Abstract
Background Individuals with chronic pain consider improved sleep to be one of the most important outcomes of treatment. Physical activity has been shown to have beneficial effects on sleep in the general population. Despite these findings, the physical activity–sleep relationship has not been directly examined in a sample of people with chronic pain.
Objective This study aimed to examine the association between objective daytime physical activity and subsequent objective sleep for individuals with chronic pain while controlling for pain and psychosocial variables.
Design An observational, prospective, within-person study design was used.
Methods A clinical sample of 50 adults with chronic pain was recruited. Participation involved completing a demographic questionnaire followed by 5 days of data collection. Over this period, participants wore a triaxial accelerometer to monitor their daytime activity and sleep. Participants also carried a handheld computer that administered a questionnaire measuring pain, mood, catastrophizing, and stress 6 times throughout the day.
Results The results demonstrated that higher fluctuations in daytime activity significantly predicted shorter sleep duration. Furthermore, higher mean daytime activity levels and a greater number of pain sites contributed significantly to the prediction of longer periods of wakefulness at night.
Limitations The small sample size used in this study limits the generalizability of the findings. Missing data may have led to overestimations or underestimations of effect sizes, and additional factors that may be associated with sleep (eg, medication usage, environmental factors) were not measured.
Conclusions The results of this study suggest that engagement in high-intensity activity and high fluctuations in activity are associated with poorer sleep at night; hence, activity modulation may be a key treatment strategy to address sleep complaints in individuals with chronic pain.
Sleep disturbance is common in people with chronic pain, with 50% to 90% of this population reporting poor sleep.1–5 Poor sleep in the general population has been shown to have a significant impact on daily function, including considerable daytime cognitive dysfunction, decreased ability to accomplish daily tasks, diminished sense of success and achievement, depressed mood, and reduced enjoyment of interpersonal relationships.6–8 A comprehensive multicenter study has revealed that individuals with chronic pain consider improved sleep to be one of the most important outcomes of treatment.9
Current educational materials recommend sleep hygiene and treatments aimed at addressing negative thoughts, mood, and stress, including relaxation therapy and cognitive-behavioral therapy (CBT), as nonpharmacological options to address sleep complaints in individuals with chronic pain.10–13 Sleep hygiene involves increasing awareness of and altering lifestyle (eg, substance use, exercise, diet), environmental (eg, light, temperature, noise), and sleep-related behavioral factors (eg, regularity of sleep schedules, presleep activities) that affect sleep quality.14 Sleep programs incorporating a combination of sleep hygiene education, relaxation training, and CBT have been shown to be effective at increasing self-reported sleep quality in a heterogeneous group of patients with chronic pain15 and in individuals with comorbid insomnia and osteoarthritis.16 In addition, Edinger and colleagues17 found that 6 weekly individual sessions of either sleep hygiene or CBT resulted in a reduction of nocturnal wake time in a group of patients with fibromyalgia. In their study, CBT was shown to be more effective than sleep hygiene at decreasing wake time (50% versus 20% reduction, respectively). Engagement in “daytime activity and exercise” or “moderate regular exercise” was recommended as part of the sleep hygiene education in the aforementioned studies. Promoting exercise engagement as a core component of sleep hygiene interventions is based on the premise that inactivity adversely affects sleep quality.14
Research with nonclinical samples does support the notion that inactivity is associated with poor sleep (see reviews18–20). Studies have shown that both short-term and long-term engagement in exercise increase total sleep time, prolong slow wave sleep, and decrease sleep onset latency in the general population.18,19 A negative association has been found between insomnia and physical activity in studies worldwide, and exercise has been shown to be as effective as hypnotic medication in decreasing sleep complaints.20
Despite these findings, the physical activity–sleep relationship has been largely ignored in chronic pain literature. To our knowledge, there are no published outcome studies that have investigated the effect of exercise or physical activity interventions on sleep disturbance in chronic pain. Only one research study has considered the association between levels of physical activity and sleep disturbance.21 In this study, objective measures of sleep and activity were used to demonstrate that a group of patients with chronic pain had more disturbed sleep compared with pain-free participants who had similar daytime activity levels.21 These results hint at the possibility that high activity levels may be associated with sleep disturbance in individuals with chronic pain. As this relationship appears to contradict current recommendations made in chronic pain educational materials, where increasing physical activity and exercise is endorsed, the empirical investigation of the direct relationship between physical activity and sleep in chronic pain is warranted.
