A portion of this work was presented as an abstract at the 18th Annual Meeting of the American Society for Bone and Mineral Research in Seattle, WA, U.S.A., September, 1996.
To examine the ability of commercially available biochemical markers of bone formation and resorption to predict hip bone loss, we prospectively obtained serum and timed 2-h urine specimens from 295 women age 67 years or older who were not receiving estrogen replacement therapy. Serum was assayed for two markers of bone formation: osteocalcin (OC) and bone-specific alkaline phosphatase (BALP). Urine specimens were assayed for four markers of bone resorption: N-telopeptides (NTX), free pyridinolines (Pyr), free deoxypyridinoline (Dpyr), and C-telopeptides (CTX). Measurements of hip bone mineral density were made at the time the samples were collected and then repeated an average of 3.8 years later. Higher levels of all four resorption markers were, on average, significantly associated with faster rates of bone loss at the total hip, but not at the femoral neck. Women with OC levels above the median had a significantly faster rate of bone loss than women with levels below the median, but there was no significant association between levels of BALP and hip bone loss. The sensitivity and specificity of higher marker levels for predicting rapid hip bone loss was limited, and there was considerable overlap in bone loss rates between women with high and low marker levels. We conclude that higher levels of urine NTX, CTX, Pyr, Dpyr, and serum OC are associated with faster bone loss at the hip in this population of elderly women not receiving estrogen replacement therapy, but these biochemical markers have limited value for predicting rapid hip bone loss in individuals.
Low bone mass is a strong risk factor for osteoporotic fracture.(1) Therefore, a test that identifies individuals at high risk for rapid bone loss could serve as a valuable clinical tool. It has been suggested that biochemical markers of bone turnover may be used to identify which subjects will lose bone most rapidly in the near future.(2–7) Indeed, some cross-sectional studies have found a weak inverse relationship between biochemical markers and bone density and that marker levels are elevated during periods of accelerated bone loss, such as early menopause.(5,8–10) However, prospective studies examining the ability of markers to predict rates of hip and spine bone loss have yielded inconsistent results.(2–7,11–14) To examine this issue further, we prospectively measured urine and serum markers of bone resorption and formation in a cohort of 295 community-dwelling women at least 65 years old and examined their association with subsequent hip bone loss.
MATERIALS AND METHODS
This study involved a subset of participants in the Study of Osteoporotic Fractures (SOF), a multicenter study of risk factors for fracture in 9704 nonblack women 65 years of age or older who were recruited from population-based listings at four clinical centers: The Kaiser-Permanente Center for Health Research, Portland, Oregon; the University of Minnesota, Minneapolis, Minnesota; the University of Maryland, Baltimore, Maryland; and the Monangehela Valley, Pennsylvania. Details of the study methods have been described previously.(15) We excluded black women because of their low incidence of hip fracture,(16) women who were unable to walk without the assistance of another person, and women with a history of bilateral hip replacement.
Fasting serum and 2-h morning urine was collected and stored at −190°C in a consecutive sample of 501 women (∼125 from each clinical center) between April and July, 1989. Bone loss was measured by hip dual-energy X-ray absorptiometry (DXA) (QDR 1000; Hologic, Inc., Waltham, MA, U.S.A.) at baseline and after a mean follow-up of 3.8 years (range 3.3–5.1 years). For these analyses, we excluded 89 women who reported oral estrogen use at baseline. Of the 412 nonusers who had an initial BMD measurement, 295 (77% of survivors) had a follow-up DXA measurement, 31 died prior to the second measurement, 4 did not return, and 82 attended both SOF visits but did not undergo a second DXA.
All serum and urine samples were stored at −190°C until assayed, and biochemical measurements were performed without knowledge of bone mass or biochemical markers results.
