Setting and population
This prospective study was conducted in two internal medicine wards (designated A and B) at the University hospital of Lund, Sweden. A LIMM-based clinical pharmacy service, including medication reconciliation at admission, was implemented in January and October, 2007, in the respective wards. All patients admitted to wards A and B after implementation of the service until the end of the year 2007, were eligible for inclusion in the study. Patients discharged or deceased before the pharmacist could conduct admission medication reconciliation were excluded from the study. There were 22 beds in each ward. The weekday staff in each ward comprised two junior physicians and two senior physicians, one clinical pharmacist, three nurses, three assistant nurses, one physiotherapist and one occupational therapist. Availability of beds alone decided the ward to which a patient was admitted. The wards used the standard hospital electronic health record (EHR) system (Melior®, Siemens Corp.); this was used in all hospital wards but was not used in primary care. The primary care centres in the region surrounding the hospital used either another EHR system or paper-based health records. Community care services used paper-based medication lists. The different levels of care (i.e. primary, secondary and community care) exchanged information about the patients' current medication lists by phone, fax, or mail. No electronic communications between the hospital and the primary care centres or community care services was possible at the time of the study. The regional ethical board of the University of Lund, Sweden, did not consider ethical approval to be necessary and had no objections to the study.
Collection of data
A two-step procedure was used to collect and classify data on medication discrepancies and errors in the medication histories. Firstly, clinical pharmacists conducted medication reconciliations and documented their work in a LIMM medication interview questionnaire form (Additional file 1). Five different pharmacists worked at the wards during the study period. Secondly, two pharmacy students and a research pharmacist classified the identified discrepancies and errors.
The medication reconciliation process in the LIMM model
The admission medication reconciliation process in the LIMM model is comprehensive and was developed over about 5 years and implemented on top of standard care [9, 12]. Following a strict protocol, clinical pharmacists identified the patient's pre-admission medication list. For patients capable of participation and willing to participate, an initial medication interview was conducted. The pharmacist asked which medications and dosages the patient had been taking before admission. Specific questions were asked about the use of painkillers, heart medications, stomach medications, sleeping pills, anti-diabetics, eye drops, inhalation drugs, over-the-counter drugs and herbal drugs, in order to increase the probability of including all the patient's medications. Sometimes, a medical interview was conducted with a close relative instead of the patient. In addition (or otherwise, if an interview could not be conducted), the pharmacist consulted all available pre-admission lists, including drug lists from primary and community care, the national pharmacy register (all drugs dispensed within the past 15 months) , and prescription forms from the medication dispensing system ApoDos (a multi-dose system where all medications that the patient should be taking on one occasion are machine-packed together in small, fully labelled plastic bags at a pharmacy dispensing centre and delivered to the patient every second week) . Based on this information, a list with the patient's prescribed medications was documented in the LIMM medication interview questionnaire, part 1 (Additional file 1). Parts 2 and 3 of the LIMM medication interview questionnaire were conducted with some patients; these parts comprised questions about knowledge of, practical handling of, adherence with, and beliefs about medications . This paper reports the results from part 1 only.
The pre-admission medication list identified by the pharmacist was regarded as the most accurate list available since it was based on all available information sources and had been compiled according to a well established, systematic method [5, 6, 20–22]. Differences between this pharmacist-compiled pre-admission medication list and the medication list in the hospital EHR were documented in the LIMM medication interview questionnaire. The pharmacist consulted the patient's EHR to establish possible reasons for the differences. Discrepancies noted in the hospital medication list which the pharmacist judged relevant and possibly requiring correction were discussed with a ward physician. The pharmacist recommended corrections for the hospital EHR, and the physician then made the final decision and was responsible for correcting the hospital EHR list when necessary. The use of over-the-counter drugs and herbal drugs by the patients was also documented in the questionnaire and discussed with the physician when considered clinically relevant. The pharmacists' recommendations and the subsequent actions by the physician or pharmacist were documented in a medication review form.
It was ward policy that a clinical pharmacist should conduct the LIMM-based medication reconciliation within one day of admission to the ward or on Mondays for patients admitted on weekends. Medication reconciliation was conducted once for each patient and took on average 32 minutes per patient if a patient interview was conducted and 15 minutes if no interview was conducted [unpublished observations, personal communication Tommy Eriksson 27/03/2012]. This time included the face-to-face discussion with a physician about discrepancies in the hospital medication list.
