We previously reported estimates of the cost burden of fractures among Medicare beneficiaries using incremental expenditures, defined as the change in healthcare expenditures in the 6 months following a fracture relative to the 6 months prior to the fracture.1 We also reported the proportions of those expenditures derived from claims with a fracture diagnosis (“Attributable”) and the proportion of incremental expenditures that were not linked to a fracture diagnosis (“Residual”). We found that hip fractures generated, on average, $31,310 in incremental expenditures, but that only $18,734 of those expenditures was directly attributable to the fracture. Fractures are associated with utilization of a wide variety of health services and also with a variety of comorbid conditions that may be either associated with fracture risk or that may be sequelae to fractures and fracture treatment; however, claims for these services do not consistently include fracture diagnoses, leading to underestimation of the cost burden of fractures.2, 3
This study builds on the earlier work in four ways. First, it decomposes health services expenditures subsequent to hip, distal radius/ulna, and vertebral fractures in order to determine the extent to which “residual” expenditures may be directly associated with fractures. Second, it describes the extent to which these expenditures are related to underlying comorbid conditions. Third, it compares baseline expenditures prior to fracture events with expenditures for age-, race-, and sex-matched control cohorts who did not experience fractures and to cohorts experiencing stroke and myocardial infarctions. Finally, the study examines trends in expenditures in the 2 years prior to fractures to shed light on how persons who experience fractures differ from those who do not and assess the potential for specific types of expenditures to serve as indicators of frailty, thus improving our ability to predict fractures.4
Materials and Methods
We used data on a 5% sample of Medicare beneficiaries obtained from the Chronic Condition Data Warehouse.5 We restricted our analysis to beneficiaries 65 years of age or older who were enrolled in Medicare Parts A and B, and not in a Medicare Advantage plan, for at least 13 months between 1999 and 2005. We used these years to replicate the results of the previous study and to allow a closer examination of the previous findings. We examined previously identified incident hip (n = 60,354), distal radius/ulna (n = 24,347), and vertebral fracture (n = 44,120) cases, and matched each of these cases with 10 beneficiaries of the same age, race, and sex who had not experienced a fracture at the site as of the fracture index date (n = 603,097, 243,429, and 441,018, respectively). Fractures were identified using diagnosis codes from inpatient claims or from outpatient claims in which the diagnosis code was paired with a procedure code specific to repair of that fracture. Our algorithm did include pathological fracture codes. We then calculated total expenditures associated with each International Classification of Diseases, version 9, (ICD-9) diagnosis code in each month from 24 months prior to and 6 months after the fracture for fracture cases and for 24 months prior to the matching index date for controls. Thus, the units of analysis are total expenditures per ICD-9 code per month among each of the case and control cohorts. For months prior to the 12th month before the fracture, some beneficiaries were not under observation and were left-censored in those months.
We used expenditures on Medicare-covered services as proxies for costs, including payments made by Medicare and copayments and deductibles covered by either beneficiaries or by supplemental insurance plans. Expenditures associated with ICD-9 diagnosis codes were examined individually and then aggregated in the following categories: pneumonia; other infections; cancer; endocrine; hematologic; psychiatric; neurological/sensory; cardiovascular, respiratory; gastrointestinal; genitourinary; obstetrical; decubitus ulcer; other integumentary; congenital; perinatal; eye; injury/trauma; osteoporosis; joint; hip fracture; distal radius/ulna fracture; spine fracture; other fractures; other musculoskeletal; aftercare; complications; and other. The main categories simply reflect the major ICD-9 coding categories. Other specific conditions were subsequently coded: diagnosis codes specific to the fractures we were analyzing; pneumonia; joint problems; osteoporosis; complications; and aftercare. Aftercare services included rehabilitation services subsequent to fractures and postoperative care and were coded using the Clinical Classification Software system algorithm developed by the Agency for Healthcare Research and Quality.6 We presumed that incremental expenditures specific to fractures, osteoporosis, joint problems, and aftercare, along with those for pneumonia, decubitus ulcers, and complications, could reasonably be related directly to the fracture event, and that those in other categories were more likely to reflect effects of comorbid conditions. We then examined expenditures prior to the fracture, comparing cases and controls, and expenditures subsequent to fractures, by diagnosis code category, for cases. Due to the large numbers or cases and controls, even trivial differences were statistically significant at p < 0.001 levels, we do not report individual tests of significance. Construction of analytic datasets was conducted using SAS version 9.2 (SAS Institute, Cary, NC, USA) and analysis was conducted using Stata version 11 (StataCorp LP, College Station, TX, USA). The study was approved by the University of Alabama at Birmingham Institutional Review Board and by the CMS privacy board.
Table 1 shows that a much larger proportion of the incremental healthcare expenditures after fractures could be directly related to the fracture event than were previously classified as attributable (88% versus 60% for hip fractures, 72% versus 36% for spine, and 80% versus 24% for distal radius/ulna). These differences arose from expenditures on services directly related to fracture care, rehabilitation, or complications arising from the fracture or its treatment. Table 1 also shows that average healthcare expenditures prior to hip and spine fractures were much higher (75.6% and 65.4%, respectively) than average expenditures for an age-, race-, and gender-matched cohort of Medicare beneficiaries who did not experience fractures. Prior expenditures for distal radius/ulna fractures were much closer (2.6% higher) to those of the matched cohort.
