Relative Contributions of Bone Density, Bone Turnover, and Clinical Risk Factors to Long-Term Fracture Prediction


  • L Joseph Melton III MD,

    Corresponding author
    1. Division of Epidemiology, Mayo Clinic and Mayo Foundation, Rochester, Minnesota, USA
    2. Division of Endocrinology, Metabolism and Nutrition, Department of Internal Medicine, Mayo Clinic and Mayo Foundation, Rochester, Minnesota, USA
    • Division of Epidemiology, Department of Health Sciences Research, Mayo Clinic, 200 First Street Southwest, Rochester, MN 55905, USA
    Search for more papers by this author
  • Cynthia S Crowson,

    1. Division of Biostatistics, Mayo Clinic and Mayo Foundation, Rochester, Minnesota, USA
    Search for more papers by this author
  • W Michael O'Fallon,

    1. Division of Biostatistics, Mayo Clinic and Mayo Foundation, Rochester, Minnesota, USA
    Search for more papers by this author
  • Heinz W Wahner,

    1. Department of Diagnostic Radiology, Mayo Clinic and Mayo Foundation, Rochester, Minnesota, USA
    Search for more papers by this author
  • B Lawrence Riggs

    1. Division of Endocrinology, Metabolism and Nutrition, Department of Internal Medicine, Mayo Clinic and Mayo Foundation, Rochester, Minnesota, USA
    Search for more papers by this author

  • Presented at the 23rd Annual Meeting of the American Society for Bone and Mineral Research, Phoenix, AZ, October 14, 2001

  • The authors have no conflict of interest


Long-term fracture prediction using bone mineral density remains controversial, as does the additional contribution from assessing bone turnover or clinical risk factors. We measured bone mineral density at various sites, along with biochemical markers of bone turnover, sex steroid levels, and over 100 clinical variables, at baseline on an age-stratified sample of 304 Rochester, MN women in 1980. The 225 postmenopausal women were subsequently followed for 3146 person-years (median, 16.2 years per subject), wherein they experienced 302 new fractures: 81% resulted from minimal or moderate trauma and 60% of these involved the proximal femur, thoracic or lumbar vertebrae, or distal forearm. Accounting for multiple fractures per subject, these osteoporotic fractures together were best predicted by baseline femoral neck bone mineral density (age-adjusted hazard ratio [HR] per SD decrease, 1.37; 95% CI, 1.10–1.70); 19 moderate trauma forearm fractures were best predicted by distal radius bone mineral content, whereas 28 hip fractures and 100 vertebral fractures were best predicted by femoral neck bone mineral density. Femoral neck bone mineral density performed comparably in predicting osteoporotic fracture risk within the first decade of follow-up (HR, 1.38; 95% CI, 1.10–1.74) as well as more than 10 years after baseline (HR, 1.39; 95% CI, 1.05–1.84). The older biochemical markers were not associated with fractures, but serum “free” estradiol index was independently predictive of short- and long-term fracture risk. Consistent clinical risk factors were not identified, but statistical power was limited. Identifying patients at increased long-term risk of fracture is challenging, but it is reassuring that femoral neck bone mineral density can predict osteoporotic fractures up to 20 years later.


THE ABILITY OF bone density measurements made at peripheral sites like the radius or calcaneus to predict long-term fracture risk is well established, as is the ability of central measurements at the hip and spine to predict fractures over shorter periods of time.(1) Indeed, bone density measurements made at the mid- and distal radius by single-photon absorptiometry are related to fracture risk more than two decades later.(2, 3) However, some have questioned the generalizability of these results to other skeletal sites, arguing that intervening patterns of bone loss might confound long-term fracture risk prediction by bone mineral density (BMD) measurements made in the proximal femur. Thus, a correlation of 0.80 between baseline and 5-year follow-up measurements could fall to only 0.51 between baseline and 15-year values, with a corresponding degradation in the accuracy of risk assessment.(4) We have found, instead, that femoral neck bone loss is essentially linear over life; there is good tracking of individual values over time, and the correlation of baseline with femoral neck BMD values in individual patients 16 years later is actually 0.83.(5) Nonetheless, the reliability of femoral BMD measurements for fracture risk prediction beyond 10 years has yet to be established.(6, 7) This is an important limitation given that hip BMD is the focus of most clinical practice guidelines.(8, 9) Using one of the first densitometers capable of measuring bone density in the hip and spine,(10) we measured BMD at a variety of skeletal sites on an age-stratified sample of 304 Rochester, Minnesota women in the early 1980s. In addition, we made baseline measurements of the biochemical markers of bone turnover that were available at that time, as well as serum sex steroid levels, and we also assessed a large number of potential clinical risk factors. These women have now been followed for fracture outcomes for up to 20 years, and despite the limited number of subjects, it is possible to provide some insight into the ability of these different approaches to predict long-term fracture risk.


