An age-stratified sample of 304 women from Rochester, Minnesota, aged 30–94 years (median 60 years) at baseline underwent measurement of femoral neck bone mineral density (BMD) over a follow-up period extending to 16 years. The average rate of change in femoral neck BMD was −1.0% per year (range −10.0% to +13.4%) and did not vary significantly with age. Because there was no marked increase in the rate of loss around the time of menopause, nor convincing evidence of there being a subset of fast losers, there was fairly good tracking of individual values over time; the correlation of baseline with femoral neck BMD values 16 years later was 0.83. Although a large number of potential determinants was assessed, the only consistent predictor of femoral neck bone loss in women of different ages was baseline femoral neck BMD (r = −0.15; p = 0.023). Otherwise, different sets of risk factors were identified for premenopausal women, women within 20 years of menopause, and women 20 years or more postmenopausal, but the predictive power of these different multivariate models was modest. Nonetheless, these data indicate that femoral neck BMD is quite predictable for extended periods of time. This is reassuring with respect to the use of statistical models that incorporate such data to estimate future fracture risk.
Hip fractures are disproportionately responsible for the morbidity, mortality, and cost associated with osteoporosis.(1) Consequently, hip fracture prevention is an important societal goal in this country and abroad. Prospects for osteoporosis control are enhanced by the novel therapeutic approaches now under development, but enthusiasm must be tempered by realization that the potential cost of osteoporosis prophylaxis could be staggering,(2) because most of the population is at some degree of risk.(3) Accurate identification of particularly high-risk patients is key to cost-effective use of the growing armamentarium of potent osteoporosis treatments, but risk assessment has been hampered by the long delay between onset of bone loss and the occurrence of hip fractures many years later. This has necessitated the development of complex statistical models to predict future hip fracture risk.(4–6) Such models would be greatly improved by better prospective data about the determinants of long-term bone loss, particularly from the proximal femur, because bone density of the hip is the best guide to hip fracture risk.(7) Specifically, it is necessary to know whether hip bone density is predictable far into the future and whether there are subsets of fast and slow losers of bone from the hip as there are for radial bone loss among perimenopausal women.(8) In addition, it is important to learn whether the determinants of femoral bone loss are the same in women of different ages, because approaches to clinical management would probably differ for premenopausal compared with perimenopausal women or women long after the onset of menopause. Because we had access to the first densitometer capable of assessing bone mineral density (BMD) in the proximal femur, we were able to address these issues in a population-based study that has now spanned 16 years.
MATERIALS AND METHODS
An age-stratified random sample of women from Rochester, Minnesota, aged 30 years and over was selected by using the medical records linkage system of the Rochester Epidemiology Project.(9) More than half of the Rochester population is identified annually in this system, and almost all are seen in any 3-year period. Thus, the enumerated population (those women seen in 1980 ± 1 year) approximated the underlying population of the community. To enroll ∼50 women per decade of age, 541 Rochester women were sampled, but 38 were ineligible for study; of the eligible women, 304 (60%) participated. After providing written informed consent, these subjects were interviewed in accordance with a standard protocol to collect clinical, demographic, and lifestyle data. There was generally good agreement between interview data and a subsequent review of each subject's complete (inpatient and outpatient) medical record(10) but, where disagreements occurred, priority was given to documented medical history. Dietary intakes of calories, protein, fat, calcium, phosphorus, and vitamin D were estimated from a 7-day dietary record.(11) At the time of interview, each subject also underwent anthropometric assessment, which included measurements of height, weight, balance, gait stability during a heel-to-toe walking test, and muscle strength.
At baseline, bone mineral density (in grams per square centimeter) was determined by dual-photon absorptiometry (DPA) of the right proximal femur, using a prototype instrument.(12) Measurement precision (coefficient of variation [CV]) was 2.2%.(13) The hip measurements were repeated over the years but, perforce, different densitometers were introduced periodically. The prototype Mayo densitometer was used until completion of 8-year follow-up, when all available measurements were reanalyzed with software version 2. At the 8-year visit, patients were crossed over to a dual-energy X-ray absorptiometer (DXA) (QDR 1000, Hologic, Waltham, MA, U.S.A.). Subsequent measurements during the 12- and 16-year visits were made on a Hologic QDR 2000 machine calibrated for compatibility with the QDR 1000 values. Precision for the QDR 2000 was 1.8% for the proximal femur measurements. Although it was possible to adjust for these changes using phantom measurements and the crossover data from patients, there were differences between DPA and DXA in the regions of interest that were assessed, and the only site that remained consistent was the neck of the femur. Therefore, this analysis is limited to an examination of bone loss from the neck of the femur.