The aim of this study, therefore, was to use a within-persons study design to examine the association between objective daytime physical activity and subsequent objective sleep for individuals with chronic pain. Furthermore, this study aimed to control for pain and psychosocial variables, including mood,4,22,23 stress,22,23 and pain catastrophizing,24 which have been associated with different aspects of sleep in cross-sectional studies utilizing samples of individuals with chronic pain. Based on existing evidence on this topic, we hypothesized that daytime activity levels would be associated with aspects of objective sleep, including sleep duration and average nocturnal awake time, after controlling for pain and psychosocial variables. No prediction was made about the direction of this relationship due to the lack of evidence on the direct relationship between physical activity and sleep in chronic pain.
Materials and Method
Participants
A sample of 50 adult patients was recruited from a multidisciplinary pain center (MPC) located in a large tertiary hospital in Australia. Inclusion criteria were: (1) outpatient of the MPC, (2) persistent non-cancer-related pain for at least 3 months, (3) generalized pain distribution affecting the participant's gross movement (ie, gross movement patterns increase the participant's pain), (4) English literate, (5) 18 years of age or older, (6) residing in the metropolitan area where the MPC is located, and (7) able to provide written informed consent. An exclusion criterion was diagnosis of a sleep disorder (eg, sleep apnea or restless legs syndrome) that was symptomatic at the time of data collection. As the activity monitors used in this study measure an individual's gross movement, only individuals who had generalized pain in body parts associated with gross movement (ie, the lower limbs or torso, or both) were recruited. The presence of pain in body parts associated with gross movement was assessed by the treating team that referred patients to the study and resulted in the exclusion of patients with pain isolated to the upper limbs, head, and face. Sixty-three patients were invited to participate in the study, with 13 declining due to other commitments, resulting in a sample size of 50 (79.4%). Demographic information is reported in Tables 1 and 2. Participants were predominantly female, married, Australian, and unemployed due to pain, with an age range of 33 to 73 years. The majority of participants reported having pain for an extended period of time (X̅=13.04 years) and numerous pain sites (X̅=5.94). The main pain complaint was lower back pain (74%).
Descriptive Information of Demographic Categorical Variables
Descriptive Information of Demographic Continuous Variables
Procedure
Over an 18-month period, patients meeting the selection criteria were identified by medical or allied health staff at the MPC. The study was explained to patients verbally, and written informed consent was obtained. Participants then completed a demographic questionnaire prior to commencing 5 days of data collection. This 5-day data collection period included at least 1 weekend day. Over the 5 days, participants wore an activity monitor and were given a Palm handheld computer (Palm Inc, Sunnyvale, California), with installed software, that administered an electronic questionnaire 6 times a day. Participants recorded the time they went to bed and the time they got out of bed each day in a diary. On completion of data collection, each participant received a $20 gift voucher for use in popular retail stores in Australia.
Measures
Demographic information.
A demographic form collected data on sex, age, pain location, number of pain sites, pain duration, marital status, level of education, and employment status.
Electronic questionnaire.
The experience sampling method, which involves responding to questionnaires on multiple occasions over a period of time, was used to measure pain, mood, stress, and catastrophizing. This method has been confirmed as a valid and reliable method for gathering data on an individual's experience and psychological state, which allows for the examination of within-person real-time associations.25,26 In this study, pain, mood, stress, and catastrophizing were measured 6 times a day over the 5-day data collection period using an electronic questionnaire. Both the average scores for each day and the final response score from each day prior to sleep onset, for each variable (pain, mood, stress, and catastrophizing), were used in analyses. The electronic questionnaire that measured these constructs was developed by the researchers with the Experience Sampling Program,27 which was installed on 8 Palm handheld computers (m100, Zire and Tungsten series). The Experience Sampling Program is an open-source software package for running questionnaires on a Palm Pilot that displays questions, receives responses, and records reaction times. The program was configured to alert participants at random intervals 6 times during their waking hours to respond to the custom-made electronic questionnaire.
