By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
Dr Orwoll serves as a consultant to Procter & Gamble, GlaxoSmithKline, Aventis, and TAP Pharmaceuticals. He also receives honoraria from Merck and grant support from Aventis, Lilly, Pfizer, and Novartis. Drs Taylor, Fink, and Ensrud are employees of the U.S. government. All other authors state that they have no conflicts of interest.
We examined determinants of nonvertebral fracture in elderly men from six U.S. communities followed an average of 4.1 years. Six clinical risk factors predicted fracture risk independent of hip BMD: tricyclic antidepressant use, previous fracture, inability to complete a narrow walk trial, falls in previous year, age ≥80 years, and depressed mood.
Introduction: There are few prospective studies of fracture determinants in men. We examined the associations between a comprehensive set of clinical risk factors and risk of nonspine fracture in older men and whether determinants of fracture risk were independent of total hip BMD.
Materials and Methods: A total of 5995 men ≤65 years of age were recruited from six communities in the Unites States and followed prospectively for an average of 4.1 years. Baseline assessments of demographic, lifestyle, medical history, functional status, anthropometry, and cognitive, visual, and neuromuscular function were assessed by questionnaire or examination. Triannual mailed questionnaires ascertained incident fracture; reported fractures were adjudicated by physicians using medical records and X-ray reports. Proportional hazards models were used to develop multivariable models, selecting variables and controlling for BMD.
Results: Of 5876 men, 4.7% (N = 275) reported an incident nonspine fracture during follow-up (11.46/1000 person-years). Tricyclic antidepressant use (hazard ratio [HR], 2.36; 95% CI, 1.25–4.46), history of fracture at or after age 50 (HR, 2.07; 95% CI, 1.62–2.65), inability to complete a narrow walk trial (HR, 1.70; 95% CI, 1.23–2.34), falls in previous year (HR, 1.59; 95% CI, 1.23–2.05), age ≤80 years (HR, 1.33; 95% CI, 1.01–1.76), depressed mood (HR, 1.72; 95% CI, 1.00–2.95), and decreased total hip BMD (HR, 1.53; 95% CI, 1.34–1.74) were independently related to increased risk. Compared with having none (48.0% of men), having three or more of the clinical risk factors (4.9% of men) increased fracture risk 5-fold, independent of BMD. Having three or more risk factors and being in the lowest tertile of BMD was associated with a 15-fold greater risk than having no risk factors and being in the highest BMD tertile.
Conclusions: Several clinical risk factors were independently associated with nonspine fractures in elderly men. The combination of multiple risk factors and low BMD was a very powerful indicator of fracture risk.
It has long been known that women of northern European ancestry are at high risk of osteoporosis, but the fact that men also are at substantial risk has not been well appreciated. In the 12 years of follow-up in the Dubbo study, for example, one in four hip fractures occurred in men, and with increasing age, the incidence rates for men approached rates in women.(1) In addition, substantial costs are incurred for health care of men with fractures. Annual direct care costs for osteoporotic fractures in white men in the United States represent approximately $3.2 billion per year in 2002 dollars, −18% of the total costs for osteoporosis.(2)
Recently, the Surgeon General's report on bone health and osteoporosis(2) urged that clinicians consider use of a formal screening tool to assess individual patient's risk for osteoporotic fracture. However, the report acknowledges that it is very unclear which risk factors should be included in these tools beyond the basic characteristics of age, sex, and ethnicity. The available tools are limited by the underlying data on which they are based, particularly the paucity of data in groups other than older white women. For older men, there are few data from prospective studies that examine multiple risk factors with fracture outcomes, particularly for nonvertebral fracture.(3–12) Because important differences could exist between groups, additional data are needed on men and other race/ethnic groups.
Recognizing the limited prospective data on determinants of fracture in men, the Osteoporotic Fractures in Men Study (MrOS) recruited nearly 6000 older men to a study designed to address important questions relevant to fracture.(13) This analysis reports on the independent, prospective predictors of non-spine fracture, including the potential effects of demographics; lifestyle; medical history; neuromuscular, visual and cognitive function; and anthropometry in men followed an average of 4 years. We examined the associations between a comprehensive set of clinical risk factors and risk of non-spine fracture in older men, while including and excluding hip BMD from the model. We also examined the associations of the number of clinical risk factors with fracture risk, independent of BMD.
