Introduction

The femur, the longest human bone, consists of a body (shaft or diaphysis) as well as the proximal and distal ends (Fig. 1). Most epidemiologic studies of fragility fractures of the femur have only considered the proximal aspects, described collectively as hip fractures, including transcervical, intertrochanteric, trochanteric, and rarely subtrochanteric fractures (not well defined in terms of location). There are many epidemiologic studies of hip fractures but few site-specific studies [1–4]. It is estimated that proximal fractures (i.e., hip fractures) are ten times more common than femur fractures below the lesser trochanter [5].

Relatively few epidemiological studies have considered the subtrochanteric femur (recently suggested to be a femur fracture within 5 cm distal to the lesser trochanter [6]), femoral shaft, and the lower femur. A Finnish study reported that the incidence of femoral shaft fractures (including open and closed) was 9.9 per 100,000 person-years between 1985 and 1994 [7]. It is commonly thought that these femoral shaft fractures are exclusively the result of high-energy trauma and that they occur almost solely in younger populations. However, in the Finnish study, 25% of these fractures occurred in the context of low-energy trauma (such as those resulting from a simple fall from standing position) and 74% of the low-energy fracture cases were in people aged 60 or older. The pathogenesis of low-energy fractures of the femoral shaft among the elderly is poorly understood [8]. Earlier epidemiologic studies suggested that some diaphyseal femoral fractures share similar age-related features with hip fractures, suggesting a relationship to osteoporosis. For example, a study of Minnesota residents in the US during 1965–1984 [9] found that the incidence of diaphyseal, subtrochanteric, and distal femoral fractures all increased exponentially with age among women, and 80% of the patients with fractures associated with modest trauma had prior evidence of osteoporosis or a predisposing condition. Studies in Sweden [10, 11] and Singapore [12] have also reported that the incidence of femoral shaft fractures increased with age. In addition, Bengner and colleagues [10] observed a significant increase in the age-specific incidence of femoral shaft fractures from the early 1950s to early 1980s among women over age 50. Our review has not discovered more current studies of the epidemiology of femoral shaft fractures.

Fig. 1
figure 1

Anatomy of the femur and fracture sites with ICD-9 codes

Full size image

Further information about femur fractures (especially below the hip) is needed at this time to evaluate recent publications of individual case reports and several small case series [1, 13–16]. These publications have suggested a possible relationship between long-term bisphosphonate use and subtrochanteric or femoral shaft fractures. The concern was raised because subtrochanteric and femoral shaft fractures are a relatively rare osteoporosis-related fracture, and some of the clinical and radiographic features seemed particularly striking, including prodromal pain and a transverse or short oblique radiographic pattern. Many cases were associated with long-term bisphosphonate use. However, these reports cannot establish a causal relationship because there are no control groups and no denominators of the number of at-risk patients and clinical information on patients prior to initiation of osteoporosis therapy is often scarce. Background data are needed in order to better understand the expected incidence of these fractures, which will serve as a foundation for future research on the putative relationship between long-term bisphosphonate use and some of these fractures.

Using two large databases, we performed an epidemiologic study of fractures of the entire hip and femur by eight specific subgroups: transcervical, pathological fractures of the femoral neck, intertrochanteric, trochanteric unspecified, hip unspecified, subtrochanteric, femoral shaft, and lower femur. For clarity, we have chosen to present results for four exclusive categories of fracture: hip, subtrochanteric, femoral shaft, and lower femur. Complete information for all eight fracture groups is presented in “Appendix 1,” “Appendix 2,” and “Appendix 3.” Secular trends of fracture discharge rates for each fracture group were evaluated using the National Center for Health Statistics’ National Hospital Discharge Survey (NHDS) data between 1996 and 2006. We also calculated the annual incidence of these fractures between 2002 and 2006 among subjects aged 50 and older and tested for linear trends over these years in a large medical claims database, along with a description of medical and prescription drug history in the previous year.