Table 3 contains details of the electronic questionnaire programmed on the Palm handheld computers. All items were measured on a 10-point horizontal visual analog scale (VAS). Studies using the experience sampling method in pain research commonly use single items or a reduced set of items from established questionnaires to measure constructs.28–30 In this study, pain, mood, and stress were assessed with single items. The single-item VAS for pain and mood have both been shown to have adequate validity.31,32
Details of Electronic Questionnaire
Catastrophizing was measured with 3 items representing each subscale of the Pain Catastrophizing Scale (ie, rumination, magnification, and helplessness).33 The 3 items were chosen based on: (1) item-total correlations from the original factor analysis of the scale,33 (2) standardized path coefficients (the square root of the percentage of variance in the item accounted for by the latent construct) from a confirmatory factor analysis using a chronic pain sample,34 and (3) authors' agreement of the single best item that represents each of the 3 Pain Catastrophizing Scale subscales. All items chosen had item-total correlations and standardized path coefficients that were ranked highly (first or second) in their respective subscale. A total catastrophizing score was calculated by adding the 3 responses. The total catastrophizing score and scores from the individual subscales (ie, rumination, magnification, and helplessness) were considered in the analyses. Empirical evidence has been supportive of the reliability and validity of the Pain Catastrophizing Scale,33,35–37 and one-item versions of other commonly used subscales of pain-related coping have been shown to be highly correlated to the parent subscale.38
Activity monitor.
The GT3X ActiGraph activity monitor (ActiGraph, Pensacola, Florida) was chosen to objectively measure sleep and daytime physical activity as actigraphy has been shown to be a practical and reliable measure of these constructs. It is effective in differentiating among various physical and sedentary activities in adults who are healthy and correlates significantly with oxygen uptake and heart rate.39 A study investigating the feasibility of actigraphy in home-based settings showed that it is easily utilized and well tolerated by participants.40 It is favored over self-report measures for quantitative assessment of physical activity in chronic pain,41 as it does not rely on patient recall. Also, compared with polysomnography, a gold standard objective sleep measure, actigraphy is a valid and reliable measure for differentiating sleep from wake in adults who are healthy (see review42).
Participants were required to wear the GT3X ActiGraph activity monitor during both sleeping and waking hours and to remove it only for showering and swimming over the 5-day data collection period. The activity monitor incorporates a triaxial accelerometer that collects changes in acceleration, 30 times each second, across 3 axes (vertical, horizontal, and perpendicular).43 The device translates this movement into a digital code that is stored in computerized form.43 In this study, activity counts per minute were recorded for each axis. This measure equates to the accumulation of filtered changes in acceleration measured during a 60-second period.43 The vector magnitude per minute (calculation of the magnitude of the vector that forms when combining activity counts per minute from all 3 axes) was then used to calculate physical activity variables. The vector magnitude per minute can be interpreted as the intensity of physical activity carried out over the course of a minute.43
Two activity variables were calculated and used in the analyses: average daytime physical activity and fluctuations in physical activity. Average daytime physical activity was calculated by finding the average vector magnitude per minute between the time participants got out of bed and when they went to bed, as indicated in their diary. Higher levels of average daytime physical activity indicated engagement in higher-intensity activities throughout the day. Engagement in high-intensity activities is a characteristic of overactivity in chronic pain.44–46 The term “overactivity” refers to engagement in high levels of activity that result in severe pain aggravation and a period of inactivity where an individual is unable to function.45 Individuals who engage in overactivity will resume daily tasks following inactive periods once their pain has subsided or frustration over inactivity stimulates new activity.47 As a result, individuals who engage in overactivity are thought to have a “sawtooth” activity pattern where their pain and activity fluctuate greatly over time.44,46 In order to capture this activity pattern, the fluctuation in daytime physical activity was calculated. The fluctuation value for each participant was obtained by adding the vector magnitude per minute over 15-minute periods from the time participants got out of bed to the time they went to bed on a given day.
