Effects of Leisure-Time Physical Activity on Vertebral Dimensions in the Northern Finland Birth Cohort 1966

Vertebral fractures are a common burden amongst elderly and late middle aged people. Vertebral cross-sectional area (CSA) is a major determinant of vertebral strength and thus associated with vertebral fracture risk. Previous studies suggest that physical activity affects vertebral CSA. We aimed to investigate the relationship between leisure-time physical activity (LTPA) from adolescence to middle age and vertebral dimensions in adulthood. We utilized the Northern Finland Birth Cohort 1966, of which 1188 subjects had records of LTPA at 14, 31 and 46 years, and had undergone lumbar magnetic resonance imaging (MRI) at the mean age of 47 years. Using MRI data, we measured eight dimensions of the L4 vertebra. Socioeconomic status, smoking habits, height and weight were also recorded at 14, 31 and 46 years. We obtained lifetime LTPA (14–46 years of age) trajectories using latent class analysis, which resulted in three categories (active, moderately active, inactive) in both genders. Linear regression analysis was used to analyze the association between LTPA and vertebral CSA with adjustments for vertebral height, BMI, socioeconomic status and smoking. High lifetime LTPA was associated with larger vertebral CSA in women but not men. Further research is needed to investigate the factors behind the observed gender-related differences.


Results
Study sample. A total number of 524 males (44.1%) and 664 females (55.9%) were included in the analyses ( Table 1). The mean age of imaging was 46.8 years for men (standard deviation, SD, 0.4 years, range 45.9-47.8 years), and 46.8 years for women (SD 0.4 years, range 45.8-48.0 years). In our sample, 34% of men and 49% of women were within the normal Body mass index (BMI) range, while others were overweight. Most subjects (72% of men and 73% of women) had attended school for 9 to 12 years and had never smoked on a regular basis (52% of men and 61% of women).

Leisure-time physical activity.
In the latent class analysis (LCA), three lifetime LTPA clusters were identified as the most appropriate solution. The clusters were named as "inactive", "moderately active" and "active"; the distributions of LTPA at 14, 31 and 46 years are shown for each cluster in Table 2. Based on the LCA, 29% of men and 24% of women were classified as active, whilst 29% and 30% were classified as inactive, respectively. Vertebral dimensions. The mean vertebral CSA was 13.25 (SD 1.67) cm 2 among males and 10.57 (SD 1.28) cm 2 among females in our sample (25.4% higher among males). The measurements of L3 and L4 showed a high correlation (Pearson's R = 0.870, p < 0.001) in the CSA of these vertebrae and the level of intra-rater reliability was high (intraclass correlation = 0.963). The values of relative measurement error (%) distributed normally around the mean of 0.0 with a standard deviation of 4.9 (n = 400 repeated measurements).
When investigating the association between LTPA and vertebral CSA using linear regression, we found that unadjusted and adjusted analyses provided similar results (Tables 3 and 4). However, the F-test p values supported statistical significance of the adjusted models (p < 0.01 in all models), whereas the unadjusted models all had a p value of > 0.05. According to the R 2 values, the unadjusted models, utilizing only the LTPA variables without adjustments, explained <1% of the variability in the CSA of the subjects. The adjusted models, in turn, explained 5.2-6.6% of the variety.
In the analysis of the association between lifetime LTPA and vertebral CSA (Table 3), those women who belonged to the "active" cluster had 0.34 cm 2 (3.2%) larger vertebral CSA, compared to those who belonged to the "inactive" cluster (p = 0.012). No statistically significant differences were detected in the CSA between women classified as "inactive" and "moderately active". Among males, no statistically significant differences were detected.
In the additional analyses regarding LTPA at different time points (Table 4), we detected that those women who were physically active ≥4 times per week at 31 years, had 0.47 cm 2 (4.5%) larger vertebral CSA than the reference group (p = 0.003). We also detected that those men who were active 2 times per week at 14 years, had smaller CSA compared to the reference group (p = 0.027). No other statistically significant findings were obtained.

