Signal Detection of Adverse Drug Reactions through LASSO Logistic Regression Using an Electronic Health Records Database：A Case-Control Study

Hiroshi Hayashi; Tatsuo Hiramatsu; Daisuke Koide; Katsuya Tanaka; Kazuhiko Ohe

Return

Signal Detection of Adverse Drug Reactions through LASSO Logistic Regression Using an Electronic Health Records Database：A Case-Control Study

VernacularTitle:Signal Detection of Adverse Drug Reactions through LASSO Logistic Regression Using an Electronic Health Records Database：A Case-Control Study
Author: Hiroshi HAYASHI ¹ ; Tatsuo HIRAMATSU ² ; Daisuke KOIDE ³ ; Katsuya TANAKA ² ; Kazuhiko OHE ¹
Author Information

1. Department of Healthcare Information Systems, School of Public Health, The University of Tokyo
2. Department of Healthcare Information Management, The University of Tokyo Hospital
3. Department of Clinical Epidemiology and Systems, Graduate School of Medicine, The University of Tokyo
Keywords: drug-related side effects and adverse reactions; electronic health records; pharmacovigilance; confounding factors（epidemiology); data mining
From:Japanese Journal of Pharmacoepidemiology 2017;21(2):51-62
CountryJapan
Language:Japanese
Abstract: Objective：The objective of this study was to apply Least Absolute Shrinkage and Selection Operator （LASSO）logistic regression to detection of adverse drug reaction （ADR） signals using an electronic health records database as a comprehensive and quantitative method to supplement the current pharmacovigilance activities in Japan.Design：case-control studyMethods：We analyzed data from 40767 inpatients using a single-institution hospital database and identified two ADRs, suspected pancreatitis and thrombocytopenia, using abnormal laboratory test results. LASSO logistic regression analysis was applied to detect ADR signals with adjustment for age, sex, comorbidities and medical procedures. The positive predictive value （PPV） was calculated using reference standard of known drug-ADR associations based on drug product labels.Results：The number of case group was 6735 for suspected pancreatitis and 11561 for thrombocytopenia. The number of ADR signals detected using LASSO logistic regression was 27 for suspected pancreatitis and 40 for thrombocytopenia. The calculated PPV was 3.7% for suspected pancreatitis and 55.0% for thrombocytopenia.Conclusion：LASSO logistic regression analysis efficiently detects ADR signals by adjusting for confounding factors such as comorbidities and medical procedures. The false positive signals may contain unknown signals and further signal assessment will be needed.