Fasting serum and 24-h urine specimens were collected from all subjects. Serum and urine calcium levels were determined by atomic absorption spectrophotometry (model 2380, Perkin Elmer, Norwalk, CT, U.S.A.). Serum phosphorus, creatinine, alkaline phosphatase, and urinary creatinine levels were determined by routine automated methods (Multistat III Plus, Instrumentation Laboratories, Lexington, MA, U.S.A.). Vitamin D metabolite levels were measured by the method of Eisman et al.(14) Serum levels of immunoreactive parathyroid hormone (iPTH) were measured by a modification of the method of Arnaud et al.(15) Serum osteocalcin (OC) concentration was measured by radioimmunoassay.(16) Hydroxyproline concentration 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. Serum estrone, estradiol, androstenedione, and sex hormone–binding globulin (SHBG) levels were measured by previously described assay techniques,(18,19) as was dehydroepiandrosterone (DHEA) concentration.(20) For premenopausal women, blood samples were obtained during the first 2 weeks of the menstrual cycle.
Adjustments between different machines and software were made by using patient crossover data, the Mayo prototype calibration data and the QDR quality-control data. Once adjustments had been made to the BMD measurements, a rate of bone loss was determined for each person. A regression equation was fit for each individual, fitting femoral neck BMD against time from baseline, and the rate of change was defined as the slope divided by the intercept (expressed as a percentage). Rates of change over comparable ages as assessed by DPA were compared with those as assessed by DXA using the two-sample t-test. Pearson correlations were used to assess univariate relationships between the rate of change and the various predictors. When appropriate, the predictors were transformed with a log transformation. Stepwise linear regression was used to select the predictors for the final multivariate models. Higher ordered terms, interactions and model assumptions were checked on the final models. S-Plus (Math Soft, Seattle, WA, U.S.A.) loess smoother functions were used in the plots to visually demonstrate relationships between the variables.
The 304 women studied ranged in age at baseline from 30 to 94 years (median 60 years). Altogether, they were followed up for 2587 person-years, or a median of 7.9 years per subject (mean 8.5 years; range 0–16.6 years). During this long period of observation, 99 women (33%) died. Because of death, declining health, etc., only 54% (58/108) of the women who at baseline were more than 20 years past menopause underwent follow-up measurements compared with 82% (96/117) of women within 20 years of menopause and 95% (75/79) of the women who were premenopausal at baseline. Changes in femoral neck bone density with aging were consistent across these three groups of women, as shown in Fig. 1, and the overall fitted curve was linear (femoral neck BMD 1.1511 − 0.0069 ∗ age; p < 0.001). Among those surviving to each assessment, 204 (70%) participated in the 4-year examination, 173 (66%) in the 8-year, 133 (57%) in the 12-year, and 118 (56%) in the 16-year assessment. The correlation (r) of baseline with subsequent femoral neck BMD, an important parameter in fracture prediction models, was 0.83 over an average (mean ± SD) of 11.3 ± 4.5 years, counting only subjects with at least one follow-up measurement. When restricted to women with the corresponding measurements, the correlation of baseline with 4-year BMD values was 0.87, baseline with 8-year values was 0.85, baseline with 12-year values was 0.82, and baseline with 16-year values was 0.83. Among those with 16-year data, 56% stayed within the same quintile ranking over the whole period of observation, and 91% remained within one quintile at 16 years compared with their quintile ranking at baseline.
Data from the 229 subjects with at least one follow-up measurement were used to estimate femoral neck bone loss as the percent change per year from baseline. Estimated rates of bone loss (percentage per year) for individual subjects are shown in Fig. 2, plotted by age at the baseline examination. The average rate of change was −1.0% per year (range −10.0% to +13.4%) and did not differ significantly over life depending on baseline age (p = 0.359). Importantly, there was no significant difference in slopes, over comparable ages, of the rate of bone loss estimated in the first 8 years by DPA and the second 8 years by DXA, although the latter had less variability. In addition, the change in femoral neck BMD over life was similar as judged from these longitudinal data and as estimated from the cross-sectional data at baseline (Fig. 3).
