Study design and population

The European Prospective Investigation into Cancer and Nutrition (EPIC)-Norfolk cohort analysed in this study is part of EPIC, a collaboration involving 10 European countries developed primarily to examine the association between diet and cancer, with additional health outcomes examined in EPIC-Norfolk. This cohort has been described in detail previously [14], but in brief, the Norfolk cohort consisted of 25,639 men and women aged 40–79 y living in the general community who participated in a baseline health examination (HE1) between 1993 and 1997 and a subsequent assessment (HE2) among 15,786 between 1997 and 2000. Both assessments were proceeded by a Health and Lifestyle Questionnaire (referred to as HLQ1 and HLQ2 respectively).

Eligible participants for the current analysis were 14,624 men and women who attended HE2 (aged 42–82 y at that time), with an available blood sample for 25(OH)D and who had data on physical activity from HLQ1.

The study was approved by the Norwich District Health Authority Ethics Committee, and all participants gave signed informed consent.

Main exposure: Blood samples

Plasma and serum samples were obtained from venipuncture blood samples during HE2. Samples were stored in liquid nitrogen tanks until 2012 when serum samples were retrieved for 25(OH)D assays. Assays were conducted by VITAS, which is a reference laboratory in Nordic countries for fat soluble vitamins [15]. Assays for 25(OH)D were based on ultraperformance liquid chromatography interfaced by atmospheric pressure chemical ionization to mass spectrometry. This method measures the 25(OH)D₃ and 25(OH)D₂. The lower limit of detection was 1–4 nmol/L. The coefficients of variation (CV) for interassay analyses were 7.6% at 25(OH)D concentrations of 47.8 nmol/L and 6.9% at 83.0 nmol/L.

Only 923 individuals had measurable 25(OH)D₂ (>1 nmol/L), the remainder were coded as zero. Population mean concentrations were 56.2 nmol/L (range 3.4–201.9 nmol/L) for 25(OH)D₃ and 5.6 nmol/L (range 3.0–82.5 nmol/L) in those with detectable concentrations for 25(OH)D₂. The total 25(OH)D concentration, was calculated as the sum of 25(OH)D₃ and 25(OH)D₂. We categorized total 25(OH)D into 5 categories by using clinically relevant cutoffs [16] as follows: <30, 30 to <50, 50 to <70, 70 to <90 and ≥90 nmol/L.

Main exposure: Physical activity

Habitual PA was assessed from HLQ1 which included the short EPIC Physical Activity Questionnaire asking about PA during the preceding 12 months. The first section was about PA at work. The second section was about the amount of time spent in hours per week for summer and winter separately in each of the following activities: walking, cycling, gardening, do-it-yourself, physical exercise and housework. A simple PA-index was constructed to allocate participants to four ordered categories of overall activity [17]: inactive (sedentary job and no recreational activity); moderately inactive (sedentary job with <0.5 h recreational activity per day or standing job with no recreational activity); moderately active (sedentary job with 0.5–1 h recreational activity per day, or standing job with <0.5 h recreational activity per day, or physical job with no recreational activity); and active (sedentary job with >1 h recreational activity per day, or standing job with >0.5 h recreational activity per day, or physical job with at least some recreational activity, or heavy manual job). For the purpose of this study participants were categorized in two physical activity levels: physically inactive (inactive and moderately inactive) versus physically active (moderately active and active). The index was validated against a combined heart rate and movement sensor (with individual calibration) in 2000 participants from 10 European countries (representative of the EPIC-population with respect to age and sex). The pooled estimate of the correlation between self-reported and objectively assessed energy expenditure was r = 0.33[18]. We also showed a high repeatability of the index (weighted kappa = 0.6, p<0.001) [17].

Ascertainment of incident fractures

Trained nosologists obtained vital status of the entire cohort based on death certificates of the United Kingdom Office of National Statistics. Individuals were also linked via their unique National Health Service number with the East Norfolk Health Authority (ENCORE) database, which identifies all hospital contacts throughout England and Wales for Norfolk residents [19]. International classification of diseases (ICD) 9 and 10 diagnostic codes were used to identify fracture events, representing all hip (120; S72), spine (225; S12,S22,S32,T08) and wrist (910; S52,S62) fracture cases in the cohort up to March 2015, an average of 15 ± 2.3 y of follow-up.

Assessment of covariates

Trained nurses carried out a health examination. Height and weight were measured according to standard protocols [14], conducted at the participant’s general practitioner’s practice. Height was determined to the nearest millimetre by using a freestanding stadiometer. Weight was recorded to the nearest 0.2 kg with the participant wearing light clothing and no shoes. Body mass index (BMI) was estimated as weight divided by the square of height (kg/m²).

Participants also completed a self-administered health and lifestyle questionnaire (assessed at HLQ2). This included a self-reported medical history of diabetes, cancer, osteoporosis, arthritis and previous fractures; menopausal status, categorized as premenopausal, perimenopausal (<1 y), perimenopausal (1–5 y), or postmenopausal; and hormone replacement therapy (HRT) status; supplement use was ascertained by the question “Have you taken any vitamins, minerals, or other food supplements regularly during the past year (such as vitamin C, vitamin D, iron, calcium, fish oils, primrose oil, betacarotene, etc)?”, categorized as yes or no; smoking status was ascertained by asking “Have you ever smoked as much as one cigarette a day for as long as a year?” and “Do you smoke cigarettes now?”, categorized as current, former, or never users; and alcohol consumption was ascertained by asking about beer, wine and spirit consumption in a week (computed as units per week).