Biochemical markers of bone formation
Serum total osteocalcin (OC) was measured with a human-specific immunoradiometric assay (IRMA) (ELSA-OSTEO; CIS BioInternational, Baglos/Ceze, France), which recognizes a large N-terminal midfragment in addition to the intact molecule.(17)
Serum bone-specific alkaline phosphatase (BALP) was measured with an IRMA using two monoclonal antibodies directed against the human bone isoenzyme and BALP purified from human SAOS-2 osteosarcoma cells as a standard (Ostase®; Hybritech, Inc., San Diego, CA, U.S.A.). This assay has a 16% cross-reactivity with the circulating liver isoenzyme.(18)
Biochemical markers of bone resorption
Urinary type I collagen cross-linked N-telopeptides (NTX) were measured with an ELISA (Osteomark®; Ostex International, Inc., Seattle, WA, U.S.A.) using a monoclonal antibody directed against the N-telopeptide–to-helix intermolecular cross-linking domain of type I collagen isolated from human urine.(19)
Urinary type I C-telopeptide breakdown products (CTX) were measured by an ELISA (CrossLaps™ ELISA; Osteometer Biotech A/S, Herlev, Denmark) based on an immobilized synthetic peptide with an amino acid sequence specific for a part of the C-telopeptide of the α-1 chain of type I collagen (CrossLaps antigen).(20,21)
Urinary free deoxypyridinoline (Dpyr) was measured by an ELISA that uses a monoclonal antibody with <1% cross-reactivity with free pyridinoline (Pyrilinks®-D; Metra Biosystems, Mountain View, CA, U.S.A.) and no significant interaction with cross-linked peptides.(22)
Urinary free pyridinolines (Pyr) were measured by an ELISA that uses a monoclonal antibody which reacts equally with free pyridinoline and free Dpyr (Pyrilinks®, Metra Biosystems) and has 2.5% cross-reactivity with cross-linked peptides.(23) All data obtained from urinary assays were corrected by the urinary creatinine concentration.
Bone mineral density
Bone mineral density (BMD) of the total hip and three subregions was measured by DXA at two visits, a mean of 3.8 years apart, using Hologic QDR 1000 bone densitometers. Details of these measurement methods and densitometry quality control procedures have been published elsewhere.(24,25) The mean coefficient of variation for the femoral neck between centers was 1.2% for two research staff who visited all centers.(24) Intrascanner coefficient of variation for the circulating femoral neck phantom ranged from 0.62% to 1.86% at the initial examination and 0.95% to 1.60% at the repeat examination.(25) The initial measurement was considered baseline for the purposes of this analysis.
All analyses were performed separately for the total hip as well as for the femoral neck and trochanteric and intertrochanteric subregions. We analyzed both the annual percentage change and annual absolute change in BMD, and the results were similar; therefore, we report only the percentage change results.
Prior to adjustment for potential confounders, we examined the relationship between marker level and change in BMD, and between age and change in BMD, with Pearson correlation coefficients. We used least squared means from linear regression models to calculate the annual change in BMD by quartiles of each urine and serum marker. Tests for linear trend in mean annual hip bone loss across quartiles of each marker were also performed. We compared the mean rate of change in BMD among women above and below the median for each marker, and compared those in the highest quartile of marker level to below the highest quartile. Bone loss models were examined with and without baseline BMD as a covariate. All regression models were adjusted for age.
To determine the sensitivity, specificity, positive predictive value, and negative predictive value of normal and elevated marker levels for rapid bone loss, we divided the cohort into tertiles of bone loss and calculated the ability of elevated baseline markers (above the median or the highest quartile) to predict the highest tertile of bone loss. The highest tertile of bone loss was chosen as there is no generally accepted clinical cutpoint that defines excessive bone loss in older untreated women.
The mean age of our participants was 73 years (range 67–89 years). Mean bone loss at the total hip was 0.6% per year. The correlations between rate of change in total hip BMD and rate of change in BMD of the femoral neck, trochanteric, and intertrochanteric subregions were 0.43, 0.84, and 0.94, respectively. Mean, median, and fourth quartile values for each marker are listed in Table 1.
Table Table 1.. Mean, Median, and Fourth Quartile Levels of Urine and Serum Markers
Baseline biochemical markers were weakly correlated with subsequent bone loss of the hip (Table 2). Correlations between baseline marker levels and rate of change in BMD ranged between –0.01 and 0.19. In general, correlations between marker levels and change in trochanteric and intertrochanteric BMD were similar to those for the total hip BMD, while associations between markers and femoral neck BMD were weaker than those for the other hip sites. By comparison, correlations between age and rates of change in BMD ranged from 0.16 for the femoral neck to 0.30 for the total hip (Table 2).