Subjects were recruited from an age-stratified random sample of Rochester, Minnesota women 30 years of age and over that was selected using the medical records linkage system of the Rochester Epidemiology Project.(11) Over one-half of the Rochester population is identified annually in this system and almost all are seen within any 3-year period. Thus, the enumerated population (those women seen in 1980 ± 1 year) approximated the underlying population of the community, including both free-living and institutionalized individuals. In an effort to enroll about 50 women in each decade age-stratum, 541 Rochester women were sampled, but 38 were ineligible for study (28 could not give informed consent, 4 were pregnant, and 6 died before they could be contacted). Of the remainder, 304 women (60%) consented to participate and provided evaluable study data. The response rate varied by age from 58% of those less than 50 years of age, to 70% of those 50–69 years of age, to 56% of those aged 70 years and older. The relatively low overall participation rate was caused mainly by the requirement that subjects agree to be studied at intervals after the initial assessment, although several refused for fear of radiation exposure. Only among women 70 years of age and older did poor health seem to be an important reason for nonparticipation.(12) The present analysis focused on the 225 women who were postmenopausal at baseline (age, 68.0 ± 13.6 years).

Bone mass was assessed at baseline at five different scanning sites in each subject. Areal BMD (g/cm2) was determined by dual photon absorptiometry of the cervical and intertrochanteric regions of the right proximal femur and the lumbar spine. Measurement precision error (CV) was 2.3% for the lumbar spine and 2.2% for the proximal femur.(6) Bone mineral content (BMC; g/cm) was measured at the midradius and distal radius (10% of radius length from the distal end) of the nondominant forearm with single-photon absorptiometry using a multiple path scanning approach with measurements averaged over five passes 1 mm apart; precision error was 1% for the midradius and 1.5% for the distal radius.

All subjects were interviewed in accordance with a standard protocol to collect clinical and lifestyle data, including information about cigarette smoking, alcohol consumption, and a history of participation in vigorous sports. With regard to medical history, there was generally good agreement between interview data and review of each subject's complete (inpatient and outpatient) medical record,(13) but where disagreements occurred, priority was given to the documented medical history; in the absence of documentation to the contrary, the patient's account was accepted. The interview and record review were completed independently of any knowledge of each subject's bone mass. Dietary intakes of calories, protein, fat, calcium, phosphorus, and vitamin D were estimated from a 7-day dietary record and converted to nutrient values using U.S. Department of Agriculture food composition tables.(14) At the time of interview, each subject also underwent anthropometric assessment, which included measurements of height and weight, as well as clinical assessments of balance, gait stability during a heel-to-toe walking test, and muscle strength. Obesity was assessed as body mass index (BMI; kg/m2; ≥30 kg/m2 inclusion criteria) on the day of interview.

Serum and urine calcium were determined by atomic absorption spectrophotometry (model 2380; Perkin Elmer, Norwalk, CT, USA). Serum phosphorus, creatinine, alkaline phosphatase, and urinary creatinine were determined by routine automated methods (Multistat III Plus; Instrumentation Laboratories, Lexington, MA, USA). Serum immunoreactive parathyroid hormone (iPTH) was measured by a modification of the method of Arnaud et al.(15) Serum osteocalcin (OC) was measured by radioimmunoassay.(16) Hydroxyproline was measured in a 24-h urine collection obtained while the patient ingested a low-gelatin diet.(17) Glomerular filtration rate (GFR) was calculated from creatinine clearance. “High” levels of the bone turnover markers (serum osteocalcin, alkaline phosphatase, and urinary hydroxyproline) were defined as those more than 2 SD above the mean for the 79 premenopausal women (mean age, 39.2 ± 5.3 years).(5) Serum estrone, estradiol, testosterone, androstenedione, and sex-hormone binding globulin (SHBG) were measured by previously described assay techniques.(18, 19) The ratio of sex steroids (e.g., estradiol) to SHBG levels was taken as an index of “free” sex steroid levels.