Discussion
In this population-based birth cohort study, our main finding was that a high level of lifetime LTPA from 14 to 46 years of age was associated with larger vertebral CSA in middle-aged women. Our additional analyses showed that the participation frequency of ≥ 4 times/week at age 31 was associated with larger vertebral CSA in women at 47 years. However, the detected CSA differences were of minor magnitude, and physically active and inactive men had similar vertebral dimensions.
In our previous study, physical activity and participation in different sports were not associated with changes in vertebral size at the age of 21 years 37 . The lack of association between physical activity and vertebral size then may be due to the young age of the study population. In the present study, we wanted to focus on the overall amount of LTPA instead of investigating individual sports. We used an older study population in which all individuals had reached their skeletal maturity and peak bone mass. The sample size we used was rather large, compared to both our previous studies and other research on vertebral dimensions 3 provided us with longitudinal data on the subjects from adolescence until middle-age. As the subjects shared their birth year, the confounding effect of age was minimized.
Several studies have analyzed the dimensions of other lumbar segments, e.g. L3, instead of L4 3 , possibly affecting the comparability of results. Previous literature has described the similarity of vertebrae in terms of dimensions and strength, concluding that the strength prediction of all thoracolumbar vertebrae can be made with high accuracy using measurements from one vertebra 38 . Our comparisons between the dimensions of L3 and L4 confirmed that the size of different lumbar segments is highly linked and so our study is comparable with others, regardless of the vertebra investigated. We decided to focus our interest on the fourth lumbar vertebra (L4), as it is situated in the inferior part of the vertebral column and therefore is amongst those vertebrae that carry the most weight. L5 too is under major strain but it is also an integral part of the lordotic curvature of the lumbar spine 39,40 , and therefore was not measured. We acknowledge that due to lumbar lordosis, the location and orientation of also L4 varies between individuals. This may have affected our orientation of the MR slices and thus added to the measurement error. Age at imaging, years; mean (SD) 46   There are many potential explanations for the observed gender differences. Given that women have smaller vertebrae, both absolute and relative to body size, there might be a higher ability or a greater demand for the vertebrae to enlarge in size. Furthermore, the men in our study sample were somewhat more physically active than women, which may prevent us from detecting a similar effect of LTPA on vertebral size. The ambiguous, inexplicable association between LTPA at a frequency of 2 times/week at age 14 and decreased vertebral CSA in men was not in line with our other results.
Another potential hypothesis is that the female gender may have vertebral compensatory mechanisms, such as periosteal apposition 3 , functioning differently from men. In males, periosteal apposition could be activated at lower LTPA levels, leaving the threshold undetectable; it might thus explain the increase in bone diameter in all men, regardless of LTPA. Interestingly, it has been suggested that androgens increase periosteal apposition, whereas estrogen levels have been shown to decrease it 41 . Intervertebral disc degeneration has also been linked with higher vertebral bone growth tendency 42 but the degree of disc degeneration was not evaluated by us when reading lumbar MR images.
We are aware that the effect of physical activity on vertebrae may be expressed in other ways than changes in vertebral dimensions which were investigated in this study. The alternative mechanisms might be e.g. cortical % (n)    shell thickening and/or changes in vertebral BMD. However, our previous study did not find any correlation between LTPA and trabecular bone density parameters in vertebrae 37 .
Our study suggests that lifetime LTPA is positively associated with vertebral size, to a small extent, among women but not men. Further research is needed to confirm our findings and shed light on the factors behind the observed gender-related differences.