Methods

We assessed the secular trends of femoral fractures in the entire US population between 1996 and 2006 using public data from the NHDS. Data were downloaded from the National Center for Health Statistics website [17], where detailed documentation of the survey design and analysis was available. Annual hospital discharge rates of femoral fractures were calculated as the number of discharges divided by the US civilian population for each of the 11 years. The International Classification of Diseases (ICD-9-CM) codes were used and femoral fracture cases of each calendar year were identified. The ICD-9 codes used for the eight subgroups of hip and femoral fractures were as follows (Fig. 1): (1) transcervical (including 820.00, intracapsular section, unspecified; 820.01, epiphysis; 820.02, midcervical section; 820.03, base of neck; and 820.09, other), (2) pathological fracture of neck of femur (733.14), (3) intertrochanteric (820.21), (4) trochanteric, unspecified (820.20), (5) hip fracture, unspecified part of neck of femur (820.8), (6) subtrochanteric (820.22), (7) femoral shaft (including 821.00, unspecified part of femur; 821.01, shaft), and (8) lower femur (including 821.20, lower end, unspecified; 821.21, condyle; 821.22, epiphysis, lower; 821.23, supracondylar; and 821.29, other). All of the ICD-9 codes above indicate closed femoral fractures since open fractures are usually related to major trauma and therefore excluded (as identified by inpatient or outpatient ICD-9 codes 821, “820.3x,” “820.9,” “821.1x,” “821.3x,” “822.1,” and ICD-9 procedural codes v54.13-v54.15; aftercare for healing traumatic fracture of leg or hip). Furthermore, we excluded those with Paget’s disease (ICD-9 731.0) or malignancy (ICD-9 140-208). Because patients with malignancy and Paget’s disease diagnosis codes were excluded, the meaning of the diagnosis of “pathological fracture of neck of femur” was uncertain but likely included patients with osteoporosis and therefore this subgroup was retained. Analyses were based on four categories: (1) hip (excluding subtrochanteric), which included five subgroups: transcervical, pathological fracture of neck of femur, intertrochanteric, trochanteric, unspecified, and hip fracture, unspecified part of neck of femur, (2) subtrochanteric, (3) femoral shaft, and (4) distal femur.

The units for the NHDS discharge rates were the number of hospital discharges instead of the number of patients; this is because the same patient may have had multiple hospitalizations with the same provider within the same year. When using the NHDS data, we were unable to exclude prevalent fractures; therefore, the discharge rates may reflect not only how fast first fractures occur (i.e., incidence) but also the incidence of recurrent fractures. We did not obtain confidence intervals for the standardized NHDS discharge rates, which were considered to be population benchmarks. Linear trends of the standardized incidence and NHDS discharge rates were analyzed by simple linear regression of the rates (on the natural log scale) on the calendar years.

Data in the present study were also derived from the 2002–2006 MarketScan® Commercial Claims and Encounters and Medicare Supplemental and Coordination of Benefits databases, which obtained administrative data from approximately 45 large employers, health plans, and government and public organizations. As of December 2006, the MarketScan data represented the medical claims experience of 37 million people of which 15 million were currently eligible for a medical claim. MarketScan data were representative of the age and geographical distribution of the whole US population, with members residing primarily in the South (40%), as well as in the West (26%), Midwest (24%), and Northeast (10%). All enrollment records and inpatient, outpatient, and drug claims were collected. Diagnosis codes in the MarketScan databases use the ICD-9-CM, and procedure codes mainly use the Current Procedural Terminology, fourth edition system. Each inpatient admission records one principal ICD-9 diagnosis code and up to 15 secondary diagnosis codes.

We evaluated fracture incidence among the MarketScan enrollees aged 50 and older that had a minimum of 1 year’s enrollment history for each of the 5 years between 2002 and 2006. Fracture incidence in each calendar year was calculated as the number of fracture cases divided by the total person-years of follow-up during that year. The person-year contribution from each individual enrollee was considered from January 1st of each year until the date of the first femoral fracture, date of disenrollment, or December 31st of that year, whichever came first.

Femoral fracture cases of each calendar year were identified from inpatient admission records, using the ICD-9 codes for the eight subgroups of hip and femoral fractures as described above. To focus on incident or new fractures of the femur, we also excluded femoral fracture cases that had a femoral fracture during the prior 12 months (ICD-9 codes 820, 821, 733.14, 733.15) since these were recurrent fractures. Fracture incidence rates were calculated by subgroups for each year between 2002 and 2006. To account for any differences in the age structure between the study subjects in the MarketScan and the US population over different years, we used the US year 2000 age distribution as the standard and calculated directly standardized rates. The following age groups were used: 50–54, 55–59, 60–64, 65–69, 70–74, 75–79, 80–84, 85–89, and ≥ 90. Confidence intervals for the standardized incidence rates were computed using Dobson’s methods [18].

