The mechanism by which thiazide diuretics might protect against hip fracture is not known. Thiazide diuretics may decrease fracture risk by preserving bone mass [10, 11, 20-23]. Cross-sectional studies have shown that users of thiazide diuretics have slightly higher cortical and trabecular bone mass than nonusers. Two randomized clinical studies of the effect of thiazide diuretics on bone loss had conflicting results [24, 25]. If thiazide diuretics reduce the risk for hip fracture by slowing bone loss, then these agents should also protect against other types of fracture that are associated with reduced bone mass [2]. On the other hand, thiazide diuretics could increase the risk for fractures by increasing the risk for falls [26]. Thiazide diuretics might do this by causing dizziness or postural hypotension [27], although the association of diuretic use with falling is not consistent [28]. More than 90% of hip fractures are related to falls [29].

No previous study has included a comprehensive analysis of the effects of thiazide diuretics on bone mass, the risk for falls, and the risk for all fractures in a single population. In our prospective study of 9704 women who were 65 years or older (the Study of Osteoporotic Fractures), we assessed thiazide diuretic use and bone mass at baseline and conducted follow-up examinations every 4 months for the occurrence of fractures and falls.

Methods

Top Methods Results Discussion References

Study Sample

From September 1986 through October 1988, women who were at least 65 years old were recruited for the Study of Osteoporotic Fractures in four areas of the United States: Portland, Oregon; Minneapolis, Minnesota; Baltimore County, Maryland; and the Monongahela Valley region near Pittsburgh, Pennsylvania. Age-eligible women were recruited through several sources: health maintenance organizations; lists of residents that had been compiled for other epidemiologic studies; jury-selection lists; voter-registration lists; and lists of driver's license holders [30]. We excluded black women because of their lower incidence of fractures, women who were unable to walk without the assistance of another person, and women who had bilateral hip replacements.

Measurement of Bone Mass

Bone mineral content (g/cm) and density (g/cm²) were measured using single-photon absorptiometry (OsteoAnalyzer, Siemens-Osteon, Wahiawa, Hawaii). Details of these methods have been reported elsewhere [30]. We scanned three sites: the distal radius, the proximal radius, and the calcaneus. The distal radius is composed of about 60% cortical bone and 40% trabecular bone, the proximal radius is about 99% cortical bone [31], and the calcaneus is about 97% trabecular bone [32]. The reproducibility of the bone mass measurements, expressed as coefficients of variation, ranged from 1.3% to 2.0% among individuals and from 0.4% to 1.2% among scanners [30].

Diuretic Use

Information on diuretic use was obtained at the baseline clinic interview. Participants were asked to bring all current medications to the clinic for verification. In addition, pictures of tablets were presented to participants to assist them in the recollection of previously prescribed diuretics. Diuretics, including combination drugs, were classified as "thiazide" diuretics according to the American Medical Association drug evaluations [33]. Separate histories were obtained from participants for use of thiazide and nonthiazide diuretics. Chlorthalidone was classified as a thiazide diuretic because its effects on calcium excretion are similar to those of thiazide diuretics. Analyses in which chlorthalidone was not considered as a thiazide diuretic yielded similar results. Women were asked whether they had ever taken diuretics; at what age they had started taking diuretics; whether they were currently using diuretics and, if not, when they had stopped; whether they took diuretics the entire time or stopped for a few years; and the name of the medication they were currently using. Duration of diuretic use was calculated as the total number of years that they reported using diuretics. No information was collected on dose at the baseline visit, but dose-related data were recorded at the first annual telephone interview. Information on duration of thiazide diuretic use was missing in 15 women. These women were excluded from all analyses.

Other Measurements

Reported health status, type of menopause, alcohol consumption, and cigarette smoking were assessed using a questionnaire that was reviewed with the participant by a trained interviewer. Women were considered to have had a surgical menopause if they reported having had a bilateral oophorectomy at the age at which they stopped menstruating. Regarding alcohol consumption, participants were asked how much they had consumed during the past 30 days and how many times they had consumed three or more drinks per day and five or more drinks per day during the past 30 days. The measure of alcohol consumption used in our study was drinks per week, which was adjusted for atypical drinking, particularly heavy drinking over the last 30 days. Dietary calcium was assessed by a food frequency questionnaire and an interview using standardized food models to estimate portion sizes [34]. In our study, total calcium intake included dietary and supplemental calcium.

