Selection of patients with an index fracture of the ankle, hip, humerus, or wrist (A) or an index fracture of the finger (B). The asterisk indicates a health plan member for less than 24 months; dagger, excluded because of health conditions or drugs.
Selection of patients in the control group without fractures. The asterisk indicates fracture of any kind before January 1, 1999; dagger, excluded because of health conditions or drugs.
Probability of remaining free of fracture during follow-up by type of index fracture. A, Ankle fracture. B, Hip fracture. C, Humerus fracture. D, Wrist fracture.
Ettinger B, Ray GT, Pressman AR, Gluck O. Limb Fractures in Elderly Men as Indicators of Subsequent Fracture Risk. Arch Intern Med. 2003;163(22):2741–2747. doi:10.1001/archinte.163.22.2741
Whether limb fracture in elderly men predicts future fracture is unknown.
Electronic health records were examined to determine fracture incidence among men 60 years or older who were members of a large health maintenance organization, experienced no fracture in the past 2 years, and experienced an ankle, hip, humerus, or wrist fracture between July 1, 1997, and August 31, 2001. Proportional hazards models were used to compare risk of new fracture (ankle, hip, humerus, or wrist) between groups. Recurrent fractures of the same type were excluded from analysis.
During the follow-up period (mean duration, 2.4 years), 0.5% of the control subjects without fractures experienced a subsequent ankle fracture; 0.6%, a hip fracture; 0.2%, a humerus fracture; and 0.4%, a wrist fracture. A limb fracture was about 4 times more likely to occur in persons who experienced a previous humerus fracture (relative risk, 3.9; 95% confidence interval, 2.5-6.0), about 3 times more likely to occur in persons who experienced a previous hip fracture (relative risk, 2.8; 95% confidence interval, 1.7-4.5), and about 2 times more likely to occur in persons who experienced a previous wrist fracture (relative risk, 2.2; 95% confidence interval, 1.4-3.5) than in controls. In contrast, persons who experienced a previous ankle fracture had no greater risk of subsequent fracture than nonfracture controls (relative risk, 1.0; 95% confidence interval, 0.5-1.9).
Among men 60 years or older, a recent hip, humerus, or wrist fracture is a statistically and clinically significant predictor of future limb fracture risk. An increased risk of future fracture is greatest after a humerus fracture and is lowest after a wrist fracture; however, among elderly men, a previous ankle fracture is not an indicator of future fracture risk.
RISK FACTORS related to osteoporotic fracture in women have been extensively investigated, but the epidemiologic features of osteoporotic fracture in men have received attention only recently. Fracture among aging men is a clinically significant health problem, as indicated by the fact that about one third of all fractures among elderly persons occur in men.1,2 The lifetime risk of fragility fracture among white men at the age of 60 years has been estimated at 21% in England and Wales1 and at 15% in Australia.2 Fracture in men results in clinically significant morbidity, mortality, and cost to society.3,4 Among people older than 75 years, the hip fracture–related mortality rate is higher in men than in women, probably for reasons related to greater incidence of comorbid conditions among aging men than among similarly aged women.5 Given the magnitude of the fracture risk among aging men, programs aimed at preventing osteoporotic fracture should receive greater emphasis.
Identifying persons at risk for future fracture is a key part of any prevention program. Several clinical risk factors for fracture are related to falls and frailty and seem to operate similarly in men and women.6,7 Low bone density6,8- 10 and the results of serum biochemical tests6 have also been shown to predict fracture risk similarly in both sexes. The previous occurrence of fracture can be determined during patient visits or from administrative electronic databases available to governmental and managed care organizations. Such data could be used to identify patients at increased risk for osteoporotic fracture and could lead to appropriate use of interventions to reduce the number of fractures in populations at high risk.