To begin to understand patient characteristics associated with different femur fractures, we randomly selected five noncases for each fracture case in the MarketScan database, matching on gender and age in the same year. For each fracture case, the date of fracture hospitalization was designated as the index date, and the five matched noncases were assigned the same index date. We then pooled fracture cases of the same subgroups from all 5 years to compare the cases and their matched noncases with respect to their medical and prescription drug history during the 1 year before the index date. Less than 0.1% of fracture cases had repeat femoral fractures more than 1 year apart; in those instances, only the earliest fracture was considered so that the case pool included unique patients. We presented a total of 12 medical conditions and nine prescription drug categories. We compared the percentages of patients having these conditions during the previous year, where the percentages were directly weighted by the age structure of all fracture cases combined. Similar data were not available from the NHDS.

Results

Table 1 shows the annual age-standardized hospital discharge rates using the NHDS data for the four major femur fracture subgroups. Specific rates for the other subgroups are found in “Appendix 1.” In women, the overall hospital discharge rates of hip fracture decreased from about 600 to 400 per 100,000 woman-years between 1996 and 2006. During the same 11 years, subtrochanteric, femoral shaft, and lower femur fracture discharge rates remained stable at approximately 20 per 100,000 women. The corresponding fracture groups in men showed similar trends, with lower rates compared to women at each femur site. The decreases in hip fracture rates occurred primarily during the late 1990s. During more recent years, 2002–2006, using MarketScan data, there were no significant trends in any of these fracture groups.

Table 1 Hospital discharge rates of closed femoral fractures, NHDS 1996–2006
Full size table

The study subjects from the MarketScan database contributed between 1.6 and 3.9 million person-years for each of the 5 years. There was a substantial overlap of the study cohorts by calendar year, especially between adjacent years, where up to 89% of the cohort could overlap. However, the overlap of fracture cases was negligible (<0.1%). The overall distribution of age and sex in these cohorts was also very similar; the mean age was about 63, and just over half (~56%) of the study subjects were women. Similar to the NHDS data, the fracture incidence rate in women was higher than in men for each fracture type. Overall, about 75% of femoral fracture cases were women and the average age was 80 years old. The femoral shaft and lower femur fracture cases were generally about 5 years younger than the other fracture subgroups; however, the subtrochanteric fracture cases had an average age similar to those in the hip region. The percentages of fracture subgroups among cases from each of the 5 years were similar; hip fractures collectively accounted for 87% of all femoral fractures. Subtrochanteric cases made up of about 3% of all femoral fractures, and femoral shaft and lower femur fracture cases each accounted for about 5% of all femoral fractures.

Table 2 shows the directly standardized annual incidence rates for the four groups of femoral fractures in the MarketScan database. Results for the other separate groups are shown in “Appendix 2.” There was no suggestion that any of the hip fracture subcategories were different from each other; therefore, analyses presented are for pooled groups. There were no statistically significant trends in either overall femur or hip fracture rates (or incidence rates for any of the eight subgroups) over the 5 years. The standardized incidence rates of all subgroups remained stable over the 5 years among men and women combined (data not shown) or separately. The overall ranking of the incidence rates of the eight subgroups of fractures is similar to the ranking of the discharge rates in the NHDS, but the absolute magnitude of the discharge rates was in general about 30% higher in the NHDS than the MarketScan incidence rates. The fracture rates for the eight fracture subgroups varied by up to 50-fold with intertrochanteric fracture being the most common type and pathological fracture of the femoral neck the least common type (see “Appendix 1” and “Appendix 2” and “Discussion” below) in both men and women. Incidence rates of femoral shaft and lower femur fractures were greater in women than in men, although this was not always the case for subtrochanteric fractures. The annual incidence for hip fractures in women was less than 296 per 100,000 and less than 182 per 100,000 in men between 2002 and 2006. The annual incidence for subtrochanteric, shaft, and lower femur fractures, respectively, was lower than 12, 14, and 18 per 100,000 in women and lower than 7, 11, and six per 100,000 in men throughout the period 2002 to 2006.

Table 2 Incidence rates of fragility fractures, MarketScan 2002–2006
Full size table

Figure 2 shows that the incidence rates of the four fracture groups increased exponentially with age (shown as nine age categories) in women and men, respectively. Assuming that fracture incidence follows Poisson distribution with a rate parameter λ, the fitted models for λ as a function of age in all eight fracture subgroups can be found in “Appendix 3.”