Women were also asked if they had ever taken the following medications: thyroid hormones, medicine for seizures, vitamin D or multivitamins containing vitamin D, TUMS, calcium supplements, antacids, prednisone, cortisone, other steroid pills, birth control pills, estrogen pills, estrogen skin patches, estrogen vaginal cream or suppositories, estrogen injections, and progestins. Information was obtained on current use, duration of use, and specific drug and dosage used for the longest period.

Functional status was measured by a respondent's answers to questions about her ability to perform six instrumental activities of daily living by herself and without using special aids or equipment. These activities were as follows: walking two or three blocks outside on level ground; climbing up 10 steps without stopping; walking down 10 steps; preparing one's own meals; doing heavy housework; and doing one's own shopping for groceries or clothes. The measure used in these analyses is the total number of activities (ranging from 0 to 6) that a woman reported she was unable to do or had difficulty doing. Additionally, women were asked to rate their health status as very good, good, fair, poor, or very poor.

During the clinic examination, body weight was measured (after removal of shoes and heavy outer clothing) using a balance beam scale. Height was measured (after removal of shoes) using a Harpenden stadiometer (Holtain, Ltd., Dyved, United Kingdom). Height and weight were used to calculate the body mass index (kg/m²).

Ascertainment of Prevalent Vertebral Deformities

The prevalence of vertebral deformities as shown by lateral thoracic and lumbar spine films was obtained on 2992 women (31% of the cohort) who were between 65 and 70 years old. The methods used to identify vertebral deformities have been described previously [35]. In summary, six points for each vertebral body, T4 through L4, were marked with a translucent digitizer. We calculated three heights (anterior, posterior, and middle) and three ratios: anterior-to-posterior height (wedge); mid-to-posterior height (endplate); and the posterior height of each vertebra to the posterior height of the adjacent vertebrae (crush). We classified a vertebral body as "deformed" if any one of the three ratios was at least 3 standard deviations below the normal mean value for that particular vertebral body [36].

Ascertainment of Incident Fractures and Falls

Details of our method for identifying new fractures during follow-up have been previously reported [37]. Briefly, we contacted participants every 4 months by postcard or telephone to ask if they had sustained a fracture or fall. More than 99.5% of these follow-up contacts were completed. We asked participants how the fracture occurred and obtained a copy of the radiographic report, which had to specifically mention the occurrence of an acute fracture. Hip fractures were also confirmed by reviewing copies of radiographs. We excluded fractures that occurred because of major trauma, such as that experienced in a motor vehicle accident. Self-reported vertebral fractures were not included because self-reporting is an insensitive and unreliable way to ascertain incident vertebral fractures. If a participant died, we interviewed her personal physician or a close contact, or both, to determine whether she had a fracture since her last contact with the clinical center.

Fractures of the distal radius or ulna were considered wrist fractures. Proximal humerus fractures were defined as those that occurred in the proximal one third of the humerus. Hip fractures were defined as those of the proximal femur. Fractures associated with low bone mass were considered "osteoporotic fractures" and included fractures of the hip, wrist, humerus, pelvis, foot, toe, leg, hand, clavicle, and rib [2]. All nonspinal fractures included "osteoporotic" fractures, as well as fractures of the ankle, elbow, finger, and face. Duration of fracture follow-up was calculated as time to first occurrence of a fracture. Follow-up for fractures ranged from 0.02 years to 4.71 years (mean follow-up, 3.32 years). All fractures that occurred up to June 1991 were included in the analysis. For women who died during the follow-up period, the date of death was used to mark the end of follow-up when fracture follow-up was not appropriate.

A fall was defined as "falling all the way down to the floor or ground". Follow-up for falls was limited to 1 year after baseline to link the exposure more closely with the event. We used two outcome variables: "1 or more falls" and "2 or more falls".

Statistical Analysis

Thiazide diuretic use was classified as never, past, and current; current users were also grouped according to duration of use. Chi-square tests of homogeneity and analyses of variance were used to compare groups by age, weight, height, obesity, reported general health status, functional status, estrogen use, smoking, alcohol consumption, and calcium intake. The two-sample median test was used to compare current short- and long-term users of thiazide diuretics regarding dosage.