Table Table 2.. Correlation Coefficients of Age and Biochemical Markers of Bone Turnover with Annual Percent Change in BMD
After adjusting for age, the mean rate of bone loss from the total hip significantly increased across quartiles of each urine marker (Fig. 1). Trends were similar and remained significant after further adjustment for baseline BMD (data not shown). Conversely, although there was a trend for increasing the mean rate of bone loss with increasing concentration of serum markers, these relationships did not reach statistical significance (Fig. 2). Women who had baseline levels of NTX, CTX, Pyr, Dpyr, and OC that were above the median had significantly higher levels of bone loss (p < 0.05) than did women whose marker levels were below the median, even after adjusting for age. The age-adjusted difference in mean bone loss between women above and below the median level was not statistically significant for BALP (p = 0.42). The age-adjusted difference in bone loss rates between women with marker levels in the highest quartile compared with the lowest three was statistically significant for NTX only (p = 0.04).
To estimate the utility of markers for the prediction of hip bone loss in an individual woman, we compared the distribution of changes in total hip BMD among women with baseline marker levels above the median to the distribution of changes in BMD among women with marker levels below the median (Fig. 3). Plots examining the distribution of change in BMD among women above and below the upper quartile of baseline marker were similar. These figures indicate that the probability of an increase or decrease in total hip BMD are similar regardless of baseline marker level.
The sensitivity, specificity, positive predictive value, and negative predictive value of above median or fourth quartile levels of each marker for identifying women in the highest tertile of total hip bone loss (loss > 1.1%/year) are listed in Table 3. Using the median marker level as a cutpoint, the sensitivity for identifying women who subsequently lost more than 1.1% of total hip BMD per year varied from 56% for Dpyr to 72% for NTX. Thus, among those women with more than 1.1% bone loss per year, 56% had Dpyr levels above the median (6.8 nmol/mmol creatinine (Cr)) and 72% had NTX levels above the median (41.8 nmol BCE/mmol Cr). The corresponding sensitivities using the upper quartile of marker as a cutpoint ranged from 28% for Dpyr to 38% for both NTX and OC. Using the median marker level as a cutpoint, the specificity for identifying women who did not lose more than 1.1% per year ranged from 46% for BALP to 56% for Pyr. The corresponding specificity's using the upper quartile of marker levels as a cutpoint ranged from 74% for CTX and BALP to 79% for Pyr.
Table Table 3.. Sensitivity, Specificity, Positive Predictive Value (PPV), and Negative Predictive Value (NPV) of Marker Levels Above and Below the Median and Fourth Quartile Levels for Identification of Women Who Lost More Than 1.1% BMD per Year from the Total Hip (Highest Tertile)
From a clinical standpoint, the sensitivity and specificity of a test are less important than the positive predictive value, which describes the probability of excessive bone loss among women with elevated marker, and negative predictive value, which describes the probability of not having excessive bone loss among women with normal markers. With a pretest probability of 33% (excessive bone loss was defined as those in the upper tertile, >1.1% per year), the positive predictive value of elevated marker levels for excessive bone loss were between 35% and 42% when the median was used as a cutpoint and between 36% and 46% when the highest quartile was used as a cutpoint (Table 3). Thus, among those women with elevated marker levels using either the median or highest quartile cutpoint, less than half lost more than 1.1% per year of bone mass.
In this large cohort of untreated older women, we found that higher levels of CTX, Dpyr, NTX, Pyr, and OC were associated with greater average rates of total hip bone loss. However, the association was modest; mean bone loss differed by only 1–2% over 4 years in the highest compared with the lowest quartiles. Furthermore, there was substantial overlap in loss rates between those with high and low marker levels, and the positive predictive value of markers predicting the highest tertile of bone loss (>1.1% per year) was <50% regardless of the cutpoint.