Table 1. Average Total Expenditures, Pre-Fracture, for Cases and Controls, and Incremental, Attributable, and Related Expenditures Post-Fracture for Cases
Associated with a primary fracture diagnosis code.
Directly related to fracture care, rehabilitation, or complications arising from fracture or treatment.
Table 2 details how expenditures were categorized by diagnosis code and classified as likely to be directly related to hip fractures or as unrelated or uncertain in relation to fracture events. For hip fracture, every diagnostic category of expenditures increased after the fracture with the exception of those related to cancer. The largest pre-post differences in directly related care were for hip fracture care and aftercare. Similar tables can be furnished on request for spine and distal radius/ulna fractures, along with the ICD-9 codes used for each diagnostic category. For both spine and distal radius/ulna fractures, expenditures increased post-fracture for every diagnostic category.
Table 2. Health Service Expenditures (% of Totals), Pre–Hip Fracture and Post–Hip Fracture, by Expenditure Category Based on Claim Diagnosis Codes
Expenditure values are in U.S. dollars.
Finally, we examined the trends in healthcare expenditures (in constant 2007 U.S. dollars) for up to 24 months prior to fracture for cases and 24 months prior to the matching index date for controls. These trends in mean expenditures are shown in Figs. 1, 2, and 3, for hip, spine, and distal radius/ulna fractures, respectively, each compared with its own control group. Expenditures overall grew over time for every cohort, and expenditures prior to fracture were higher for cases than controls. For hip and spine fracture cases, however, the difference in expenditures was much greater than for distal radius/ulna, and the rate of increase was considerably greater, starting at approximately 12 months prior to the fracture event. We examined trends in cost by diagnostic category for hip fracture. Although cardiovascular disease and cancer accounted for the largest proportion of expenditures, expenditure rates increased the most for decubitus ulcers, pneumonia and other infections, respiratory problems, and rehabilitation care, all of which more than doubled over the 2 years prior to hip fracture.
The results of this study confirm that incremental expenditures provide a more reasonable estimate of the avoidable costs of fractures than other alternatives, such as using attributable costs or comparing fracture cases to controls. Between 72% (spine) and 88% (hip) of the difference in expenditures before and after fractures resulted from direct fracture care, aftercare, treatment of joint problems, or complications and sequelae likely to have been caused by the fracture. The remaining differences in expenditures appear associated with comorbid conditions, some of which might contribute to fracture risks.
Other studies have used administrative data to estimate the costs for fractures. The approaches used have included summing all costs after a fracture,7–9 using diagnosis codes to identify costs,10, 11 and comparing costs among fracture cases with matched controls.11–14 Our study described monthly trends in health service expenditures for incident fractures, using an analytic approach where expenditures post-fracture were compared to expenditures prior to the fracture, essentially treating cases as their own controls.1
Higher expenditures prior to fracture for cases compared to matched controls indicate that people about to experience fractures at sites most often associated with osteoporosis are much heavier users of health services than the average Medicare beneficiary of the same age, race, and sex. It is noteworthy that the differences for spine and hip fractures cases compared with controls is much greater than that found among those experiencing distal radius/ulna fractures. Moreover, the trends in expenditures over time prior to fracture suggest that an increasing trend in Medicare expenditures may provide a measure of frailty that might be of use in predicting hip fractures as well as for risk adjustment particularly related to acute exacerbations and serious consequences of chronic diseases.
This study is subject to some of the same limitations reported previously. Administrative data is subject to errors in coding and misclassification. We have taken great care in developing algorithms to identify fractures, but other diagnosis codes could be inaccurate. By restricting the analysis to paid claims, we hope to minimize this source of error. We used expenditures for Medicare covered health services as proxies for costs, thus excluding most outpatient prescription drug costs as well as costs associated with long-term care. We hope to examine these costs in a future study using claims data for Medicare Part D and from Medicaid. Another issue concerns fractures related to high levels of trauma. Despite considerable effort, we have not been successful in developing an algorithm accurate enough to exclude such fractures from the analysis. However, Mackey and colleagues15 found that even in the presence of high trauma, fractures were associated with low bone density. We did exclude open fractures of the hip or vertebrae as likely to result from major trauma, but some of the cases in our analysis a likely of traumatic origin. Finally, the difficulty in timing the precise date of fracture occurrence led to some services related to the fracture being counted in the month prior. The mean potentially attributable costs in the prior months were quite small, and we excluded the month prior to the fracture from our baseline analysis. Further research is needed to determine the extent to which a history of health services utilization and expenditures, and particularly specific patterns of expenditure, can be used to improve risks scores for fracture and for other illnesses.
MLK has provided consulting services and received research support from Amgen. KGS has received research support and has provided consulting services to Amgen, Eli Lilly, Merck, and Novartis.
This research was supported by a contract between UAB and Amgen. Only the authors from UAB had access to the Medicare data used. The analysis, presentation, and interpretation of the results were solely the responsibility of the authors. JRC is supported by the NIH (AR 05331).
Author roles: MLK, ED, JRC, DJB, KGS, and MAM contributed to study conception and design. MLK, ED, JRC, and TA collected data and and performed analyses. MLK, ED, TA, DJB, and MAM drafted the manuscript. All authors critically revised the manuscript and approved the final version. MLK takes responsibility for the integrity of the data analysis.