Analysis of covariance was used to compare the age-adjusted mean bone mass at all three sites across categories of thiazide use. Tukey's range tests were carried out for the post hoc comparisons. Multivariable linear regression models were used to assess the relation between the duration of thiazide diuretic use (those who had never used thiazide diuretics were coded as zero) and bone mass at the three sites while controlling for age, body weight, duration of estrogen use, and calcium intake. Body weight was used in the analyses because weight was more strongly associated with bone mass than the Quetelet index (body mass index) [38].

The proportion of women in each category of thiazide with prevalent vertebral deformities was calculated according to the method of Fleiss [39]; 95% CIs were also calculated. Chi-square tests were used to compare rates of prevalent vertebral deformities and incidence rates for fractures and falls by category of thiazide use; rates are expressed as number per 1000 person-years. For prevalent vertebral deformities, logistic regression models were used. For incident fractures, proportional hazard regression models were used to assess the independent relation between thiazide diuretic use and fracture after adjustments were made for age, weight, functional status, self-reported health status, calcium intake, and oral estrogen use. The relative risk for fracture (with the 95% CI) was calculated. To test the hypothesis that bone mass accounts for the effect of thiazide diuretics on fracture risk, we subsequently added the bone mass of the distal radius to the models. Comparisons were made between current users of thiazide diuretics who had used these drugs for more than 10 years and two control groups: One control group included participants from all the other categories (those who had never used thiazide diuretics, those who were past users, and current users with a duration of use of 10 years); the second control group included only those who had never used thiazide diuretics. Results of the comparisons were similar for both control groups. Thus, we present only the results of the comparison between current users of thiazide diuretics and those who had never used these drugs.

For the analysis of falls, we used logistic regression analysis and calculated the risk for 1 or more falls and for 2 or more falls in the first year after baseline in current users of thiazide diuretics, past users of thiazide diuretics, and current users of nonthiazide diuretics. Women who had never used diuretics served as the reference group. We calculated the age-adjusted odds ratio and 95% CI. All analyses were done using SAS software (SAS Institute, Inc., Cary, North Carolina).

Results

Top Methods Results Discussion References

Twenty-seven percent of women (n = 2637) in the cohort were taking thiazide diuretics at the time of the baseline interview, and 9% (n = 851) had used thiazide diuretics previously (Table 1). Among current users, 153 reported chlorthalidone use. The mean duration of thiazide diuretic use was 11.2 ± 8.9 years (median, 10 years) in current users. The median daily dose of thiazide was 37.5 mg/d for current users of thiazide diuretics who had been receiving therapy 10 or fewer years and 50.0 mg/d for current users who had been receiving therapy for more than 10 years (P = 0.08). Compared with women who had never taken thiazide diuretics, users were slightly older and heavier, had poorer functional status, had lower calcium intakes, were more likely to have had a surgical menopause and to report current estrogen replacement, and less likely to report excellent health (Table 1). The median duration of estrogen replacement therapy was 7 years for long-term thiazide users, 5 years for past and short-term current users, and 4 years for those who had never used thiazide diuretics. Three hundred ninety-eight women died during the follow-up period.

Bone Mass

After adjusting for age, we found that women currently using thiazide diuretics had higher bone mass at all three measurement sites than did women who had never used thiazide diuretics (Table 2). A trend was observed toward greater preservation of bone mass with increasing duration of use; however, the magnitude of the effect was small. The age-adjusted mean bone mass for current users of thiazide diuretics who had used these drugs for more than 10 years was 0.03 to 0.04 g/cm² (7% to 10%) greater than that of women who had never used thiazide diuretics. Additional adjustment for body weight, calcium intake, and duration of estrogen replacement therapy tended to reduce these differences by half (that is, to 4% or 5%). In linear models, bone mass at all three sites was significantly associated with years of thiazide diuretic use, even after adjustments were made for age, body weight, duration of estrogen replacement, and calcium intake. The regression equations estimate (average ß = 0.0009) that after adjusting for important confounders, 10 years of thiazide diuretic use would prevent loss of about 0.01 g/cm² of radial bone mass, about 2% of the mean radial bone mass of a 70-year-old woman.

