Development of a detection algorithm for statin-induced myopathy using electronic medical records

Demonstration of the utility of electronic medical records (EMRs) for pharmacovigilance (PV) has been highly anticipated. Analysis using appropriately selected EMRs should enable accurate estimation of adverse drug event (ADE) frequencies and thus promote appropriate regulatory actions. Statin-induced myopathy (SIM) is a clinically important ADE, but pharmacoepidemiological methodology for detecting this ADE with high predictability has not yet been established. This study aimed to develop a detection algorithm, highly selective for SIM using EMRs.

We collected EMRs on prescriptions, laboratory tests, diagnoses and medical practices from the hospital information system of Kobe University Hospital, Japan, for a total of 5109 patients who received a statin prescription from April 2006 to March 2009. The current algorithm for extracting SIM-suspected patients consisted of three steps: (i) event detection: increase in creatine kinase (CK) and subsequent statin discontinuation, (ii) filtration by exclusion factors (disease diagnosis/medical practices) and (iii) refinement by the time course of CK values (baseline, event and recovery). A causal relationship between the event and statin prescription (probable/possible/unlikely) was judged by review of patient medical charts by experienced pharmacists. The utility of the current algorithm was assessed by calculating the positive predictive value (PPV). In a comparative analysis, subjects screened in step 1 were extracted by the diagnostic term/code for 'myopathy/rhabdomyolysis', and the PPV of this diagnostic data approach was also estimated.

Five subjects with suspected SIM were identified using our proposed algorithm, giving a frequency of 0·1% for the adverse event. Review of the medical charts revealed that the causal association of SIM with statin use was judged as 'Likely (probable/possible)' for all five suspected patients; thus, the PPV was estimated as 100% (95% confidence interval: 56·6-100%). The higher utility of the current algorithm compared with the diagnostic data approach was also shown by assessing the PPV (100 vs. 33·3%).

We report on a detection algorithm with high predictability for SIM using EMRs. Combined use of exclusion criteria for disease, medical practice data and time course of CK values contributes to better prediction of SIM. The utility of the proposed algorithm should be further confirmed in a larger study.