Finally, we compared the fracture cases with their matched noncases from the MarketScan databases, focusing particularly on four categories of fracture cases: hip, subtrochanteric, femoral shaft, and lower femur. Table 3 shows the age-adjusted percentages for each of the medical conditions and prescription drug use during the 1 year before the index date. Overall, the fracture subgroups appeared similar in comorbidities and prescription drug use in the year before fracture. Conditions known to predispose to fracture (such as previous fractures, osteoporosis, rheumatoid arthritis, other musculoskeletal diseases, diabetes, chronic obstructive pulmonary disease, renal disease, Alzheimer’s and other mental illness, cardiovascular disease, and use of glucocorticoids, antidepressants, and proton pump inhibitors) were quite common for all cases and appeared to be more common than their matched noncases in both women and men. The percentage of bone density tests (approximately 10% in women and 4% in men) were similar between cases and noncases (data not shown).

Fig. 2
figure 2

Age-specific incidence rates in men (X) and women (O) for the four fracture groups (MarketScan)

Full size image
Table 3 Medical and prescription drug history (percentage) during the 1 year before index date
Full size table

Discussion

In females, hospital discharge rates for hip fracture decreased from 1996 to 2006, from about 600 to 400 per 100,000 person-years, whereas subtrochanteric, femoral shaft, and lower femur discharge rates remained stable during this time, each at about 20 per 100,000 person-years. Similar trends existed in males, but, as expected, the corresponding rates were lower as compared to females. In the more recent years of 2002 to 2006, using the medical claims database, there were no trends in the incidence rates of fragility fractures for any of the femur fracture categories. The incidence of all femur fracture types increased exponentially with age in both men and women. Medical conditions and use of medications that predispose to fracture were seen more commonly in fracture cases as compared to controls.

The strengths of the present study include the large sample size that enabled us to calculate incidence rates for specific fracture subgroups. To the best of our knowledge, our study is the first to directly compare the incidence rates of all major subgroups of femur fractures. The decreases seen in overall hospital discharge rates of hip and femur fractures in the US over the 11 years from 1996 to 2006 is similar to the findings of Gehlbach and colleagues [19] who reported a decreasing secular trend in the age-adjusted rates of hip fracture hospitalizations in the US during 1993–2003. In contrast, we found that, during the most recent 5 years (2002–2006), the overall age-standardized incidence rates of these fractures remained stable among patients aged 50 and older in a large claims database. The relative frequency across subgroups was also stable over these years. Our study did not find any increasing trend, either in the incidence or hospital discharge rates for any of the subgroups of femoral fractures, in particular shaft or subtrochanteric fractures. Recent case studies raised the question of whether there had been an increase of subtrochanteric or femoral shaft fractures, particularly those with a transverse or short oblique pattern [5, 13–16]. Although transverse or short oblique type femoral shaft fractures may account for about 50% of all femoral shaft fractures (oblique 37% and spiral 23%) among skeletally mature patients [7], we could not evaluate the number with this specific radiographic pattern in this study. If we assume that the transverse pattern had increased consistently over the time periods we evaluated and that all other patterns had either remained stable or increased consistently, we could have detected a significant trend of 8% increase overall (20% increase in transverse or short oblique type) in this database. Our data provided evidence that the overall fracture rates at these sites did not increase from the time of the introduction of bisphosphonate drugs in the mid-1990s until most recently. However, without specific radiologic or clinical information about the fractures in either MarketScan or NHDS databases, we were not able to infer pattern-specific (e.g., transverse versus spiral radiologic pattern) trends within the same fracture subgroups. For example, if the incidence of the transverse fracture pattern had increased while the spiral fracture pattern had decreased, these opposing trends of different patterns may have been masked leading to an overall stable rate in subtrochanteric and shaft fractures. In addition, the potential inconsistencies in coding for the different nonhip sites (subtrochanteric versus femoral shaft versus lower femur) can also lead to a difficulty in finding any trends.

On the other hand, each of the four subgroups of femoral fractures increased exponentially with age, despite the variable incidence among subgroups. Therefore, these femoral fractures may overall share common etiology with osteoporosis. Diseases that increase with age exponentially may also involve more complicated etiologies, such as general frailty and other comorbidities [20]. It was apparent from Table 3 that the fracture cases overall shared medical and prescription drug history compared with noncases.

Although we found that intertrochanteric fractures were the most common of all femoral fractures, if we combined hip fracture, unspecified (ICD-9 820.8), with transcervical fractures, then these would be the most common (“Appendix 2”). One previous study found that most hip fractures, unspecified (>80%), were in fact of the transcervical type [2, 21]. We did not combine these subgroups together because we could not verify the actual location of the unspecified fractures. Besides ICD-9, other classification systems have also been used to classify femoral fractures; the AO system in particular has been recommended for planning treatment and predicting outcome [22]. However, radiologic data were not available in order to use these alternative classification systems. Although misclassification of the fracture subgroups is possible, the overall relative frequency of the subgroups was similar to these previous studies [2, 21, 22].