These women were then followed for the occurrence of any new fracture (prospective cohort study). Fractures were assessed by periodic interview and by review of each subject's medical record at every local medical care provider that attended her. Mayo Clinic records, for example, contain the details of every inpatient hospitalization at its two affiliated hospitals (Saint Marys and Rochester Methodist), every outpatient or office visit at the Clinic, emergency room, or local nursing homes, as well as all radiographic and pathology reports, including autopsies. Fractures were recorded whether they occurred before or after the date of the baseline examination, but only those that occurred afterwards (incident fractures) were considered as outcomes in this analysis. The medical records contained the clinical history and the radiologist's report of each fracture, but the original roentgenograms were not available for review. Consequently, the diagnosis of vertebral fracture was accepted based on a radiologist's report of compression, wedging, or collapse of one or more thoracic or lumbar vertebrae. We previously showed that clinically evident fractures typically represent “severe” vertebral deformities; 80% have reductions in vertebral height ratios (e.g., anterior/posterior) that are 4 SD or more below vertebra-specific means by morphometry.(20) Similarly, pathological fractures were those so characterized by the attending physicians. Because we reviewed all inpatient and outpatient records for each subject, ascertainment of clinically evident fractures is believed to be complete, including some fractures which were found incidentally on roentgenograms taken for another purpose.

The hazard ratio (HR) from the Andersen-Gill model was used to estimate the relative risk of fracture, accounting for multiple fractures per person.(21) This is an adaptation of the Cox proportional hazards model, wherein each subject is treated as a multievent counting process with essentially independent increments; the variance was adjusted for these multiple events per person. These hazard ratios represent the risk of fracture (whether it be first, second, third, etc.) in contrast to the standard Cox hazard ratios that represent the risk of a first fracture only. A stepwise model selection process was used to build multivariate models of the independent predictors. Because of the large number of potential predictors examined, they were grouped into sets of similar variables (e.g., BMD variables, bone turnover variables, etc.), and the stepwise process was applied to each set of variables. Independent predictors from each set were then examined together along with age to determine the final multivariate models. The cumulative incidence of new fracture (1 − survival-free-of-fracture) was projected for up to 20 years using product-limit life table methods.(22) Observed and expected cumulative incidence curves were compared using the one-sample log-rank test statistic.(23)


The 225 postmenopausal women included in this analysis ranged in age at baseline from 37 to 94 years (median, 68 years). Altogether, they were followed for 3146 person-years, or a median of 16.2 years per subject (mean, 14.0 years; range, 54 days to 20.6 years). During this period of observation, 126 women experienced 302 new fractures of various kinds (Table 1). Of these, 56% involved the proximal femur, thoracic or lumbar vertebrae, or distal forearm, the skeletal sites traditionally associated with osteoporosis. Eight-one percent of all new fractures resulted from minimal to moderate trauma (by convention, the equivalent of a fall from standing height or less), while 17% were caused by severe trauma (e.g., motor vehicle accidents and falls from heights); six fractures (2%) resulted from some specific underlying pathology, usually metastatic malignancy. Altogether, 110 women had one or more new fractures caused by minimal or moderate trauma: 24 women suffered 28 moderate trauma hip fractures, while 67 women experienced 100 vertebral fractures and 18 women experienced 19 distal forearm fractures caused by moderate trauma during this period of observation. The remainder of the analysis focused on the osteoporotic fractures, that is, fractures of the hip, spine, or distal forearm caused by minimal or moderate trauma.

Table Table 1.. New Fractures Among an Age-Stratified Sample of 225 Postmenopausal Rochester, Minnesota Women Followed for Up to 20 Years
original image

Table 2 compares the ability of bone density measurements made at different skeletal sites to predict specific incident fractures. As assessed by Andersen-Gill model analysis, and after adjusting for age, each 1 SD decline in lumbar spine BMD was associated with a 1.5-fold increase in the risk of a vertebral fracture caused by moderate trauma. Relative risks for spine fractures were less with the other densitometry measurements, but they were not significantly different one from another. Similarly, hip fracture risk was increased about 2-fold per 1 SD decrease in femoral neck or trochanter BMD. Confidence intervals around each estimate were fairly wide, however, and the different estimates were indistinguishable given the small number of moderate trauma hip fractures observed thus far, only 28. Each 1 SD decrease in BMC at the distal radius was associated with a 2.1-fold increase in the risk of a forearm fracture. In multivariate analyses including age and the various bone density measurements, femoral neck BMD was the best predictor of osteoporotic fractures generally (HR, 1.37; 95% CI, 1.10–1.70), as well as hip and vertebral fractures specifically, while distal radius BMC was the best predictor of distal forearm fractures. Femoral neck BMD was as good a predictor of the osteoporotic fractures that occurred 10 or more years after baseline (HR, 1.39; 95% CI, 1.05–1.84) as it was of those that occurred within the first 10 years of follow-up (HR, 1.38; 95% CI, 1.10–1.74). Because there was no age × femoral neck BMD interaction (p = 0.318), fracture predictability did not seem to vary by patient age at baseline.

