The American Journal of Managed Care August 2009
Distal Upper and Lower Limb Fractures Associated With Thiazolidinedione Use
This study demonstrates that patients with diabetes taking thiazolidinediones have higher proportions of distal upper and lower limb fractures than those not taking the drug.
Objective: To determine if patients with diabetes mellitus taking a thiazolidinedione experienced higher proportions of distal upper and lower limb fractures compared with those not taking a thiazolidinedione, as recent US Food and Drug Administration safety alerts suggested.
Study Design: This 3-year cross-sectional study used medical and pharmacy claims from a large southeastern managed care organization for continuously enrolled members from January 1, 2004, through December 31, 2006.
Methods: A total of 29,284 patients with type 2 diabetes mellitus aged 18 to 64 years were allocated to mutually exclusive study groups of thiazolidinedione users versus thiazolidinedione nonusers and thiazolidinedione type (pioglitazone hydrochloride, rosiglitazone maleate, or a combination). χ2 Tests were used to determine if fracture proportions for thiazolidinedione users differed from those of thiazolidinedione nonusers and if thiazolidinedione type was significant. Multivariate logistic regression models and backward stepwise elimination algorithms were constructed to evaluate associations of fracture proportions with age, sex, and chronicity of drug use for 7462 members using a thiazolidinedione.
Results: The mean (SE) fracture proportions were significantly higher for thiazolidinedione users (5.1% [0.5%]) versus nonusers (4.5% [0.3%]) (P = .03). Fracture proportions did not differ by thiazolidinedione type (P = .86). Overall, women experienced a higher mean (SE) proportion of fractures compared with men (6.0% [0.4%] vs 3.5% [0.3%]) (P <.001), regardless of thiazolidinedione use. On average, the odds of experiencing a fracture for women using a thiazolidinedione increased 2% for every year increase in age.
Conclusions: Patients with diabetes using thiazolidinediones, regardless of type, had higher proportions of distal upper and lower limb fractures compared with those not using thiazolidinediones. Fracture proportions were higher among women and increased with age.
(Am J Manag Care. 2009;15(8):491-496)
Using administrative data, we confirmed that more limb fractures occur in patients with diabetes using thiazolidinediones versus those not using this drug class.
- There were no significant differences among patients using pioglitazone hydrochloride versus rosiglitazone maleate.
- Fracture proportions were higher among women and increase with age.
- Fracture proportions for thiazolidinedione users were nonsignificant among men.
- “Time on drug” was nonsignificant in explaining variation in fracture occurrences.
A higher proportion of fractures among patients with type 2 diabetes mellitus is well documented in the recent literature.4,5 Adjusting for age, sex, smoking status, body mass index, and stroke history, patients with diabetes have an increased risk for fractures compared with patients without diabetes.6 In addition, recent research suggests that thiazolidinediones may cause bone loss by increasing bone marrow adiposity and decreasing mature osteoblasts.7,8 These facts raise concerns about the use of the thiazolidinedione drug class among patients already at increased risk for fractures.
The FDA recently released safety alerts concerning drugs in the thiazolidinedione class. Rosiglitazone use was linked to increased fractures of the upper arm, hand, and foot among female patients.9 Of 645 women in A Diabetes Outcome Progression Trial (ADOPT)10 taking rosiglitazone, 60 (9.3%) experienced a fracture during the 2-year study period. This proportion of fractures was significantly higher compared with that in female patients randomized to receive metformin hydrochloride or glyburide (P <.05). A nonsignificant increased proportion of fractures was found in male patients. In addition, pioglitazone use has been linked to increased rates of distal upper and lower limb fractures among women.11 In that clinical trial involving 15,599 patients, results showed an increased proportion of fractures for female patients taking pioglitazone versus a comparator female group receiving placebo or active nonpioglitazone treatment. Most fractures occurred in locations different from those normally associated with postmenopausal osteoporosis (eg, foot, spine and tailbone, and hip, pelvis, and upper leg). As in ADOPT, a nonsignificant increased proportion of fractures was observed
among male patients.
Although the aforementioned clinical trial results indicate a sex predilection for increased risk of fractures, they do not address the influence of age and chronicity of drug use among patients taking thiazolidinediones. In addition, it is unknown if the proportion of fractures differs for patients using pioglitazone versus rosiglitazone or if this phenomenon is simply a drug class effect. To address these concerns, 3 years of relevant medical insurance claims data were examined. The objective of this study was to determine if the proportion of distal upper and lower fractures differs for patients with diabetes using a thiazolidinedione versus those not using a thiazolidinedione. We also examined fracture rates among patients using pioglitazone, rosiglitazone, or a combination to determine if fracture proportions differ within the drug class. Finally, we determined if age, sex, and chronicity of drug use influenced the proportion of fractures among thiazolidinedione users and nonusers.