Although the association between fracture occurrence and future fracture risk has been adequately studied in women, this association in men requires further elucidation. A meta-analysis11 of 34 studies published through September 1999 concluded that previous fracture approximately doubled the risk of future fracture in men and women; the researchers commented on the uniformity of risk between sexes and for various fracture sites. However, this summary of 34 good-quality studies contained only 4 studies12- 15 designed to examine fracture risk in men and 4 other studies16- 19 that yielded usable information on fracture risk in men after studying both sexes. Thus, a need exists for data on fracture risk after various types of fracture in aging men and on the strength of associations for different types of limb fracture.
We hypothesized that limb fracture would signal increased skeletal fragility and, thus, increased risk of future osteoporotic fracture. Therefore, we studied the associations between 4 limb fractures considered osteoporotic and the subsequent risk of similar fracture among elderly men. The database for members of a large health maintenance organization was analyzed to compare the risk of subsequent fracture among men with each of these 4 sites of index fracture with the risk among men without such a history. The analysis included adjustment for clinical confounders related to frailty and falling.
The Kaiser Permanente Medical Care Program is an integrated, group-model, nonprofit health maintenance organization serving approximately 3 million members throughout Northern California. The Kaiser Permanente Northern California Institutional Review Board approved the study protocol.
Kaiser Permanente hospital diagnoses were extracted from the health plan's Admission Discharge and Transfer system, in which are maintained records for all hospital encounters (inpatient and ambulatory surgery) and all associated diagnoses. Diagnoses for emergency department visits and office visits were extracted from Kaiser Permanente's Outpatient Summary Clinical Record. At each visit, physicians and nurse practitioners select from relevant diagnoses listed in specialty-specific forms; diagnoses checked off on these forms populate the Outpatient Summary Clinical Record database. Diagnoses for episodes of hospitalization and for emergency claims made at non–Kaiser Permanente sites were extracted from the Outside Medical Services systems, which capture diagnostic information included on billing statements.
The evolution of our study cohort is depicted in Figure 1. From our automated databases, we extracted all cases of fracture diagnosed in men between July 1, 1995, and August 31, 2001. By using these data, we identified all closed fractures of the ankle (talus or distal tibia), hip (proximal femur), humerus, or wrist (Figure 1A). We assumed that a true hip fracture would have resulted in a hospitalization; thus, to exclude spurious diagnoses, we required that hip fracture be identified from inpatient hospital stays. Similarly, we required that an ankle, humerus, or wrist fracture be diagnosed either in an orthopedics department or at a second visit for the same fracture within 60 days after the initial fracture event. The study included only men 60 years or older at the index fracture. The index fracture was defined as the first nonpathologic fracture of the ankle, hip, humerus, or wrist not preceded by any fracture of any kind during the preceding 2 years. To ensure a minimum 2 years of prior observation, we required that subjects be members of the health plan continuously during the 2 years before the index fracture. Patients who had multiple fractures occurring on the same day were excluded from the analysis to preclude the possibility of miscoding as a new fracture one that occurred concurrently with an index fracture. We also excluded from analysis persons with a history of pathologic fractures occurring before the index date. Pathologic fractures were identified by International Classification of Diseases, Ninth Revision (ICD-9) code.
The primary comparison group was randomly selected from all men who were 60 years or older on January 1, 1999, and who had continuous health plan membership but no fracture of any kind during the previous 2 years (Figure 2). The index date for these men was January 1, 1999. A second comparison group was composed of men who sustained a closed finger fracture (Figure 1B). This group was selected using the same inclusion and exclusion criteria as used for the 4 limb fracture groups, but because a closed fracture of the finger rarely requires multiple visits or a visit to an orthopedist, a single diagnosis from any source was considered sufficient for selection of the second comparison group.
Among the fracture and comparison cohorts, we excluded persons who, before the index date, received treatment known to affect risk of osteoporosis. Therefore, we excluded 3 groups of men: those diagnosed as having Paget disease, those who used the equivalent of 2 g or more of prednisone-equivalent glucocorticoid agents in the year before the index date, and those who used osteoporosis prevention–related drugs (alendronate sodium, calcitonin, etidronate disodium, or risedronate sodium) during the year before the index date. The level of prednisone equivalent was chosen because intake of 5 mg/d or more for more than 3 months is considered by experts a level of use likely to increase the risk of osteoporotic fracture.20
Our final cohort consisted of 3576 men 60 years or older with an index fracture as defined for this study in the following areas: ankle (n = 1067), hip (n = 1114), humerus (n = 470), and wrist (n = 925). The comparison group with a finger fracture consisted of 841 men, and the nonfracture comparison group consisted of 173 402 men, of whom 86 408 (about half the sample) were randomly selected for study.