Vertebral Deformities

The prevalence of vertebral deformities among current users of thiazide diuretics who had taken these drugs for more than 10 years (25%; CI, 19% to 30%) was similar to that among women who had never taken thiazide diuretics (24%; CI, 22% to 26%), past users (20%; CI, 15% to 25%), and current users who had used thiazide diuretics for 10 or fewer years (22%; CI, 18% to 26%). The odds of having at least one vertebral deformity was not higher in long-term (> 10 years) thiazide users (odds ratio, 0.98; CI, 0.73 to 1.33) compared with those who had never used thiazide diuretics.

Fracture

During an average follow-up of 3.32 years, 1113 incident nonspinal fractures occurred. These included 137 hip fractures, 250 wrist fractures, and 113 proximal humerus fractures. No evidence was found that long-term use of thiazide diuretics reduced the incidence of either all fractures or those previously found to be related to low bone mass (Table 3). Women who had used thiazide diuretics for more than 10 years had a somewhat lower incidence of hip and wrist fractures and a somewhat higher incidence of proximal humerus fractures than did women who had never used thiazide diuretics (Table 3). None of these differences was statistically significant. The age-adjusted relative risk among current thiazide users (regardless of duration of use) relative to those who had never used thiazide diuretics was 0.95 (CI, 0.83 to 1.09) for nonspinal fractures; 0.94 (CI, 0.81 to 1.10) for osteoporotic fractures; 0.82 (CI, 0.56 to 1.12) for hip fracture; 0.90 (CI, 0.67 to 1.2) for wrist fracture; and 1.02 (CI, 0.67 to 1.54) for humerus fracture.

Thiazide users differed from other women in several ways that might influence the risk for fracture (Table 1). Thus, we adjusted for age, weight, functional status, health status, total calcium intake (diet plus supplement), and duration of estrogen replacement therapy (Table 4). No association was found between thiazide diuretic use or duration of use and the risk for either nonspinal fracture (relative risk, 0.99; CI, 0.81 to 1.20) or osteoporotic fracture (relative risk, 0.98; CI, 0.79 to 1.22). To determine whether use of thiazide diuretics alters the risk for specific types of fracture, we analyzed the risk for fractures of the hip, wrist, and proximal humerus. The risk for hip fracture was about 35% lower in current users of thiazide diuretics who had used these drugs for more than 10 years, but the results were not statistically significant (Table 4). A modest trend toward a protective effect for wrist fractures was also observed (Table 4). In contrast, the risk for proximal humerus fracture was elevated slightly (Table 4).

If thiazide diuretics protect against fracture by preserving bone mass, then statistical adjustment for bone mass should substantially reduce the association between thiazide use and the risk for fracture. However, adjusting for bone mass at the distal radius had no appreciable effect on estimates of the association between use of thiazide diuretics and all osteoporotic fractures and fractures at the hip and humerus (Table 4). After adjustments were made for age and other potential confounders, the relative risk for wrist fracture among women who had been using thiazide diuretics for more than 10 years was 0.66 (CI, 0.40 to 1.08); after further adjustment for bone mass at the distal radius measured near the site of wrist fracture, the relative risk was 0.73 (CI, 0.45 to 1.20), representing an 11% change in the relative risk.

Subgroup Analyses

To determine whether effects of thiazide diuretic use on fracture may be confounded or masked by the use of other drugs known to affect bone mass, we excluded all women currently receiving estrogen, calcium supplements, corticosteroids, anticonvulsants, and thyroid replacement therapy. After these women were excluded, 4442 (46%) remained in the analysis. Results were similar to those derived from analysis of the whole cohort, but the CIs were wider because of the smaller number of women. Specifically, we found no association between long-term (> 10 years) thiazide diuretic use in current users and either nonspinal fracture (relative risk, 1.03; CI, 0.78 to 1.36) or osteoporotic fracture (relative risk, 0.99; CI, 0.72 to 1.35). Long-term use of thiazide diuretics was associated with a trend toward a reduction in the risk for hip fracture (relative risk, 0.70; CI, 0.24 to 2.03) and wrist fracture (relative risk, 0.85; CI, 0.45 to 1.62) after adjustment for age. Little association was found between long-term use of thiazide diuretics and proximal humerus fracture (relative risk, 1.03; CI, 0.45 to 2.33).

