The influence of self-reported leisure time physical activity and the body mass index on recovery from persistent back pain among men and women: a population-based cohort study

Settings and study population

This longitudinal study is based on the Stockholm Public Health Cohort (SPHC, n=23 794). The cohort includes information from two Public Health Surveys performed in Stockholm County, Sweden.

The source population was residents, 18 to 84 years old, of Stockholm County, an urban region consisting of 26 municipalities. Selected subjects (n=49 914) received the baseline postal questionnaire between October 2002 and March 2003. The 31 182 subjects responding were sent a follow-up questionnaire between March and August 2007 which 23 794 (76%) subjects completed. Out of those, 1982 subjects reported persistent BP at baseline. After exclusion of 146 subjects with missing data on exposure and outcome the study population included 1836 participants (Figure 1).

Both questionnaires included questions regarding demographic characteristics, physical and psychological health, work related factors, lifestyle and socioeconomics. Data on marital status were retrieved from a Swedish national register [14].

Having persistent back pain (PBP) at baseline was equal to answer “Yes, every day” to the question: “During the previous six months, have you experienced pain in your lower back”? This question was modified from the Standardized Nordic Questionnaire [15]. The term “persistent” is suggested for back pain present on at least half the days in a 6-months period [16].

Prognostic factors

Potential prognostic factors were self-reported regular leisure time physical activity (PA) and the body mass index (BMI), reported at baseline.

PA was categorized into four levels using a measure based on the question: “During the previous 12 months, how physically active have you been during leisure time? If your activity differs between e.g. summer and winter, please estimate the average activity”. The PA levels were: (a) Sedentary; “You mostly devote yourself to reading, TV, movies or other sedentary activity during leisure time. You walk, cycle or are active in other ways less than 2 hours a week”, (b) Low; “You walk, cycle or are active in other ways at least 2 hours a week, mostly without sweating. Also include walking or cycling to and from work, Sunday walks, ordinary gardening, fishing, table tennis and bowling”, (c) Moderate; “You are physically active regularly, 1-2 times a week at least 30 minutes each session with running, swimming, tennis, badminton or other activity that makes you sweat”, (d) High; “You devote yourself to e.g. running, swimming, tennis, badminton, aerobic exercise or similar on average at least 3 times a week, each session lasting at least 30 minutes”.

In a working population with musculoskeletal complaints this PA measure was found to have moderate criterion validity compared to a structural personal interview. Moreover, there were no substantial sex differences in the validity of the measure [17]. Based on the same measure, PA was dichotomized into: Sedentary; alternative a, and Active; alternative b-d.

BMI (kg/m²) was calculated using self-reported weight and height and categorized into normal weight (BMI < 25) and overweight (BMI ≥ 25) [18]. The normal weight category included 31 underweight participants (BMI < 18.5) and the overweight category included 270 obese participants (BMI ≥ 30) [18].

Potential confounders

We did not treat self-rated health (SRH) as a potential confounder as poor SRH is shown to be a consequence of, rather than the cause for, pain in chronic pain conditions such as BP [24]. If that is true SRH could be regarded as a collider [25]. Baseline SRH was assessed by using an item from the Short Form 36: “In general, would you say your health is?” (very good, good, fair, poor, very poor) [26]. We dichotomized SRH into good (very good, good) and poor (fair, poor, very poor) to be used in additional analyses.

Outcome

The follow-up questionnaire supplied information about recovery from PBP using a combination of two questions: “During the last 5-year period have you had back pain for at least 3 consecutive months that has disturbed you considerably?”, and “During the last 5-year period have you had back pain, in at least 7 consecutive days but less than 3 consecutive months, that has disturbed you considerably?” We defined participants as recovered if they answered “No” on both questions. Consequently our final definition of recovery from PBP was equal to reporting: “No periods of considerably disturbing back pain lasting for 7 days or more, during the latest 5-year period”.

Statistical methods

Main results

Compared to sedentary leisure time, the chance of recovery from PBP was greater for women that were physically active during leisure time. The adjusted RR among women was 1.46 (95% CI: 1.06, 2.01) for low PA, 1.51 (95% CI: 1.02, 2.23) for moderate PA, and 1.67 (95% CI: 1.08, 2.58) for high PA. No analyses indicated that PA was associated with recovery from PBP among men or that BMI was associated with recovery from PBP, either among men or among women.

Additional results

Comparison with other studies

We found only few studies with an aim similar to ours. However, these studies have differences concerning design, outcome, study sample as well as definition of disease which may explain the somewhat conflicting results [4]. In a community-based study with subjects reporting BP the previous month, neither leisure activity nor sports were found to be predictive for BP one year later [29]. Hurwitz and colleagues found increased levels of recreational physical activity to reduce pain and disability among BP patients randomized to chiropractic or medical care [6]. In contrast, Mortimer and colleagues found no effect from recreational exercise on pain and disability among men or among women, five years after seeking care for BP [30]. A cohort study, following sick-listed BP patient’s showed no or moderate physical activity to be associated with worse disability and pain at one-year follow-up [31]. The authors observed no relation between BMI and disability or pain. Further, a review of prognostic factors for patients sick-listed with acute BP found moderate evidence for BMI and no evidence for physical fitness or sports to be prognostic factors for duration of sick leave [13].

