A structured search strategy was developed using relevant articles on file. We searched MEDLINE for English-language articles published from 1966 through April 2005 using the following MeSH (medical subject heading) terms: “medication history taking,” “medication errors,” “physicians,” “pharmacists,” “prescription medications,” “pharmaceutical preparations,” “hospital medication systems,” “hospital pharmacy services” and “medical records.” The search strategy was deliberately broad to ensure inclusion of the maximum number of relevant articles (details of the search strategy appear in on online appendix, available at www.cmaj.ca/cgi/content/full/173/5/510/DC1
). All bibliographies of papers identified in the search were screened for additional articles, and this was done subsequently for all papers retrieved. We searched the EMBASE and CINAHL databases using a similar search strategy.
Two of us (V.C.T. and R.M.) identified relevant articles for retrieval by screening the titles, abstracts and subject headings of the MEDLINE citations for the following inclusion criteria: primary research article; comparison of physician-acquired medication histories (chart notes, admission orders or medication administration record) with comprehensive medication histories; adult inpatient population; and sample size of at least 30 patients. Full-text versions of the identified papers were retrieved and screened again by the 2 independent readers for the inclusion criteria.
All included articles were independently reviewed by 3 of us (V.C.T., N.F. and E.E.E.) for methodological features and results. Any discrepancies were resolved through discussion.
The reviewed studies were analyzed on the basis of their explicit descriptions of prospective or retrospective design, use of consecutively enrolled patients and adequate blinding. Quality grades were assigned as follows: grade A studies had prospective enrolment of consecutive patients and a sample size of at least 100 patients; grade B studies had prospective enrolment of consecutive patients and a sample size of less than 100; grade C studies included all other designs.
We sought data on prescription medication histories obtained by physicians at the time of hospital admission. Such data could include physician admission notes, admission medication orders or medication administration records. We also recorded data on the main comparative measure, which was usually a comprehensive medication history completed by a pharmacist. The comparative measure could have included a patient interview, a review of the physician's admission notes or admission medication orders, a review of medication lists, and contact with community pharmacists and physicians. An error in a prescription medication history was defined as a discrepancy between the medication history obtained by the physician and the comprehensive medication history. We also recorded discrepancies for nonprescription medications, allergy history and prior adverse drug reactions, when reported. Discrepancies between physician-acquired medication histories and comprehensive medication histories are not necessarily errors. Some “discrepancies” may be intentional therapeutic adjustments of the patient's usual medications by the treating physician. Physicians may choose to discontinue a specific medication, or adjust its dose, without documenting a reason in the chart. Therefore, we sought evidence from the studies of discussions with ordering physicians to distinguish intentional from unintentional discrepancies.
Certain medication history errors have more potential for harm than others. A reduction in laxative dose may have less consequence than the abrupt discontinuation of a β-blocker, for example. Therefore, we sought data from each study regarding the clinical importance of the errors.
We calculated the proportion of patients with 1 or more prescription medication history errors and the mean number of medication discrepancies per patient from each study whenever possible. The study methods and results were heterogeneous, and therefore we made no attempt to combine results for a meta-analysis.