MATERIALS AND METHODS
MrOS is a multicenter prospective study of risk factors for vertebral and all nonvertebral fractures in older men. The design, measures, and recruitment methods used by the study have been previously described.(13,14) In brief, 5995 men ≤65 years of age were recruited from March 2000 to April 2002 from the populations of Birmingham, AL; Minneapolis, MN; the Monongahela Valley near Pittsburgh, PA; Palo Alto, CA; Portland, OR; and San Diego, CA. Approval of the conduct of MrOS was obtained from the institutional review boards of the participating institutions, and written informed consent was obtained from all study participants.
Inclusion criteria were designed to result in a cohort of men representative of the communities from which they were recruited. Men were excluded from the study if they could not provide informed consent or self-reported data, could not walk without the assistance of another, had bilateral hip replacements, did not live in or planned to move from the area surrounding the study site, or if in the judgment of the investigator they had a severe medical condition that would preclude participation in follow-up.
Of the 5995 men enrolled in the study, we excluded from this analysis 112 men who reported taking osteoporosis medications at baseline (bisphosphonates, calcitonin, fluoride, or raloxifene); 3 men who reported both receiving testosterone injections and a history of osteoporosis; and 4 who were missing follow-up data. This analysis includes the remaining 5876 men (98.0%) who were followed for an average of 4.1 ± 0.9 years.
Information on demographic, lifestyle, personal and family medical history, functional status, anthropometric, and cognitive, visual, and neuromuscular function data were obtained by self-report, interview, or examination by trained and certified staff.(13) Questions included those on age, race/ethnicity (white/white, black, Asian, Hispanic, Native Hawaiian/Pacific Islander, American Indian/Alaskan Native, and multiracial), socioeconomic status (subjective socioeconomic status(15) and education), and marital status. Also reported were weight and height at age 25. Physical activity was reported with the Physical Activity Scale for the Elderly(16) (PACE) and additional questions on daily walking for exercise, and sedentary activity (sometimes/often sit >4 h/day). Personal history of fracture at or after age 50 years (any fracture; or hip, wrist, or spine fracture), parental history of fracture (any fracture or hip fracture), falls in the previous 12 months, and specific medical conditions (e.g., diabetes mellitus, hyperthyroidism, stroke, osteoporosis, and glaucoma) were assessed. General health status was rated as excellent/good versus fair/poor/very poor. Functional status was assessed by summing the amount of difficulty (on a 0–3 scale) with five instrumental activities of daily living (IADLs consisting of difficulty with walking two to three blocks outside on level ground, climbing 10 steps without resting, preparing meals, heavy housework, and shopping for groceries or clothes [score range = 0–15]); modified physical and mental function summary scales from the 12-item Medical Outcomes Study (MOS) Short Form 12 (SF-12)(17) were also reported.
Lifestyle risk factors included alcohol consumption (average number of drinks per week), smoking (current, past, never), and dietary intake. The Block 98 semiquantitative food frequency questionnaire (FFQ; Block Dietary Data Systems, Berkeley, CA, USA) was specifically modified for MrOS to capture the most important sources of calcium and vitamin D and other nutrients associated with osteoporosis in older men in the United States. The nutrient composition was calculated using the USDA Database for Standard Reference, Version 12, and the 1994–1996 Continuing Survey of Food Intake in Individuals (CSFII) database. For this analysis, we used usual intake of daily caffeine (mg), total calcium (mg) and vitamin D (IU) intakes from diet and supplements. Participants were asked about use of medications and asked to bring in current medications for data collection. Specific classes of medications of interest were coded by trained staff (e.g., αblockers, nonbenzodiazepine anticonvulsants, corticosteroids [inhaled and oral], diuretics [thiazides, loop, other], serotonin reuptake inhibitors, thyroid hormones, and tricyclic antidepressants).