For men in the fracture and nonfracture groups, we identified ankle, hip, humerus, and wrist fractures that occurred after the index dates. Follow-up from the index date ended when 1 of 4 conditions was met: (1) the first ankle, hip, humerus, or wrist fracture was sustained; (2) health plan membership was discontinued; (3) death; or (4) the end of the study period was reached (August 31, 2001). Without examining written medical records, we could not always distinguish between continuing care for a fracture and a new fracture of the same type; therefore, we considered only subsequent fractures that differed in site from the index fracture. For example, for men who had an index hip fracture, only a subsequent ankle, humerus, or wrist fracture counted as an event; subsequent diagnoses of hip fracture were ignored.
One of us (B.E.) reviewed the radiographic reports of all men with an index fracture who subsequently had a new fracture during the study period. Cases were excluded if electronically available radiology reports failed to confirm a subsequent fracture or if the presumed subsequent fracture was, in fact, concurrent with the index fracture but had not been included in the Outpatient Summary Clinical Record or the Admission Discharge and Transfer system until a later time. Most cases of incorrect coding of a presumed subsequent fracture were because of continuing care for the index fracture and seemed to be caused by physicians or nurse practitioners inadvertently checking a different fracture type on the Outpatient Summary Clinical Record diagnosis form. Therefore, errors of this type were associated with the index fracture, and we did not expect to find them for the comparison group with no index fracture.
We used a separate proportional hazards model to estimate the increased hazard of a subsequent fracture for each index fracture type. The outcome of interest was the first fracture after the index date. Data for any person were censored if the person died or otherwise discontinued health plan membership before the end of follow-up. Separate models compared persons with each type of index fracture with persons who had no previous fracture and with persons whose index fracture was of the finger.
Because our analysis excluded subsequent fractures of the same site as the index fracture, the same fracture was necessarily excluded as an outcome in the nonfracture and finger fracture comparison groups. Thus, for example, a person in the nonfracture comparison group who had a hip fracture on March 1, 2000, and an ankle fracture on August 1, 2000, was counted as having an ankle fracture when compared with men who had an index hip fracture and was counted as having a hip fracture when compared with men who had an index ankle fracture.
All proportional hazards models were adjusted for the following factors: (1) age, (2) cardiac-related drug use, (3) central nervous system–related drug use, (4) diabetes mellitus–related drug use, (5) hospitalizations, and (6) office visits. Each drug use variable was modeled dichotomously and refers to any use in the year before the index fracture. Hospitalization and visit counts were each modeled as a continuous variable and were measured in the 395 to 30 days before the index fracture. When deriving hospitalization and office visit measures, we excluded the period 30 days before the index date so as not to include health care use that might have been associated with the fracture event.
On average, men who had an index ankle, hip, humerus, or wrist fracture differed from each other and from men in the nonfracture comparison group. Men with a limb fracture tended to be older than men without an index fracture, had more hospitalizations and outpatient visits, and were more likely to have received a prescription for one of the high-risk drugs (Table 1). More men with an index hip or humerus fracture died during follow-up compared with men without an index fracture. When follow-up was calculated for persons who did not have a subsequent fracture (because follow-up
"artificially" ended for them at that point), follow-up ranged from 15 to 23 months for persons with an index fracture but was 29 months for men without an index fracture. This difference was in part explained by a higher mortality rate among some of the fracture groups (particularly those with a hip or humerus fracture) but was also due to the study chronology. Follow-up for the nonfracture group began on January 1, 1999, whereas follow-up for patients with an index fracture started at a later date (ie, when the index fracture occurred). Figure 3 shows the proportions of these groups who were free from fracture over time; the hazard function was linear for all types of fracture, and no excessive risk was observed during the early months after the index fracture occurred. The proportion of men with 1 or more subsequent fractures of a type different from that of the index fracture was higher for men after an index fracture of the hip (2.0%), humerus (5.1%), or wrist (2.5%) than for men who had a similar fracture outcome but no index fracture. Among those men, a subsequent fracture at a different location occurred in 1.1% of those who previously had a hip fracture, 1.5% of those who previously had a humerus fracture, and 1.3% of those who previously had a wrist fracture. These results, however, are based on differing lengths of follow-up and on differing baseline risk factors.
