バイアスと交絡：医療情報データベースを使った薬剤疫学研究

抄録

  The provision of health care frequently creates digitalized data such as hospital-based electronic data, medication prescription records, and claims data collectively termed “administrative database research”. The data source and analytical opportunities for study create risks that can lead to misinterpretation or bias the results. This review serves as an introduction to the concept of bias and confounding to help researchers conduct methodologically sound pharmacoepidemiologic research projects using administrative databases. Beyond general considerations for observational study, there are several unique issues related to database research that should be addressed. The risks of uninterpretable or biased results can be minimized by: providing a robust description of the data tables used; focusing on why and how they were created; measuring and reporting the accuracy of diagnostic and procedural codes used; and properly accounting for any time-dependent nature of variables. The hallmark of good research is rigorously careful analysis and interpretation. The promise for value of real world evidence using databases in medical decision making must be balanced against concerns related to observational inherited limitations for bias and confounding. Researchers should aim to avoid bias in the design of a study, adjust for confounding, and discuss the effects of residual bias on the results.