Depression was assessed with self-reported mood, using one question from the SF-12 about mood during the previous 4 weeks. Reported use of medications with antidepressant effects (taking a tricyclic [TCA], selective serotonin reuptake inhibitors [SSRI], or trazodone) was examined separately, because there was little evidence of overlap between medications and between medications and symptoms, and there are other uses of these drugs besides treatment of depression. Trazodone use was rare (<1%) and not further examined. Among all men, 95.1% were not taking either a TCA or SSRI, 1.5% were taking a TCA only, 2.5% were taking an SSRI only, and 0.1% were taking both. Among the 156 SSRI users, 7 (4.5%) were also taking a TCA; this was 7.2% of the 97 TCA users. Only 9.3% of men classified as having depressed mood reported taking either drug, whereas 89.0% men who reported TCA or SSRI use did not report depressed mood.
Examination measurements included anthropometry, cognitive function, visual function, neuromuscular function, and BMD. Body weight (kg; in indoor clothing without shoes) was recorded with a calibrated balance beam or electronic scale. Height (cm) was measured using a wall-mounted Harpenden stadiometer (DyFed). Radial pulse was palpated for 30 s after a 5-minute rest; two measurements were averaged for analysis. The modified mini-mental state (3MS) examination was conducted to assess cognitive function (scored 0–100).(18) Corrected visual acuity (Baily-Lovie chart), contrast sensitivity (vision contrast test system; Visitech Consultants, Dayton, OH, USA), and depth perception (Frisby stereo test; Richmond Products, Boca Raton, FL, USA) were assessed. BMD of the total hip was measured by DXA (model QDR 4500; Hologic, Waltham, MA, USA, at each site). Extensive quality assurance protocols were used throughout, including central training and certification of technicians and regular phantom scans within and across centers.
We examined three neuromuscular function variables in the analysis. Participants were asked to stand from a chair without using their arms; those who were unable to do a single chair stand were classified as “unable” to complete the test. All men who were able to complete the single chair stand were asked to complete the repeated chair stand test. The ability and time required to complete five stands without using the arms were recorded. If they were unable to do five chair stands, used their arms at any time during the test, were unable to complete the test, or refused to do the repeated chair stand test, they were also classified as “unable.” Participants were asked to complete a narrow walk trial over a 20-cm-wide, 6-m course; two trials were required of all participants. A third trial was performed if the participant deviated at least twice during either trial. Inability to compete any of these trials successfully was recorded. Grip strength (kg) was measured twice by a hand held dynameter (Jamar) in both the right and left arms; the average of right and left was used in analysis.
After recruitment, men were followed for incident fracture with a triannual questionnaire administered by mail or telephone. Reports of fracture were followed up by study staff to determine date, description of how the fracture occurred, and any trauma that resulted in the fracture. Fractures were verified by physician adjudication of medical records and X-ray reports. All fracture sites (hip, wrist, skull/face, cervical, thoracic and lumbar vertebrae; shoulder; arm; hand/finger; rib/chest/sternum; pelvis/tailbone; leg; and ankle/foot/heel/toe) were included, but unconfirmed fractures were not considered in the analyses. Excessive trauma was determined by the physician adjudicators; generally, moderate or severe trauma other than a fall or a fall from more than standing height were considered excessive trauma. Pathologic fractures were excluded from analysis. Because exclusion of fractures resulting from excess trauma has been reported to underestimate the contribution of osteoporosis to fractures in women,(19) we included fractures regardless of trauma level in the principal analyses, but repeated analyses censoring participant follow-up at time of traumatic fracture.
