The CEOR study is an ongoing prospective, population-based cohort of 912 women aged ≥50 years residing in the Jeddah area of Western Saudi Arabia who were recruited in 2004. The study investigated the determinants of bone loss and fractures among healthy postmenopausal women. The rationale and design of the study as well as sampling and data collection have been described previously.[5,17] The baseline evaluation included 3,015 women. At enrollment, women were eligible for the study if they were aged ≥50 years, had been postmenopausal for >1 year (with serum FSH levels >15 mIU/L), were independently mobile and on an unrestricted diet. Exclusion criteria included a cancer history (n=11), regardless of type and stage) and current diagnosis of or prior therapy for osteoporosis (n=15). Women were also ineligible at enrollment if they had diseases or conditions that could interfere with bone metabolism or if they used anti-osteoporotic treatments (n=36), bisphosphonates, estrogen, calcitonin, fluoride salts, calcitriol, strontium or selective estrogen-receptor modulators). Women were also excluded if they had fever (oral temperature ≥38.0℃) or abnormal findings on complete blood cell counts including platelets and/or leukocytes (n=305). Women with abnormal hepatic, renal or thyroid function tests were also excluded (n=1,736). All women provided written informed consent for study participation. The study was approved by the Human Research Ethics Committee of CEOR, King Abdulaziz University. During the follow-up, of the 912 women initially recruited (fulfilling inclusion criteria), 148 were lost to follow-up, 20 died and 37 withdrew. The present study investigated 707 healthy postmenopausal women (mean age, 61 years; range, 53-93 years) from the CEOR cohort at the sixth annual follow-up for whom blood and urine samples were obtained at the baseline and follow-up visits. Women were followed with the intention to treat from the baseline visit to fracture, completion of the follow-up period, withdrawal from the study, or loss to follow-up or death, if no fracture had occurred. The participants received vitamin D and calcium supplementation to maintain the serum 25-hydroxy-vitamin D (25[OH]D) concentration >50 nmol/L and eura parathyroidism. Among the women who completed the study, plasma S1P and BTM levels, along with other analytes, were evaluated during the baseline visit, the 1-year follow-up visit, and annually thereafter (Fig. 1). Additional baseline assessments (using a questionnaire) included information on medical history, menopausal age, years since menopause, socioeconomics, education level, family history of fragility/fracture(s), and occurrence of a fall(s) during the past year. Information on smoking, physical activity (PA), vitamins and medication use were also recorded. Height and weight, as well as hip and waist circumferences, were measured while the women were wearing lightweight clothing with no shoes. BMI and the waist-to-hip ratio were also calculated. Cumulative PA scores from 0 to 27 were constructed based on the 7-day PA record described previously.[1] Women were classified into either sedentary (PA-score ≤14) or moderate/high activity (PA-score >14) groups. Hand-grip strength was measured using a JAMAR-Plus hand dynamometer (Sammons Preston, Bolingbrook, IL, USA) on both hands. Dietary intake of calories, protein, calcium and vitamin D were assessed using a food-frequency questionnaire.[17]

Research physicians at CEOR monitored the enrolled women for incident fractures, which were recorded using CEOR's information system network. Two research physicians classified fracture events independently according to the International Statistical Classification of Diseases and Related Health Problems, 10th revision (ICD-10).[18] A radiologist and/or an orthopedic surgeon reviewed all coded events for final fracture classification. Only low-trauma fractures (i.e., falls from standing height or less) were considered and classified as incident fragility fractures. Incident fractures were reported during annual follow-ups, and all peripheral fractures were confirmed by radiographs or a surgical report, as described previously.[17]

We excluded all non-ORFs (i.e., accident-related fractures such as a traffic road accident or a fall from a height higher than a chair or the first rung of a ladder or severe trauma; and all finger, toe, skull, and facial fractures). The specific fractures and fracture follow-up times were recorded at different sites during a follow-up period when a woman sustained an incident first fracture. The follow-up period was calculated as the time from the baseline visit to the first fracture, death, or the end of the study, whichever occurred first.[17] For women lost during the follow-up, the period was computed as the time from enrollment to the date of last contact. For incident vertebral fractures, lateral X-ray films of the thoracic and lumbar spine were obtained at baseline and annual visits for all women, and at the end of the study. All prevalent and incident vertebral fractures were identified by a radiologist as described by Genant et al.[19]. A prevalent fracture was defined as a vertebral body with a semi-quantitative grade ≥1. Any new vertebral fracture was identified by an increase of at least one grade in a vertebral body that was defined as normal at baseline. We excluded vertebral fractures due to major trauma, and/or vertebral deformities with non-osteoporosis causes (e.g., osteoarthritis and/or bone diseases). All vertebral and non-vertebral fractures were included in the final analyses.