Participants’ occupation (assessed at HLQ1) was classified according to the Registrar General’s occupation-based classification scheme into 6 main categories with social class I representing professionals, social class II representing managerial and technical occupations, social class III representing subdivision into non-manual (III.N) and manual (III.M) skilled workers, social class IV representing partly skilled workers, and social class V representing unskilled manual workers. We re-categorized social class into manual (III.m-V) and non-manual (I-III.N) social classes [20]. Educational status (assessed at HLQ1) was based on the highest qualification attained and was categorized into 4 groups as follows: degree or equivalent, A-level or equivalent, O-level or equivalent, and less than O-level or no qualifications. O-level indicates educational attainment to the equivalent of completion of schooling to the age of 15 y, and A-level indicates educational attainment to the equivalent of the completion of schooling to the age of 17 y. Educational level was categorized into a binary variable: “at least O-level” (which includes O-level, A-level, and degree) versus “no qualifications” [20].

Statistical analyses

All analyses were conducted using the Statistical Package for the Social Sciences for Windows version 20 (SPSS Inc, Chicago, IL, USA). A p-value < 0.05 was considered statistically significant.

We examined risk-factor distributions in men and women by 25(OH)D category. Potential confounders such as, age, BMI, social class, educational level, calcium intake, alcohol intake, smoking status, menopausal status, HRT use, supplement use, month of blood drawn, history of diabetes, cancer, osteoporosis, arthritis and previous fractures were evaluated for inclusion in the Cox regression models. Concentrations of 25(OH)D showed marked seasonal variations, and thus, all results were adjusted for the month of assessment as well as for age. The direction of the risk factors between men and women were similar and thus, analyses were performed for men and women together.

Cox proportional hazard models were used to determine hazard ratios (HR) of fractures for the two main exposures (A) by categories of plasma 25(OH)D with adjustment for: 1) age, sex and month of blood sampling, 2) model 1 and BMI, smoking status, alcohol intake, supplement use and history of fractures and 3) Model 2 and physical activity and (B) by physical activity where model 1 excluded month of blood sampling.

We tested the potential interaction between physical activity levels (active vs. inactive) and categories of 25(OH)D serum concentrations and between age (< = 65 y vs. >65 y) and 25(OH)D serum concentrations by including these as an additional model (model 4) and using the likelihood ratio test (LRT) to test for significance compared to model 3.

We also estimated HRs for fracture incidence up to 2, 4, 6, 8, 10 and 12-years of follow-up, per 20-nmol/L increase in serum 25(OH)D using model 1, 2 and 3.

Strengths and limitations

The present study has several limitations as well as strengths. Regarding limitations, firstly the physical activity index used was a combined index of leisure time and occupational activity. While this index has been objectively validated against heart rate measurements, and has demonstrated utility in being a strong predictor of mortality and cardiovascular disease [34], this index did not take into account the type of exercise undertaken (e.g. weight bearing, high impact) and this could partially explain why we did not find associations between our measure of physical activity and fracture risk [35]. Also, those individuals who exercise outdoors may get additional benefits from sun exposure than those who exercise indoors [36].

Some have suggested that while increasing physical activity may improve bone health, it may increase fracture risk by increasing falls [37]. Lastly, it must be recognized that the opportunity for falling probably increases as people become more physically active, particularly in community dwelling elderly [38].

There is some evidence suggesting that physical activity habits during childhood may have long-lasting benefits on bone health [39]. Unfortunately, we did not have information on physical activity when peak bone mass was attained in this cohort, and therefore we could not investigate the relationship between physical activity in early adulthood, adolescence or childhood with fracture risk.

Misclassification of physical activity levels due to under- or over-reporting is inevitable, although the physical activity questionnaire has shown good reliability in classifying individuals into physical activity levels [18]. In contrast, we did not keep track of physical activity level over time using similar assessment methods and those individuals who were physically active when data were collected could have decreased their activity and consequently being more prone to fractures as has been suggested [9].

Unfortunately, we did not keep track of the place while exercising if indoors or outdoors. Moreover, the causes of fractures, whether by fragility or high impacts, were not registered and this could have explained why we did not observe any interaction between physical activity and 25(OH)D, because fractures could be due to high impacts while exercising.

As previously discussed, a single baseline blood sample to assess vitamin D status may not have adequately characterised an individual’s exposure over a long follow up time period, especially over an era of changing clinical and public health practice in vitamin D supplementation, particularly to groups perceived as most vulnerable to fractures such as older women [40]. Moreover, while ELISA assay for 25(OH)D₃ are generally accepted the LC/MS methods would be more precise, as it allows detection and measurement of other recently described mono-xydroxymetabolites that are biologically active [41–44].

Potential strengths of our study are the large population sample including both men and women, large pool of fractures (1183 cases), long duration of follow-up and detailed assessment of potential confounders.