Next, the difference among these 15-minute periods was found by subtracting the value for each 15-minute time period from that of the 15-minute time period directly before it. The root mean square of these difference values was then calculated to express the magnitude of these differences. This calculation was done by squaring each score, calculating the mean difference value, and taking the square root from this mean value. This method for calculating fluctuations in physical activity has been used in previous studies.47–50 Higher values indicate greater fluctuations in activity levels throughout the day. The last 15 minutes of each day was not included in calculations if a full 15 minutes of daytime activity recording was not available.
Sleep measures were generated with ActiLife software version 4.4.1 (ActiGraph), using the sleep scoring function. This function uses the Sadeh algorithm,51 which determines an individual's sleep state by examining the actigraph activity over an 11-minute sliding window.43 For any given window, a “sleep score” (whether the person is asleep or not) can be determined by applying the algorithm. Time in bed and time out of bed, as indicated in each participant's diary, were entered in order to calculate sleep scores. The following variables were then generated using the sleep scoring function: (1) sleep duration—the total number of minutes the algorithm indicates “asleep,” (2) number of awakenings—the number of different times the algorithm scores “awake,” (3) average awake time—the number of minutes the algorithm indicates “awake” divided by number of awakenings, and (4) sleep efficiency—sleep duration divided by total time in bed.43 The Sadeh algorithm has also been shown to have agreement rates with polysomnography scoring ranging between 91% and 93%.51 Table 4 reports the descriptive statistics for activity and sleep variables.
Descriptive Information of Continuous Experimental Variables
Data Analysis
The Statistical Package for Social Sciences (SPSS) GradPack version 18.0 (SPSS Inc, Chicago, Illinois) provided advanced analytical techniques, including generalized linear mixed modeling, which was used to analyze the results of this study. Generalized linear mixed models cover a wide variety of modeling, including multilevel models. In this study, multilevel modeling was applied to examine both within-person and between-persons variance in sleep measures.
Three levels of data were collected in this study (time points nested within a day nested within a person). Time point data resulted from the electronic questionnaire, which was administered 6 times a day. The data were arranged so that the average scores and the final response score from each day could be examined in relation to subsequent sleep that night. This process resulted in a 2-level hierarchical structure with daily observations and responses nested within individuals. Measurements of sleep, activity, pain, mood, stress, and catastrophizing were collected each day for all individuals and were defined as response-level, or level 1, variables. Demographic variables are variables that were considered stable across the 5 days of data collection. As such, data for these variables were collected on one occasion for each individual and were defined as person-level, or level 2, variables.
A series of 2-level hierarchical linear regression analyses were produced to examine significant predictors of sleep. When applied to the data structure used in this study, at the first level of this analysis, coefficients were estimated for an equation within each person that expresses sleep variables as a function of response-level variables for that person. Individual parameter estimates then become the dependent variables in level 2 equations that model how the relationship between level 1 variables and sleep variables varies among people. The effects of person-level variables are estimated at this level.26
A sample size of 5 was used at level 1 (ie, 5 records for each response level variable per person relating to the 5 days of data collection), and a sample size of 50 (ie, 50 participants) was used at level 2. This procedure resulted in a total sample of 250 (maximum) records. Estimating the sample size required for the analyses undertaken was complicated given the multiple levels and parameters of interest.52 As no data from similar studies were available to estimate values needed for power calculations, an accurate estimate of sample size could not be determined prior to data collection.52–54 Simulation research, using the same modeling as that produced in this study, suggests that the sample size at level 2 has a greater impact on increasing power than the sample size at level 1, with level 2 sample sizes greater than 30 having a minimal impact on the accuracy of the standard error for fixed effects.52–56 As estimates of fixed effects were the primary interest of this research, a sample size of 50 participants at level 2 was chosen.
A data screen was conducted prior to the analyses to detect the most appropriate covariates to include in models in order to reduce multicollinearity. This screen was done by computing Pearson correlation analyses on level 1 variables (response level variables) to examine the strength and direction of relationships. The strength of correlations between the catastrophizing subscales (ie, magnification, rumination, and helplessness) and the total catastrophizing scores was strong, and the correlation coefficients between these variables and sleep variables were comparable. Therefore, only the total catastrophizing scores were used in further analyses. Overall, visual comparisons of associations among the average daytime scores for pain, mood, total catastrophizing, and stress generally suggested stronger correlations with sleep variables compared with the final response of that day for these variables. As a result, final daytime responses prior to sleep onset were discarded, and average daytime scores were retained.
