The acute effects of aerobic exercise on sleep in patients with depression: study protocol for a randomized controlled trial

This trial does not exclude patients with psychiatric comorbidities in order to preserve external validity. Therefore, we perform an extensive baseline characterization of the sample (Fig. 2). The characterization consists of questionnaires on psychiatric, somatic, and insomnia-specific symptom severity as well as sleep-related variables such as chronotype and sleep-related cognitions. Furthermore, baseline polysomnography will provide an extensive characterization of sleep.

Sleep disorders are of complex etiology and have various psycho-physiological consequences. Therefore, it is recommended to measure the effects of interventions on insomnia symptoms as well as other factors such as daytime functioning and mood [81]. Insomnia can be classified into sleep-onset, sleep-maintenance, and sleep-offset subtypes. Meta-analysis of polysomnographic data has shown that patients with depression and other mental health disorders have mixed (i.e., sleep-onset and sleep-maintenance) insomnia [82]. Sleep efficiency best captures both of these aspects and is therefore chosen as the primary outcome. Objective quantification of the primary outcome is vital in this study since the blinding of participants is not possible, thereby potentially affecting subjective measurements.

Secondary outcomes which inform clinical decision making include further polysomnographic variables, subjective pre-sleep arousal, cardiovascular autonomic modulation, subjective sleep quality, daytime sleepiness, and adverse events. Wake after sleep onset and number of awakenings identify sleep-maintenance insomnia, while sleep-onset latency characterizes sleep-onset insomnia.

Patients with depression show characteristic changes in rapid eye movement (REM) latency (time between sleep onset and first rapid eye movement sleep episode), REM-density (frequency of rapid eye movements/REM episode), and duration of REM sleep (collectively known as REM pressure) [82]. More importantly, these alterations have a negative impact on treatment and increase the risk of relapse [83]. Exercise has been shown to reduce REM sleep and increase slow wave sleep [54]. The discrepancy between objective and subjective sleep measurements is well documented [81, 84, 85]. Therefore, we measure the subjective sleep quality of the previous night in addition to the objective sleep assessment. Patient views influence the treatment of sleep disorders [86, 87]. Hence, it is crucial to ascertain the credibility of the intervention and the expectancy of participants. Adverse events are underreported in sleep trials but highly relevant to clinical decision making [88]. The case report form specifically captures adverse events (see below) to gauge the benefit to harm ratio.

In accordance with the International Consortium for Health Outcomes Measurement (ICHOM) Depression and Anxiety working group, depression symptoms are assessed with the German version of the Patient Health Questionnaire-9 (PHQ-9) including the additional question on functioning [95]. The nine symptom items are scored on a four-point Likert scale (not at all to nearly every day; Cronbach’s α = 0.89). The cut-offs of the aggregated sum score have been validated and allow for classification of mild to severe depression [96]. Anxiety is assessed using the Hospital Anxiety and Depression Scale. This questionnaire measures depression and anxiety with seven items each on a four-point Likert scale [97]. A meta-analysis has found Cronbach’s α to be 0.83 and 0.82 for the anxiety and depression subscales, respectively [98]. Diagnostic test accuracy of the depression and anxiety subscale is high [99]. More recently, a meta confirmatory factor analysis has suggested the presence of a general distress factor explaining most of the covariance between items [100]. The German version has been shown to have adequate psychometric properties [101]. We measure stress with the German version of the ten-item Perceived Stress Scale, which has demonstrated good reliability (Cronbach’s α = 0.84) and validity [102]. Therein, stress is operationalized as the degree to which life is experienced as unpredictable, uncontrollable, and overloaded in the past months on a five-point Likert scale (never to very often). Somatic multimorbidity is measured using the self-administered Patient Health Questionnaire Somatic Symptom Scale (PHQ-15), which has good reliability (Cronbach’s α = 0.8) and validity [103]. This questionnaire measures the presence and severity of somatic symptoms during the past 4 weeks on a three-point Likert scale (not bothered at all to bothered a lot). The items assess symptoms clusters which account for more than 90% of physical complaints reported in out-patient settings [103]. Lastly, the German version of the Modified Cumulative Illness Rating Scale provides physician-rated scores of multimorbidity. This scale has a good inter-rater agreement and concurrent and predictive validity. It measures the presence and severity multimorbidity over 14 organ systems on a five-point scale (no problem to extremely severe problem) [104].