All analyses were conducted using SAS version 9.1 (SAS Institute, Cary, NC, USA). To develop a parsimonious, multivariable prediction model from a large number of variables collected on the cohort, we generated a list of potential risk factors based on the literature, as described above. Continuous variables were transformed (log10 for contrast sensitivity) or categorized (PASE and age, for which quadratic terms were significant) when necessary because of their distributions or nonlinear associations with fracture risk. Because of the quadratic relationship with fracture risk, we examined age in 5-year increments. Men who were in the 70- to 74-year age range were at lowest risk and men ≤80 years of age were at highest risk. We used a single age cut-point of 80 years in model building because this seemed to be most clinically useful. We compared the baseline characteristics of men who did versus those who did not have an incident nonspine fracture using χ2 tests, t-tests, Fisher's exact test, and Wilcoxon rank sum tests, as appropriate. Variables were parameterized such that the reference is the lowest risk group. Univariate analyses tested for associations with non-spine fractures; those related at the level p < 0.10 were further tested for association in age-adjusted proportional hazards models.
Because of the age of the cohort, it was necessary to consider missing data resulting from inability of the participant to complete the neuromuscular function tests. We used the following classifications in analysis: (1) grip strength, categorized into quartiles, with men who were unable or refused to do the test placed into a separate category (−2% of the cohort); (2) narrow walk, categorized as men who could not complete any of the trials (−9% of the cohort) compared with men who successfully completed one or more trials; (3) chair stand, categorized as men who were able to stand from the chair five times without using their arms versus those who could not (−3% of the cohort), including those who were unable or refused to do so in the latter group.
Among the variables associated with fracture in the univariate analyses, some were related and highly correlated with each other, for example, self-reported history of hypothyroidism and use of thyroid supplement. In some cases, we chose the variable that seemed most specific for a condition, for example, thyroid medication use rather than self-report of thyroid disease. In other cases, we chose to analyze the variable that was collected most completely, for example, using whether the participant was able to complete the chair stand test rather than time to completion (in which case those unable to perform the test would be missing). Several variables reflecting history of specific fractures at or after age 50, for example, hip fracture, and the report of any fracture at or after age 50 were significantly related to risk of an incident fracture in MrOS, the latter inclusive fracture history variable was chosen for further modeling. The remaining variables were tested in age-adjusted models for relationship to fracture (p < 0.10); independent variables still related to fracture were not correlated with each other. Associations adjusted only for age and BMD were also examined with proportional hazards models.
Potential risk factors from age-adjusted models (associations of p < 0.10) were entered into stepwise proportional hazards models, using p < 0.10 to enter and p < 0.05 to stop. Models were built using forward stepwise selection and backward elimination (using p < 0.05 to enter and to stop) and a best subsets selection method (using the top three χ2 scores for models with varying numbers of predictors [six, seven, and eight predictors]). Similar results were obtained with these different methods; therefore, stepwise selection results are shown. Models were first constructed using clinically available variables without including BMD among those considered (Table 2, model 1). Once developed, BMD was added to the model to assess whether associations were independent of BMD (Table 2, model 2). Another model was constructed including BMD in the variable selection to determine whether including BMD influenced the selection of other variables in the final model (Table 2, model 3). All multivariable models were adjusted for race/ethnicity and clinic site by forcing these into the model.
We next created a score for each participant based on the number of independent clinical risk factors he had. We used the final multivariable model that did not include BMD in variable selection. We compared the risk of non-spine fracture among men with zero (reference), one, two, and three or more risk factors using proportional hazards models with and without adjustment for BMD. We examined the absolute rates of fracture related to risk factor score within BMD tertiles (cut-points: <0.898; ≤0.898 and <1.013; and ≤1.013 g/cm2). To facilitate clinical relevance of the risk score, we grouped continuous variables based on their means, distributions, and univariate relationships to fracture risk. For example, age cut-points were ≤80 versus <80 years, reflecting the strength of the association of age 80 years and above with fracture relative to younger ages.