The final dataset was assessed for normality, linearity, constant variance, and outliers. Five variables with multiple outliers were identified: number of awakenings, average daytime physical activity, fluctuations in daytime physical activity, average daytime total catastrophizing, and pain duration. For number of awakenings, 9 of the identified outliers were determined to be secondary to data error and, therefore, were deleted. The remaining outliers were thoroughly reviewed, and deletion was not justified. Seven significantly skewed variables were detected. Number of awakenings, average awake time, average daytime total catastrophizing, average daytime physical activity, number of pain sites, and pain duration were positively skewed, and sleep efficiency was negatively skewed. Box-Cox transformation is a procedure that identifies the most appropriate exponent to use to transform data into a normal shape and, as such, was used to transform skewed variables. Sleep efficiency could not be transformed to a normal distribution and, therefore, was not included in the analyses, as normality of the dependent variable is one of the assumptions of the models produced.57
The data also were assessed to identify any patterns to missing data. If more than 2 hours of data were missing on a given day or night due to removal of the ActiGraph, activity data for that day or night were considered invalid and classified as missing. On inspection, there was no observable pattern to the missing data. In addition, a series of independent-sample t tests were conducted using a dummy coded variable for missing data and level 1 variables as dependent variables. There was no difference between the means of the 2 groups for any of the level 1 variables in these analyses. As such, missing data resulted in exclusion of that case from the analyses. The amount of data missing for each variable is presented in Tables 1, 2, and 4. A significance level of .05 was set for statistical tests. As recommended by Streiner and Norman,58 a correction was not used to account for multiple analyses due to the exploratory nature of this study. If a Bonferroni correction were used, the significance level would be reduced to .017.
Three 2-level hierarchical linear regression models were produced, with 1 of the 3 sleep variables (sleep duration, average awake time, or number of awakenings) entered as dependent variables. All independent variables were first centered before being entered into the models to ensure interpretable and meaningful zero points. As the primary interest of this study was to examine the predictive influence of response-level variables, level 1 variables were group mean centered (scores were deducted from the person's mean score for that variable), and level 2 variables were grand mean centered (scores were deducted from the sample mean) to produce unbiased estimates of beta (β) at level 1.59 Level 1 variables that were entered into models were: sleep duration of the previous night, average daytime physical activity, fluctuations in daytime physical activity, average daytime total catastrophizing, average daytime mood, and average daytime stress. Patient demographics, including age, sex, pain duration, and number of pain sites, were the level 2 variables entered. Residuals from each analysis were saved, and normality assumptions were examined. R2 change values of significant independent variables were then calculated. This calculation was done by individually removing a significant variable from each model and examining the resultant change in covariance parameters, as described by Heck and colleagues.60 The removed variable was replaced prior to the removal of another significant variable.
Role of the Funding Source
The equipment used in the study was funded by the Professor Tess Cramond Multidisciplinary Pain Centre. Ms Andrews was supported by a Royal Brisbane and Women's Hospital Foundation scholarship, an Occupational Therapy Board of Queensland Novice Researcher Grant, and the Cramond Fellowship in Occupational Therapy and Pain Management at the Royal Brisbane and Women's Hospital.
Results
The results of the analyses are presented in Table 5 and are discussed below. The residuals were approximately normally distributed in all models, meeting the assumptions of linear regression.
Two-Level Hierarchical Regression Analyses of Variables Predicting Sleep
Variables Predicting Sleep Duration
Fluctuations in daytime physical activity made a significant contribution to the prediction of sleep duration (β=−.0002, t81.36=−2.05, P=.04, 95% confidence interval [95% CI]=−0.0004 to −0.000006). The results suggested that greater changes in activity levels throughout the day were linked to a reduced amount of sleep at night. Fluctuation in daytime physical activity accounted for about 3% of the variability in sleep duration within individuals. None of the other experimental or demographic variables made a significant contribution to explaining the variation in sleep duration.
