In this large study of one thousand elderly women, bone turnover was profiled over ten years using six bone turnover markers capturing different aspects of bone metabolism. The study demonstrated the longitudinal change in BTM values during ageing and confirmed the ability of BTMs to predict fracture risk, particularly in the short-term. The ability to predict fracture attenuated over time and age. In concordance with the recent meta-analysis [5], CTX appears to be the most robust marker for future fracture risk.

Measured at age 75, high levels of CTX were consistently associated with fracture risk, including major osteoporotic fractures, across all time-frames both short and long. Although there is some attenuation of the HR’s over time, CTX appears to be a robust marker for fracture prediction, albeit not specifically for hip or vertebral fractures. This inability to discriminate is consistent with existing data [33,34,35], but could also be explained by an insufficient number of hip fractures, or that vertebral fractures were not clinically defined. While attempting to capture as many vertebral fractures as possible we can presume this to be an underestimate [36]. Particularly for hip fractures it is likely that factors such as predisposition to fall or general frailty attenuates the importance of the instantaneous bone metabolic status. In contrast, PINP was not consistently associated with fracture risk over time in this cohort of aging women, which could reflect a generally reduced bone formation rate at this stage in the life-course. This finding suggests that it may be more relevant to use PINP when investigating osteoporosis in younger patients.

For other bone turnover markers, although directionally in concordance while not consistently statistically significant, resorption markers were clearly more commonly associated with increased fracture risk, particularly in the short term; with the exception of BALP. For example, TRAcP5b appears to be robust in predicting any fracture within a three year time frame. Of interest is also that high levels of urinary osteocalcin were consistently associated with elevated short-term fracture risk. This included major osteoporotic fractures, but also vertebral. While replication in other studies is necessary, this speaks for its potential value, given the difficulty to identify vertebral fractures which are often asymptomatic, while degenerative changes, common in the elderly, can misleadingly indicate higher BMD [37].

We did not systematically adjust for BMD, since it is undoubtedly a stronger predictor of fracture risk than bone turnover markers [38] and has been shown previously in this cohort [6, 7]. In the sensitivity analyses, fracture risk was, as expected, attenuated. However, even if bone turnover markers are not independent of BMD, this does not necessarily undermine their value. On the contrary it provides an instrument capturing underlying metabolic aspects of overall skeletal health. While low BMD at a given assessment captures the end-point of earlier bone loss, high BTMs indicate high global skeletal turnover that may predispose to future bone loss and deterioration of bone architecture or quality, particularly the trabecular network prominent in vertebrae. Being more dynamic than BMD, bone turnover markers are routinely employed for rapid monitoring of both anti-resorptive and anabolic treatment [39] and our analyses indicate that fracture incidence in those with high or low marker levels begins to differ already within the first twelve months of measurement. This suggests that the ‘snap-shot’ of bone turnover reflected by BTM assessment provides valuable evidence of skeletal health, which has clinical value for patient management.

Like most biomarkers there is attenuation with age, and here the association between bone turnover markers and fracture obviously attenuates [7] with our data showing that by age 80, BTM assessment is most likely not useful for fracture prediction. Furthermore, fracture risk prediction by bone markers does not extend over longer time-frames. For predicting long term fracture risk, bone turnover measured at a given time point appears to be less important than other accumulated risk factors, in elderly women well beyond menopause. Given this attenuation with age and over time, incorporation of BTMs, with the possible exception of CTX, into FRAX would be unlikely to benefit 10 year risk assessment. However, biomarkers might be usefully applied at transition stages in health during aging. The results from this study suggest that the ‘snap-shot’ of bone turnover reflected by BTM assessment does in fact provide valuable information of current skeletal health and could provide additional information as an adjuvant tool with other risk assessment methods. In the very old, low bone mass, propensity to fall and composite measures of health are more relevant indicators of fracture risk [40,41,42].

The major strengths of this study include, firstly the clinically relevant cohort of women at demographically high fracture risk, large sample size and homogeneity. This includes the single age of all participants, which helps minimize variation in marker levels associated with chronological age, even if biological age differs. The six bone turnover markers capture all aspects of bone turnover, and measured at three time-points over 10 years, capture the remaining-lifetime perspective of elderly women. Recommended reference markers, CTX and PINP, were assessed at all time-points and in conjunction, we had confirmed fracture data for up to 15 years post-baseline. We envisage these data are appropriate to include in future meta-analyses to estimate effect sizes of BTMs for fracture prediction.

