High-impact exercise in adulthood and vertebral dimensions in midlife - the Northern Finland Birth Cohort 1966 study

The Northern Finland Birth Cohort 1966 (NFBC1966) is a prospective population-based birth cohort study [22]. The population initially consisted of pregnant women living in the northern provinces of Finland (Oulu and Lapland) with expected dates of delivery in 1966 (n = 12,068 mothers, n = 12,231 children, 96% of all births). The mothers and their children have been followed since.

Figure 1 presents a flow-chart of the study. At age 31, i.e. in 1997–1998, a postal questionnaire enquiring about the participants’ lifestyle and health were sent to all those whose addresses were known. The response rate was 75% (n = 8767). At age 46, i.e. in 2012–2014, participants with known Finnish addresses (n = 10,282) were again invited to fill in questionnaires and also to take part in clinical examinations, with a total number of 5861 (57%) attendants. Those who attended the clinical examinations and were living in the Oulu region (n = 1988) were further invited to lumbar magnetic resonance imaging (MRI). The MRI population consisted of 1540 participants, as 448 participants did not participate in the imaging due to 1) no show (n = 409), 2) claustrophobia (n = 35), 3) severe obesity (n = 3), or 4) a pacemaker (n = 1). After the imaging, 517 participants were further excluded from this study due to 1) vertebral pathologies (segmentation error, severe disc degeneration, endplate erosions, presence of spondylodesis or Schmorl’s nodes; n = 159), 2) bone-affecting medication (calcium supplements, osteoporosis medication; n = 42), and 3) missing exercise or covariate data (n = 316). The final eligible population was 1023 participants.

Sports participation was self-reported at 31 and 46 years of age via identically formulated questionnaires at both time points. The participants were asked about their participation in walking, cycling, cross-country skiing, swimming, running, gym training, aerobics and gymnastics. They were also asked about their participation in ballgames such as badminton, volleyball, tennis and squash (Group #1), and floor ball, ice hockey, soccer, rink ball, and basketball (Group #2). The answer options in each activity were 1) never, 2) once a month, 3) 2–3 times a month, 4) once a week, 5) 2–3 times a week, and 6) ≥ 4 times a week. Those who reported participating ≥1/week were considered ‘participants’ of the respective sport, and others were considered ‘non-participants’.

To estimate high-impact exercise activity, we assigned a longitudinal impact classification to each participant, based on their high-impact sports participation at the ages of 31 and 46 (Fig. 2). For this purpose, a distinction between high-impact and low-impact sports was made a priori; ballgames, running and aerobics were considered high-impact sports [5, 6, 20, 23]. Gymnastics was excluded from the list, as the Finnish term used in the questionnaires was vague. An individual who reported participating ≥1/week in at least one high-impact sport at the respective time point was considered a high-impact sport ‘participant’; the others were considered ‘non-participants’. Individuals who were consistently high-impact sport participants at both 31 and 46 years were placed in the ‘high’ group. Correspondingly, individuals with consistently little activity (< 1/week) in any of the high-impact sports at both time points were placed in the ‘low’ group. The rest of the population, with some high-impact activity, were placed in the ‘mixed’ group.

We performed magnetic resonance imaging scans using a 1.5-T imaging system (Signa HDxt, General Electric, Milwaukee, WI). The imaging sequences followed routine lumbar spine protocol including T2-weighted fast-recovery fast spin-echo (frFSE) images in sagittal (TR/effTE 3500/112 ms, 4 averages, FOV 280 × 280 mm, acquisition matrix 448 × 224, slice thickness 3 mm with 1 mm interslice gap) and transverse planes (TR/effTE 3600/118 ms, 4 averages, FOV 180 × 180 mm, acquisition matrix 256 × 224, slice thickness 4 mm with 1 mm interslice gap).

We measured eight dimensions (in millimetres, mm) of the corpus of the fourth lumbar vertebra (L4) to calculate its axial cross-sectional area and mean height (Fig. 3). L4 was chosen since it is located caudally in the spine but is more stabile than L5 [24]. L4 has been commonly used in studies similar to the present one [19, 25‐27]. Height dimensions (anterior height, posterior height, minimum height) were measured using the sagittal view and the midsagittal MR slice. Width dimensions, i.e. minimum mediolateral width and maximum mediolateral width, were measured using the appropriate axial slices, which varied between subjects. Depth dimensions, i.e. anteroposterior dimensions, were also measured using axial slices; the superior depth dimension was measured from the most superior and the inferior depth dimension from the most inferior slice possible. The middle depth dimension was measured from the slice existing halfway.

CSA values were calculated using the acknowledged formula CSA = π * a * b, where a = vertebral width/2 and b = vertebral depth/2 [28]. We used the mean of maximum and minimum mediolateral dimensions as the width dimension, and the mean of superior, inferior and middle anteroposterior dimensions as the depth dimension. Vertebral height was calculated as the mean of anterior, posterior and minimum heights.