Overall, the mean rate of longitudinal bone loss was somewhat greater for the women within 20 years of menopause (−1.2% per year) compared with the premenopausal women (−1.1% per year) or the women 20 or more years past the menopause (−0.6% per year), but these differences were not statistically significant (p = 0.515). The mean rate for the 119 postmenopausal women with no history of hormone replacement therapy was −1.2% per year compared with only −0.3% per year for the 35 women who had been on hormone replacement therapy for at least 6 months (p = 0.197). The latter group included 3 osteoporotic women with the highest rates of bone gain (+13.4%, +8.3% and +6.9% per year, respectively) that were observed during the course of the study. Excluding all women with a history of hormone replacement therapy at baseline, 59% of the postmenopausal women were slow bone losers (<−1.5% per year), whereas 33% were intermediate and only 8% were fast losers (>−3.0% per year). The proportion of fast losers was 4% for those within 20 years of menopause compared with 14% for the women who had experienced menopause more than 20 years previously. This is partly an artifact of the smaller average number of measurements per subject in the latter group (median 2 scans) compared with the former (median 4 scans), resulting in less precise estimates of the rate of bone loss and more scatter in the data as can be seen in Fig. 2.
More than 170 different variables were assessed with respect to their relationship with the rate of bone loss from the femoral neck. Those that showed a substantial association in either univariate or multivariate analyses are listed in Table 1. This analysis excluded the 3 osteoporotic women undergoing hormone replacement therapy who had many risk factors but anomalously low bone loss. As anticipated from Fig. 2, there was no overall association of bone loss with baseline age (median 60 years; r = −0.04; p = 0.574), or age at menopause (median 48 years; r = −0.07; p = 0.393). Higher baseline femoral neck BMD was associated with a greater rate of bone loss overall (median 0.76 g/cm2; r = −0.15; p = 0.023), particularly among the younger women (Table 1). Baseline bone density measured at the lumbar spine, radius, or femur trochanter did not predict the rate of femoral neck bone loss. Bone loss was greater in those with higher levels of bone turnover as assessed by serum levels of alkaline phosphatase (median 25 U/L; r = −0.15; p = 0.028) and, among the oldest women, serum levels of osteocalcin (median 6.9 ng/mL; r = −0.41; p = 0.002). There was no association with higher bone turnover as assessed by urinary hydroxyproline concentration (median 26.5 μg/100 ml GFR; r = −0.06; p = −0.396), and prediction was not improved by use of a “uncoupling index” (hydroxyproline Z score minus osteocalcin Z score). Other biochemical parameters, including serum calcium, phosphorus, parathyroid hormone, alkaline phosphatase, and protein levels and total vitamin D and urinary calcium levels, were not associated with the rate of femoral neck bone loss. Higher levels of dehydroepiandrosterone were associated with a reduced rate of bone loss in premenopausal (median 3.9 ng/mL; r = 0.31; p = 0.030) but not postmenopausal women (median 1.8 ng/mL; r = −0.09; p = 0.261). None of the other sex steroids tested was associated with bone loss nor was sex hormone binding globulin. There was no association in any age group or overall between the rate of femoral neck bone loss and biologically available (ratio with SHBG) estrone (median 4.8; r = 0.03; p = 0.656), estradiol (median 2.6; r = 0.07; p = 0.302), androstenedione (median 74.6; r = −0.01; p = 0.885), DHEA (median 0.36; r = 0.02; p = 0.799) or testosterone (median 30.3; r = −0.03; p = 0.649) and this was so even when the analysis was restricted to women not on hormone replacement therapy. In particular, the rate of bone loss was not statistically significantly greater among the postmenopausal women with estradiol levels in the lowest quartile compared with those with higher levels (−1.3% per year versus −0.9% per year; p = 0.891). However, hormone replacement therapy was associated with reduced bone loss in the oldest women (r = 0.33, p = 0.010).
