Bidirectional associations between chronic low back pain and sleep quality: A cohort study with schoolteachers

S Aims: Although both chronic low back pain (cLBP) and sleep problems are prevalent among active workers, the relation between these variables is not well established. This study aimed to examine the bidirectional association between cLBP and sleep in schoolteachers. Methods: The Pittsburgh Sleep Quality Index (PSQI) and cLBP were self-reported by 530 schoolteachers in Londrina, Brazil, at baseline and after 2 years of follow-up. Generalized estimating equations were adjusted for sociodemographic, lifestyle and mental health variables. Results: Poor sleep quality at baseline was associated with cLBP at follow-up after adjusting for sociodemographic and lifestyle variables (OR = 1.61; 95% confidence interval [95% CI] = 1.06, 2.47). Changes in the PSQI score over time were also associated with a higher likelihood of cLBP at follow-up (OR = 1.13; 95% CI = 1.07, 1.20 for each 1- point increase in the PSQI score), regardless of mental health condition. cLBP at baseline was associated with worse sleep quality at follow-up after adjusting for sociodemographic and lifestyle variables (OR = 1.56; 95% CI = 1.02, 2.37). The presence of cLBP also changed the PSQI score over time ( ß coefficient = 1.153; 95% CI = 0.493, 1.814). Conclusions: Worse sleep quality was prospectively and bidirectionally associated with cLBP. Concretely, changes in PSQI values after 2 years of follow-up increased the likelihood of reporting cLBP, and baseline cLBP was associated with sleep quality worsening (i.e., higher score in the PSQI). Mental health conditions such as self- rated health, depression and anxiety play a relevant confounding role in the bidirectional associations between sleep and chronic low back pain.


Introduction
Both chronic pain [1] and sleep disorders [2] are highly prevalent in the general population and have been associated with work-related psychosocial factors [3]. Although the association between these two conditions is not well established, the concomitant presence of chronic pain and sleep complaints might compromise biological and behavioral aspects with harmful consequences for adequate performance at work and in private life [4].
Longitudinal studies have shown that poor sleep quality reduces pain thresholds [5,6], perpetuates pain symptoms [7], and affects the transition from relatively localized to generalized [8]. Likewise, chronic pain has been related to worse sleep quality [9]. In this context, some studies have aimed to understand the direction of the relation between sleep and pain [5,10]. A review pointed out that sleep problems are a more reliable predictor of some specific types of pain than pain is of sleep problems and highlighted that it remains to be determined whether the bidirectional association of sleep and pain varies across different chronic pain disorders [11]. For musculoskeletal pain, a bidirectional relationship with sleep problems has been suggested [12]. Studies suggest a deteriorating cycle starting either with sleep problems or with pain, in which the two components maintain or increase each other [13].
Chronic low back pain (cLBP) is a major public health problem throughout the world, with a lifetime prevalence of 20.1% [1]. Regarding its relationship with sleep, a systematic review found evidence of an increased frequency of cLBP associated with poor sleep quality, short sleep duration and worse ability to fall asleep [14]. However, it remains unclear whether sleep problems are a cause or an effect of cLBP. Some prospective studies in patients with low back pain have suggested a role of sleep problems on pain recovery [15] and on the development of troublesome LBP [16].
In this context, schoolteachers are commonly affected by musculoskeletal disorders, possibly because of a cumulative effect of workload on the musculoskeletal system and other psychosocial (e.g., high perceived stress level and monotonous work) [17] and ergonomic (e.g., prolonged periods of standing or sitting, head-down posture and lifting head) risk factors. Furthermore, teachers are also more likely to have poor sleep quality and insomnia symptoms, mainly due to long working hours in the classroom, overtime work preparing for lectures and a high prevalence of work-related stress and burnout syndrome [18][19][20]. Thus far, only a cross-sectional study exploring the association between sleep and musculoskeletal pain in teachers has been found. The authors reported that those with poor sleep quality were approximately twice as likely to present pain in several body parts [21].
Therefore, the objective of this study was to examine the bidirectional association between sleep quality and cLBP in schoolteachers at a 2-year follow-up.

