INTRODUCTION: Prior studies evaluating International Classification of Diseases (ICD) codes for identifying medication-related osteonecrosis of the jaw (MRONJ) have reported poor validity but were largely limited to postmenopausal women, small samples, or patients receiving bisphosphonate or denosumab for cancer-related complications rather than fracture prevention.
PURPOSE: To evaluate the performance of ICD-9 and ICD-10 codes for identifying MRONJ among adults treated with bisphosphonates or denosumab for fracture prevention.
METHODS: We included two retrospective U.S. cohorts of adults age ≥ 50 years: the national Veterans Health Administration (VHA) and HealthPartners (HP), a large nonprofit health system. All patients had dental benefits and filled ≥ 1 prescription for a bisphosphonate or denosumab for fracture prevention from 10/1/1999-12/31/2022, with follow-up through 12/31/2023. In VHA, a previously validated natural language processing algorithm (F-Measure:0.96) ascertained MRONJ events. In HP, all electronic medical records associated with candidate ICD-9/10 claims underwent clinician review. MRONJ was defined per the American Association of Oral and Maxillofacial Surgeons criteria. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value were calculated for each code in VHA; PPV was calculated in HP.
RESULTS: Among 46,408 VHA Veterans and 4,625 HP participants (64 undergoing review), 47 and 10 MRONJ cases were identified, respectively. The only code with PPV>0.40 in both systems was ICD-9733.45 (osteonecrosis of the jaw), with PPVs of 0.43 (95% confidence interval [CI]: 0.1.0-0.82) in VHA and 0.50 (95% CI: 0.07-0.93) in HP. The highest-performing ICD-10 code in both cohorts was M87.10 (osteonecrosis due to drug, unspecified bone), with PPVs of 0.27 (95% CI: 0.06-0.61) and 0.67 (95% CI: 0.09-0.99) in VHA and HP, respectively. Sensitivity ranged from 0.00 to 0.08.
CONCLUSION: Available ICD-9/10 codes demonstrate low sensitivity and PPV for MRONJ, limiting accurate ascertainment in administrative data and highlighting the need for improved hybrid ascertainment algorithms or dedicated MRONJ registries for epidemiologic study of this rare outcome.