After adjusting for follow-up time, age of the patient, and baseline measures of health status, a subsequent fracture was more likely to occur in men with an index hip, humerus, or wrist fracture than in men who had either no fracture or an index fracture of the finger (Table 2). Compared with men who had no index fracture, men with an index fracture of the humerus were nearly 4 times more likely to have a subsequent nonhumerus fracture (ankle, hip, or wrist). Compared with men who had no index fracture, men with an index fracture of the hip were about 3 times more likely to have a subsequent fracture, and men with an index fracture of the wrist were about 2 times more likely to have a subsequent fracture. In contrast to men with a history of these sites of limb fracture, men with an index fracture of the ankle had no greater risk of subsequent fracture than men in the nonfracture control group.
When the ankle, hip, humerus, and wrist fracture cohorts were compared with those who had an index fracture of the finger, results were similar. Men with a fracture of the humerus or hip were still more likely to have a subsequent fracture, although the hazard ratios for this group were somewhat lower and the confidence intervals were wider. Men with a wrist fracture were also more likely to have a subsequent fracture than those with an index fracture of the finger, but this difference was not statistically significant (P = .13).
In our study of aging men, we found a strong association between a hip, humerus, or wrist fracture and the risk of a subsequent fracture, but this association did not exist for an ankle fracture. Compared with either men who had no recent fracture or men with an index fracture of the finger, men in our study who had a previous humerus, hip, or wrist fracture were about 2- to 3-fold more likely to have a subsequent limb fracture. A recent summary11 of 34 studies found a 2- to 3-fold increase in the risk of a subsequent fracture for both sexes. Most published studies on the subject have reported an association in women or in men and women, but relatively few men were studied. These studies focused primarily on a fracture of the hip or spine.
Compared with community-derived fracture rates, 243 men 35 years or older in Rochester, Minn, who had an index wrist fracture had a 2.7-fold greater risk of a subsequent hip fracture.17 A similar epidemiologic study16 of fracture rates among 212 Swedish men 40 years or older who had an index wrist fracture compared this group against a population registry of men free of fracture and found a relative risk (RR) of 2.3 of subsequent hip fracture in the study group. In cohort studies12,13 of aging men, the RR of prior fracture was estimated after multiple statistical adjustments, including an adjustment for bone mineral density. A study13 of 654 Swedish men compared those with and without a previous fracture and found that 82 men with a previous wrist fracture had an RR of 1.8 of a subsequent fragility fracture of the vertebra, wrist, humerus, pelvis, tibia, or hip, whereas 39 men with a previous hip fracture had a 1.6-fold higher RR (neither RR was statistically significant). Another similar cohort study12 of men 60 years or older living in Australia found that those with a history of nontraumatic or low-trauma fracture in the previous 5 years had a 1.4-fold increased risk of subsequent nontraumatic or low-trauma fracture during a 5-year follow-up period. In a study19 designed to examine the risk of recurrent hip fracture, men with a previous hip fracture had an RR of 3.2 of a subsequent hip fracture. Beyond hip and wrist fracture data, to our knowledge, predictive values of other specific limb fractures have not been described in men.
