Osteoporosis and associated fragility fractures are a major health concern and a source of substantial morbidity and mortality around the world. It was estimated that, in 2006 in the United States, the economic burden associated with hip fractures alone might have been in excess of $20 billion1-9. As the world's population increases and ages, the number of hip fractures will continue to grow. In addition to the increased mortality associated with hip fractures, the related loss of independence and function and the increased burden on family and caregivers can be onerous10-13. Given the enormous social and monetary costs of hip fractures, their prevention is a pressing concern.
It is well established that patients who have had a single fragility fracture are at significantly increased risk of having a second fracture in the future14-16. Proximal humeral fractures represent a unique subset of osteoporotic fractures; up to 73% of these fractures occur in women, making them the second most common upper-extremity fractures17-19. A proximal humeral fracture also appears to be a risk factor for the subsequent occurrence of other incident fractures, including those at the hip20,21. Johnell et al. evaluated the timing of subsequent fragility fractures after an individual had sustained an incident fracture of the spine, proximal part of the humerus, or hip. They found that the risk was highest immediately following the incident fracture and that the risk decreased with time22. However, that study was limited by the fact that the authors did not control for many important risk factors, such as bone mineral density. Still, these data suggest that a fracture of the proximal part of the humerus may be predictive of an increased risk of a subsequent hip fracture.
Interestingly, the mechanism of proximal humeral fractures tends to be more similar to that of hip fractures than to that of any other osteoporotic fracture; i.e., they occur when the individual is unable to break his or her forward or oblique fall and therefore lands directly onto the shoulder or hip23,24. Given the similar mechanisms of the fractures, it is intuitive that the timing of a hip fracture would be relatively close to the timing of a proximal humeral fracture and likely would place the patient at higher risk for sustaining a subsequent hip fracture.
The purpose of this study was to further evaluate the relationship between fractures of the proximal part of the humerus and fractures of the hip in an elderly female patient population while adjusting for bone mineral density and other important covariates. We hypothesized that individuals who sustained a proximal humeral fracture would be at higher risk for a subsequent hip fracture and that the hip fracture would tend to occur within five years after the fracture of the proximal part of the humerus.
Participants
The Study of Osteoporotic Fractures is a prospective multicenter cohort study of 9704 women sixty-five years of age and older who were enrolled from September 1986 to October 1988 in four separate geographic areas of the United States (Baltimore, Maryland; Minneapolis, Minnesota; Portland, Oregon; and Monongahela Valley, Pennsylvania). The institutional review boards at the individual study sites approved the study, and all of the women provided written informed consent. Women were recruited if they were sixty-five or older, community dwelling, and able to walk and had not had a bilateral hip replacement. A complete description of the methods of patient recruitment as well as of the inclusion and exclusion criteria have been provided previously25,26. The women were followed prospectively at regular intervals for up to ten years. The participants returned for seven examinations at approximately two-year intervals and were contacted by telephone or postcard every four months to ascertain fracture history; the follow-up rate was 99%, and the accuracy of the patients' self-report of fractures was 90%27. From 1997 to 1998, a cohort of black women was added to the study; this cohort was excluded from the present study because of the low risk of osteoporotic fractures in these women and the considerably shorter follow-up time28.
The original cohort of the Study of Osteoporotic Fractures included 9704 women, of whom 1655 (17%) were excluded from our study because of missing data regarding prior fracture status or age (116 women), lack of complete follow-up (269 women dropped out before the second examination performed for the Study of Osteoporotic Fractures), a history of a hip or humeral fracture prior to the second examination (481 women), or missing bone-mineral-density data at the time of the second examination for the Study of Osteoporotic Fractures (789 women). The second examination was used as the baseline for analysis as bone mineral density, an important risk factor, was measured at that visit but not at the first examination. A total of 8049 women (83%) were entered into our present study, and their information was used in the univariate Cox regression analyses. Fewer than 8049 women were included in some univariate Cox regression analyses because of missing values for the predictor variable. A total of 1128 (12%) were excluded from the final multivariate analysis because of missing data regarding one or more of the covariates listed below; this left 6921 women (71%) to be analyzed in the multivariate Cox regression model (Fig. 1).