Table Table 1. Predictors of Femoral Neck Bone Loss in Univariate Analyses Among Women from Rochester, Minnesota, Who at Baseline Were Premenopausal, Within 20 Years of Menopause, or More than 20 Years Postmenopausal
There was no overall association of the rate of bone loss with various lifestyle factors, including cigarette smoking in 126 women (41.6%) or consumption of alcohol in 72.0% or caffeinated beverages in 21.4%. Likewise, there was no overall association between femoral neck bone loss and reproductive variables including marital status (53.6% ever married), age at menarche (median 13 years), gravidity (median 3), parity (median 2), history of miscarriage (27.3%) or stillbirth (3.0%) or infertility (3.0%), duration of breastfeeding (median 1 month) or use of oral contraceptives (31.3%). There was significantly less bone loss among premenopausal women with a history of infertility (r = 0.51; p < 0.001), but this was accounted for by one outlier. There was also more bone loss (r = −0.21; p = 0.050) among the early-postmenopausal women with a later age at first delivery (median 24 years; range 18–40 years). Although there was no overall correlation with measures of height (height, arm span) or obesity (weight, relative weight, body mass index), there was an indication that greater height (median 1.6 m) was associated with reduced bone loss among the early-postmenopausal women and, in the same group, that increased skin transparency (9.4%) was associated with greater bone loss. However, the latter was also negatively correlated with height (r = −0.31; p < 0.001). Among the women more than 20 years postmenopausal, there was less bone loss among those with greater skin-fold thickness (median 1.6 mm) measured on the back of the wrist with a caliper. In none of the groups was there any influence of dietary calcium, phosphorus, protein, fat or total calories. Total vitamin D intake (median 125 IU) was inversely correlated with bone loss among the late postmenopausal women (r = −0.40; p = 0.002), but this was due to higher rates of loss among a small number of women with intakes of more than 400 IU per day (−2.8% per year) or more than 800 IU per day (−5.3% per year). There was no significant correlation among the women with vitamin D intakes of 400 IU per day or less. Most of the subjects in this study were reasonably active, and there was no association of bone loss with limited activity or a subjective assessment of impaired muscle strength. However, femoral bone loss was greater among the subgroup of women with impaired coordination on the heel to knee test (r = −0.29; p < 0.001), abnormal Romberg test (r = −0.17; p = 0.012) or impaired gait (r = −0.19; p = 0.004), but these abnormalities together affected only 23 women.
A score for the number of conditions that have been associated with secondary osteoporosis that were present (median 0) did not predict the rate of bone loss overall, but there was greater bone loss with increasing comorbid conditions among the women more than 20 years postmenopausal (r = −0.30; p = 0.027). Individual conditions associated with a statistically significant decrease in bone loss included a history of urolithiasis in 2 early-postmenopausal women (r = 0.32; p < 0.001) and a history of cholecystectomy in 7 premenopausal women (r = 0.42; p < 0.001), whereas bone loss appeared to be greater for 2 premenopausal women with arrhythmia (r = −0.25; p = 0.031) and one with hypotension (r = −0.26; p = 0.023); all of these associations affected small numbers of subjects, appeared in the context of multiple comparisons, and may not be meaningful. There was also greater femoral neck bone loss among 5 women with a history of hyperthyroidism (r = −0.15; p = 0.028) that was not accounted for by thyroidectomy or subsequent thyroid hormone–replacement therapy. There was no significant association with corticosteroid use, but only 8 women were exposed. There was also no association of bone loss with a history of osteoporotic fractures (hip, spine or wrist) in 12 subjects or a family history of osteoporosis (mother, father, sister, or brother) in 69 women.
In a multivariate model, the only independent predictors of the rate of femoral neck bone loss among the premenopausal women were greater baseline femoral neck BMD (p = 0.040) and DHEA (p = 0.049), as shown in Table 2. However, this model accounted for only 17% of the variance in premenopausal bone loss (R2 = 0.170). Among the women within 20 years of menopause, the independent predictors of the rate of femoral neck bone loss were greater baseline BMD (p < 0.001), impaired coordination (p < 0.001), impaired gait (p < 0.001), age at first delivery (p < 0.001) and height (p = 0.004). The model R2 was 0.571. Finally, among the women who were more than 20 years beyond menopause at baseline, the independent predictors of femoral neck bone loss were greater baseline BMD (p = 0.004), elevated levels of osteocalcin (p = 0.005) and greater dietary vitamin D intake (p < 0.001). Hormone replacement therapy for more than 6 months was associated with reduced bone loss (p = 0.006), as also shown in Table 2. This model appeared to account for 45% of the variance in the rate of bone loss from the femoral neck in the oldest age group (R2 = 0.448).