Previous reports concerning the association of markers with bone loss have been inconsistent. Christiansen et al. measured forearm bone mineral content by single photon absorptiometry (SPA) every 3 months for 2 years in 178 early postmenopausal women. They found that a baseline measurement of urinary calcium, urinary hydroxyproline, and serum alkaline phosphatase identified 79% of “fast bone losers,” defined as >3% loss per year.(2,7) Uebelhart et al. measured forearm bone mineral content by SPA every 3 months for 2 years in 57 early postmenopausal women and found that baseline levels of OC, urinary hydroxyproline, Dpyr, and Pyr correlated with bone loss rates.(5) In another study, OC levels were found to correlate with bone loss at the forearm (measured by SPA) and lumbar spine (measured by dual photon absorptiometry [DPA]).(6) In a 3-year study of 6 female and 17 male subjects over age 65 years, Dresner-Pollak et al.(4) performed annual hip DXA measurements and measured urinary NTX and serum OC at the end of the study. They found that bone loss rates of the total hip, but not the femoral neck, were correlated with marker levels. In a large prospective study, McClung et al.(11) recently analyzed the correlation between baseline values of urinary NTX and serum OC values and change in DXA of the lumbar spine and proximal femur over 2 years in 1609 early menopausal women. They found no differences in the loss of BMD at either skeletal site by tertile of NTX or OC. Keen et al.(12) measured lumbar spine and femoral neck BMD using DPA annually for 4 years in 141 early postmenopausal women. They found no significant correlation between the rates of change in bone density with OC, urine Pyr, Dpyr, hydroxyproline, or calcium assessed by first generation assays. Cosman et al.(13) measured lumbar spine and hip BMD by DXA or DPA every 6 months for 3 years in a heterogeneous group of 81 women, and measured baseline serum OC, total alkaline phosphatase and BALP, carboxy-terminal propeptide of type I collagen (ICTP) and tartrate-resistant acid phosphatase (TRAP), and urine hydroxyproline, calcium, total Pyr, and total Dpyr. This study found that some markers correlated with bone loss but that the sensitivity of high marker levels for identifying rapid bone loss did not exceed 60% for any marker.(13)
Thus, previous studies have demonstrated inconsistent relationships between biochemical markers and subsequent bone loss. The differing results of these previous studies may be explained by several factors: the use of different marker assays (first generation vs. later generation); measurement of different sites of bone mass with different bone densitometry technology; difference in the age and menopausal status of the study populations; and difference in the length of follow-up and study size.
In our study, the correlations between marker levels and hip bone loss were low (r = −0.01 to 0.19). These low correlations may in part reflect the poor precision of marker measurements. Furthermore, the correlation between the rate of bone loss of the total hip and subregions of the hip was high except for the femoral neck. The lower correlation at the femoral neck may be at least partly due to the poorer precision of this measurement of BMD. Interestingly, we found no significant correlation between marker levels and bone loss at the femoral neck. This discordance in the relationship of marker levels to femoral neck bone loss vs. bone loss at other hip subregions has been previously reported.(4) The strengths of our study included its prospective design, large number of subjects with prolonged follow-up, and careful measurement of bone mass and biochemical markers. Nonetheless, our study had several limitations. Most markers of bone resorption, and to a lesser extent markers of bone formation, exhibit high day to day variability. We obtained only a single baseline urine and serum specimen for measurement of markers, and theoretically, precision may be increased by averaging the results of repeated baseline measurements. However, in most clinical settings only a single measurement of bone markers is feasible. In addition, we measured BMD only at the beginning and at the end of the study, and annual measurements may assess rates of bone loss more accurately.(26) However, it is likely that this effect was in part mitigated by our careful attention to participant positioning(25) and our relatively long follow-up (nearly 4 years).(26) We studied older ambulatory Caucasian women, and the results may not be generalizable to other populations, such as perimenopausal women or those treated with antiresorptive therapy. Lastly, our markers were measured in previously frozen sera and urine; however, we have found that concentrations of peptides and other biochemical substances are highly stable in sera that is stored at −190°C.
In conclusion, we found that higher levels of NTX, CTX, Pyr, Dpyr, and OC are associated with somewhat faster total hip bone loss in elderly women. However, the predictive value of these markers for bone loss in an individual woman is quite limited and therefore their clinical utility remains uncertain. Our data do not support the use of currently available markers to identify older women at risk of rapid bone loss. It is likely that future assays for markers of bone resorption and formation will have improved biologic and analytic precision and perhaps will better reflect bone turnover and bone loss in older women.
This work was supported by grants (1-R01-AG05407, 1-R01-AR35582, 5-R01-AG05394, 1-R01-AR35584, and 1-R01-AR35583) from the Public Health Service.