The Use of Administrative Data and Natural Language Processing to Estimate the Incidence of Statin Related Rhabdomyolysis FLOYD
The use of administrative data and natural language processing to estimate the incidence of statin- related rhabdomyolysis James Floyd, MD, MS HMORN Conference May 3, 2012
Background: Statins and Rhabdomyolysis• Statins – Reduce the risk of cardiovascular events and death – Can cause a spectrum of muscle injury• Rhabdomyolysis – Other causes: immobility, arterial ischemia, surgery – Rhabdomyolysis related to statin use occurs about once per 10,000 person-years of statin use
Background: Simvastatin• SEARCH:1 Secondary prevention trial comparing simvastatin 80mg/day vs 20mg/day – Rhabdomyolysis RR 26• FDA safety announcement: June 8, 2011 1. Lancet. 2010;376:1658.
Background: Study of Rare ADRs• Spontaneous adverse event reports: FDA AERS1 – Incomplete reporting of cases – No information about denominators• Administrative data in large health plans2 – Difficult to identify “statin-related” cases – Among statin users in 11 health plans, only 24/194 (12%) of potential cases were validated• Rhabdomyolysis ICD-9 code introduced in 2006 1. Staffa JA. NEJM. 2002;346:539. 2. Graham DJ. JAMA. 2004;292:2585.
Aims• Aim #1: Evaluate use of the new ICD-9 code for rhabdomyolysis as a method of identifying cases of statin-related rhabdomyolysis• Aim #2: Determine whether the markedly increased risk of rhabdomyolysis associated with high-dose simvastatin use can be detected using these methods
Methods• Setting: Group Health Cooperative, 2006-2010 – Electronic medical record introduced in 2005• Statin use estimated from computerized prescription data• Statin-related rhabdomyolysis: Muscle symptoms with peak creatine kinase (CK) level ≥ 10x ULN, no other cause
Methods• Rhabdomyolysis ICD-9 code (728.88)• Other methods: – ICD-9 code for adverse event of a lipid agent – CK level > 5x ULN in GHC laboratory database – Natural language processing (NLP)• Incidence rates estimated from cases divided by person-years of statin use – One set of cases identified only by rhabdo ICD-9 code – Second set of cases validated by EMR review
Results: Case Identification Validated Reviewed CasesCase identification method N N %Rhabdomyolysis ICD-9 292 22 8%Other criteria AE of lipid agent ICD-9 30 1 3% CK > 1000 IU/L 39 1 3% Natural language processing 438 5 1%Total, all methods 799 29
Results: Characteristics of Cases Validated Cases N=29Age, median (range) 73 (53-87)Female 18 (62%)Hospitalized 26 (90%)Renal failure 8 (29%)Hemodialysis 2 (7%)Death 0 (0%)Creatine kinase, median (range) 7,450 (1,477-150,510)
Summary of Findings• Poor positive predictive value of rhabdomyolysis ICD-9 code: 8%• NLP detected additional cases• Use of administrative data without medical record review may fail to detect important harms• Confirmed in a community setting the increased risk with high dose simvastatin in SEARCH trial
AcknowledgementsCardiovascular Health Research Unit, University of Washington Bruce Psaty Susan Heckbert Noel WeissGroup Health Research Institute, Group Health Cooperative David Carrell Eric LarsonNHLBI T32 Training Grant PI David SiscovickNHLBI grants HL078888 and HL085251 PI Bruce Psaty