Secondly, we used Z-scores (rather than concentrations) when analyzing HR per SD change in markers which minimizes fluctuations in BTM measures over time (due to methodological changes and sample storage). It also provides evidence that the association between fracture risk and (some or all) bone turnover markers is not necessarily linear, rather that BTMs must be above a threshold before they are associated with fracture risk. By also analysing BTM tertiles we could more clearly demonstrate the ability of bone markers to predict short-term risk of osteoporotic fracture, which would have otherwise been missed. For the majority, the same tertile is maintained across sampling times [43].

Z-scores do not provide generalized diagnostic thresholds for bone turnover markers. While thresholds and ROC analysis have been used in intervention studies to identify responders and non-responders to anti-resorptive therapies [8, 39], employing specific cut-points for bone turnover may be sub-optimal in a trauma-dependent fracture prediction setting. Finally, prior fracture is known to influence bone turnover markers levels, therefore this was included in the adjusted models (along with smoking and bisphosphonate use) and did not alter the results.

As with all observational studies, the results should be interpreted cautiously and we acknowledge the limitations of this study. First are the collection of serum in non-fasted state and single measurement. However, we show that CTX (the BTM most affected by feeding) is very highly correlated in non-fasted and fasted samples and that the majority of women are similarly classified in tertiles from both samples. Overall, this demonstrates that non-fasted/fasted status per se is less relevant; most important is consistency in sampling. We have previously shown that serial assessment of turnover (four samplings over 5 years), is more strongly associated to bone loss than a single BTM measurement [43]. We also showed that the prediction of 9 year fracture risk was more consistent using the average of two resorption marker measurements taken within one year [7]. Second, a confounding effect on BTMs from fractures sustained during the study, is possible, since BTMs increase after fracture [9, 44], remaining higher than pre-fracture levels up to a year or beyond [45]. However, adjustment for fractures sustained prior to baseline led to fracture risk being generally more pronounced. Even adjusting for fracture within the previous 2 years (between ages 73–75) instead of prior osteoporotic fracture (between ages 50–75) had minimal effect on 1, 2 or 3 year HRs (data not shown). Furthermore, in a sub-analysis using CTX, the risk of any fracture over 1, 2 and 3 years was elevated in women without a fracture prior to baseline, although the confidence intervals were wide due to low number (data not shown). Therefore, while recognizing that a previous fracture is one of the strongest risk factors for subsequent fractures [36], our data suggests that even so, markers can pick up a risk signal in the elderly.

Third, while bone active medications may influence marker profiles, in the OPRA cohort their use was rare (3%, 7%, 12% at respective visits) since the study started in the late 1990’s when use of bone active medications was relatively low. Warfarin use was also rare (6% at 5 year; 11% at 10 year). Fourth, inherent in the longitudinal investigation of an already elderly cohort, there is an increase in comorbidities and medications as well as fewer attending follow-up, although cause of mortality is unknown. To address the loss to follow-up, we also reported on those women attending all three visits and also analyzed CTX and PINP using death as a competing risk, with little difference in HRs (Supplemental Table 2). Being cautious, it is possible that different trajectories of bone loss may influence the predictive ability of bone turnover markers at age 75, while conversely by age 80 there may be a selection bias towards healthier individuals who were able to participate; alternately, it reflects reduced power with follow-up. Finally, we acknowledge that multiple tests have been performed, although the confidence intervals, in conjunction with consistent direction of BTM effect, suggests biological plausibility. We cannot fully explain the attenuation of association with age, although sample size is one possibility; other factors may include fall propensity, comorbidity and frailty. The results should be interpreted with these in mind and cannot be extrapolated to younger women, elderly men or other ethnic groups.

In conclusion, the principal contribution from this comprehensive analysis of bone markers in older women, is the demonstration of their ability to inform on fracture risk in a short term, one to three year, perspective, whereas notably, in the long-term or above age 80 years, bone markers appear less valuable. Bone resorption markers, particularly CTX, were more consistently associated with fracture risk than formation markers in the very elderly.