The BMD (lumbar spine [L1-4] and the mean of the proximal right and left femur [total and sub-regions]) was measured by dual energy X-ray absorptiometry using the Lunar Prodigy Model (Lunar Corp., Madison, WI, USA). Quality-control procedures were performed according to the manufacturer's recommendations.[17]

During the baseline and follow-up visits, venous blood samples were collected between 9:00 and 11:00 a.m. after an overnight fast under standardized conditions. Second-void morning urine samples were collected on the same day of blood sampling. The blood was centrifuged, and plasma (sodium citrate- or EDTA-treated), serum, and urine samples were stored in liquid nitrogen at −190℃ at the Biobank Unit of CEOR until they were analyzed. Plasma (treated with sodium citrate) S1P was measured using a competitive ELISA kit (Echelon Biosciences, Salt Lake City, UT, USA) according to the manufacturer's instructions.[20] All plasma samples showing hemolysis or clotting were discarded. However, no sample was collected immediately following any ORF. For women with ORFs, samples (n=9 women) were collected at least 3 months following the incident, and adjustments were made for the collection time. [17] The lower detection limit was 0.06 µmol/L with intra-assay and inter-assay coefficients of variations equaling 6.3% and 5.9%, respectively. Duplicate samples were assayed, and all results were reported as means. Plasma samples were diluted by a factor of 1/10. Further validation studies were performed for this assay. Linearity was assessed by serially diluting samples with sample diluents (dilutions 1:15, 1:20, and 1:30) and comparing observed values with expected values (observed recoveries range, 101%-109%) (data not shown). Additionally, recovery of spiked standards was tested by adding different S1P concentrations (three different concentrations) to eight different human plasma samples presented with various levels of endogenous S1P. Spiked recovery ranged from 90% to 106% (data not shown). No sample tested had S1P values below the lower limit of detection or above the upper limit of quantitation corresponding to 2.0 µmol/L. Other biochemical tests were performed at the Diagnostic laboratory of CEOR according to standard methods as described previously [21,22].

BMD (g/cm²) was determined as described before.

Part of the withdrawn fasting blood sample was used for serum preparation to measure the biochemical BTMs. Moreover, urine samples were obtained in morning. Serum bone alkaline phosphatase (s-bone ALP), serum procollagen type 1 N-terminal propeptide (s-P1NP) , urinary cross-linked C-terminal telopeptide of type 1 collagen (CTX) and serum N-terminal telopeptide of collagen type I (NTx) were measured using competitive inhibition enzyme linked immunoassay by the commercially available kits.

Serum follicle stimulating hormone (FSH), luteinizing hormone, E2 and intact-parathyroid hormone were also, measured by commercially available immunoassays using Elecsys autoanalyzer (Roche Diagnostics GmbH, D-68298. Mannheim, Germany). The intra- and inter-assay coefficient of variations (CVs) were less than 4.0%. Serum 25(OH)D was measured by direct competitive chemiluminescence immunoassay using LIASON autoanalyzer (DiaSorin Inc., Stillwater, MN,. USA). The intra- and inter-assay CVs were 7.8% and 3.8% respectively.

Results are presented as means±standard deviation (SD), and categorical variables are expressed as frequencies. Data were analyzed using SPSS version 17.0 software (SPSS Inc., Chicago, IL, USA). Analysis of variance or the Mann-Whitney test was used, as appropriate, to examine differences among the groups for different variables and χ² test for categorical variables to test for significant differences among groups. Correlation analysis was performed using Pearson's test. Partial correlation analysis was performed to adjust for various possible confounders. The relationship between incident fracture risk and various plasma S1P levels was evaluated using SD score changes and quartile-based analysis. The SD score was calculated using the formula (S1P_i-S1P_m)÷SD, where S1Pi is the individual S1P level, S1P_m is the mean cohort S1P level, and SD is the SD of cohort S1P levels. This calculation was used to determine the change in incident fracture risk for each incremental increase of 1-SD in S1P levels. Similar calculations were made for BTMs. Plasma S1P and BTMs were categorized into Quartiles, and the lowest Quartiles were compared to the fourth (highest) or other Quartiles. Quartiles were calculated for the following variables: age, years since menopause, dietary calcium intake, serum 25(OH)D, PA score, hand-grip strength, and BMD measurement at all sites. Cox proportional hazards regression analysis was used to estimate fracture risk. Data were used to estimate the hazard ratios (HRs) by Quartile and per SD increase in plasma S1P and other BTMs, and results were reported as HRs with estimation of 95% confidence intervals (CIs). In the Cox model, the follow-up time describes the time from the baseline visit to (a) the first fracture; or (b) completion of the follow-up period; or (c) death, if no fracture had occurred. The first fracture was considered the analysis endpoint. HRs were computed by comparing women with and without a fracture. Cox regression analysis was performed for different times: 2.0, 3.0, 5.0, and 6.0 years after collecting samples of the baseline visit. All fracture risk estimates were adjusted for age, BMI, and other confounders as appropriate. The population attributable risk (PAR) proportion was calculated according to the formula {Px(RR-1)÷ [1+Px(RR-1)]}×100, where P is the percentage of the population exposed, and RR is the relative risk.[23]