Methods
Study population. Individuals whose expected date of birth fell between January 1 st and December 31 st 1966 in Northern Finland (96.3% of all 1966 births, n = 12,058 live births) were included in the prospective NFBC1966 study. Since their mothers' recruitment during their first visit to the maternity health centers, data have been collected on the subjects' health, lifestyle and social status. The study was conducted according the Declaration of Helsinki and approved by the Ethical Committee of the Northern Ostrobothnia Hospital District in Oulu, Finland. Cohort members and in youth also their parents provided written informed consent for the study. All personal identity information was encrypted and replaced with identification codes, providing full anonymity for the whole study population. We confirm that the study was carried out in accordance with the approved guidelines.
At the age of 46 years, subjects who were living at known addresses in Finland (n = 10,282) were invited to attend clinical examinations (Fig. 1). A total of 5,861 (57%) subjects participated. Of these, we invited all participants living within 100 km of the city of Oulu (n = 1,988) to lumbar magnetic resonance imaging (MRI). Of them, MRI was not performed to 448 subjects due to 1) not showing up, 2) claustrophobia, 3) severe obesity preventing the imaging, or 4) a pacemaker. The final MRI study population consisted of 1540 participants.
Assessment of leisure-time physical activity at 14, 31 and 46 years. Physical activity was self-reported at 14, 31 and 46 years of age 43 . At the age of 14, subjects were asked how often they participated in sports outside school hours with the following alternatives: 1) daily, 2) every other day, 3) twice a week, 4) once a week, 5) every   other week, 6) once a month, and 7) generally not at all. The subjects having answered either 5, 6 or 7 were combined into one category ("every other week or less"), as the groups were small. Other categories were kept as-is. At the ages of 31 and 46, subjects were asked how often they participated in brisk physical activity/exercise during their leisure-time. The term 'brisk' was defined as physical activity causing at least some sweating and getting out of breath, corresponding to moderate-to-vigorous intensity. The six response alternatives were 1) daily, 2) 4-6 times a week, 3) 2-3 times a week, 4) once a week, 5) 2-3 times a month, and 6) once a month or less often. The subjects having answered either 1 or 2 were combined into one category ("≥4 times/week"), as the number of respondents was small, and other categories were kept as-is.
Covariates at 46 years. At the age of 46, subjects underwent clinical examinations, where their height and weight were systematically measured by a trained study nurse. Body mass index (BMI) was then calculated for each subject (kg/m 2 ). Socioeconomic status was evaluated based on the number of years the subject had attended school for (≤9 years, 9-12 years, >12 years). This was determined by asking: "What is your basic education?" 1) Less than 9 years of ground school, 2) ground school, or 3) matriculation examination. Smoking history and current smoking status were inquired with two questions: 1) "Have you ever smoked cigarettes (yes/no)?" and 2) "Are you currently smoking (yes/no)?" According to these questions, three categories were formed: 1) non-smoker, 2) former smoker, and 3) current smoker.
Lumbar magnetic resonance imaging. Magnetic resonance imaging scans were performed with a We measured 8 dimensions from the corpus of the L4 vertebra ( Fig. 2) to calculate the axial cross-sectional area and mean height of L4. Vertebral height dimensions (anterior height, posterior height, minimum height) were measured using the sagittal view and the most medial slice that was available. Width dimensions, i.e. minimum mediolateral width and maximum mediolateral width, were measured using the appropriate axial slices that varied between subjects. Typically the minimum width was encountered near the middle part of the vertebra and the maximum width near either the superior or inferior end of the vertebra. Depth dimensions, i.e. anteroposterior dimensions, were measured using axial slices. The superior depth dimension was measured using the most superior appropriate slice just before the intervertebral disc. Correspondingly, the inferior depth dimension was measured using the most inferior slice possible. In order to measure the middle depth dimension we chose the slice that existed halfway between the superior and inferior ends of the vertebra. CSA values were calculated by using the acknowledged formula CSA = π * a * b, where a = vertebral width/2 and b = vertebral depth/2 44 . The mean of maximum and minimum mediolateral dimensions was used as the width dimension, and the mean of superior, inferior and middle anteroposterior dimensions was used as the depth dimension (Fig. 2). We additionally used vertebral height as a covariate in the analysis. This height dimension was calculated as the mean of anterior, posterior and minimum height dimensions of L4.
All MRI measurements were performed by the same researcher, using the NeaView Radiology software (Neagen Oy, Oulu, Finland) version 2.31, which is collectively in use on clinical workstations in Oulu University Hospital. The measurements were made prior to gathering or analyzing any other data on the subjects. To explore the correlation between dimensions of different vertebrae, we also measured the same dimensions from the L3 vertebra from a subsample (n = 110).
Intra-rater reliability and measurement error. In order to investigate intra-rater reliability, 50 MR images (equivalent to n = 400 measurements) were randomly selected for a second measuring three weeks after the first one, conducted by the original measurer. Based on the original and repeated measurements, intraclass correlation coefficient was calculated. Additionally, measurement errors were calculated. As the height, width and depth dimension values were of different magnitudes, we considered relative error (%) more informative as opposed to investigating absolute error.
Statistical analyses. Latent Class Analysis (LCA) was used to obtain clusters, i.e. groups in which the subjects had a similar profile of LTPA during their lifetime (14-46 years of age). In LCA, the number of clusters is increased until the most appropriate model is found 45,46 . In this study, the LCA was conducted according to the self-reported frequency of LTPA at 14, 31 and 46 years of age. Models with cluster numbers from one to seven were assessed, and the best-fitting cluster model for these subjects was determined by calculating the Bayesian Information Criterion (BIC), where lower values indicate a better fit. Clinical interpretability of the classification, the conceptual meaningfulness of the models, and the sizes of the subgroups were also assessed while choosing the best model. In order to obtain equal definitions of the LTPA clusters in both men and women, and thus ensure the comparability of results between genders, the clustering was not stratified by gender.
We used linear regression analysis to reveal the association between LTPA and vertebral size. Vertebral size, i.e. the CSA of L4, was regarded as the dependent variable in all analyses, whereas LTPA variables acted as the explanatory variables. As both CSA 12 and LTPA 47,48 were known to differ between genders, a significant gender interaction was expected and therefore all regression analyses were performed separately for men and women. The models were adjusted for 1) height of the L4 vertebra, continuous variable; 2) BMI at 46 years, continuous variable; 3) socioeconomic status determined by education years, categorical variable; and 4) lifetime smoking Scientific RepoRts | 6:27844 | DOI: 10.1038/srep27844 status determined at 46 years, categorical variable. These determinants were known to associate with changes in vertebral size 3,49,50 . First, we attempted to test our main hypothesis of high lifetime LTPA associating with increased vertebral dimensions. In order to investigate this, we analyzed the association between lifetime LTPA, represented by the LTPA cluster variable, and vertebral CSA. The "inactive" cluster was chosen as the reference category, and the other two clusters were compared to it. Two analyses were conducted (for both genders separately), one with and one without adjustments for vertebral height, BMI, education years and smoking.
We also conducted additional analyses to assess the most important time point in terms of LTPA and increased vertebral dimensions. In order to investigate this, two linear regression analyses were conducted for both genders with the following variables included: 1) LTPA at 14, 31 and 46 years, 2) LTPA at 14, 31 and 46 years, vertebral height, BMI, education years and smoking. The LTPA variables of all three time points were included in every analysis and they thus acted as covariates for each other and their confounding effect was minimized. In the LTPA variables, the category with lowest LTPA frequency was chosen as the reference category, and the other 4 categories were compared to it.
Statistical analyses were conducted using the SPSS software (IBM, Armonk, NY, USA) version 22, 64-bit edition.