Detection of statin-induced rhabdomyolysis and muscular related adverse events through data-mining technique
Issued Date
2024
Copyright Date
2020
Resource Type
Language
eng
File Type
application/pdf
No. of Pages/File Size
ix, 131 leaves: ill.
Access Rights
open access
Rights
ผลงานนี้เป็นลิขสิทธิ์ของมหาวิทยาลัยมหิดล ขอสงวนไว้สำหรับเพื่อการศึกษาเท่านั้น ต้องอ้างอิงแหล่งที่มา ห้ามดัดแปลงเนื้อหา และห้ามนำไปใช้เพื่อการค้า
Rights Holder(s)
Mahidol University
Bibliographic Citation
Thesis (M.Sc. (Data Science for Health Care))--Mahidol University 2020
Suggested Citation
Patratorn Kunakorntham Detection of statin-induced rhabdomyolysis and muscular related adverse events through data-mining technique. Thesis (M.Sc. (Data Science for Health Care))--Mahidol University 2020. Retrieved from: https://repository.li.mahidol.ac.th/handle/20.500.14594/99479
Title
Detection of statin-induced rhabdomyolysis and muscular related adverse events through data-mining technique
Author(s)
Abstract
Rhabdomyolysis (RM) is a life-threatening adverse drug reaction (ADR) with the syndrome of myopathy, myalgia, and dark urine. Its mortality rate worldwide from 2004 to 2018 ranged from 7.1 to 13.0 percent per year. Statins or a combination with other drugs which increase the level of statins are the common drugs that cause RM and muscular related adverse events. Detecting RM in the early stage by finding the relationship between drugs and RM from electronic health records (EHRs) will be beneficial to patients. The study aimed to explore and estimate the association between using a statin alone or combined with other drugs and RM or muscular related adverse events by using Ramathibodi Hospital data. A retrospective cohort of statin users aged older than 18 years who had received at least one form of statin from any clinic during 2012 to 2019 were recruited. Patients were grouped based on a lipophilic property, i.e., hydrophilic (pravastatin and rosuvastatin) and lipophilic (atorvastatin, fluvastatin, pitavastatin, and simvastatin) statins. The outcome was confirmed based on 2 steps: (1) identifying with clinical diagnosis or the elevation of CK level; and (2) ascertaining by manual chart review. The dataset was splited into training and test datasets and they were separately imputed by the multiple imputations by chained equations technique (MICE). All features were transformed into categorical variables and selected by multiple processes: (1) logistic regression (LR) with bootstrapping; (2) expert's opinions; and (3) univariate analysis. The study proposed LR and Bayesian network (BN) to estimate the association between drug(s) and outcome. The study found that the overall prevalence of current outcome was 15 per 10,000 persons. Taking statin alone reduced the prevalence to 9 per 10,000 persons, whereas, taking statin combined with other drugs increased the prevalence to 19 per 10,000 persons. Common statins related to the outcome were simvastatin, atorvastatin, and rosuvastatin. LR and BN had high performance of sensitivity with 0.850 and 0.900, respectively. LR (0.802) had a statistically significant higher area under the ROC than BN (0.731). Although the BN had lower performance, it was able to overcome the limitations of the LR model with the ability to handling missing values and it was easier to analyze interaction effects. The study suggested applying the LR for confirmation of the occurred outcome, and BN for screening the outcome before happening. In the future, with the higher number of training data, the more accurate clinical reporting, and the automatic application data preparation could improve model performance.
Description
Data Science for Health Care (Mahidol University 2020)
Degree Name
Master of Science
Degree Level
Master's degree
Degree Department
Faculty of Medicine Ramathibodi Hospital
Degree Discipline
Data Science for Health Care
Degree Grantor(s)
Mahidol University