In our study, an ankle fracture among aging men was not associated with a subsequent limb fracture. And unlike the strong age-related relation we observed between hip and humerus fractures, we did not find any age-related increase in the rate of ankle fracture among men at our health maintenance organization. Other researchers21- 23 have suggested that ankle fractures may not indicate osteoporosis. Lauritzen and Lund21 failed to find a statistically significant association between ankle fracture and subsequent hip fracture in women. In that study,21 elbow and knee fractures showed a statistically significant association with hip fracture. Moreover, an ankle fracture in women does not show the usual relation to age and bone mineral density22,23 seen for other fragility fractures.
A major strength of our report is its large sample size, which included hundreds of fractures of each type. In addition, we were able to validate all fracture outcomes among persons with an index limb fracture. We also were able to adjust for measures of medical use, general health status, and prescription use, which we interpreted (from existing medical literature) to be related to falls or to low bone mineral density. Certain drugs are associated with a high risk of dizziness, a condition that may lead to falls and, therefore, to fracture. The risk of falls and fracture may be increased about 2-fold in persons who use psychoactive medications, including antidepressant agents24 and benzodiazepine drugs.25 However, we found that adjusting for these health conditions and medications did not consistently reduce the strength of the association between fracture and subsequent fracture, perhaps because the index fracture variable subsumed factors relating to frailty and falls.
A limitation of our study was that we did not validate fracture outcome in the large group of nonfracture controls. This omission could produce a conservative bias because possible incident fracture was removed from the limb fracture groups but not from the nonfracture controls. In addition, we found it necessary to examine only fractures differing from the index fracture because of the difficulty in distinguishing between continuing care for the original fracture and care for a new fracture of the same type; this problem is especially difficult during examination of a hip fracture, a condition for which the expected annual rate of readmission for initial fracture (about 10%) is considerably higher than the risk of new hip fracture (about 1%). Another limitation of our study was its relatively short follow-up period, which was mandated by the availability of historical data in our database; however, a linear hazard function constructed for patients in the study suggests that the RR remains stable for 1 to 2 years.
Our study was limited to limb fracture because we were unable to adequately document cases of incident vertebral fracture treated at our health maintenance organization. Postmenopausal women with a recent vertebral deformity are well known to be at substantial risk for subsequent fracture, especially in other vertebrae.26 However, only about 1 in 4 vertebral deformities observed radiologically will manifest clinically as fracture events,27 and a diagnosis of spinal fracture is frequently overlooked by those reading radiographs in the clinical setting.28 Moreover, among hospitalized women 60 years or older in whom radiographs showed moderate to severe vertebral deformity, only 8% had this diagnosis noted at discharge from the hospital.28 Thus, unless patients with a vertebral deformity are carefully scrutinized, this condition will not be discovered by clinicians or included in administrative databases. This fact is true of our health plan also and was the reason for our exclusion of spinal fractures from the present epidemiologic study. A prospective study29 of vertebral deformity in men and women based on radiologic assessment has been performed.
We conclude that among elderly men, a recent history of a humerus, hip, or wrist fracture is a statistically significant and clinically relevant predictor of future osteoporotic fracture risk but that a previous ankle fracture is not. Use of fracture history can provide a simple way to identify patients at increased risk for future fracture. Further definition of future fracture risk could be based on other clinical risk factors and on bone density. Identification and risk assessment could ultimately lead to more appropriate use of fracture risk reduction strategies, including medication. Bisphosphonates30 and parathyroid hormone31 have enhanced bone mineral density in men and hold promise for reducing the risk of subsequent fracture in men. Other types of interventions designed to reduce fracture risk—calcium with vitamin D32 or use of hip protectors33—have proved as effective in men as in women. Thus, programs aimed at identifying men with limb fractures may provide a simple low-cost means of focused fracture prevention in these high-risk, but frequently overlooked, patients.
Corresponding author: Bruce Ettinger, MD, Division of Research, Kaiser Permanente Medical Care Program, 3505 Broadway, Oakland, CA 94611 (e-mail: firstname.lastname@example.org).
Accepted for publication January 24, 2003.
This study was supported by an unrestricted educational grant from Merck & Co, Inc, West Point, Pa.
We thank the Medical Editing Department, Kaiser Foundation Research Institute, Oakland, Calif, for editorial assistance.