Multiple sleep-related variables are measured to characterize the impact of sleep disorders in the population. The German version of the Pittsburgh Sleep Quality Index [105] is used to quantify subjective sleep disturbance. This 18 item scale assesses subjective sleep quality, sleep latency, sleep duration, habitual sleep efficiency, sleep disturbances, use of sleeping medication, and daytime dysfunction. It has adequate psychometric properties (Cronbach’s α = 0.83) [106]. The corresponding global score with a cut-off value of ≥ 5 has been shown to distinguish good and poor sleepers [107]. We measure sleep reactivity, i.e., the degree to which an individual experiences disrupted sleep due to stress, with the Ford Insomnia Response to Stress Test [108, 109]. The nine-item self-report questionnaire is answered on a four-point Likert scale (not very likely to very likely) with higher total scores indicating a higher likelihood of stress-induced insomnia. The reliability (Cronbach’s α = 0.80) and validity of the German version have been demonstrated [110]. Dysfunctional sleep-related thoughts and attitudes are related to the development and trajectory of sleep disorders [111]. The reliability and validity of the German short form of the dysfunctional beliefs and attitudes about sleep scale have recently been demonstrated (average Cronbach’s α = 0.71) [112]. Sixteen items measure four domains (consequences, worry/helplessness, expectations, medication) on a ten-point Likert scale (strongly disagree to strongly agree). Excessive daytime sleepiness is associated with insomnia and depression [113, 114] as well as reduced quality of life and work performance [115]. We assess chronic daytime sleepiness using the German version of the Epworth Sleepiness Scale. This questionnaire has good reliability (Cronbach’s α 0.83 in patients) and validity [116]. The likelihood of dozing off in eight daily situations (e.g., sitting and reading) is assessed on a four-point Likert scale (would never doze to high chance of dozing). Chronotype, i.e., peak alertness throughout the day, is associated with more severe depression, insomnia, and suicidality in patients with major depression. Furthermore, eveningness substantially increased the risk of non-remission independently of insomnia severity [117]. The Morningness-Eveningness Questionnaire assesses chronotype with 19 multiple-choice questions (four- or five-point scale) on sleep habits and propensity for performance throughout the day. The sum ranges from 16 to 86 and can be translated into chronotypes (< 42, evening type; 42–58, neither; > 58, morning type) [118]. Criterion validity has been established [119]. Validity and reliability (test-retest reliability > 0.96) have been confirmed for the German version [120].

Polysomnography is performed with the SOMNOscreen™ plus RC (Somnomedics, Randersacker, Germany) using the following montage: one EEG channel (Fp2-A1, 512 Hz), two EOG channels (1 cm below and 1 cm lateral of the outer right canthus as well as 1 cm above and 1 cm lateral of the outer left outer canthus, 512 Hz), one EMG channel (Chin1-Chin2, 512 Hz), one ECG channel (modified lead II, 512 Hz), thoracic respiratory effort channel (inductance plethysmography belt, 32 Hz), finger photoplethysmography (non-dominant arm, 128 Hz), body position (stored every 30 s), movement (32 Hz), and ambient light (stored every 30 s). Relevant skin areas are prepared with Nuprep (Weaver & Co., Aurora, CO, USA) before electrode placement to reduce skin impedance. Although this montage does not comply with AASM standards [128], the use of a single channel EEG, two EOG channels, and one EMG channel has been shown to have good validity when assessing sleep stages [129].

We apply low- and high-frequency filters according to AASM guidelines [128]. Oxygen desaturation index (ODI) is defined as the number of oxygen desaturations (≥ 4%) measured by photoplethysmography per hour of total sleep time [128]. We analyze polysomnography data visually in epochs of 30 s by two trained, independent, and blinded raters according to AASM guidelines [128]. If sleep efficiency differs by ≤ 5% between raters, we use the mean of each quantitative sleep parameter of both raters for analysis. If sleep efficiency differs by > 5% between raters, a consensus scoring is done by a third rater who is blinded against allocation and previous ratings. Inter-rater reliability (intraclass correlation) based on the scoring of the first two raters will be provided for the primary outcome. Sleep parameters are calculated with the proprietary DOMINO software (Somnomedics, Randersacker, Germany) (Table 2).