Of the 5876 men included in this analysis, 275 (4.7%) suffered an incident nonspine fracture during 4.1 ± 0.9 years of follow-up, with an incidence rate of 11.46/1000 person years. Of these men with fractures, 217 (78.9%) experienced nontraumatic fractures. Considering all fractures, the most common were ribs (18.6%); hip (16.4%); wrist (13.1%); and ankle (7.6%). Without traumatic fractures, the most common were hip (18.9%); ribs (17.5%); wrist (13.8%); and ankle (7.8%). Men who sustained a nonspine fracture during follow-up differed from those who did not on a number of characteristics, including mean age (75.5 versus 73.5 years, respectively, p < 0.001), weight (81.7 versus 83.4 kg, p = 0.045), total hip BMD (0.894 versus 0.964 g/cm2, p < 0.001), having fallen in the past year (33.1% versus 20.3%, p < 0.001), having had a fracture at or after age 50 (43.6% versus 23.0%, p < 0.001), TCA use (3.6% versus 1.6%, p = 0.024), depressed mood (5.5% versus 2.9%, p = 0.014), and the inability to complete any narrow walk trial successfully (18.9% versus 8.7%, p < 0.001).
Several variables were not associated with risk of nonspine fracture in unadjusted analyses. Some of these were uncommon exposures, and we had little power to examine their associations with fracture, including use of antiandrogen medications (reported by 3.3% of participants), use of inhaled or oral steroids (5.3%), and cigarette smoking (3.5% were current smokers). Other characteristics that did not differ included socioeconomic status indicators, cognitive function, alcohol consumption, usual dietary intake of caffeine, total calcium, and vitamin D, physical and mental function summary scales (SF-12), radial pulse, depth perception, and visual acuity (20/40 or worse in either eye).
After controlling for age, variables that were significant predictors of fracture risk included demographics (race/ethnicity), body size (weight and height), history of fracture at or after age 50, any fall in the previous 12 months, family history (maternal history of hip fracture), history of a number of medical conditions (including hypothyroidism and osteoporosis, stroke and heart attack), use of certain medications (including SSRI, thyroid hormone, and TCA), depressed mood, functional status difficulty, neuromuscular function indicators (inability to perform chair stand without arms, inability to complete a narrow walk trial, and grip strength), and total hip BMD (all p < 0.10; Table 1). The associations with physical activity and age were nonlinear. Risk of fracture was greatest at the lowest and highest quartiles of PASE score (quadratic term: p = 0.03). For age (quadratic term: p = 0.02), men 75–79 (hazard ratio [HR], 1.40; 95% CI, 1.00–1.99) and ≤80 years of age (HR, 2.3; 95% CI, 1.64–3.21) were at increased risk of fracture compared with men 70–74 years of age; there was little difference with those 64–69 years of age (HR, 1.03; 95% CI, 0.73–1.46). After adjustment for both age and BMD (Table 1), weight was significantly inversely related to fracture, as expected. Also significantly associated were fracture at or after age 50; history of hypothyroidism and heart attack; inability to rise from a chair and to complete the narrow walk; grip strength; falls; physical activity; daily sitting; difficulty with IADL; and use of loop diuretics, TCA, and thyroid hormone.
Table Table 1.. Risk Factors for Non-Spine Fracture After Adjustment for Age and BMD (Including Variables With Age-Adjusted p < 0.10)*
Multivariable models were constructed first with clinical variables obtainable through a medical history and physical examination, excluding BMD (Table 2, model 1). In the initial model, age at least 80 years, any fracture at or after age 50, any fall in past year, TCA use, unable to complete any narrow walk trial, and depressed mood were significantly associated with risk. SSRI use was not included in the multivariable model.
Table Table 2.. Multivariable Models of Risk Factors for Non-Spine Fracture With and Without Including BMD in Variable Selection [HR (95% CI)]
BMD was significantly associated with fracture risk, and when added to the model (Table 2, model 2), associations were attenuated for some variables (e.g., past fracture [HR, 2.35] without versus [HR, 2.07] with BMD in the model, depressed symptoms [HR, 1.98 versus 1.72]), but were not for others (e.g., use of TCA [HR, 2.39 versus 2.36]). When BMD was included in the variable selection (model 3), the results were similar to those obtained when BMD was controlled for after variables were selected (model 2).
