Material and methods
Participants
We conducted a 2-year prospective cohort study including 11–13-year-old participants who, at baseline, reported never having had spinal pain. The definition was based on three identical questions asked for the three spinal regions (neck, mid back and low back) separately. The first question was ‘Have you ever had neck pain?’ with the response options ‘often’, ‘sometimes’, ‘once or twice’ and ‘never’. This was repeated for the mid back and low back. A diagram with the spinal areas clearly shaded and labelled was shown alongside the questions. The inclusion criterion was self-report of ‘never’ in all three spinal regions at baseline. Participants who (1) were lost to follow-up; (2) did not wear the accelerometer; (3) had <3 days with valid accelerometer data; or (4) had missing data in physical measurements were excluded.
This study was nested within the SPACE study,10 which was a school-based cluster-designed randomised controlled trial aimed at investigating how physical environment combined with organisational initiatives could promote engagement in physical activity in children aged 11–13 years. It involved 14 schools in the Region of Southern Denmark. All 1348 fifth and sixth grade students at these schools were invited to participate. Participation did not require parental consent, but the parents were informed that they could decline their child's participation at any time. The interventions had no effects in the SPACE study,11 and therefore we have treated the study sample as a cohort study.
According to Danish laws, a study that does not contain invasive tests or interventions aimed at individuals does not require ethics approval,12 but the Regional Ethics Committee for Southern Denmark was advised about the study and data collection. Approval from the Danish Data Protection Agency was obtained (#2010-41-5147).
Data collection
Baseline data were collected from April to June 2010 and follow-up data from April to June 2012. The questionnaires (e-survey) were completed individually with teachers observing in order to ensure that there was no interaction between participants. The questionnaires were completed both at baseline and follow-up. Accelerometers were handed out in the classroom with instructions on how to use them. All anthropometrical assessments and physical fitness tests were performed during school time in a sports hall close to the school that the participant attended. Assessors who were members of the research staff and who were trained in the test procedures provided instruction in the use of the accelerometer and performed all physical measurements. The accelerometer data and physical measurements were collected at baseline only. Detailed information on the SPACE protocol can be found elsewhere.10
Accelerometer data
Physical activity was measured using the Actigraph GT3X Triaxial Activity Monitor.13 Participants were asked to wear the accelerometer on their waist for seven consecutive days, except during activities in water. In order to increase compliance, participants and their parents received an SMS text reminder every morning.
The Actigraph registers and records accelerations ranging from 0.05G to 2.5G in the vertical, horizontal and transverse axes. Currently, there are no available calibration studies for the Actigraph Triaxial Activity Monitor, but the vertical dimension of the instrument has been compared with uniaxial accelerometers. Only the vertical axis was used in the analyses. However, activity in all three axes was used to calculate the non-wear time because the triaxial accelerometer records more activity in the anteroposterior direction during sedentary activities and is therefore more sensitive to differentiating between non-wear time and sedentary activity.14
To ensure valid measures, a minimum wear-time of the accelerometer was set to 10 hours a day between 6:00 and midnight. This was required for a period of at least 3 days, which has been shown to give a reliable estimate of physical activity in children aged 7 years.15 In preliminary analyses, there were no statistical differences in the overall physical activity between weekdays and weekend days, and therefore the required days needed neither to be consecutive nor include both weekdays and weekend days. Non-wear time was defined as no activity measured by the accelerometer for at least 60 consecutive minutes as recommended by Toftager et al14 and was excluded from the data. Activity was summarised for every 10 s (epoch length) with counts per minute (cpm) as output. We categorised the total time spent at different activity levels using cut-points recommended by Evenson et al:16 sedentary was between 0 and 100 cpm; light was between 101 and 2295 cpm; moderate was between 2296 and 4011 cpm; and vigorous was between 4012 and 50 000 cpm. More than 50 000 cpm was considered non-physiological and replaced by 0. Processing of the accelerometer raw data was undertaken using the software Propero Actigraph Data Analyzer V.1.1.2 (RICH, University of Southern Denmark, Denmark).
Variables
Spinal pain (outcome)
The same questions used to define the cohort as described above in the participant section were used to define the outcome. An incident case was defined as a report of pain ‘often’, ‘sometimes’ or ‘once or twice’ in at least one of the questions addressed to neck pain, mid back pain or low back pain at follow-up. The questions were developed and tested for feasibility, content validity and item agreement between questionnaire scores and interview findings in 9–11-year-olds.17
Objective measure of physical activity (exposure)
We calculated the proportion of the day spent for different activity levels (ie, sedentary, moderate and vigorous combined, and vigorous alone) by dividing the total time spent at the specific level by the total wear time across all valid accelerometer days. Then, we dichotomiszed the activity at the 90th, 75th and 50th centiles in order to explore cut-points for potential risk. For the 90th centile cut-point, we compared the top 10% with the remaining 90%.
Overall physical activity was defined by taking the mean cpm divided by 100 across all valid accelerometer days. The variable was investigated for outliers, defined as three SDs above and below the mean. In cases with outliers, raw data were visually inspected for abnormal activity patterns. No suspicious movement patterns were registered, and therefore all data were included.
Potential confounders
Sex, height (cm), weight (kg), body mass index (BMI) (kg/m2), waist-to-height-ratio, participation in contact and collision sports, 10×5 m shuttle run test (sec), handgrip strength test (kg) and Andersen test (m) were included as potential confounders in the analysis. A more detailed description of these is presented in online supplementary appendix 1. Psychological factors were included in the questionnaires but not included as potential confounders because of insufficient numbers of participants who felt low, were irritable/in a bad mood, felt nervous or had sleeping difficulties.
Other variables
We adjusted for the mean cpm for time spent in other activity levels. For example, we adjusted for non-sedentary activity, that is, the mean cpm of light activity and above (101–50 000 cpm) in sedentary models. The rationale for this was that adolescents may have tolerated sedentary activity better if they had higher exposure levels of physical activity the rest of the time and vice versa.
Statistical analyses
Attrition bias was investigated by comparing who responded to the follow-up questionnaire against those who did not with regard to sex and BMI. In order to investigate selection bias, we compared the excluded participants to the final sample with regard to sex, but due to missing assessments of height and weight, a comparison with regard to BMI was not possible.
Descriptive statistics with frequencies were calculated for the categorical and ordinal variables. Mean, SD, median with 95% CIs, and range were calculated for interval and ratio variables.
To determine if physical activity predicted the development of spinal pain, a generalised Poisson regression model for data with underdispersion was fitted.18 This was because the variance (0.22) was smaller than the mean (0.67). The unadjusted and adjusted relative risks (RRs) for the incidence of spinal pain were calculated for each physical activity exposure variable. Each potential confounder was tested by including it in the unadjusted model and determining if the β coefficient of the physical activity exposure variable changed by more than 10%.19 This approach was preferred rather than forcing all confounders into the model because the theoretical basis for choice of variables is weak. All confounders that met the above criteria were entered into the final model for each physical activity exposure variable. Multicollinearity was checked and if variance inflation factors of each variable exceeded 10, the variable was transformed by squaring the variable. Model assumptions were checked.
Sensitivity analyses were performed with a 30 s epoch because choice of epoch length is debatable and because a longer epoch length might lead to an underestimation of moderate-to-vigorous physical activity.20 All analyses were performed using STATA V.11.2 (Stata Corporation, College Station, Texas, USA).