Participants and Study Groups
A 3-year (January 1, 2004, through December 31, 2006) cross-sectional member sample was established using medical and pharmacy claims data extracted from a large southeastern managed care organization’s commercial line-of-business claims data warehouse. Administrative claims data have been shown to be effective and valuable in related research,12 as they allow researchers to leverage statistical power through large sample sizes. Member enrollment data were extracted to attribute demographics, including age, sex, and length of enrollment. Age was calculated as the member’s age as of December 31, 2005 (the study period midpoint). Members aged 18 to 64 years within a commercial health plan and continuously enrolled during the study period were included for further analyses. Members older than 64 years were excluded for the following 2 major reasons: (1) a significant number of these members are covered by Medicare, whose data are incomplete and are not comparable with those of our commercial line of business, and (2) concentration on a younger population assists the analysis by partially controlling for known associations of age with bone fragility. Medical and pharmacy claims data were mined for diabetes status using McSource ViPS software (version 6.0; McSource ViPS, Inc, Baltimore, MD).13 Members were identified as having diabetes if any of the following individual conditions were met within a 24-month period: (1) 2 face-to-face encounters on different dates of service in an ambulatory or nonacute inpatient setting with a diagnosis of diabetes, (2) 1 acute inpatient or emergency department encounter with a diagnosis of diabetes, or (3) at least 1 prescription for insulin or an oral hypoglycemic or antihyperglycemic agent in an ambulatory encounter.
We excluded members with chronic corticosteroid use, type 1 diabetes mellitus, or chronic kidney disease (International Classification of Diseases, Ninth Revision [ICD-9] codes beginning with 585). Chronic corticosteroid use was defined as having filled a 6-month supply or more of any corticosteroid powder, elixir, capsule, solution, reconstituted solution, suspension, tablet, or syrup during the 3-year study period. A member was determined to have type 1 diabetes mellitus if he or she had a type 1 diabetes mellitus–related episode of care during the study period. After these criteria were applied, 29,284 members with type 2 diabetes mellitus were included in the study. We initially created the following 2 mutually exclusive study groups to determine if members with diabetes taking a thiazolidinedione of any kind experienced a higher proportion of fractures compared with those not taking a thiazolidinedione: (1) thiazolidinedione users (7462 members with diabetes who filled a pioglitazone- or rosiglitazone-containing prescription during the study period) and (2) control subjects with diabetes (21,822 members with diabetes who did not fill a thiazolidinedione prescription of any kind during the study period).
We then parsed the group of thiazolidinedione users into the following 3 mutually exclusive study groups based on the type of thiazolidinedione prescription filled: (1) pioglitazone group (2589 members with diabetes who filled only a pioglitazone- containing prescription during the study period), (2) rosiglitazone group (3908 members with diabetes who filled only a rosiglitazone-containing prescription during the study period), and (3) combination group (965 members with diabetes who filled both a pioglitazone- and rosiglitazone-containing prescription during the study period).
To ensure data accuracy, data validation methods are executed monthly for the entire data warehouse claims information and for McSource ViPS data transfer methods. If any issues arise from these validations, problems are immediately rectified, and corrected data are posted.
Distal upper and lower limb fractures were identified using 3-digit ICD-9 codes 813.x through 817.x (distal upper limb fractures, including radius and ulna, carpal and metacarpal, and phalanges and hand bones) and 823.x through 826.x (distal lower limb fractures, including tibia and fibula, ankle, tarsal and metatarsal, and phalanges and foot bones). A member was considered to have a fracture if any of these codes were included on a medical claim having service dates during the study period. It was our intent to perform tests of proportions across the study groups (fracture vs no fracture) rather than to examine the number of fractures a member may have experienced during the study period.
Chronicity of Drug Use
Pharmacy claims data were examined for chronicity of drug use by thiazolidinedione type and were represented per 100 days of drug use, where drug possession was derived from the number of supply days. “Supply days” represents the number of days a member’s prescription lasts and serves as a proxy for drug use amount over time. The duration of the study period was 1095 days; therefore, the theoretical maximum number of “per 100 days of drug use” (hereafter, time on drug) equaled 10.95.
Statistical analyses were performed using SAS 9.1 (SAS Institute, Cary, NC) with α = .05. Values are reported as the mean (SE), and proportions of fractures and odds ratios (ORs) are given with 95% confidence intervals (CIs), where applicable.
Comparison of Fracture Proportions Across Groups. To address the specific concerns of the FDA warning and to validate these events in our member population, we constructed a contingency table using a X2 test to determine if the proportion of fractures among members with diabetes using a thiazolidinedione was significantly different from that among those not using a thiazolidinedione. We then tested for significant differences across the 3 thiazolidinedione study groups (pioglitazone, rosiglitazone, and combination). This was done to address our specific hypothesis that the fracture phenomenon is not limited to a single thiazolidinedione type but rather is a drug class effect.
Age, Sex, and Chronicity of Drug Use. Multivariate logistic regression models were constructed to evaluate associations of fracture proportions with age, sex, and chronicity of drug use among 7462 members using a thiazolidinedione. We first constructed a fully-fitted model that included all main effects (age, sex, and time on drug) and interaction effects. Once it was determined from our X2 test that fracture proportions did not differ across thiazolidinedione types, this class variable was excluded from the logistic regression modeling. A backward stepwise elimination algorithm was used to remove variables from the fully-fitted model, where P >.05. If any of the interaction terms were significant, we constructed subsequent logit models to calculate appropriate adjusted ORs. Results for these models are given as ORs with 95% CIs. Model fit was tested by evaluation of the area under the receiver operating characteristic curve, where values range from 0.5 to 1 (with 1 being a perfect model fit), as well as by a Hosmer-Lemeshow goodness-of-fit test,14 where P >.05 suggests that the fitted model is an adequate model if making future predictions.