Thiazide diuretics are known to decrease urinary excretion of calcium and to improve calcium balance [6]. Therefore, the effect of thiazide diuretics on calcium balance might depend on calcium intake. To explore this possibility, we stratified total calcium intake by the recommended dietary allowance, 800 mg/d. Results were similar in both groups: The age-adjusted relative risk for all fractures and for osteoporotic fractures in current thiazide diuretic users who had taken these drugs for more than 10 years was about 1.0 in both the low-calcium (n = 4176) and high-calcium (n = 5540) groups. However, a trend toward a protective effect of thiazide diuretics against hip and wrist fracture was observed in both groups (relative risk for hip fracture: low-calcium group, 0.68 [CI, 0.26 to 1.76]; high-calcium group, 0.54 (CI, 0.25 to 1.19); relative risk for wrist fracture: low-calcium group, 0.87 (CI, 0.46 to 1.66); high-calcium group, 0.44 [CI, 0.22 to 0.92]). A total of 1295 women reported receiving less than 400 mg calcium per day. In this subgroup, the age-adjusted relative risk among current users of thiazide diuretics who had used these drugs for more than 10 years was 1.09 (CI, 0.67 to 1.77) for nonspinal fracture and 1.23 (CI, 0.12 to 2.13) for osteoporotic fracture, relative to those who had never used thiazide diuretics. There was no suggestion that thiazide diuretics influenced the risk for all fractures, even in women with minimal calcium intake.

Users of thiazide diuretics were significantly more obese than nonusers, and because the degree of obesity is inversely associated with the risk for hip fracture [40], the effect of thiazide diuretics on fracture risk may be obscured by inclusion of the extremely obese. To test this hypothesis, we excluded obese women (body mass index 31.5 kg/m², which corresponds to the 85th percentile). Results were similar to those derived from the analysis of the whole cohort. None of the relative risks was statistically significant, but they did show a trend toward a protective effect of thiazide diuretics against hip and wrist fracture.

Nonthiazide Diuretics

We analyzed the association among use of nonthiazide diuretics, bone mass, and risk for fracture. Five hundred three women (5.2%) reported current use and 92 (1%) reported past use of nonthiazide diuretics at baseline; 99 women (1%) were current users who had used nonthiazide diuretics for more than 10 years. The mean duration of nonthiazide diuretic therapy for current users was 6.8 ± 7.7 years (median duration, 4 years).

Little association was found between current use of nonthiazide diuretics and bone mass after adjustments were made for age, obesity, estrogen replacement, and calcium intake. In current users of nonthiazide diuretics, the multivariate adjusted bone mass was 0.366 (g/cm²) for the distal radius; 0.638 g/cm² for the proximal radius; and 0.404 g/cm (²) for the calcaneus. Corresponding values in those who had never used nonthiazide diuretics were 0.361, 0.633, and 0.403 g/cm². Age-adjusted relative risks among current users of nonthiazide diuretics, compared with women who had never used thiazide diuretics, were 1.04 (CI, 0.80 to 1.35) for nonspinal fracture; 1.04 (CI, 0.78 to 1.40) for osteoporotic fracture; 1.07 (CI, 0.56 to 2.05) for hip fracture; 1.38 (CI, 0.84 to 2.26) for wrist fracture; and 0.30 (CI, 0.07 to 1.24) for humerus fracture.

Falls

Eight hundred seventy-three women experienced at least 2 falls during the first year after baseline; 2608 women fell at least once during the first year after baseline. Women who were current users of thiazide diuretics had no increased risk for falling. The annual incidence rate for experiencing at least 2 falls was 91.6 per 1000 person-years in women who had never used thiazide diuretics; 118.1 per 1000 person-years in past users of thiazide diuretics; and 98.4 per 1000 person-years in current users of thiazide diuretics. In each age group, past users tended to have the highest incidence of falls (Figure 1). The age-adjusted odds ratio for experiencing at least 2 falls in the first year after baseline in current users of thiazide diuretics compared with women who had never used diuretics was 1.06 (CI, 0.90 to 1.25); however, the odds ratio was significantly elevated (1.48; CI, 1.15 to 1.90) when past users of thiazide diuretics were compared with those who had never used the diuretics. Similarly, the age-adjusted odds ratio for experiencing at least 1 fall was 1.29 (CI, 1.08 to 1.54) in past users of thiazide diuretics and 1.06 (CI, 0.95 to 1.19) in current users of thiazide diuretics, when compared with women who had never used diuretics. Current use of nonthiazide diuretics was not associated with an increased risk for falling relative to women who had never taken diuretics: The odds ratio was 1.11 (CI, 0.86 to 1.44) for at least one fall and 1.29 (0.90 to 1.85) for experiencing at least 2 falls.