Table Table 2. Independent Predictors of Femoral Neck Bone Loss in Multivariate Analyses Among Women From Rochester, Minnesota, Who at Baseline Were Premenopausal, Within 20 Years of Menopause, or More than 20 Years Postmenopausal
The present prospective investigation confirms cross-sectional studies that have indicated that bone loss from the femoral neck is approximately linear across life in white women. Indeed, the first such report, a cross-sectional analysis based on a small number of Rochester women, demonstrated this pattern,(13) which has since been confirmed by the Third National Health and Nutrition Survey in a study of 1541 non-Hispanic white women.(21) The rate of bone loss across life from our longitudinal data compared well with a cross-sectional estimate using the baseline DPA measurements from this study. The cross-sectional estimate from the Third National Health and Nutrition Survey, based on DXA measurements of the femoral neck, is also shown in Fig. 3 and reveals somewhat lower bone density values among younger women and a slightly less steep rate of loss thereafter. Over shorter periods of time, some(22–25) but not others(26,27) have shown an apparent acceleration of bone loss with age. We might have underestimated the rate of bone loss among elderly women if those with the greatest rates of bone loss are more likely to die,(28,29) but this would affect all prospective studies. Discrepancies probably relate more to differences in statistical power, as we saw a great deal of scatter in results among the older women that were based mainly on a comparison of only two scans. This was also evident in other reports(22,23,26) and contrasts with the greater precision of our estimated rates of bone loss among younger women who underwent four scans on average.
The rate of femoral neck bone loss in this study was −1.0% per year overall and −1.2% per year among women aged 65 years or over. In comparison, the mean rate of femoral neck bone loss was −0.5% per year among 5698 elderly women in the Study of Osteoporotic Fractures who were measured twice by DXA an average of 3.6 years apart.(24) Among 385 elderly Australian women in the Dubbo Study who were measured twice by DXA over an average of 2.5 years, the mean rate of change in femoral neck BMD was −1.0% per year.(22) Femoral neck bone loss averaged −0.6% per year among 2452 elderly Dutch women in the Rotterdam Study who were followed up for 2 years.(25) Among 85 elderly women who underwent DXA measurement 1 year apart, femoral neck bone loss was −0.8% per year, which was not statistically significant.(23) Likewise, femoral neck bone loss averaged −0.2% per year among 288 postmenopausal women followed up for 2 years by DPA; this was not significant either, but half of the women were taking calcium supplements.(26) Another group of postmenopausal women with DXA measurements had femoral neck bone loss that averaged −1.0% per year.(27) Studies among pre- and perimenopausal women have found rates of femoral neck bone loss ranging from −0.3% to −1.3% per year(30–32) compared with our estimate of −1.1% per year for premenopausal women.
Because proximal femur BMD is the strongest predictor of hip fracture risk,(7) it is the assessment favored for clinical decision making about osteoporosis prevention and treatment. Consequently, femoral BMD has been factored into many of the models used to predict treatment cost-effectiveness(4,6,33,34) which, in turn, provides a foundation for evidence-based therapeutic recommendations.(34) Such models generally incorporate parameters describing the rate of bone loss with aging (or else the correlation of BMD values at different ages) and the risk of fracture at different levels of BMD. A large amount of prospective data has documented a very consistent relationship between femoral neck BMD and fracture risk,(7) but less is known about the long-term predictability of femoral BMD. Despite the obvious potential for errors introduced by changing densitometers over time, the long-term correlation of baseline with 16-year femoral neck BMD was 0.83 in this study. This is quite close to correlation coefficients used in the model developed for the Office of Technology Assessment(35) and subsequently used in the National Osteoporosis Foundation's cost-effectiveness analysis,(34) e.g., a 15-year correlation of 0.77 between ages 50 and 65 and 0.86 between ages 65 and 80 years.