The incidence rates from the MarketScan data were lower than the US NHDS for the same calendar years. The reasons for this are unknown. One possible reason may be that the MarketScan population with private insurance may be healthier than the general US population. Another reason may be that we excluded prevalent fracture cases and applied additional exclusion criteria (malignancy, Paget’s disease, and <1-year enrollment history) in the MarketScan database. Therefore, the MarketScan incidence rates could only measure how many new femoral fractures occurred among subjects free of previous or recent femoral fractures during a specified time, while the NHDS counted the total hospitalizations of femoral fractures among all subjects, including new and repeat fractures, as well as multiple hospitalizations for the same fracture. In the MarketScan data, we may have underestimated the incidence of (new) contralateral fractures that occurred in the same patients within 1 year; we did not have detailed data as to which leg (i.e., left or right) was fractured. Early studies reported that up to 83% of noncontemporary bilateral fractures of the hip were of the same type as the initial fracture [23]. About 7% of patients admitted for a hip fracture have in the past sustained a fracture of the contralateral hip [24]. Furthermore, refracture rates may differ between fracture subgroups. Dretakis and colleagues [24] found among patients with bilateral fractures that 92% of the trochanteric and 68% of the cervical fractures were followed by a second hip fracture of the same type. However, the impact should be small because the average time between the bilateral hip fractures in different studies ranged from 3 years to up to 7 years [24].

Although we described fracture cases in the particular subgroups of interest with respect to a list of recent (within 1 year) medical and prescription drug history (Table 3), the comparisons across subgroups were descriptive and exploratory, and we did not conduct statistical tests for any particular etiological hypothesis. For example, although bisphosphonate use in the past year appeared more common among cases of any of the fracture subgroups compared with the noncases, it cannot be presumed that these drugs were causing fractures, based on the descriptive nature of the data. In addition, we did not examine the duration or type of bisphosphonate used for the cases or noncases. A more prolonged enrollment history that allows a more complete analysis would be required to evaluate any association. The likely explanation may be that fracture cases had been more prone to fractures in the first place and initiated bisphosphonate therapy to reduce the risk of future fractures. A formal analysis of these subgroups of fracture cases in terms of long-term medical and drug history is beyond the scope and intent of the present study.

However, our findings provide useful information for the planning of future studies of the possibly unique clinical and radiographic femur fractures reported in recent case studies. The overall incidence of subtrochanteric and shaft fractures combined was less than 30 per 100,000 person-years. Given the fact that femoral shaft and subtrochanteric fractures are related to age and possibly osteoporosis, many potential confounders need to be considered. Confounding by indication is a particular challenge.

Previous studies have suggested that the specific site of femoral fractures may be related to patient-level variables, such as nutrition [25] and morphology of the femur [26, 27], including hip axis length and cortical thickness. A review article by Mautalen and colleagues [26] showed that women with trochanteric hip fractures may be older, thinner, and shorter, have lower bone mass at the proximal femur and spine, and have a more severe alteration of trabecular bone properties than women with femoral neck hip fractures [26]. In addition, in that cohort, previous vertebral fractures were up to twice as common in women with trochanteric fractures compared to those with femoral neck fractures. Subtrochanteric/diaphyseal femur fractures were found to share the epidemiology and treatment response of classical hip fractures and the authors concluded that they are best classified as osteoporosis-related fractures [28]. More research is needed to identify distinctive characteristics of different femur fracture subgroups.

To our knowledge, this is the first study to estimate the incidence and secular trends of all specific femoral fracture sites, as well as describe some of the patient characteristics. This information should be useful for future etiologic research. In summary, our study did not find any increasing trend in the incidence and hospital discharge rates in any individual or subcategory of femoral fractures. Hospital discharge rates either remained stable (subtrochanteric, femoral shaft, or lower femur) or decreased (overall hip fractures) in recent years. As expected, fracture cases had a greater rate of known conditions predisposing to osteoporosis than noncases. Although some conditions appeared more common in specific subgroups, all fracture subgroups displayed an exponentially increased rate with age, and predominance in women suggesting a possible common osteoporosis-related etiology. Furthermore, the incidence of subtrochanteric and shaft femur fractures is very low, accounting for less than 10% of all femur fractures. No real conclusions can be drawn about additional predisposing factors at this time.