Table Table 2.. Age-Adjusted Hazard Ratio (HR) of Osteoporotic Fractures per 1 SD Decrease in Baseline Bone Mineral Density (BMD) or Bone Mineral Content (BMC) at Five Measurement Sites Among an Age-Stratified Sample of 225 Postmenopausal Rochester, Minnesota Women Followed for Up to 20 Years
original image

Of the biochemical markers of bone turnover that were available when the baseline data collection was performed, a 1 SD rise in serum osteocalcin was associated with a modest 1.09-fold increase in osteoporotic fracture risk (95% CI, 0.91–1.31), which was not statistically significant, whereas a 1 SD increase in serum alkaline phosphatase was associated with a similar small increase (HR, 1.15; 95% CI, 0.92–1.44) and a comparable change in log urinary hydroxyproline with a negligible 1.03-fold increase in risk (95% CI, 0.85–1.24) in univariate analyses. After adjusting for age, vertebral fractures were negatively associated with serum osteocalcin and urinary hydroxyproline levels, but there was no relation between increased levels and any specific fracture type or all osteoporotic fractures combined (Table 3). There was no relation of fracture risk before or after 10 years of follow-up with either the markers themselves or “high” levels of bone turnover (>2 SD above the premenopausal mean) or the ratio of serum osteocalcin to urinary hydroxyproline or the osteocalcin Z-score minus the hydroxyproline Z-score as an “uncoupling” index (data not shown).

Table Table 3.. Age-Adjusted Hazard Ratio (HR) of Osteoporotic Fractures Per 1 SD Change in Baseline Biochemical Tests Among an Age-Stratified Sample of 225 Postmenopausal Rochester, Minnesota Women Followed for Up to 20 Years
original image

Among the other biochemical assessments of bone metabolism, a 1 SD decrease in 24-h urine calcium was associated with the risk of vertebral fractures (HR, 1.66; 95% CI, 1.21–2.30), but there were no other statistically significant associations (data not shown). Lower serum total sex steroid levels at baseline (log estradiol, log estrone, and log testosterone) did not predict overall osteoporotic fracture risk after adjusting for age, even when the lowest quartile was compared with the others. However, higher SHBG levels were associated with significant increases in vertebral fractures, proximal femur fractures, distal forearm fractures, and all osteoporotic fractures together (Table 3). Lower free estradiol index (estradiol/SHBG) and free testosterone index were associated with an increased risk of vertebral fractures, proximal femur fractures, and osteoporotic fractures generally, whereas free estrone index was associated with vertebral, distal forearm, and all osteoporotic fractures combined. In a full multivariate analysis (see below), free estradiol index was the best predictor of fracture risk. Each age-adjusted 1 SD decrease in free estradiol index was associated with vertebral fractures (HR, 1.48; 95% CI, 1.18–1.85), proximal femur fractures (HR, 1.55; 95% CI, 1.06–2.27), and osteoporotic fractures generally (HR, 1.46; 95% CI, 1.21–1.75), but not with distal forearm fractures (HR, 1.24; 95% CI, 0.86–1.78). Comparing the first and second decades of follow-up, both SHBG (HR, 1.36; 95% CI, 1.20–1.55 vs. 1.34; 95% CI, 1.12–1.61) and free estradiol index (HR, 1.38; 95% CI, 1.13–1.68 vs. 1.37; 95% CI, 1.10–1.69) predicted osteoporotic fracture risk generally after adjusting for age. There was no age-adjusted association with fracture risk of log androstenedione or of serum protein, total vitamin D, creatinine, or creatinine clearance (data not shown).

Subjects also underwent a series of anthropomorphic assessments at the baseline examination, including an evaluation of coordination, balance, and gait, and a subjective assessment of muscle strength of the lower extremity. In univariate analyses, increased height was associated with a reduction in the risk of vertebral fractures, and all osteoporotic fractures together, whereas an increase in weight was associated with a decreased risk for all fracture types. After adjustment for age, only the negative association of height with vertebral fractures remained significant (HR, 0.78; 95% CI, 0.63–0.96). Increasing weight and/or BMI was not protective for any fracture type after adjusting for age. Conversely, impaired coordination, balance, gait, and lower extremity muscle strength generally increased fracture risk in univariate analyses, but age-adjusted relative risks were statistically significantly increased only for decreased coordination and the risk of all fractures combined (HR, 2.40; 95% CI, 1.18–4.90).