Ascertainment of Incident Fractures
The study subjects were asked to notify their local clinical site after the occurrence of any fracture. Additionally, participants were contacted every four months by letter or telephone to ask if a fall or fracture had occurred in the preceding four months. When fractures were reported, the treating physician was contacted and the anatomic location and circumstances were established. Fractures were then confirmed by central review of the community radiologist's report or of preoperative radiographs as previously described by Nevitt et al.27. A fracture of the proximal part of the humerus was defined as one in the proximal one-third of the humerus. For the purposes of our study, a non-humeral fracture, non-hip was defined as a fracture of the wrist, pelvis, finger, clavicle, elbow, rib, ankle, hand, face, toe, foot, heel, leg, knee, or distal part of the femur.
Ascertainment of Variables
Candidate variables for inclusion in the multivariate analyses were identified from among common risk factors for hip and humeral fractures reported in prior studies25,29-32. At the first examination for the Study of Osteoporotic Fractures, participants were asked to self-report age, height, and nonpregnant weight at the age of twenty-five years; parental history of fractures; physician-diagnosed fractures since the age of twenty-five; estrogen therapy, currently and in the previous year; current usage of long-acting benzodiazepines; overall health status; number of falls within the last year; and number of hours spent standing or walking each day. Depth perception was assessed with use of the Howard-Dolman device and scored as the standard deviation, in centimeters, of four trials33.
At the second examination for the Study of Osteoporotic Fractures, hip bone mineral density was measured with use of dual x-ray absorptiometry (QDR-1000; Hologic, Waltham, Massachusetts). Current height (as measured with a stadiometer) and weight and the ability to rise from a chair without using the arms were documented during the physical examination.
For the purposes of our study, the second examination was considered to be the baseline examination for analysis and all covariates that were available at multiple time points, including health status as self-rated on a scale of 1 (excellent) to 5 (very poor), use of the arms to rise from a chair, walking for exercise, number of hours standing or walking each day, visual depth perception, weight gain since the age of twenty-five years, total hip bone mineral density, estrogen use, and history of falls within the last year. This was done in order to include bone mineral density, which was not measured until the second examination. All patients who sustained a fracture of the hip or proximal part of the humerus between the first and second examinations were excluded from the analysis. An average of 2.04 years elapsed between the first and second examinations. Information about the use of long-acting benzodiazepines, height at the age of twenty-five, and maternal history of hip fracture (after the age of fifty) was collected only at the first examination, and this information was used for the patients included in our study.
Statistical Analysis
Cox proportional hazards models were used to quantify the association between incident humeral fracture and the risk of subsequent hip fracture. All models were adjusted for current age and total hip bone mineral density. Each observation in the Cox regression was left-censored at the age at which the subject entered the study and either ended at the time of hip fracture or was right-censored at the end of the follow-up period.
Humeral fracture was handled as a time-dependent variable. Each participant's study period was divided into two intervals: one prior to a humeral fracture, and one following a humeral fracture and ending at the time of a hip fracture or at the end of the follow-up period. Patients who did not sustain a proximal humeral fracture were evaluated from the time of the second examination through the end of the follow-up period. Non-humeral fracture, as a time-dependent variable, was treated similarly.
Control variables that were available at multiple times were also handled as time-dependent variables. These included self-rated health status, use of the arms to rise from a chair, walking for exercise, number of hours standing or walking each day, visual depth perception, weight gain since the age of twenty-five, total hip bone mineral density, estrogen use, and history of falls within the last year. As mentioned, information about use of long-acting benzodiazepines, height at the age of twenty-five, and maternal history of hip fracture (after the age of fifty) were collected only at the first examination, and these were used as time-constant covariates. The data setup of the time-dependent variables was handled by the traditional method used for survival analysis with time-dependent variables34. Each subject's time in the study was partitioned into intervals such that the value of all variables remained constant within each time interval.
Multivariate Cox regression analysis was used to test the association between incident humeral fracture and subsequent hip fracture, with adjustment for selected control variables. Control variables were selected by forward stepwise regression (p < 0.05 for inclusion), with total hip bone mineral density and age always being included in the model. The control variables included in the final multivariate proportional hazards model were self-reported health status, height at the age of twenty-five, history of recent falls, depth perception, and prior non-humeral fracture. After selection of control variables for and running of the final model without reference to the incident humeral fracture, the incident humeral fracture variable was added into this model and the model was rerun.
The potential confounding role of a non-humeral fracture was evaluated with two methods. First, we calculated the risk due to the humeral fracture in a multivariate model that included a covariate indicating a prior non-humeral fracture after the second examination. This method controlled for the non-humeral fracture by modeling—i.e., by use of a covariate. We also calculated the risk due to the humeral fracture by using a multivariate model but excluding time intervals following any non-humeral fracture. This method controls for the non-humeral fracture by eliminating any periods of observation after such a fracture has occurred.