View larger version (59K): [in this window] [in a new window]   Figure 1. Age-specific incidence rate for 2 or more falls by thiazide diuretic use. White bars denote women who had never used thiazide diuretics; black bars, past users of thiazide diuretics; and gray bars, current users of thiazide diuretics.

Discussion

Top Methods Results Discussion References

We did not find that long-term use of thiazide diuretics led to a substantial reduction in a woman's risk for all fractures. Like other investigators [10, 11, 20-23], we found that thiazide diuretics preserve appendicular bone mass but that the magnitude of the effect is small. In our cohort, each decrease of 1 standard deviation in bone mass results in approximately a 1.65 age-adjusted increased risk for "osteoporotic" fractures (that is, those fractures found to be related to low bone mass) [2]. We estimate that 10 years of thiazide diuretic use preserves only about one tenth of one standard deviation in bone mass. Such a small change in bone mass would have a negligible effect on the risk for fractures.

Most fractures in older women result from falls [29]. Thiazide diuretics occasionally cause postural hypotension [27], and concern has arisen that they might increase the risk for falls and fractures. The incidence rate for falls was similar to that observed in previous studies [28]. However, we found no association between current use of thiazide diuretics and the risk for falling. This finding is consistent with most previous research [9]. The increased risk for falling in past users of thiazide diuretics may have occurred by chance, but it may also reflect confounding by other factors. Most of the previous studies of diuretics and the risk for falling have been confined to current users of thiazide diuretics. Past users may have discontinued therapy with thiazide diuretics because of the development of dizziness or other symptoms that increased their risk for falls. Given the increased risk for falling in past users, it is somewhat surprising that we found no increased risk for fracture in this group.

We also found that long-term and current use of thiazide diuretics had no statistically significant effect on the risk for all fractures and no effect on fractures that we previously found to be associated with reduced bone mass [2], despite the large number of fractures. We found no association between long-term use of thiazide diuretics and the presence of vertebral deformities. Although our study had limited power to evaluate individual types of fractures, we observed that current and long-term users of thiazide diuretics have about a one-third reduction in the incidence of hip fractures. These results are consistent with those of previous cohort studies [15-17]. We noted a similar trend for wrist fracture but not for proximal humerus fracture. Similar results were found in women who were not taking medications that affect bone mass, in women with a low calcium intake, and in women who were not obese.

None of the previous cohort studies included bone mass in their models. If thiazide diuretics reduce the risk for fracture by preserving bone mass, then statistical adjustment for bone mass should attenuate the relation between thiazide use and the decreased risk for fractures. However, the inclusion of the distal radius bone mass did not substantially change the relation between thiazide use and the reduced risk for most fractures, including hip fracture, despite the fact that the distal radius bone mass predicts hip fracture [30] and osteoporotic fracture [2]. For wrist fractures, a modest change was observed in the relative risk when the distal radius bone mass measured near the site of wrist fracture was included in the model. It is possible that inclusion of bone density at the hip may have attenuated the effect of thiazide diuretics on hip fracture. Nevertheless, it is doubtful that the only mechanism by which thiazide diuretics prevent fracture is through preservation of bone mass because thiazide diuretics are only one of many determinants of bone mass. Like other investigators, we found that women who use thiazide diuretics are heavier, are somewhat more likely to take estrogen, have poorer functional status, and have a lower calcium intake than do women who have never used thiazide diuretics. Adjusting for these differences did not substantially alter the association between long-term use of thiazide diuretics and the risk for hip or wrist fractures. Nevertheless, it is possible that the apparent association between thiazide diuretics and a reduced risk for hip and wrist fractures is also due to confounding factors that we did not measure. Heidrich and colleagues [19] recently presented evidence that patients who take thiazide diuretics differ in many ways from those who do not. They found that the crude odds ratio for hip fracture in users of thiazide diuretics was 1.1 and not statistically significant. However, when they adjusted for the potentially confounding effects of several variables, a significantly increased risk for hip fractures was found among women who reported current use of thiazide diuretics.