Baseline characteristics of the participants are presented in Table 1. During the mean follow-up period of 5.2±1.3 years, 138 women sustained ORFs: 44 vertebral fractures and 104 non-vertebral fractures (52 wrist, 12 hip, 21 rib, 8 humerus, 4 ankle, 5 metatarsal, and 2 pelvic) were documented. Women who had no fracture (n=569) were used as age and BMI matched control group in all analyses. As expected, women with ORFs were older, had higher number of years since menopause, lower values for BMD and lower hand-grip strength and PA-score values. Plasma S1P levels were significantly higher in women with ORFs than in women without fractures (7.23±0.79 vs. 5.02±0.51 µmol/L; P<0.001) with no significant difference in BTMs except for plasma CTX (p-CTX) (P=0.002). The mean plasma S1P values at baseline and at different time-point measurements are shown in Table 2. The plasma S1P levels remained stable during the follow-up period with only minor but significant increases (1.94%/year; P<0.001) for all women studied. Approximately 76% to 93% of women remained in the same plasma S1P Quartile at baseline and at the 5-year interval (data not shown). Of those at the highest Quartile at baseline, 89% remained at the highest Quartile also 5 years later (data not shown). Women who had fractures during the follow-up period had constantly high plasma S1P at baseline and at different time-point measurements (Table 2).

Stratifying the participants into three groups according to the number of incident fracture (i.e., no fracture, single or multiple fractures) demonstrated that plasma S1P levels (µmol/L) were significantly higher in women with single (7.03±0.71) or multiple fractures (7.92±0.64) than in those without fractures (5.02±0.51), (P-value for trending P<0.001). After adjustment for BMD value of the lumbar spine (L1-4) or total hip, statistical significance persisted. Plasma S1P levels in vertebral fractures (7.12±0.87) was not significantly different from women with non-vertebral fractures (7.27±0.75).

High plasma S1P levels were associated with increased incident fracture risk (Table 3). After adjusting for age and other potential confounders, the ORF risk for each increment increase of 1-SD of the S1P level was greater than 6-fold among postmenopausal women. Women in the highest Quartile (risk group) had an increased incident fracture risk, so the risk was more than 10-fold as high as the risk in the lower three Quartiles combined (reference group) (Table 3). Significant negative correlations were observed between the plasma S1P levels and BMD at different sites studied, and the adjustments for controlling factors strengthened the statistical significance (Table 4). When we included BMD in multivariate regression analyses, risk estimates were not significantly affected (Table 3). Figure 2 shows the cumulative incidence of fractures in the cohort according to the specific Quartile of the plasma S1P level.

Circulating S1P levels were positively correlated with age (r=0.339; P<0.001), age at menopause (r=0.103; P=0.006), BMI (r=0.212; P<0.001), and years since menopause (r= 0.189 P<0.001). Furthermore, in the women of the present cohort, plasma S1P showed a significant positive correlation with BRMs: p-CTX (r=0.166; P<0.001) and urinary NTx (u-NTx) (r=0.253; P<0.001) (data not shown). Conversely, bone formation markers (BFMs) s-P1NP (r=−0.014; P=0.708) and s-bone ALP (r=0.028; P=0.457) were not associated with plasma S1P, respectively, before or after adjustment for covariates (Table 4).

Circulating S1P levels could predict ORFs 1 year from the baseline visit with HR (95% CI, 6.36 [5.08−7.94]) for a 1-SD increase and 10.23 (2.87-34.46) for levels lower than the highest Quartile (data not shown). HRs for incident fracture prediction was of similar magnitude or slightly lower but remained significant when the follow-up was extended to 2, 3, and 5 years after enrollment (Table 5). There was no apparent increase in the risk of incident fractures in Quartile 2 as compared with quartile 1, but it became significant for Quartiles 3 and 4 (Fig. 3). There were positive significant correlations between plasma S1P levels measured at baseline visit and at 1 year (r=0.812; P<0.001), 2 years (r=0.851; P<0.001), 3 years (r=0.802; P<0.001), and 5 years (r=0.846; P<0.001), after the basal visit.

Baseline BRMs were associated with fracture risk: the HRs per SD for p-CTX, and u-NTx were 1.29, and 1.41, respectively (data not shown). These values persisted after adjustment for age and other potential covariates. When fracture risk was analyzed for Quartiles, women in the highest Quartiles of BRMs exhibited higher fracture risk. These values persisted following adjustment for age and other potential covariates (data not shown). However, BFMs s-P1NP and s-bone ALP were not associated with fracture risk either when analyzed per SD increase or in quartiles of the BFMs and none was able to predict incident fractures (data not shown). Plasma CTX, and u-NTx levels were able to predict ORFs 1 year from baseline visit with HRs (95% CIs, 1.28 [1.26−1.54]; 1.36 [1.08−1.50]) for 1-SD increase and 1.46 (1.10-1.97), and 1.38 (1.10-1.80) for levels lower than the highest quartiles, respectively (data not shown). The HRs for incident fractures prediction was of similar magnitude or slightly higher but remained significant when the follow-up period was extended to 2, 3, and 5 years (data not shown).

The incident fracture risk attributable to plasma S1P levels (in the highest Quartile) was estimated at 31.5%. The PAR for incident fracture independent risk factors among women studied is detailed in Table 6.