Occasionally, it took longer than one day for the pharmacist to conduct the medication reconciliation. This was attributed to time constraints, lack of personnel, or temporarily closed wards because of an infection outbreak among the patients. If a clinical pharmacist was not available, physicians and/or nurses occasionally corrected errors in the medication history (standard care), but there were no instructions or forms for these changes, in contrast to the LIMM-based structured medication reconciliations.
Definition and classification of medication discrepancies and errors
The identified differences between the pharmacist-acquired medication list and the medication list in the EHR were classified retrospectively by reviewing the LIMM medication interview questionnaires and the EHR. A medication discrepancy was defined as an addition or withdrawal of a drug, or a change to the dose or dosage form. An incorrect dosage interval was not defined as a discrepancy if the total dosage/24 h had not been changed. Changes to an equivalent generic drug or withdrawal of drugs with a long dosage interval, e.g. once monthly, were also not regarded as medication discrepancies. The medication discrepancies were further classified by type: drug omitted (the drug had not been registered in the hospital EHR drug list), additional drug (a drug had been erroneously added to the hospital EHR drug list), dosage too high, dosage too low, or wrong dosage form.
Medication discrepancies for which the reviewing pharmacists could not identify any clinical reason (unintentional changes) were deemed to be medication errors. Our definition of a medication error was based on the definition proposed by Leape : "A medication error is any error in the process of prescribing, dispensing or administering a drug, whether there are adverse consequences or not". There were two exceptions to this: that only errors in the medication history were included, and that discrepancies corrected before reaching the patient were not considered medication errors. For example, discrepancies concerning weekly doses that were identified before the dosing occasion or involving omission of drugs that were to be given as needed and which the patient had not yet required were not counted as errors. Over-the-counter drugs and herbal drugs were not included in the drug list in the EHR and hence could not result in medication discrepancies or errors.
Two pharmacy students (doing their Masters theses) were responsible for the classification of medication discrepancies and errors as described above. They were informed by a research pharmacist (ÅB) about the classification procedure and they continually discussed any lack of clarity with the research pharmacist. To evaluate the percent of cases which the raters agreed upon, 30 patients were classified independently by the two students and one research pharmacist (LH). The agreement with regard to the number and type of medication errors was 83%.
An open source software based on the R language and environment for statistical computing (http://www.R-project.org) was used for all statistical analyses . Descriptive statistics are shown as medians (interquartile range), means (95% confidence intervals, CI) and frequencies or percentages (95% CI) when appropriate. The denominator for calculating the error rate was the number of prescribed medications in the hospital EHR before medication reconciliation, plus any medications omitted. A multivariable binary logistic regression analysis was conducted where the response variable was the presence or absence of medication errors. The following model variables were pre-specified potential predictors: number of drugs at admission (every increase by one drug), age (every 10 years' increase), sex (0 = Female, 1 = Male), type of care service before admission (0 = Living in own home with no care service, 1 = Living in own home but enrolled in community home care services, 2 = Living in care home), and directly admitted to the study ward without transferral from another ward (0 = No, 1 = Yes). In addition, the extent to which standard care identified medication errors was evaluated by including the number of days from admission to the ward until medication reconciliation (0 = 0-1 days, 1 = 2-3 days, 2 = 4 days or more) in the model. This variable was not considered a potential predictor but was included in order to evaluate if the medication errors remained undetected by standard care if the LIMM admission medication reconciliation was delayed. The variables "2-3 days" and "4 days or more" were not targeted controls; the medication reconciliation was delayed because of time constraints or other factors and never deliberately. No variables were eliminated from the regression model. Eleven percent of the data for the variable "directly admitted to the study ward", and 10% for "number of days from admission to the ward until medication reconciliation" were missing. Data were complete for the other variables. In the multivariable regression model, missing data was imputed using a multiple imputation method . The le Cessie-van Houwelingen-Copas-Hosmer unweighted sum of squares test for global goodness of fit was used to assess the fit of the model . The significance level in the analysis was set to 0.05.