Premium
Potential bias caused by control selection in secondary data analysis: Nonaspirin nonsteroidal anti‐inflammatory drugs and hemorrhagic stroke
Author(s) -
Choi NamKyong,
Hahn Seokyung,
Yoon ByungWoo,
Park ByungJoo
Publication year - 2010
Publication title -
pharmacoepidemiology and drug safety
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.023
H-Index - 96
eISSN - 1099-1557
pISSN - 1053-8569
DOI - 10.1002/pds.1903
Subject(s) - medicine , odds ratio , stroke (engine) , case control study , pharmacoepidemiology , conditional logistic regression , selection bias , confidence interval , logistic regression , neurology , nonsteroidal , emergency medicine , medical prescription , psychiatry , pharmacology , mechanical engineering , pathology , engineering
Background This study investigated the potential for bias introduction when selecting controls for secondary analysis of case‐control study data. Methods We used a data set previously collected for an acute brain bleeding analysis (ABBA) study, which was designed to investigate the risk of hemorrhagic stroke (HS) resulting from the use of phenylpropanolamine in Korea. Cases in that study had experienced an HS. Each HS case was matched with age‐ and gender‐based hospital and community controls. Information was obtained on drug exposures including nonaspirin nonsteroidal anti‐inflammatory drugs (NANSAIDs). Odds ratios (OR) for, and 95% confidence intervals (CI) of, experiencing an HS were calculated using conditional logistic regressions for each control group. Results A total of 940 patients were matched with 1880 controls. The OR of HS occurring in NANSAID users was 1.18 (95%CI, 0.80–1.73) in community controls and 0.67 (95%CI, 0.45–0.98) in hospital controls. The majority of the hospital controls were selected from patients who had visited neurology, neurosurgery, or orthopedic departments. Conclusion The difference between OR values estimated from hospital and community controls could be the result of selection bias. The study data were originally obtained for a different purpose than this study, and NANSAID use was not considered when the hospital controls were selected. When performing secondary analyses, extra care is needed to note whether the results are consistent across control groups and whether there are indications of bias related to the selection of those controls. Copyright © 2010 John Wiley & Sons, Ltd.