The same researcher took all measurements using the NeaView Radiology software version 2.31 (Neagen Oy, Oulu, Finland) prior to data analyses. In order to calculate the intraclass correlation coefficient and measurement errors, 400 MRI measurements were taken for a second time by the original measurer, as described in our previous paper [19].

The number of education years and smoking habits were elicited at 46 years using questionnaires. Socioeconomic status, represented by education years, was classified on a three-point scale (≤ 9 years, 9–12 years, > 12 years). This was determined by asking ‘What is your basic education? 1) Less than nine years of elementary school, 2) elementary school, 3) matriculation examination’. Smoking habits were elicited using two questions: 1) ‘Have you ever smoked cigarettes (yes/no)?’ and 2) ‘Do you currently smoke (yes/no)?’ Three categories were formed on the basis of the answers: 1) non-smoker, 2) former smoker, and 3) current smoker. Body mass index (kg/m²) was calculated for each participant using height and weight values that were systematically measured by a trained study nurse at the clinical examinations at age 46.

As our main approach, we used linear regression analysis to investigate the association between vertebral CSA and impact categories. The CSA of L4, normally distributed among each sex, was the outcome variable, and the impact category variable acted as the predictor in the models. Both crude and adjusted regression analyses were run. Since both CSA [19, 29] and exercise levels [19, 30, 31] were a priori known to differ between sexes, the potential interaction between sex and impact exercise was investigated in the models. The assumptions of linear regression were fulfilled. We gathered beta estimates (β) of the predictors with their 95% confidence interval (CI). The low-impact category was chosen as the reference group.

As a confirmatory analysis, we also used generalized estimating equation (GEE) modelling to investigate the association between vertebral CSA and high-impact exercise. We did not construct any longitudinal impact categories for the GEE models, as the method itself took into account the non-independence of the high-impact exercise data recorded at two time points. In the GEE models, CSA was the outcome variable with the dichotomous high-impact sport participant/non-participant variable (at 31 and 46 years) as the predictor. The participant/non-participant variable was considered a repeated measurement with both 31 and 46-year data in the models. We ran both crude and adjusted analyses, and gathered β estimates with their 95% CIs. The non-participants formed the reference group.

The following variables were assessed as potential covariates in the models: 1) height of L4 (continuous variable) 2) BMI at the age of 46 (continuous variable), 3) education years (categorical variable), and 4) smoking status (categorical variable). We chose these variables for assessment because the literature has suggested they are potential confounders [3, 32‐34] and because they were available to us. Each potential confounder was analysed a) individually in relation to vertebral CSA, and b) as part of the linear regression/GEE models. The confounders that were not significantly associated with the outcome and did not alter the models were excluded from the final analyses.

The study adhered to the Declaration of Helsinki and was carried out in accordance with the relevant guidelines and regulations. Approval for the study was obtained from the Ethics Committee of the Northern Ostrobothnia Hospital District. Participation was voluntary and members signed their informed consent at all stages. All data were handled anonymously.

Our sample consisted of 459 (44.9%) men and 564 (55.1%) women with the mean imaging age of 46.8 (standard deviation 0.4) years (Table 1). Most participants had attended school for 9–12 years and had a BMI ≥ 25. Current or previous regular smoking was reported by 50% of men and 39% of women.

In our sample, the most frequent types of activity were traditional Finnish outdoor activities such as walking, running, cycling, and cross-country skiing (Table 2). Participation in one or more high-impact sports was reported by 43% of men and % of women at the age of 31, and by 37% of men and 29% of women at the age of 46 (Table 1). A minority of individuals remained active in high-impact sports at both time points (Table 1).

The mean CSA of L4 was 13.3 (standard deviation 1.7) cm² among men and 10.5 (1.3) cm² among women. The level of intra-rater reliability was high (intraclass correlation coefficient = 0.963), and the values of relative measurement error (%) were distributed normally around the mean of 0.0, with a standard deviation of 4.9.

A significant sex*impact exercise interaction was detected in the linear regression/GEE models, and all analyses were therefore stratified by sex. Assessment of the potential confounders revealed that vertebral height and BMI were significantly associated with vertebral CSA in the models, and that education years and smoking were not. Education years and smoking did not alter the estimates of the models in any direction and were thus not included in the final models.

The linear regression models showed that the women in the ‘mixed’ and ‘high’ impact groups had larger vertebral CSA than those in the ‘low’ group (Table 3). According to the β estimates of the adjusted models, the mixed group had 36.8 mm² (3.5%) larger CSA and the high group 43.2 mm² (4.1%) larger vertebral CSA than the low group. No statistically significant differences in vertebral CSA were detected among men. In the GEE, participation in high-impact sports was also associated with larger vertebral CSA among women but not among men (Table 4).