Based on a 7-day diet record, each 1 SD decrease in dietary protein intake was associated with an increased risk of vertebral fractures (HR, 1.44; 95% CI, 1.15–1.82) and all osteoporotic fractures combined (HR, 1.30; 95% CI, 1.08–1.57) after adjusting for age. Protein intake from animal sources was highly correlated with total dietary protein (r = 0.92; p < 0.001) and was essentially interchangeable in the analysis. Likewise, dietary protein was strongly correlated with dietary phosphate intake (r = 0.90; p < 0.001), and the later variable performed similarly in the analyses. A decrease in total fat intake was linked to an increased risk of vertebral fractures (HR, 1.31; 95% CI, 1.09–1.57), as was a decrease in overall caloric intake (HR, 1.34; 95% CI, 1.09–1.65). Each 1 SD decrease in log dietary calcium intake was associated with an age-adjusted increase in all osteoporotic fractures combined (HR, 1.29; 95% CI, 1.06–1.56). Additional adjustment for weight did not change these results, and none of the other relationships was statistically significant.

A host of behavioral and comorbidity variables was also assessed. In the context of over 100 comparisons, only a few of these were associated with fracture risk after adjusting for age so some of them may have resulted from chance alone. There was no association of osteoporotic fracture risk with age at menarche, age at menopause, or history of bilateral oophorectomy, but gravidity was associated with an increased risk of proximal femur fractures (HR, 1.42; 95% CI, 1.04–1.93). There was no association with fracture risk of parity, age at first birth, or duration of breast-feeding. Use of oral contraceptives for at least 6 months was associated with a reduced risk of all osteoporotic fractures combined (HR, 0.38; 95% CI, 0.17–0.85), but there was no relation with hormone replacement therapy at baseline (HR, 1.10; 95% CI, 0.70–1.74). Likewise, we found no association of fracture risk with measures of physical activity or with a history of cigarette or alcohol use. In univariate analyses, a history of prior osteoporotic fracture at baseline was associated with subsequent osteoporotic fractures generally (HR, 1.91; 95% CI, 1.14–3.21) and with vertebral fractures specifically (HR, 2.39; 95% CI, 1.27–4.49), but neither relationship persisted after adjusting for age. Diseases linked with secondary osteoporosis were, in aggregate, associated with increases in the risk of vertebral (HR, 1.30; 95% CI, 0.82–2.05), proximal femur (HR, 1.34; 95% CI, 0.62–2.87), distal forearm (HR, 1.68; 95% CI, 0.67–4.19), and all osteoporotic fractures together (HR, 1.35; 95% CI, 0.91–1.99), but none of these differences was statistically significant. Likewise, there was no significant increase in fracture risk associated with any risk factor for falling. An overall historical risk factor score was positively associated with vertebral fractures (HR, 1.59; 95% CI, 1.32–1.92) and all osteoporotic fractures combined (HR, 1.35; 95% CI, 1.15–1.58), but not with proximal femur or distal forearm fractures alone. Individual scores ranged from 0 to 89.

Multivariate models combining the independent predictors from all of these risk factor domains were created for each fracture type (Table 4). The independent predictors of osteoporotic fractures generally were age, femoral neck BMD, free estradiol index, and dietary protein intake. Proximal femur fractures were associated with femoral neck BMD and free estradiol index, whereas the only independent predictor of distal forearm fractures was distal radius BMC. Vertebral fractures were associated with lumbar spine BMD, free estradiol index, dietary protein intake, and 24-h urine calcium levels. The independent predictors of any osteoporotic fracture in the first 10 years of follow-up were decreased femoral neck BMD (HR, 1.46; 95% CI, 1.19–1.80), free estradiol index (HR, 1.36; 95% CI, 1.13–1.64), and dietary protein intake (HR, 1.37; 95% CI, 1.12–1.68). For the osteoporotic fractures that occurred in the second decade, the independent predictors were lower femoral neck BMD (HR, 1.35; 95% CI, 1.02–1.79) and free estradiol index (HR, 1.34; 95% CI, 1.08–1.65).