Study design, population and location
This study is a longitudinal analysis of elementary and secondary schoolteachers as a part of the PRO-MESTRE study conducted between 2012 and 2014 with the objective of analyzing aspects of health, lifestyle and work. In brief, the study included a census of teachers from the 20 largest elementary and secondary public schools (i.e., those with more than 70 teachers) in Londrina, a large city in southern Brazil. Data were collected through a self-administered questionnaire, and personal interviews were carried out by trained undergraduate and graduate students [22].
The inclusion criteria were teaching in a classroom for at least one period in a week and being responsible for one or more subjects. At baseline (2012), among the 1126 teachers who met the inclusion criteria, 63 (5.6%) refused to participate, and 85 (7.5%) could not be located after six contact attempts on different days and times. Thus, the final analysis at baseline included 978 (86.9%) teachers (Fig. 1). After 24 months (2014), we repeated a self-administered questionnaire and interviews with the participants. However, 385 individuals were lost to follow-up interviews due to a labor strike of those professionals. Among the 591 participants remaining, 22 refused to participate, and one provided incomplete answers. Thus, the final sample size at follow-up for this cohort study was 530 participants (response rate: 54.2%).
The project was approved by the local Human Research Ethics Committee. All participants were informed about the study goals, were assured of their anonymity, and signed a consent form.

Sleep variables
In both the baseline and follow-up interviews, sleep quality was assessed with the Pittsburgh Sleep Quality Index (PSQI), a selfadministered tool used to evaluate sleep quality and possible disorders in the previous month [23]. The Brazilian version of the instrument was previously validated, and scores above 5 points indicate poor quality of sleep [24].

Pain variables
Chronic pain information was obtained at the baseline and follow-up interviews with the question "Do you have any type of chronic pain (i.e., pain symptoms that you have felt for six months or more)?", the temporal reference recommended for research purposes by the International Association for the Study of Pain when the study was designed (2012) [25]. If the answer was "yes", the question "In which part of your body do you feel this pain?" was asked, and a figure with a human body was presented for the teacher to point out the location, allowing for multiple sites. cLBP was deemed to be present when the participant mentioned that chronic pain was felt in the low back area.

Covariates
Information on the following variables that may influence sleep quality and pain perception was also obtained through the interview and considered potential confounders: sex (male or female), age (continuous, years), self-reported weight (kilograms) and height (meters), which were used to calculate the body mass index (BMI, kg/m 2 ), leisure time physical activity (yes or no), smoking (yes or no), coffee intake (yes or no), alcohol intake (yes or no), self-rated health (optimal: excellent/ very good/good, or suboptimal: regular/poor/very poor), anxiety and depression (both based on the report of medical diagnosis or use of specific medication).

Statistical analyses
To analyze the association between sleep quality (continuous PSQI score or a PSQI score >5) and cLBP (yes or no), generalized estimating equation (GEE) regression models were used to estimate the odds ratios (ORs) (binary outcomes: PSQI >5 and cLBP) or the ß coefficient (linear outcome: continuous PSQI score) and their respective 95% confidence intervals (CIs). This method can account for specificities of the longitudinal design used and allows analysis of potentially correlated data at baseline and at follow-up. Therefore, the GEE method considers information on the within-participant variability regarding changes in exposure, outcome and covariates over time [26]. Crude models (Model 1) are presented, and in sequence, sex, age, BMI, physical inactivity, smoking, coffee intake and alcohol intake were included as potential confounders (Model 2). Finally, the last model included the previous covariates plus optimal self-rated health, depression and anxiety (Model 3). These analyses were performed based on the following parameters: binary or normal distribution of the outcome, as appropriate, log link, exchangeable correlation and robust variance.
Considering that the association between cLBP and sleep quality could be affected by pain from other body regions, we performed a sensitivity analysis by repeating the main analyses stratified between those with only cLBP and those with also other chronic pain. Of the 56 teachers with cLBP, 32 (57.1%) reported chronic pain in one or more of the other regions, mostly in the dorsal region and upper extremities.
The significance level was established as p<0.05. The statistical analyses were performed using STATA SE software, version 15 (StataCorp, College Station, TX, USA).