We considered other explanations for our findings. Misclassification of thiazide use would bias the study toward finding no association, and self-reported medication use might be inaccurate. However, we confirmed medication use by interview, review of medication bottles brought from home, and presentation of pictures of diuretic tablets to participants; we also rechecked the classification of medication use by reviewing a random sample of questionnaires. Misclassification of fractures is an unlikely source of error because all fractures were documented by radiographic reports and, in cases of hip fracture, by copies of the radiographs themselves. Users of thiazide diuretics might have a lower risk for hip fracture because of a selection bias. Physicians may be less likely to prescribe thiazide diuretics for women who fall frequently or who are thought to be too frail to benefit from the treatment of hypertension, the usual indication for thiazide use. For patients able to take thiazide diuretics for hypertension, it is possible that the hormones or other physiologic mechanisms that control blood pressure may be related to the risk for fracture. Yet, we found no association with nonthiazide diuretics, which may have been prescribed for hypertension as well. We do not have information on use of broader antihypertensive agents.

Results from the Framingham Study suggest that the thiazide dose may be important [17]. No protective effect of thiazide diuretics was observed in individuals using combination thiazide diuretics, which generally contain only 25 mg of thiazide. It is possible that the lack of an association between thiazide diuretics and fracture risk may have reflected a higher prevalence of low-dose users. This is unlikely, however, because the mean thiazide dose in our study was 50 mg/d.

Our study had several limitations. Women in our cohort were ambulatory volunteers who had a somewhat lower incidence of hip fracture. Our patients may have been less frail than those of Ray and colleagues [15] and LaCroix and colleagues [16]. We excluded men and black women, but our results can be generalized to the large population of community-living elderly white women. The average calcium intake in our cohort was higher than that reported for the general population of women in the United States. However, we found no evidence of a protective effect in women who were consuming less than the recommended calcium allowance of 800 mg/d or in women who had a minimal intake of calcium. Our baseline measurements included only measurement of appendicular bone mass; measurement of bone mass in the hip and spine were added to later examinations, and the effect of thiazide diuretics on bone mass in these sites will be the subject of future reports.

In summary, we found that thiazide diuretics had only a small effect on bone mass and had no effect on the risk for falls in older women. Consistent with these findings, long-term thiazide diuretic therapy was found to have no overall protective effect against all fractures. The protective trend for hip fracture that we observed is similar to that reported by previous studies, but this trend could not be explained by the higher bone mass of long-term users or by confounding factors that we assessed in this study. We observed a similar trend toward prevention of wrist fractures. We believe that a randomized clinical study of the effect of thiazide diuretics on the incidence of fractures is the only way to resolve this important issue [41]. Until such a study is done, physicians should not rely on thiazide diuretics to prevent fractures in older women.

Appendix. Study of Osteoporotic Fractures Research Group

University of California, San Francisco (Coordinating Center): S.R. Cummings, M.C. Nevitt, D. Black, H.K. Genant, C. Arnaud, C. Auburn, W. Browner, K. Faulkner, C. Fox, C. Gler, S. Harvey, S. Hulley, P. Steiger, L. Palermo, S. Sanchez, and D. Seeley.

University of Maryland: R. Sherwin, J. Scott, K. Fox, J. Lewis, L. Emerson, S. Trusty, M. Bahr, L. Finazzo, D. Rebar, B. Hohman, and E. Oilner.

University of Minnesota: R. Grimm, Jr., K. Ensrud, C. Bell, E. Mitson, P. Frank, K. Jacobson, and D. Thomas, S. Jackson, and L. Stocke.

University of Pittsburgh: J. Cauley, L. Kuller, L. Harper, M. Nasim, C. Bashada, L. Buck, A. Githens, D. Medve, and S. Rudovsky.

The Kaiser Permanente Center for Health Research, Portland, Oregon: T. Vogt, H. Glauber, W. Vollmer, E. Orwoll, J. Blank, R. Bright, J. Downing, and B. Mastel-Smith.