The prevalence of the six clinical risk factors identified in the final multivariable model, not including BMD, were report of any previous fracture at or after age 50 years (24.0% of men), any falls in the past year (20.9%), age at least 80 years (17.6%), unable to complete any narrow walk trials (9.1%), depressed mood (3.0%), and TCA use (1.7%). Forty-eight percent of the men had none of these risk factors, 33.6% had one risk factor, 13.4% had two risk factors, and 4.9% had at least three of the risk factors. When not considering BMD, men with one of these risk factors had a 66% increased risk of a nonspine fracture compared with men who had no risk factors (Table 3). The risk was over 3-fold higher in men with two and >6-fold higher in men with three or more risk factors. The increased risk associated with a 1 SD decrease in BMD was similar to the increased risk of having one clinical risk factor compared with having none. Including BMD in the model attenuated slightly the hazard ratios associated with the number of clinical risk factors; however, risk was still markedly increased with multiple clinical risk factors. Within each tertile of BMD, fracture rate increased with the number of clinical risk factors, particularly with three or more (Fig. 1). The rate of non-spine fractures in men with three or more risk factors and BMD in the lowest tertile was nearly 15-fold higher than that of men with no risk factors in the highest BMD tertile and nearly 6-fold higher than that of men with no risk factors and BMD in the lowest tertile.
Table Table 3.. Risk of Non-Spine Fracture by Number of Clinical Risk Factors With and Without Total Hip BMD (1 SD Decrease)
When fractures caused by excessive trauma were not included, the final model controlling for BMD (analogous to model 2, Table 2) again included history of fracture at or after age 50, age, falls, tricyclic use, inability to complete any narrow walk trials, and depressed mood with similar hazard ratios. Additional variables included were grip strength (unable or refused to perform [HR, 2.35; 95% CI, 1.10–5.04]; lowest quartile [HR, 1.58; 95% CI, 1.01–2.47] compared with the highest quartile), and visual impairment (decreased log contrast sensitivity per SD [HR, 1.14; 95% CI, 1.01–1.29]). The risk of a non-spine fracture associated with the number of these risk factors was similar to the model containing all fractures: risk for fracture with three or more risk factors was 7-fold higher (HR, 7.21; 95% CI, 4.84–10.73) without adjusting for BMD and nearly 6-fold higher (HR, 5.64; 95% CI, 3.76–8.48) when adjusting for BMD compared with having no risk factors. Fewer men had no risk factors, 37.0% when censoring traumatic fractures versus 48.0% when including all fractures, and more men had three or more risk factors, 13.7% versus 4.9%, respectively.
Our study of elderly men recruited from six U.S. communities found several clinical measures were independently associated with non-spine fractures over 4 years of follow-up. When not considering BMD and including all fractures in the analysis, these risk factors included history of fracture at or after age 50, TCA use, age, falls in the previous year, depressed mood, and poor neuromuscular function, as measured by the inability to complete a narrow walk trial. In models that included BMD in the initial variable selection, BMD was included in the final model, as well as all the clinical risk factor variables previously selected except for depressed mood, with little major change in the strength of association for the other variables. Because of the high prevalence and potential for intervention, previous fracture at or after age 50 and falls in the past year are of greatest public health importance. Personal habits such as smoking, alcohol, or calcium intake were not significantly associated with risk.
We constructed a risk factor score from our final model; this model predicted increasing risk with an increasing number of the six clinical risk factor variables and increased risk independent of BMD. Having three or more of the clinical risk factors predicted a 6-fold increased risk when not considering BMD and a 5-fold increased risk when also considering BMD in risk prediction. Having three or more risk factors and being in the lowest tertile of BMD was associated with 15-fold greater risk than having no risk factors and being in the highest BMD tertile. Finally, our results were similar whether we included all fractures or censored fractures caused by more than minimal trauma.
Whereas risk for osteoporotic fracture is multifactorial, bone mass measurement was an important determinant in previous studies of both women and men. Previous analyses from MrOS by Cauley et al.(20) documented cross-sectional associations of femoral neck and lumbar spine BMD with a number of demographic, anthropometric, physical function, health, lifestyle, and behavioral risk factors. Age (for femoral BMD), white (compared with black) race, and previous fracture were important independent risk factors for both lower BMD and fracture. Grip strength was modestly related to BMD but not independently related to all fractures in our analysis. The ability to perform the chair stand trial without arms was related to femoral neck and not to lumbar BMD, but was not independently related to fracture. Whereas the narrow walk trial was not analyzed for BMD, it was a strong risk factor for fracture and considerably more important than the chair stand trial. Weight was related to BMD but was not independently related to fracture in multivariable models including other clinical risk factors.