Results
The sociodemographic characteristics and lifestyle and health characteristics of the participants at baseline (2012) are described in Table 1. Poor sleep quality was more frequent among participants with suboptimal self-rated health and the presence of depression and anxiety. Moreover, the mean body mass index was higher in those with cLBP, and the frequency of this pain was significantly higher in males and in those with suboptimal health.
Regarding the relationship between cLBP (exposure) and a PSQI >5 (outcome), the presence of cLBP at baseline was associated with poor sleep quality at follow-up, even after controlling for sociodemographic and lifestyle variables (OR= 1.56; 95% CI: 1.02, 2.37). When controlling for mental health conditions, self-rated health, depression and anxiety, the association lost statistical significance (OR=1.39; 95% CI: 0.89, 2.16) ( Table 3). When the PSQI score was treated as a continuous outcome, cLBP predicted worse sleep quality regardless of all confounders (ß coefficient=1.153; 95% CI: 0.493, 1.814).
As presented in the Supplementary Material (Table S1), in the sensitivity analysis stratified by the concomitant presence of chronic pain in another body region, the results were generally very similar, Table 2 Association between sleep quality (exposure) and chronic low back pain (outcome) after a 2-year follow-up.  except for some situations where the confidence intervals were quite wide due to the low number of subjects in the group without other chronic pain.

