occupational Functional Assessment Scale; EDEQ, Eating disorder examination questionnaire total score; CUXOS, Clinically Useful Anxiety Outcome Scale self-rating scale total score; rRSTQ, Revised Reinforcement Sensitivity Theory Questionnaire total score; CTQ, the Childhood Trauma Questionnaire total score of Items 1–7.

*p-value 0.01 to 0.05.

**p-value <0.01.

Discussion

This is the first study to describe rest:activity rhythms in individuals with EDs using an objective measure, wrist actigraphy. We found that late and variable phasing of sleep associated robustly with an irregular pattern of eating and eating frequency. The associations of current ED symptoms, mood, or anxiety with eating patterns were separate from the associations between sleep and eating patterns. However, an association of reported early trauma was seen with the sleep patterns. Notably, the association between eating and sleep was very robust considering the presence of several other characteristics that could affect both eating and sleep, most importantly, primary diagnosis, medication, phase of treatment, BMI, mood, or anxiety.

We are the first to present a detailed analysis of both sleep and eating patterns in a sample of ED patients. Our work provides the basis for tools for clinical interventions and research to further understand causal relations between eating and sleep patterns. Yet, this is a proof of concept study where the size of the sample is small, and the results await replication. However, the findings are statistically strong given the size of the data, and they repeat consistently with similar measures. A major strength is that we used sleep scores based on actigraphy data, which is well validated to objectively track a person’s rest:activity patterns [30, 46]. The wrist-accelerometry device used is affordable, noninvasive, waterproof, and transportable, enabling a month of continuous recording without a need for being recharged. It, thus, permits the patients to adhere to their daily routines as opposed to PSG, which is much more invasive [47]. The observed late and variable phasing of sleep and the associated irregular meal patterning could be due to internal timing system perturbations. Alternatively, it can result from external factors, such as social and workplace-related demands or medication. However, even external factors could exert their effect on sleep and eating via internal timer perturbation. In future studies, evaluating central pacemaker phasing more directly by measuring dim-light melatonin onset or core body temperature will likely shed more light on the true causes of the observed simultaneous aberrations in eating and sleep.

Definitions for the pattern of eating are used inconsistently in research on EDs and this has to be taken into account when comparing our work to previous literature about meal patterns. One of the commonly used measures for patterns is the Eating Disorder Examination (EDE), which is based on the subjective and retrospective frequency of predefined eating events, that is, the average occurrence of breakfast, lunch, and an evening meal, as well as snacks during the 28 days immediately prior to assessment. Using this approach, in a sample of ED patients, dinner was the most and breakfast the least commonly consumed meal, snacking was most often reported during the evening time, and nocturnal eating was infrequent [26].

In treatment, regular eating patterns are defined as three meals and two snacks per day and refer to the frequency and content of the meals. Psychoeducation toward regular eating is a routine focus of therapy to treat ED [48]. A regular eating frequency has shown beneficial effects on symptom severity, especially, the frequency of bingeing in BN [25]. Accumulating evidence also supports the idea that manipulating eating patterns can have an impact on circadian rhythm and sleep, while optimal eating pattern for health remains open [49]. For instance, a shift of three evenly spaced daily meals toward later hours appeared to cause a corresponding delay in the phasing of the central circadian pacemaker [50]; this seems to suggest that clock phasing is sensitive to feeding cues. A circadian misalignment as seen, for example, in shift work results in dysbalanced energy homeostasis [51], which can partly be compensated for by maintaining an eating rhythm that is aligned with the day:dark cycle.

The size of our data set was not sufficient to reliably exclude differences between diagnostic subgroups in sleep or eating patterns. People with AN report an increased prevalence of sleep disturbances, especially, delayed sleep onset, early morning awakening, and poorer sleep quality [1, 4, 52]. While the previously published data on sleep is very limited in binge-purge AN and BN patients, results remain inconsistent [1, 4, 52]. A recent PSG study with 23 BN and AN patients reported disrupted sleep in both groups when compared to controls [53]. Patients with AN also typically follow rigid dietary behaviors, such as fixed meal times [26, 54, 55], while BN and binge eating tend to be associated with more chaotic and inconsistent dietary behaviors, as well as greater intraindividual variability as compared to AN [56–58].

We selected a transdiagnostic approach to ED. This theory suggests that common core psychopathology is the central component across a range of EDs and has been used for phenotypes, such as restraint model of binge eating [59]. Future studies should replicate our findings in larger data sets to show whether specific features of the sleep pattern represent a transdiagnostic finding reflecting core psychopathology or are specific to subgroups or certain phenotypes, such as purging.

Limitations of the study include the lack of a structured evaluation of sleep disorders. Most importantly, we did not differentiate current and chronic problems with the timing of sleep and, thus, cannot exclude primary circadian sleep–wake disorders. We also did not have an age-matched control group. However, large data sets with noncomparable methods have described characteristics of rest:activity rhythm in the general population; for sleep, see, for example, Mitchell et al. [60] and, for eating patterns, see Kant and Uzhova et al. [61, 62]. Our approach to eating occasion recording does not make any assumptions on the type of food intake at a given eating occasion. Patients with an ED have a tendency for restriction most of the time, and occasional extreme energy content, often followed by purging. Thus, the energy content of eating occasions is highly variable and not comparable to the general population. We cannot exclude the enrichment of patients with sleep problems. The patients received psychoeducation as part of the treatment, including advice to eat frequently. Thus, the findings about the characteristics of sleep and eating pattern might not be epidemiologically generalizable to all ED patients. However, these factors can be expected to rather weaken the association between sleep and eating patterns.

In conclusion, irregular eating was strongly associated with later and variable sleep phasing in our sample of ED patients. Our results suggest that, in addition to cognitive and/or social influences, dysregulated eating pattern in EDs could be partly explained by internal dysregulation of rest:activity rhythms.

Supplementary Material

Supplementary material is available at Annals of Behavioral Medicine online.

Figure S1. Hourly journal for eating and sleep used to define eating occasions and subjective onset and offset of sleep.

Figure S2. Correlation between calculated (based on actigraphy records) and patient reported sleep on- (left) and offsets (right) for the cohort of 29 patients with an ED. Shown are individual daily on- and offsets (blue) and their averages (red) for a given patient.

Figure S3. Inactograms of 29 ED patients with eating events indicated. Plots are ordered by increasing variability of the timing of sleep midpoint (CenDI).

Figure S4. Inactograms of 29 ED patients with eating events indicated. Plots are ordered by increasing variability of the timing of eating occasions (index I^TIM).

Compliance with Ethical Standards

Authors’ Statement of Conflict of Interest and Adherence to Ethical Standards The authors report no conflicts of interest.

Funding This work was supported by The Fonds de Recherche du Québec – Santé (FRQS) #8400863 and #252872 (OL); The Canadian Institutes of Health Research (CIHR) #366858 and The Natural Sciences and Engineering Research Council of Canada (NSERC): #RGPIN-2015-04034 (K-FS).

Authors’ Contributions O.L., H.S., and K.F.S. planned the study protocol; A.B. and D.S. participated in planning the data collection methods; O.L., H.S., O.C., and D.S. were in response of data collection and processing clinical data; C.B., K.F.S., and O.L. planned the methods for dimensional measurement of sleep and eating patterns; O.L. and O.C. conducted the statistical analysis; O.L. and K.F.S. wrote the first draft; all authors approved the final manuscript.

Ethical Approval The protocol was accepted by the Douglas Mental Health University Institute ethical committee (protocol 15/47).

Informed Consent All participants provided an informed consent.