Interestingly, SSRI use was related to BMD in MrOS but not TCA use,(20) whereas TCA use was strongly related to fracture in this analysis. In addition to TCA use, we found an independent association of fracture risk with depressed mood, as assessed with a single question, in multivariable models including other clinical risk factors. However, the association was attenuated when BMD was also included. Because there was little overlap between depressed mood and antidepressant use, it is entirely possible that both TCA use and depressed mood predicted fractures. On the other hand, report of depressive symptoms was not an important predictor of BMD in MrOS.(20) It is possible that these variables work through different mechanisms in ways that could account for these findings or that physician prescribing practices may result in indication, or contraindication, biases. For example, TCA use could be related to fracture because of side effects such as sedation or cardiac dysrhythmias leading to falls but might not have any intrinsic or indirect effect on bone. A case-control study of male and female Medicaid enrollees ≤65 years of age found an increased odds of hip fracture in current users of hypnotics/anxiolytics with long half-lives, TCAs, and antipsychotics.(21) The odds ratio for TCA use was similar to our findings, but the study predated widespread use of SSRIs. The Study of Osteoporotic Fractures (SOF) found an increased risk of nonspine fractures in women who reported use of some drugs with central nervous system effects, including antidepressants (TCA and SSRI).(22) The increased risk was more strongly related to hip than to all nonspine fractures, as well as to TCA than to SSRI use, and was independent of several risk factors, including BMD. BMD did not differ between those who did and those who did not report antidepressant use in SOF. Consistent with our findings, depressive symptoms, as measured by a standardized scale, increased fracture risk in SOF, but not independently. Limited power because of the small number of fractures and the low frequency of these exposures precludes our ability to further examine these complex issues.
Falls are common in the elderly and are an important risk factor for osteoporotic fracture; −1 in 10 result in a serious injury, including fracture.(23) Our analysis found that any fall in the previous year was an independent risk factor for fracture, similar in risk to that associated with being ≤80 years of age. Other risk factors associated with fracture in our analysis have been associated with increased fall risk,(23) including depressive symptoms, balance and gait impairment (measured with narrow walk trials in our study), and use of SSRIs and TCAs, perhaps explaining at least in part their relationships with fracture, although these factors remain significant with fall history in the model. Finally, the presence of multiple risk factors consistently increases fall risk.
Small body size has been advocated as a risk factor for determining screening of elderly women by BMD testing. The Surgeon General's report, for example, identifies low body weight as a “red flag,” indicating the need for further evaluation,(24) although the report acknowledged the inherent problem of confounding between weight and BMD. Our analysis did not find that weight was an important independent risk factor for fracture. Analyses of osteoporotic fractures in men in the Dubbo Osteoporosis Epidemiology Study found that weight was not independent of BMD.(5) Analyses of women in SOF found that weight was a risk factor for certain fractures (hip, pelvis, rib) but not others (wrist, humerus, ankle, and foot), and was not independent of BMD.(25) Further follow-up of MrOS with additional fracture events will allow us to examine this association more closely.
Consistent with other studies,(26–29) there was a marked reduction in risk for blacks compared with whites in univariate analyses. Whereas the age- and BMD-adjusted association with fracture was not statistically significant because of wide confidence intervals, the magnitude of the hazard ratio remained strong. The small number of men from nonwhite ethnic groups, and the even smaller number of fractures in them, limited study power. We were thus unable to examine associations of race/ethnicity in detail and have only adjusted for it in analyses.