With the high correlation between BMD values measured at various intervals, there was fairly good tracking of individual values of femoral neck BMD: more than 90% of subjects were within one quintile of their starting rank as assessed 16 years earlier. Both the tracking and the high correlations are due partly to the fact that bone loss from the femoral neck was generally linear over life. As in other prospective studies, where the rate of early-postmenopausal femoral bone loss also averaged about −1.0% per year,(30–32) there was not the dramatic acceleration of femoral neck bone loss at the menopause generally reported from other skeletal sites. This could be due partly to inaccurate dating of the menopause, as we saw no correlation with years since onset of menopause, but it is also evident that patterns of bone loss at the forearm and spine differ considerably from that at the hip.(36) Thus, both cross-sectional(21) and longitudinal(37) studies indicate that femoral bone loss begins at an earlier age. Also, we found no strong association of bone loss with bioavailable estrogen levels, although this may reflect the limitations of these older assays which had limited sensitivity for the low levels found to be predictive of bone loss from the total hip in the Study of Osteoporotic Fractures.(38) We also may have had inadequate statistical power to detect the small effects that were observed in that study, i.e., a 0.2% reduction in the rate of femoral bone loss for every standard deviation increase in serum total estradiol level. However, as in the Study of Osteoporotic Fractures,(24) we did find a positive influence on femoral neck bone loss of hormone replacement therapy in the oldest age group.
Other than baseline femoral neck BMD, there were no other consistent predictors of bone loss across women in the three age groups. The influence of higher baseline BMD is related in part to peak bone mass, a major determinant of subsequent bone density,(39) but may also represent some degree of regression to the mean.(40) Thus, the greater the baseline BMD, the greater the loss, both in absolute and relative (percentage) terms. With respect to other influences on femoral neck bone loss, our study had limited statistical power for identifying weak risk factors or disentangling highly correlated ones, but it seems clear that some risk factors which were quite important in one age group had no effect whatsoever in another. Other risk factors had a strong effect on affected individuals but were uncommon in the population generally. These observations reflect the fact that osteoporosis is a “Gompertzian” condition, like osteoarthritis and other common chronic diseases.(3) Almost everyone in the population is at risk of these disorders, which lack a single “cause”; instead, their promotion is influenced by numerous risk factors.(41) This helps explain why it has been difficult to delineate an unequivocal set of risk factors for osteoporosis. In the Dubbo longitudinal study, for example, independent predictors of femoral neck bone loss included age, baseline BMD, weight, weight change, and physical activity but not hormone replacement therapy or cigarette smoking.(42) By contrast, in the Rotterdam Study, femoral neck bone loss among women was associated with cigarette smoking and lower body mass index but not with disability, calcium intake, alcohol consumption, thiazide or corticosteroid use, or hormone replacement therapy.(25) On the other hand, hormone replacement therapy was associated with significantly lower femoral neck bone loss in the Study of Osteoporotic Fractures as noted above,(24) and dietary calcium intake was an independent determinant among the women in Dubbo,(22) but body mass index was not an independent predictor of femoral neck bone loss in the Boston study.(23)
Similarly, some studies indicate that biochemical markers of bone turnover can predict femoral bone loss,(43,44) but markers did not add much to the prediction of bone loss in this investigation. This may reflect the fact that we found no subset of fast losers of femoral bone at the menopause as seen for perimenopausal bone loss from the radius.(8) Although our statistical power to exclude a small subset of fast losers was limited, other longitudinal studies of young(29) as well as older women(22) also did not find evidence for a specific subpopulation of fast losers of femoral neck BMD. In addition, the baseline studies for the present cohort were performed in the early 1980s. Consequently, it was necessary to rely on total serum alkaline phosphatase and osteocalcin as markers of bone formation and on urinary hydroxyproline as an indicator of bone resorption. Urinary hydroxyproline is not a specific marker for bone resorption,(45) and newer markers may be better for this purpose.(46) For example, the correlation with total hip BMD was twice as great for urinary cross-linked N-telopeptides of type I collagen as for serum osteocalcin in a more recent study among Rochester women.(47) Additional studies will be needed to quantify the improvement in bone loss prediction that might be gained through the use of modern markers. Moreover, it remains the task of the large epidemiologic studies that are now underway to resolve a set of variables that can optimize bone loss estimates. However, our data are reassuring insofar as femoral neck bone loss is relatively uniform over life and can be used for fracture prediction in the models available today.
We thank Linda Richelson, Brenda Mickow, Vicki Gathje, and Margaret Holets for assistance with data collection, Cindy Crowson for help in data analysis, and Mary Roberts for aid in preparing the manuscript. Supported in part by research grants AR 27065 and AR 30582 from the National Institutes of Health, U.S. Public Health Service.