Abstract / Summary
Aims: This study aimed to identify patients with statin intolerance and estimate the prevalence of statin intolerance among patients prescribed statins for primary prevention of cardiovascular disease using electronic health record (EHR) data. It also aimed to identify, implement, and evaluate the performance of EHR-detectable electronic phenotyping algorithms for statin intolerance. Methods: This study used the Electronic Practice Based Research Network (ePBRN) dataset. The methodology took place in four stages: (1) a literature review to identify EHR-detectable phenotyping algorithms for statin intolerance, (2) implementation of the identified algorithms in the ePBRN dataset, (3) development and implementation of a reference standard, and (4) evaluation of algorithm performance against the reference standard. Results: Six EHR-detectable electronic phenotyping algorithms for statin intolerance were identified, including the Minnesota Combined Rule-Based algorithm, Japan statin induced myopathy (SIMs), USA SIMs, Singapore SIMs (algorithms A, B, C, and D), Japan statin-associated muscle toxicity (SAMT), and NHS-UK statin intolerance pathway. The prevalence of statin intolerance in the ePBRN dataset was 5.09%. No algorithm performed consistently well across all performance measures. The Singapore SIMs-B algorithm showed the highest accuracy (57.05%), sensitivity (92.95%), negative predictive value (43.43%), and F1 (71.51%) scores, while the Japan SAMT algorithm showed the highest specificity (99.13%), and positive predictive value (76.19%). Overall correspondence with the reference standard was weak, with correlation coefficients ranging from approximately -0.02 to 0.05. Conclusion: The prevalence of statin intolerance in ePBRN dataset was at the lower end of the 5-15% estimates reported in Australia and internationally. The performances of the evaluated algorithms varied substantially across screening and classification measures, and no single algorithm demonstrated consistently strong performance. These findings highlight the challenges of identifying statin intolerance using routinely collected EHR data and support further development and validation of EHR-detectable phenotyping approaches before their use in clinical decision-support.