Prospective studies of fracture in men including assessment of a number of risk factors have recently begun to appear in the literature. In a study of three Norwegian counties by Meyer et al.,(30) >27,000 men 35–49 years of age were followed for hip fracture from the mid-1970s through 1990, a cohort considerably younger than MrOS. The analysis excluded fracture in bone metastases and caused by high-energy trauma and collected outcome data from hospitals on 64 hip fractures in men. Significant risk factors included older age, greater height, lower BMI, history of diabetes mellitus and stroke, receiving a disability pension (not for diabetes or stroke), and marital status.
An analysis of hip fracture risk in men and women in the Dubbo Osteoporosis Epidemiology Study found that, in addition to low femoral neck BMD, age, postural instability (sway), quadriceps weakness, a history of fall, and a prior fracture predicted fracture independent of BMD in sex-adjusted analyses.(31) Consistent with our study in men, men and women with multiple (three or more) risk factors had the highest risk of fracture at any level of BMD. In the male Dubbo cohort, age, weight, BMD, prior fracture, a fall in the previous 12 months, postural sway, and quadriceps strength were all significantly related to hip fracture in univariate analyses; BMI, home physical activity, calcium intake, duration of smoking, and alcohol consumption were not associated.
The NHANES I Epidemiology Follow-up Study examined risk factors for hip fracture in a cohort of nearly 3000 white men 45–74 years of age followed for an average of nearly 14 years.(4) Independently and significantly related risk factors included only presence of one or more chronic diseases, weight loss from maximum of ≤10%, and phalangeal BMD by radiographic absorptiometry. Age, previous fracture other than of the hip, BMI, smoking, alcohol consumption, and several dietary variables were not related. NHANES had limited power based on the small number of hip fractures (71 overall) and the numbers of participants with complete risk factor measurements.
This study has many strengths, including the large number of men involved, prospective follow-up, a comprehensive set of baseline measurements, and standardized data collection including adjudication of fractures ascertained through frequent direct surveillance of the MrOS cohort and near complete follow-up rates.
Nevertheless, our study does have limitations. Because our population was restricted to relatively healthy, ambulatory older men, as exemplified by the small number of current smokers, results may not be generalizable to other groups. Even though great efforts were made to achieve ethnic/racial diversity consistent with the communities included in MrOS, the numbers of minority participants was substantially lower than that of whites, and their number of fractures was lower, thus there was little power to examine these groups specifically. We therefore adjusted our multivariable models for race/ethnicity as well as clinic site. Because the results in Fig. 1 are obtained from models that adjust for race-ethnicity, the absolute risks shown are a weighted estimate of the results for all participants and likely slightly underestimate the risk of fracture in whites and somewhat overestimate risk in blacks.
Despite our large sample size, power was somewhat limited for analyses of less common risk factors, such as smoking (4% of men were current smokers). Although we relied on self-report of fracture for case finding, false-negative reporting is unlikely to be significant because there were no unreported fractures in the medical records of a subgroup of women in SOF.(32) As stated, we did not use a validated scale to assess depressive symptoms. Our risk score was based solely on the number of risk factors and does not weight them according to the relative risks of each. We did not have sufficient numbers of fractures to examine risk factors specific for different fracture types. Thus, a different mix of fracture types may have resulted in findings of different risk factors for nonspine fracture.
In conclusion, whereas there is considerable literature on predictors of fractures in women, there have been few comprehensive, prospective studies in men. In men ≤65 years of age, we found several clinical risk factors that were independently associated with nonspine fractures, regardless of hip BMD. The combination of several risk factors and low BMD was a very powerful indicator of fracture risk. These data add to our knowledge of osteoporotic fracture risk in men and will assist clinicians in screening for those at high risk for these events. They should be considered in the formulation of updated screening recommendations for older men, an area where there is currently no consensus.
The Osteoporotic Fractures in Men (MrOS) Study is supported by National Institutes of Health funding. The following institutes provide support: the National Institute of Arthritis and Musculoskeletal and Skin Diseases, the National Institute on Aging, and the National Cancer Institute, under the following grant numbers: UO1 AR45580, UO1 AR45614, UO1 AR45632, UO1 AR45647, UO1 AR45654, UO1 AR45583, UO1 AG18197, and M01 RR000334.