Sleep parameters can be grouped into three domains:

Sleep hygiene recommendations include daily exercise. Two caveats of this recommendation are that exercise should be performed before 2 pm and that no strenuous exercise ought to be done close to bedtime. One reason for this recommendation is that exercise in the afternoon or evening might increase arousal and thereby prevent sleep [59]. However, evidence from numerous epidemiological, observational, and experimental studies have repeatedly failed to show such an adverse effect or have found that the opposite is true [54, 130]. The issue of nocturnal autonomic modulation following exercise is of central importance for three reasons. Firstly, arousal is integral to one of the most widely accepted etiological theories of insomnia [15, 16]. Secondly, heart rate variability (HRV), as a marker of autonomic arousal, has been proposed as a potential pathophysiological mechanism linking depressive disorders [131, 132] and insomnia [133] with cardiovascular disease. Thirdly, nocturnal blood pressure and nocturnal blood pressure dipping are of prognostic relevance for cardiovascular disease [134]. Therefore, sleep quality, as well as nocturnal autonomic activity, should be assessed when evaluating interventions in sleep research. Measurements of heart rate, HRV, and pulse transit time (and thereby calculated blood pressure) are collected to quantify the effect of exercise on parameters of autonomic cardiovascular arousal.

We record the ECG while subjects are lying in bed. Patients were instructed to lie in the supine position and refrain from speaking as well as moving during the two short-term (i.e., 5-minute) recordings. No instructions concerning breathing were given. All ECG measurements were sampled at 512 Hz using modified lead II.

Three sections of HRV data are analyzed:

The time- and frequency-domain measures which we calculate are described in Table 3. Low (LF) and high frequency (HF) power in normalized units (LFnu and HFnu) will not be reported. The rationale for this deviation from guidelines is that LFnu, HFnu, and LF/HF ratio have been shown to carry algebraically and therefore physiologically redundant information [142, 144]. Reporting of data collection, analysis, cleaning, and calculation will follow the GRAPH guidelines [145]. The percentage of beats identified as artifacts will be reported for both groups for each segment (median and inter-quartile range). Measurements with > 5% artifacts will be excluded from analysis.

An ongoing debate in the literature concerns the choice of method for spectral analysis. Fast Fourier transform (FFT) using Welch’s periodogram and autoregressive modeling (AR) seem to be the most frequently used methods for power spectral analysis [135, 143]. Another method to estimate power spectral density is the Lomb-Scargle periodogram (LSP) [146, 147]. LSP has numerous advantages when compared to FFT and AR. LSP is designed to estimate the power density spectrum directly from the unevenly sampled tachogram. AR and FFT, on the other hand, need to be interpolated and resampled to fulfill the prerequisite of evenly sampled data. Resampling leads to over- and under-estimation of LF and HF, respectively [148, 149]. FFT and AR require a trade-off between frequency resolution and time resolution (statistical stability) when choosing window length [150] and model order [151], respectively. In contrast, LSP makes no assumptions of models. Investigations have shown that LSP is more accurate [152‐154], is less noisy [154], has higher reliability [155], and is more sensitive to physiological changes [155‐157] when compared to FFT. Consequently, multiple authors have suggested LSP as the method of choice for spectrum analysis of HRV [152, 158]. For these reasons, we estimate power spectral density using LSP smoothed with a moving average filter (width 0.02 Hz) in this trial. To enable comparison with other studies, we also calculate Welch’s FFT and will report it as a sensitivity analysis.

We perform ECG pre-processing and HRV analysis using Kubios HRV (University of Eastern Finland, Kuopio, Finland) [159]. This software has been shown to have perfect intra-class correlation coefficients (ICC = 1.000) across HRV variables measured during different postures when compared to two other software [120]. QRS detection is based on the Pan-Tompkins algorithm [160], including bandpass filtering. Beat detection is visually inspected. Erroneously detected RR fiducial points are corrected by manual editing of R-wave. Artifacts are identified using the automatic artifact correction algorithm [161]. This algorithm has been shown to detect ectopic beats with an accuracy of 97%. Ectopic beats are replaced by phantom beats using cubic spline interpolated RR values. Correction of aberrant RR intervals using cubic spline interpolation has been shown to perform as well as other correction methods for frequency analysis [162]. Detrending, i.e., removal of slow-trend and non-linear trend components is performed using the smoothness priors approach [163] with λ = 500 and f_c = 0.035 Hz (thus not affecting the lower band of LF). Frequency-domain variables estimated by LSP are based on de-trended RR series. Frequency-domain variables estimated by FFT with Welch’s periodogram are based on de-trended as well as interpolated (i.e., resampled) RR series. We employ the following parameters for power spectral density estimation using FFT. RR series are resampled to obtain an evenly sampled time series using a cubic spline interpolation with a rate of 4 Hz. A Hann window with a width of 60 s and 50% overlap is used (corresponding to 240 samples). These window parameters are chosen to balance the requirement of stationarity and frequency resolution resulting in an average of 9 and 119 FFT spectra for 5 min and hourly segments, respectively, with a frequency resolution of 0.025 Hz. Singh et al. [150] have shown that these approximate parameters produce a good spectral estimate (smooth with clear peaks).