In our study only 21 percent of the participants recovered from BP when other studies report recovery proportions of 50 to 60 percent [2, 4]. This discrepancy may reflect the recurrent pattern characteristic for BP as we defined recovery to be free of pain for five years and the other studies reported recovery one year after onset [32]. Despite our stringent definition of recovery we found a positive effect of PA among women, supporting its importance as a prognostic factor.

Why would PA have a positive effect on recovery from PBP in general and why among women only? These mechanisms have not yet been fully elucidated and we can only speculate. PA could for example; (1) give general increase in circulation and production of endorphins suggested to reduce back pain [6], (2) reverse connective tissue fibrosis and neurally-mediated inflammation probably linked to back pain [7], (3) be of benefit in rheumatoid arthritis, fibromyalgia and osteoporosis, health problems possibly resulting in back pain and more common among women [33, 34] and (4) reduce pain more among women than men as indicated by experimental studies [35].

Strengths and limitations

We believe that our study has some strengths to mention. First, the longitudinal study design supports a causal relationship between PA and recovery from PBP among women [27]. Second, the large number of potential confounders assessed strengthens the internal validity, although we cannot rule out residual or unmeasured confounding. Third, we consider our overall sample size to be large for a study concerning every day BP for six months or more. Still the analyses of some exposure categories (e.g. 27 recovered women with high PA) were somewhat hampered by lack of statistical power and should therefore be interpreted with caution. Fourth, the baseline definition of PBP indicates back pain severe enough to have negative consequences for the affected individual as well as for the society and therefore important to study.

One problem with studies concerning BP is the lack of consensus for the definition of BP episodes as well as definition for recovery from BP [36, 37]. In a Delphi study from 2011 aiming to standardize the definition of recurrent BP the definition for recovery was “at least 30 days pain-free” [38]. This definition was also incorporated in the definition of an episode of low back pain suggested by de Vet in 2002 [39]. To our knowledge, this is as close to a consensus on a definition of recovery from BP that there is today.

We believe that our recovery outcome (“No periods of considerably disturbing back pain lasting for 7 days or more, during the latest 5-year period”) might be even more relevant. First it incorporates not only pain but also disability which is recognized to be clinically important when defining BP and recovery from BP [36]. Second, it accounts for the fact that BP is a recurrent disorder and therefore being free of disabling pain for 5 years is a very stringent definition.

The present study also has limitations. We had no baseline information on pain intensity or back pain prior to inclusion into the cohort why we could not test these factors as potential confounders. If women with more severe PBP had lower baseline levels of PA than women with less severe PBP, bias due to reversed causation may be present [27, 40]. Then our results may have been overestimated. However this effect may be counterbalanced if some women with more severe PBP were more active than women with less severe PBP. The later could be a result of advice to stay active given by health care providers. Furthermore, the additional analyses stratified on self-rated health (as a substitute for severity of BP) indicated that being physically active is beneficial for women no matter the severity of the PBP. There may also be a problem with unmeasured confounding from health care utilization.

When using self-reported BMI, overweight participants may be classified as having normal weight [41]. Such misclassification would be non-differential and tend to dilute the strength of association [27]. This could explain the lack of association with recovery from BP in our study. Nevertheless, BMI is widely used and recommended as a baseline exposure measurement in cohort studies with BP as outcome [42]. Non-differential misclassification may also concern the measurements of PA with a dilution of potential associations as a consequence. If men tend to misclassify their PA to a higher degree than women, non-differential misclassification may partly explain the different findings between the sexes. However, our measurement method for PA has been reported to be useful for categorizing adults into different levels of PA, based on the physical activity in the different groups as measured by accelerometer [43].

Due to the five-year recall period recovery from PBP could have been misclassified [27]. If for example physically active participants remember their periods of back pain better as it may interfere with their activities, misclassification could either exaggerate or underestimate a true effect. We regard the risk for such differential misclassification as small but if present it would most probably underestimate the effect size.

The follow-up definition of PBP incorporates a dimension of disability different from the baseline definition of PBP. This may cause differential misclassification of outcome if active participants found BP more disabling because it prevents their PA. Again the most probable consequence would be dilution of the estimated association.

Attrition during the sampling process of the cohort and the fact that we select a subsample from the general population could bias our findings and affect external validity. However, some facts contradict this. First, public health survey data from extensive questionnaires with participants blinded to the study hypothesis supports missing at random [44]. Second, given the associations studied are causal we believe that our results are valid also for subjects that dropped out or were excluded.

We believe in summary, considering the strengths and limitations that the associations found in this study are valid.