Validation of Diagnostic Codes for Subtrochanteric, Diaphyseal, and Atypical Femoral Fractures Using Administrative Claims Data

Abstract

Administrative claims databases have large samples and high generalizability. They have been used to evaluate associations of atypical femoral fractures with bisphosphonates. We developed and assessed accuracy of claims-based algorithms with hospital and physician diagnosis codes for these fractures. Medical records and radiology reports of all adults admitted at University of Alabama at Birmingham Health System from 2004 to 2008 with International Classification of Diseases, Ninth Revision hospital discharges and surgeons’ fracture repair codes for subtrochanteric femoral fractures and random sample of other femoral fractures were reviewed. We identified 137 persons with suspected subtrochanteric femoral fractures and randomly selected 50 persons with either suspected diaphyseal femoral fractures or hip fractures other than subtrochanteric and diaphyseal femoral fractures (typical hip fractures). Eleven patients had radiographic features indicative of atypical femoral fractures. The positive predictive value (PPV) of claims-based algorithms varied with primary or secondary positions on discharge diagnoses and the sources of diagnosis codes. The PPV for fractures ranged 69–89% for subtrochanteric femoral, 89–98% for diaphyseal femoral, and 85–98% for typical hip fractures. The PPV of administrative codes for defining a femoral fracture as atypical was low and imprecise. Claims-based algorithms combining hospital discharges with surgeon’s diagnosis codes had high PPV to identify the site of subtrochanteric or diaphyseal femoral fractures vs typical hip fractures. However, claims-based data were not accurate in identifying atypical femoral fractures. These claims algorithms will be useful in future population-based observational studies to evaluate associations between osteoporosis medications and subtrochanteric and diaphyseal femoral fractures.

Introduction

Bisphosphonates remain the first-line therapy for most patients with osteoporosis (1). These drugs increase bone mineral density, reduce biochemical markers of bone turnover, and decrease the number of osteoporotic fractures at both vertebral and nonvertebral sites 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13. An increasing number of reports suggest association between bisphosphonate use and fractures of the subtrochanteric and proximal diaphyseal regions of the femur after minimal or no trauma. The occurrences of these fractures with minimal trauma, their uncommon anatomic position, and the other unusual radiographic features of these fractures have helped coin the term “atypical femoral fractures.” Multiple case reports and small case series have raised concern about a possible association of such fractures with long-term bisphosphonate use 14, 15, 16, 17, 18, 19, 20, 21, 22. On the other hand, a registry-based cohort study in Denmark that used administrative data showed no difference in the incidence rate of hip and subtrochanteric/diaphyseal femoral fractures between alendronate users and a matched comparison group (23). Secondary analyses using results of 3 large randomized bisphosphonate trials also demonstrated very rare occurrences of subtrochanteric or diaphyseal femoral fractures and showed no significantly increased risk even among women who were treated with bisphosphonates for nearly 10 yr (24). Recently, 2 population-based case-control studies in Canada and Sweden, also relying on administrative data, demonstrated an association between long-term bisphosphonate use and subtrochanteric or femoral shaft fractures 25, 26.

Because of cost, feasibility, and generalizability concerns of conducting long-term safety studies using randomized controlled trials, large observational studies are commonly used to clarify these associations. Administrative claims databases often provide a unique population base useful for such studies. However, the accuracy of administrative claims to identify certain fracture types is not well established. Although administrative claims data have been used to define some common types of fractures 27, 28, 29, 30 and those with radiographic features considered atypical 17, 20, 31, to our knowledge, few studies have assessed the positive predictive value (PPV) of subtrochanteric and diaphyseal femoral fracture codes. Ray et al (32) reported a PPV of 75% for the diaphyseal femoral fracture International Classification of Diseases, Ninth Revision (ICD-9), Clinical Modification code. In contrast, Spangler et al (33) recently reported PPV for the various diaphyseal femoral fracture ICD-9 subcodes that ranged from 20% to 83% and a PPV of 50% for subtrochanteric femoral fracture ICD-9 code. In another cohort of women aged 68 yr or older, ICD-10 codes had a 90% PPV (95% confidence interval [CI]: 88–92) for patients hospitalized with subtrochanteric and diaphyseal femoral fractures (25). To develop and validate claims-based algorithms for identification of closed subtrochanteric and closed diaphyseal femoral fractures, closed hip fractures other than subtrochanteric and diaphyseal femoral fractures (typical closed hip fractures), and atypical femoral fractures, we assessed the accuracy of hospital and physician diagnosis codes to identify these fractures.