Pulse transit time is calculated using ECG and pulse waveform from photoplethysmography. Blood pressure (BP) is calculated using the pulse transit time. This method has been validated [164] according to the European Society of Hypertension International Protocol (ESH-IP) revision 2010 criteria [165]. Accordingly, we perform a single initial calibration measurement (manual cuff-based method, contralateral arm of photoplethysmography, sitting position). Two significant advantages of this method are the continuous measurement of BP and elimination of cuff inflations. The latter is poorly tolerated by patients, causes awakenings, and may affect the validity of BP measurements [166]. BP levels differ between non-REM and REM sleep [167]. Therefore, mean systolic, mean diastolic, and mean arterial pressures will be reported separately for total sleep time, non-REM sleep, and REM sleep.

Patients allocated to the intervention group perform a single bout of supervised aerobic exercise. The starting time is approximately 1645 h. The exercise mode is a bicycle ergometer (ergoselect 200, Ergoline, Bitz, Germany). After a warm-up period of 5 min, during which the intensity is gradually increased, patients maintain an intensity of 80% of the individual anaerobic threshold for 30 min. The intensity level is chosen based on clinical experience that this corresponds to an approximate rate of perceived exertion of 13 (on a scale from 6 to 20) in this population. The duration of exercise corresponds to physical activity recommendations [181].

At the same time as individuals performing the exercise intervention, individuals allocated to the control group are placed in a room which is comparable to that of the exercise group concerning light, temperature, and absence of music. The control group is asked to remain seated and read magazines.

The rules and schedules of the inpatient rehabilitation clinic (e.g., timing of meals, consumption of multimedia, and alcohol) limit the variability of many behavioral aspects which could influence sleep. Occasional smokers are asked to refrain from smoking after dinner on the days of polysomnographic assessment. Chronic smokers are not asked to abstain from smoking as this might be an additional stressor.

We assess the implementation of the intervention with continuous measurement of Watts and heart rate using a Polar® H7 chest strap (Polar OY, Finland). We measure heart rate throughout the intervention period, including 3 min post-exercise. We measure perceived exertion using the Borg scale (6–20) [93]. All subjects complete a questionnaire on their current mood immediately before and at the end of the control condition as well as the exercise intervention. The Befindlichkeitsskala has adequate psychometric properties [182] and is considerably more economical than other comparable measures [183]. The questionnaire consists of 40 items on a five-point Likert scale (not at all to very much). Items load onto eight subscales (with five items each): activity (Cronbach’s α = 0.82), elation (Cronbach’s α = 0.81), contemplation (Cronbach’s α = 0.70), calmness (Cronbach’s α = 0.78), fatigue (Cronbach’s α = 0.88), depression (Cronbach’s α = 0.80), anger (Cronbach’s α = 0.86), and excitement (Cronbach’s α = 0.73) [182]. Contamination through any or additional physical activity (depending on the allocation) is assessed using a wrist-worn accelerometer (on non-dominant hand) on the days prior to and after the sleep assessments. The wrist-worn accelerometer vivofit®2 (Garmin, Schaffhausen, Switzerland) validly assesses steps in various walking conditions [184]. Although adequate blinding cannot be achieved by design in exercise studies, this allows for a partial assessment of performance bias.

The trial takes place in the first 5 days of the patient’s psychosomatic in-patient rehabilitation. The rehabilitation programme entails different therapies, including exercise therapy. Patients included in the study are asked to refrain from exercise except as defined by the protocol on the days of testing. Patients are explicitly made aware of this aspect before enrollment. Ancillary and post-trial care is provided throughout the in-patient rehabilitation program, i.e., on average for 4 weeks after completion of the study.

We assess adverse events through a questionnaire. We ask patients in both groups whether they experience adverse effects on a five-point Likert scale (not at all to very) using the following categories immediately after the intervention and the following morning:

Unplanned termination of participants and the reasons thereof (if participant proactively gives one) will also be reported.