Table Table 4.. Multivariate Predictors of Osteoporotic Fractures Among an Age-Stratified Sample of 225 Postmenopausal Women Followed For Up to 20 Years
original image


This study confirms previous work showing that radial BMC measurements are a strong predictor of overall fracture risk(1) and extends earlier studies(6, 7) by showing that BMD measurements at the hip can predict osteoporotic fractures beyond 10 years. Femoral neck BMD was the best predictor of overall osteoporotic fracture risk and was as good a predictor of the fractures that occurred after 10 years as those experienced earlier in follow-up. This would seem to contradict the expectation of Kanis et al.,(4) who assumed that individuals would lose bone at different rates, and therefore, the correlation between baseline and follow-up BMD measurements would degrade over time. However, we and others have found that bone loss from the femoral neck is essentially linear over life, with relatively little inflection at the menopause.(5,24–27) Moreover, we found little evidence for “fast” or “slow” bone losers at the hip, and the correlation between baseline and follow-up femoral neck BMD measurements in individual patients was still high even after 16 years.(5) On the other hand, Kanis et al.(4) estimated that the degradation in risk prediction would be modest if the gradient in fracture risk were small: With a relative risk of 1.5 per 1 SD change, baseline measurements would be expected to overestimate fracture risk 20 years later by just 12%. In the present study, overall fracture prediction did not seem to degrade over time, but the risk of osteoporotic fracture per SD change in femoral neck BMD was in this lower range.

The early biochemical markers of bone turnover that were available for this study did not predict osteoporotic fracture risk either in the first 10 years or the second 10 years of follow-up, whether assessed per 1 SD increase or as “high” values (more than 2 SD above the premenopausal mean) compared with the others or as an “uncoupling index” (the hydroxyproline Z-score − the osteocalcin Z-score). Bone density was a better predictor of fractures than were any of the biochemical markers of bone turnover, but the tests used in 1980 were inferior to those available today with respect both to sensitivity and to specificity.(28) This suggests that any problems lie with technical deficiencies of the older tests, although we also measured actual rates of femoral neck bone loss,(5) and the rate of bone loss over the first 4 years of follow-up failed to predict any of the osteoporotic fracture types either. This differs from earlier work indicating an increased risk of fracture in women with the greatest rates of bone loss.(29)

Serum sex steroid levels per se were not significantly associated with fracture risk, after adjusting for age, but increasing levels of SHBG were associated with fractures in both time periods, as were lower levels of free estradiol index. The association of free estrogen with fracture risk 10 or more years later extends the work of Cummings et al.,(30) who showed that low serum estradiol concentrations or high SHBG concentrations are associated with an increased risk of hip and spine fractures within six years of baseline. It should be noted, however, that ours was an older, less precise radioimmunoassay for estradiol, although only five subjects had unmeasurable values. Most,(30–35) but not all,(36, 37) recent work with more sensitive assays suggests that bone loss and fracture risk are greatest for women with the very lowest levels.

We were unable to identify consistent clinical risk factors for osteoporotic fractures, but our statistical power was quite limited. The only additional independent predictor of osteoporotic fractures in the multivariate model was low dietary protein intake, although protein intake was closely correlated with dietary phosphate intake, and the latter variable performed comparably when substituted into the model. A recent report from the Framingham Study demonstrated that low dietary protein is associated with increased bone loss from the hip and spine in elderly women,(38) although we did not find such a relation in a cross-sectional analysis of the smaller group of postmenopausal women studied here.(14) Investigators have found inconsistent results with respect to fracture risk, including reports that high protein consumption, particularly from animal sources, is associated with an increased risk of distal forearm(39) and hip fractures,(40) or conversely, that it is protective for hip fractures.(41) In this analysis, both animal and total protein intake had a similar relationship to fracture risk.

In conclusion, fracture prediction by femoral neck BMD did not degrade after 10 years, but the relative hazard for osteoporotic fractures was modest (1.37 per SD decrease). Older biochemical markers of bone turnover did not predict fracture risk, but low free estradiol index levels were associated both with short- and long-term fracture risk. A clinical risk factor score did not contribute to overall fracture risk prediction, although statistical power was limited. Predicting fractures over the long-term remains a challenge, but it is reassuring that low femoral neck BMD measurements at baseline are significantly associated with the risk of fracture up to 20 years later.


The authors thank Linda Richelson, Brenda Mickow, Veronica Gathje, Margaret Holets, Karen Nelson, Roberta Soderberg, and Susan Bonde for assistance with data collection, and Mary Roberts for help in preparing the manuscript. This work was supported in part by research grants AR-27065 and AR-30582 from the National Institutes of Health, U.S. Public Health Service.