Variables Predicting Average Awake Time
Two variables predicted average awake time: average daytime physical activity and number of pain sites. For average daytime physical activity (β=.29, t88.84=2.09, P=.04, 95% CI=0.0015 to 0.06), individuals who engaged in higher-intensity activities throughout the day spent more time awake when lying in bed at night. For number of pain sites (β=.28, t30.53=2.40, P=.02, 95% CI=0.04 to 0.52), individuals who reported a greater number of pain sites had longer periods of wakefulness at night. The R2 change values indicated that average daytime physical activity accounted for about 3% of the variability within individuals and approximately 2% of the variance among individuals for average awake time. Number of pain sites had a large effect on average awake time, explaining approximately 19% of the variance among individuals.
Variables Predicting Number of Awakenings
Participants' sex was the only variable to make a significant contribution to number of awakenings (β=−.48, t27.25=−2.61, P=.02, 95% CI=−0.86 to −0.10), with female participants waking more often during the night compared with male participants. Participants' sex accounted for about 36% of the variability in number of awakenings among individuals.
Exploring Sex Differences
As participants' sex made a large contribution to explaining the between-persons variance in number of awakenings, further analyses were undertaken to explore differences in sex with the objective of providing further insight into this relationship. A series of independent t tests were conducted to examine whether demographic and experimental variables differed significantly with sex. Female participants tended to have higher average daytime pain (X̅=5.54, SD=2.03, t233=−3.14, P=.01, 95% CI=−1.37 to −.37) and higher average daytime stress (X̅=3.58, SD=2.18, t233=−2.2, P≤.001, 95% CI=−0.77 to −0.04) compared with male participants (X̅=4.67, SD=2.03, and X̅=3.08, SD=2.5, respectively). In addition, female participants reported a higher number of pain sites (X̅=7.53, SD=5.29, t47=−2.13 P=.04, 95% CI=−6.05 to −0.17) and higher mean activity levels throughout the day (X̅=39.00, SD=10.94, t206=−2.3, P=.03, 95% CI=−6.51 to −0.43) compared with male participants (X̅=4.42, SD=4.44, and X̅=35.53, SD=10.56, respectively).
Discussion
The present study utilized an innovative ambulatory monitoring product and a within-person study design to examine the association between objective daytime physical activity and subsequent objective sleep for individuals with chronic pain while controlling for pain and psychosocial variables. The association between physical activity and sleep in chronic pain had not been directly examined, and this was the first study to use a within-person design to examine the association between daytime psychosocial variables and subsequent sleep in individuals with chronic pain. The application of a prospective within-person design and activity monitoring increased the reliability and validity of findings by: (1) objectively measuring the association between activity and sleep; (2) allowing for the measurement and documentation of thoughts, feelings, and behaviors as they occurred; (3) allowing the examination of the relationship between these factors and subsequent sleep; and (4) providing estimates of both within-person and between-persons variability in sleep.
The results indicate that objective daytime activity does predict subsequent objective sleep in adults with chronic pain above and beyond measures of daytime pain intensity, catastrophizing, stress, and mood. Higher mean daytime activity levels (indicative of engagement in higher-intensity activities throughout the day) predicted longer periods of wakefulness at night both within and among individuals. In addition, fluctuations in daytime activity was a significant predictor of sleep duration, with higher fluctuations (greater changes in activity levels from one 15-minute period to the next) linked to a shorter sleep duration within individuals. These results support the inclusion of activity modulation interventions as treatment options to address sleep complaints in individuals with chronic pain. In line with findings from previous cross-sectional studies,23,61,62 participants' sex and number of pain sites also explained the variance in different aspects of sleep among individuals. Sex accounted for a large proportion of the variability in number of awakenings in this study, with female participants waking more frequently than male participants. Female participants in this study reported a higher number of pain sites and had higher mean activity levels, which might explain some of the variance. Possible explanations for the association between activity variables and sleep are discussed below.
