Abstract
Background/Aims Despite the problem of misclassification, many studies of adverse drug reactions (ADRs) rely on administrative data rather than events validated by medical record review. We evaluated the new ICD-9 code for rhabdomyolysis (728.88) and natural language processing (NLP) as methods to identify cases of statin-related rhabdomyolysis, and estimated the incidence with various statins and doses.
Methods We conducted a population-based study of statin users in Group Health Cooperative from 2006–2010. Person-years of statin use among all enrollees by statin and dose were estimated using computerized pharmacy records. Trained abstractors reviewed the medical records of selected statin users to identify cases of statin-related rhabdomyolysis and myopathy, defined as muscle injury with a peak creatine kinase level = 10x and 5–10x the upper limit of normal, in the absence of other likely etiologies.
Results Review of medical records for 361 statin users with a qualifying administrative code yielded 24 cases of statin-related rhabdomyolysis and 12 of myopathy. The positive predictive value of the rhabdomyolysis ICD-9 code was only 7.5% (22/292). NLP methods identified another 5 cases of rhabdomyolysis and 6 of myopathy. Using validated cases, the rhabdomyolysis incidence rate (IR) per 100,000 person-years of statin use was 13.2 for simvastatin compared with 5.2 for users of other statins (IRR 2.61; 95% CI, 1.03–7.84), and 64.8 for the 80 mg/day dose of simvastatin compared with 5.3 for the 20 mg/day dose (IRR 12.2; 95% CI, 3.6–52.3). In an analysis using only the ICD-9 code for rhabdomyolysis, the IRR for simvastatin compared with other statins was 1.03 (95% CI 0.80–1.34), and for 80 mg/day dose of simvastatin compared with 20 mg/day it was 1.77 (95% CI 1.05–2.88).
Discussion Use of the administrative diagnostic code for rhabdomyolysis was highly nonspecific for this ADR and resulted in weaker associations than methods that verified ADRs with medical record review. The use of administrative data alone in surveillance studies of other medication ADRs with multiples causes, such as drug-induced liver injury, may fail to detect actionable and clinically important harms.