In order to examine whether the risk of a subsequent hip fracture attributable to an incident humeral fracture changes over the time elapsed after the humeral fracture, two multivariate models categorizing time after the humeral fracture as a time-dependent variable were used. The three post-humeral fracture intervals were less than one year, one to five years, and more than five years after the humeral fracture, with subjects not sustaining an incident humeral fracture utilized as the reference group for all analyses. The first of the two multivariate models adjusted only for age and bone mineral density, whereas the second model adjusted for age, bone mineral density, and the control variables selected in the forward stepwise selection process (described above). Each of these two multivariate models provided an estimate of the hazard ratio for hip fracture specific to each post-humeral fracture time interval (less than one year, one to five years, and more than five years) relative to the risk in subjects without a humeral fracture.
Source of Funding
No outside funding was used in support of this research. The Study of Osteoporotic Fractures is supported by National Institutes of Health funding. The following institutes provide support: the National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS) and the National Institute on Aging (NIA) under the following grant numbers: AG05407, AR35582, AG05394, AR35584, AR35583, R01 AG005407, R01 AG027576-22, 2 R01 AG005394-22A1, and 2 R01 AG027574-22A1.
In this cohort of older, community-dwelling women, an incident proximal humeral fracture significantly increased the risk of a subsequent hip fracture, with the risk being six times higher within the first year following the proximal humeral fracture after we controlled for other important risk factors in a multivariate analysis. The association between proximal humeral and subsequent hip fractures was not significant during time intervals of greater than one year after the incident humeral fracture. Although our study could have missed a modest persistent association between incident humeral and subsequent hip fractures occurring at more than one year after the humeral fracture, it was clear that the increased risk of hip fracture attributable to a prior humeral fracture sharply waned after one year of follow-up.
Few investigators have examined the risk of hip fracture after a proximal humeral fracture or, importantly, the timing of those subsequent fractures. In studies in which this relationship was evaluated20,22,35, sample sizes were small, the subjects were from a single geographic region, or the authors did not control for important risk factors when evaluating the times of subsequent fractures.
Lauritzen et al. found that women between sixty and seventy-nine years of age who had previously sustained a fracture of the proximal part of the humerus had a relative risk of sustaining a hip fracture of 2.520. However, this study included women from only one geographic area and the follow-up period was relatively short (3.7 years). This is of importance because of the geographic variability in hip-fracture incidence noted by other authors17,36. In contrast, women from four different geographic regions were analyzed in the current study. This likely increased the heterogeneity of the evaluated patient population and may make the results more representative when applied to the women of the United States as a whole.
Johnell et al. reported an increased risk of hip fracture following an incident humeral fracture, with the increased risk persisting for up to five years after the incident fracture22. The absence of a significant increase in hip-fracture risk beyond one year after the humeral fracture in our study may be related to our adjustment for additional important covariates that have been shown to increase the risk of a subsequent hip fracture, especially bone mineral density; it may also be due to either an absence of risk or a lack of power to detect a modest risk. We previously showed a modest increase in the risk of subsequent hip fracture following a non-spine, non-hip fracture after controlling for age and bone mineral density35. The adjusted hazard ratio was 1.70 (95% confidence interval = 1.30 to 2.22) during the first five years of follow-up. The risk was attenuated but still significantly increased during the five-to-ten and more-than-ten-year follow-up periods (hazard ratios of 1.32 and 1.21, respectively) in that study. The reason for the nonsignificant association between humeral fractures more than one year old and subsequent hip fracture in the current study, a finding that is in contrast to those of the previously published studies, is not clear, although our study may have been underpowered to detect a modest persistent association between humeral fractures and hip fractures occurring after more than one year of follow-up. Although the associations between prior humeral fractures as well as non-spine, non-hip fractures and subsequent hip fractures were also noted to wane with follow-up time in the study by Johnell et al.22 and our previous study35, the associations were still significant after one year in both of those studies. One can speculate that, in the initial time following a proximal humeral fracture, the effect of both environmental and medical interventions has not yet been fully realized, so the patients are still at risk for additional falls and subsequent fractures. Additionally, one can conjecture that a proximal humeral fracture and subsequent immobilization may affect the patient's ability to walk and perform activities of daily living safely. Furthermore, decreased balance and reaction time may predispose patients to future falls and fractures.