Both high fluctuations in activity and engagement in high-intensity activities are characteristics of overactivity.44–46 As defined previously, overactivity refers to engagement in high levels of activity, resulting in severe pain aggravation.45 During the pain exacerbation, the individual experiences a period of inactivity and resumes activity once pain has subsided or frustration over inactivity stimulates new activity.45,47 These findings represent a “sawtooth” pain and activity pattern where activity and pain fluctuate greatly over time.44,46 Both pacing education and activity scheduling are common interventions to address patterns of overactivity in individuals with chronic pain.63 Pacing is a strategy used to divide a person's daily activities into smaller, more manageable portions, which allows the individual to participate in activities in a way that should not exacerbate his or her pain while facilitating planned and calculated increases of activity.46,63 High fluctuations in daytime activity may reflect incidences of overactivity or the use of an ineffective pacing strategy (ie, using prolonged rest periods).
One possible explanation for the associations between activity variables and sleep is that pain exacerbations, caused by overactivity, led to poorer sleep at night due to increased pain at the end of the day and at night. A delayed exacerbation in pain following a period of concentrated physical activity has been demonstrated in chronic back pain research,64 and the data screen in this study showed a positive correlation that approached significance between the last pain report prior to sleep (which was measured at a random time point in relation to sleep onset) and both mean daytime activity and fluctuations in daytime activity (r=.12, P=.10, and r=.13, P=.06, respectively). Although average daytime pain was not associated with subsequent objective sleep in this study or in previous research,65,66 higher fluctuations in pain levels during the day (suggestive of pain exacerbations) have been shown to predict large fluctuations in nighttime activity.65 A carefully designed time-series mediation analysis investigating the associations among activity, subsequent pain levels, and the succeeding sleep period is needed to further explore this notion.
Links between activity and sleep may be further explained by individual differences in coping. Individuals who engage in overactivity persist with activity despite pain and usually attempt to ignore pain and use distraction as a coping strategy.67,68 Subsequently, they may be more aware of their pain or daily stresses while lying in bed at night due to limited distractions, which, in turn, may affect sleep. A study that looked at the diurnal variation of pain perception in patients with chronic pain supports the notion that engagement in work and productive tasks can distract individuals from their pain.69 The authors found that patients who worked reported less pain during working hours, but their pain escalated when they returned home. In contrast, participants who stayed at home reported pain levels that rose at the start of the day but remained relatively stable thereafter.69 Individuals who engage in overactivity also may wake early to attend to productive tasks in the morning, resulting in shorter sleep duration.
Modification of sleep physiology may provide an alternate explanation for the observed results. To date, there is no evidence of links between fluctuations in daytime activity and sleep, with either nonclinical or chronic pain samples. As such, the physiological effect of high fluctuations in daytime activity on the normal sleep cycle is unknown. Physical activity has been shown to exert an influence on sleep physiology by altering endocrine and metabolic functions during sleep.70 Alterations of these functions through habitual changes in activity pattern may alter the sleep-wake cycle.70 Hence, links between activity and reduced sleep duration in this study may have been influenced by these mechanisms.
Various considerations should be acknowledged when interpreting the findings of the present study. Participants had generalized pain in body parts associated with gross movement, which limits the ability to generalize the results of this study to individuals with other types of pain. In addition, all participants were sourced from a tertiary pain clinic and were selected by medical and allied health staff, thus introducing selection bias and potentially affecting the external validity of the study. A retrospective power analysis conducted using Power Analysis in Two-Level Designs (PinT) software,71 after accounting for missing data, revealed that the study's sample size has adequate power (>80%) to detect a moderate effect size for independent variables in the models produced. Nevertheless, the study sample size (N=50) may limit the generalizability of findings. In addition, missing data may have led to overestimations or underestimations of effect sizes. Producing 3 models increases the chance of making a type I error, and the precision of estimates for significant level 2 variables was low. As a result, the results of this study warrant replication. Furthermore, a self-report measure of sleep quality was not administered in this study to validate objective sleep measures and provide insight into an individual's perception of sleep quality. Inclusion of such a measure would strengthen future studies. As measures of pain and psychosocial variables relied on participants' self-reports, social desirability responding also existed. Finally, a number of additional factors that may be associated with sleep, such as menopausal symptoms, medication usage, and environmental factors (eg, exposure to light or noise), were not considered in this study.