Discussion
In this 2-year follow-up study of Brazilian schoolteachers, poor sleep quality was associated with a higher likelihood of reporting cLBP and vice versa, regardless of the main confounders. Mental health conditions such as self-rated health, depression and anxiety played a relevant confounding role in the bidirectional association between sleep and back pain.
Our results suggest that poor sleep quality predicts the maintenance of chronic pain symptoms in the low back region in individuals with cLBP at baseline and increases the incidence of this pain over time. Few studies have investigated the prospective influence of sleep on cLBP. Consistent with our results, subjective sleep disturbances predicted troublesome LBP after a 4-year follow-up in Swedish workers with occasional LBP [16]. In addition, in a 6-month follow-up study, a significant relationship between reported sleep problems and pain intensity was found in patients with LBP [15]. In agreement with our study, other authors investigated adult women with and without any chronic pain at baseline and found that disrupted and nonrestorative self-reported sleep increased the risk of pain persistence and worsening as well as of chronic pain onset over the course of 17 years [27]. Skarpsno et al. (2020), in the HUNT study, reported that sleeplessness and the number of insomnia symptoms reduced the recovery of cLBP in a 10-year follow-up period [7]. Since these studies analyzed data from only individuals with LBP [7,15,16] or only women [27], our results extend the knowledge showing this association in a sample of both female and male workers, including those without cLBP at baseline. The current study also expands on these findings by showing that the PSQI score, a widely used and easily applicable test, is linearly associated with cLBP.
Associations lost statistical significance when controlling for selfrated health, depression, and anxiety for poor sleep quality (PSQI >5), which suggests that those mental health parameters might explain part of the relationships between cLBP and sleep. Evidence shows depression as a proxy for worse sleep [28] and for pain symptoms [29]. Self-rated health can also influence sleep or pain experience, since this variable might encompass chronic diseases, social context and important physiological dysregulations, including inflammatory processes [30], which exceeds the scope of the present study. Agmon and Armon (2014) found an association between insomnia symptoms and an increased risk of back pain, even after adjusting for, among other variables, self-reported health [31]. In the present study, the linear relationship between continuous PSQI (exposure) and cLBP (outcome) remained statistically significant even after adjusting for self-rated health, depression, and anxiety, reinforcing the influence of sleep on pain experience. Each 1-point increase in the PSQI score from baseline to follow-up increased the cLBP odds by 13% at follow-up.
Interestingly, the predictive effect of sleep disorder on cLBP was observed when using a continuous PSQI score. However, when the standard cutoff point of worse sleep quality (PSQI> 5) was used, the association lost statistical significance after controlling for mental health conditions. This finding suggests that whether sleep quality is treated as a continuous or dichotomous variable plays a decisive role in understanding sleep measures and their consequences on health outcomes. Therefore, it seems that the cutoff points, although widely used, are less useful than the continuous scales for analyzing discrete variations in sleep quality over time. Regarding the PSQI, the cutoff point of 5 in a 0 to 18 range may make it difficult to detect changes in sleep quality in the poor sleep group. For instance, having a PSQI score of 18 is certainly worse than having a score of 6.
Recent experimental studies have shown that insufficient sleep may lead to heightened sensitivity to pain, [32,33] and the mechanism might involve inflammatory cytokines released as a consequence of sleep deprivation [7]. Inadequate sleep appears to diminish the pain threshold by hyperactivating the somatosensory cortex of the brain, in addition to diminishing nucleus accumbens activity, which results in lower dopamine release, a neurotransmitter that causes pleasure and relieves pain. In conclusion, poor sleep quality appears to worsen pain experience and block natural analgesia [32].
Our study also suggests a relationship between cLBP and worsening sleep quality over time. Tang et al. (2015) found a dose-response association between baseline pain and insomnia 3 years later [9]. This association is intuitive when considering the need to eliminate distractions and other influences when lying down with the intention to fall asleep. Pain is certainly a distraction and a source of discomfort that might disturb sleep and, prospectively, result in worse sleep parameters. However, other studies have not found bidirectionality in the relationship between sleep and pain. Agmon and Armon (2014) found an association between insomnia symptoms and future back pain; however, the reverse association was not found [31]. Although categorical associations lost statistical significance after controlling for depression, anxiety and self-rated health, analysis using a continuous PSQI score remained significant in all models, reinforcing that 2 years may be a sufficient period to detect minor changes in sleep quality.
The strengths of this study include the use of a highly educated population exposed to working conditions that may favor cLBP and sleep problems. In addition, our analyses are robust since the sample size was adequate to consider a number of potential confounders, and the GEE method allowed us to explore the bidirectional effects between sleep and pain symptoms considering changes that occurred over the period for both variables and for the adjustment variables. Additionally, it was possible to control for correlations between the responses of the teachers at two different times and mitigate the effect of missing values [26].
However, some limitations should be acknowledged. First, we had a substantial loss to follow-up, which could lead to attrition bias. However, since missing data were generated completely at random, this bias is possibly less important [34]. Second, although the analyses considered chronic low back pain, since chronic pain can also occur concomitantly in other regions of the body, it cannot be affirmed that the relationship with sleep is exclusively due to cLBP. However, as in our additional analyses no substantial differences were observed between those with only cLBP or also pain in other body regions, it is reasonable to consider that the observed associations are due to cLBP. It is also true that it would be interesting to explore other aspects of chronic pain, such as intensity and severity, but these data were not collected at baseline in the present study. Third, for reasons of limited sample size, the covariates physical activity and alcohol, tobacco and coffee consumption were adjusted in a dichotomous format (i.e., yes or no), whereas ideally, they should discriminate continuously for frequency, intensity and dose of exposure. This limitation was partially explored when we considered the covariates physical activity and alcohol and coffee consumption as ordinal variables, from lowest to highest frequency, and the results were practically identical to those presented. Nonetheless, future studies on the present subject should consider samples of sufficient size to adjust for covariates with a greater level of detail than those used in this study. Last, although our analyses controlled for many potential confounders, residual confounding (e.g., due to the use of sleep, pain and mental disorders medications) is still possible.

Conclusions
In conclusion, sleep quality was bidirectionally associated with cLBP over time. Therefore, we suggest that sleep may play a role in the control of cLBP symptoms over time. Thus, sleep-focused interventions, such as those based on sleep hygiene promotion, may prevent the development of cLBP among teachers. In addition, addressing cLBP at early stages may also be protective in preserving sleep quality in this population. Finally, because depression and anxiety played a crucial confounding role in the bidirectional relationship between pain and sleep, mental health interventions could be equally critical in preventing the development of both cLBP and poor sleep in teachers.

Funding statement
This work was supported by the Brazilian National Research Council (CNPq), Brazil (Grant number 459671/2014-6), Araucaria Foundation (Grant number 168/2014), Parana, Brazil, and the Coordination for the Improvement of Higher-Level Personnel (CAPES), Ministry of Education, Brazil.

Declaration of Competing Interest
The authors report no conflict of interest.

Data Availability
Data will be made available on request.

Supplementary materials
Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.physbeh.2022.113880.