The results of the current study have important implications for the clinical evaluation, treatment, and prevention of future fractures in patients who have sustained a proximal humeral fracture. They demonstrate that the most dangerous time with regard to the risk of a subsequent hip fracture is within a year after the proximal humeral fracture and therefore intervention following a humeral fracture should be initiated without delay to reduce the risk of subsequent fractures. Studies have suggested that oral bisphosphonates begin to reduce the risk of fractures within three to six months after they are started37. In addition to the initiation of medical treatment of osteoporosis, steps should be taken to prevent falls in the at-risk population, as nearly 80% of proximal humeral fractures and 90% of hip fractures are related to falls from a standing height38,39. A recent meta-analysis demonstrated the need for a multifaceted approach to the prevention of falls in hospitals and nursing homes and that no single intervention had a significant effect in a hospital setting40. Although the patients in that study were evaluated in a hospital or nursing home setting (unlike our patients, who were community dwellers), the patients in our study also likely needed a multifaceted approach to the prevention of additional falls, whether it be assistive devices at home, adjustment of medications, or evaluation of environmental factors that lead to falls as well as the initiation of medical therapy for osteoporosis. Recent guidelines for the prevention of falls by the elderly formulated by the American Geriatrics Society, the British Geriatrics Society, and the American Academy of Orthopaedic Surgeons serve as a useful resource in the evaluation and prevention of falls in the geriatric population41. It is also important to note that the risk of a subsequent fracture is increased after a proximal humeral fracture not only in women but also in men, as noted by Ettinger et al.42.
This study has numerous strengths. To our knowledge, we were the first to assess how the association of humeral fractures with subsequent hip fractures changes over time, in an analysis adjusted for hip bone mineral density and other covariates. Second, because we used humeral fractures occurring after the second examination for the Study of Osteoporotic Fractures as a time-varying predictor, we were able to very accurately assess the time since the predictor fracture. Third, this study was of a large cohort of elderly women, in whom incident hip fractures are ascertained with 99% accuracy28. Additionally, the women in our study were from four different geographic regions of the United States so the results are likely more generalizable to the older white female population of the United States than are results of women from a single geographic region.
This study also has important limitations. Participants in the Study of Osteoporotic Fractures were community-dwelling white women sixty-five years of age and older who had volunteered for inclusion in the study. They likely represent a healthier population in comparison with similarly aged individuals living in nursing home or assisted-living environments, and our conclusions may not apply to individuals living in those settings. Because the risk of proximal humeral fractures in people of other ethnicities is typically lower than that in the white population, generalizations regarding fracture risk may not be applicable to individuals of other racial backgrounds. In addition, sex-related differences in fracture incidence make the results less meaningful for the male population.
Fracture occurrence was determined initially on the basis of self-report and was confirmed by radiographs or radiographic reports. However, this possible limitation is mitigated by the fact that ascertainment of incident hip and humeral fractures in the Study of Osteoporotic Fractures has been shown to be highly accurate27. The number of potential risk factors considered in the multivariate analysis was limited in comparison with the numbers of risk factors for hip fractures reported in previous studies. The variables were chosen in the hope of defining easily identifiable and clinically relevant risk factors that could be ascertained quickly in a clinical setting. This may have led to an overestimation of the effect of proximal humeral fractures. However, the variables used in the analysis were based on the most predictive variables reported in the current literature, and both age and bone mineral density were controlled for in all analyses. Those two variables have consistently been shown to be the most significant risk factors for osteoporotic fractures.
The data retrieved from the Study of Osteoporotic Fractures is based on self-reported questionnaires and is subject to error and patient bias. Recall bias may be of particular importance with regard to questions about height and weight at the age of twenty-five, given the time that had elapsed between when the participants were twenty-five and when they enrolled in the study. Although the rate of follow-up of patients enrolled in the Study of Osteoporotic Fractures was excellent, approximately 1700 patients were excluded from our univariate analysis because of missing or incomplete data, and an additional 1100 were excluded from the multivariate analysis. However, the remaining number of participants with complete follow-up was still quite large and allowed meaningful evaluation and statistical analysis.
In conclusion, the current study supports our hypothesis that a proximal humeral fracture is an independent risk factor for subsequent hip fracture. Importantly, the time of greatest risk is the first year following the proximal humeral fracture, with the risk of an incident hip fracture attributable to a prior humeral fracture waning sharply after that. This small window of time provides an opportunity to implement medical and environmental interventions that may decrease the risk of subsequent hip fractures and their cost to the patient and to society. 