Despite these limitations, the results of this study offer the first empirical evidence that objective daytime activity is associated with subsequent objective sleep in adults with chronic pain above and beyond measures of daytime pain intensity, catastrophizing, stress, and mood. This finding is important because educational material currently recommends sleep hygiene and treatments aimed at addressing negative thoughts, mood, and stress, including relaxation therapy and CBT, as nonpharmacological options to address sleep complaints in individuals with chronic pain.10–13 Reference to recommendations for physical activity in these publications is absent or limited to “engagement in daytime activity and exercise” as part of the sleep hygiene education. These recommendations are based on evidence that supports the association between increased physical activity or exercise and improved sleep quality in the general population.18–20
The results of this study suggest that when individuals with chronic pain engage in high-intensity activity and have high fluctuations in their activity throughout the day, they experience poorer sleep (ie, a shorter sleep duration and longer periods of wakefulness) at night. These results do not negate the beneficial effects of all types of exercise for sleep in this population or imply that inactivity should be promoted, but do suggest that engagement in high-intensity exercise that may severely aggravate pain is detrimental. As such, recommending increased daytime physical activity and exercise that are unguided and not supervised as part of sleep programs in this population may be insufficient and could be contraindicated. Given the observed associations in this study, activity modulation may be a key treatment strategy in addressing sleep complaints in individuals with chronic pain. Introducing pacing education, activity scheduling, and guided exercise sessions (based on graded activity principles46) into sleep programs for individuals with chronic pain may be of value. There are also potential benefits from incorporating education on the effects of overactivity on sleep into pain education programs. Finally, activity monitors may be beneficially applied to clinical practice to monitor activity levels and sleep, which would assist with individually tailored treatment.
Future research should focus on replicating the results of this study while controlling for additional variables such as medication usage. Using self-report measures of sleep also would provide insight into individuals' perceptions of sleep quality, which would be of value clinically. Further research is needed to improve our understanding of the relationship between activity levels and sleep in chronic pain. Research investigating the effects of activity on sleep physiology, and possible moderators or mediators of the relationship between daytime activity and sleep, is warranted to continue to improve treatment strategies.
The Bottom Line
What do we already know about this topic?
Fifty percent to 90% of people with chronic pain report sleep disturbances, and improved sleep is considered to be one of the most important outcomes of treatment for this population. Physical activity has been shown to have beneficial effects on sleep in the general population, but the relationship between physical activity and sleep has not been directly examined in a population with chronic pain.
What new information does this study offer?
The results of this study indicate that daytime activity does predict subsequent sleep in adults with chronic pain above and beyond measures of daytime pain intensity, catastrophizing, stress, and mood. The results suggest that when people with chronic pain engage in high-intensity activity and have high fluctuations in their activity throughout the day, they experience poorer sleep at night.
If you're a patient/caregiver, what might these findings mean for you?
Given the observed associations in this study, activity modulation may be a key treatment strategy in addressing sleep problems in people with chronic pain. Interventions such as pacing education, activity scheduling, and guided exercise sessions may have a beneficial effect on sleep quality.
Footnotes
All authors provided concept/idea/research design and writing. Ms Andrews provided data collection. Both Ms Andrews and Ms D'Arrigo provided data analysis. The authors acknowledge Dr Asad Khan, The University of Queensland, for his assistance with the statistical processes of some parts of this research and the staff and patients of the Professor Tess Cramond Multidisciplinary Pain Centre for their contribution to data collection.
The study was approved by The Royal Brisbane and Women's Hospital Human Research Ethics Committee and The University of Queensland Behavioural and Social Sciences Ethical Review Committee.
The results of this study were presented at the Australian Pain Society 33rd Annual Scientific Meeting; March 17–20, 2013; Canberra, Australia.
The equipment used in the study was funded by the Professor Tess Cramond Multidisciplinary Pain Centre. Ms Andrews was supported by a Royal Brisbane and Women's Hospital Foundation scholarship, an Occupational Therapy Board of Queensland Novice Researcher Grant, and the Cramond Fellowship in Occupational Therapy and Pain Management at the Royal Brisbane and Women's Hospital.
- Received July 15, 2013.
- Accepted November 8, 2013.
- © 2014 American Physical Therapy Association
References
Issue highlights
