Does electronic medication reconciliation at hospital discharge decrease prescription medication errors?

Abstract

Purpose -

Medication errors are an important patient safety issue. Electronic medication reconciliation is a system designed to correct medication discrepancies at transitions in healthcare. The purpose of this paper is to measure types and prevalence of intravenous antibiotic errors at hospital discharge before and after the addition of an electronic discharge medication reconciliation tool (EDMRT).

Design/methodology/approach -

A retrospective study was conducted at a tertiary hospital where house officers order discharge medications. In total, 100 pre-EDMRT and 100 post-EDMRT subjects were randomly recruited from the study center’s clinical Outpatient Parenteral Antimicrobial Therapy (OPAT) program. Using infectious disease consultant recommendations as gold standard, each antibiotic listed in these consultant notes was compared to the hospital discharge orders to ascertain the primary outcome: presence of an intravenous antibiotic error in the discharge orders. The primary covariate of interest was pre- vs post-EDMRT group. After generating the crude prevalence of antibiotic errors, logistic regression accounted for potential confounding: discharge day (weekend vs weekday), average years of practice by prescribing physician, inpatient service (medicine vs surgery) and number of discharge mediations per patient.

Findings -

Prevalence of medication errors decreased from 30 percent (30/100) among pre-EDMRT subjects to 15 percent (15/100) errors among post-EDMRT subjects. Dosage errors were the most common type of medication error. The adjusted odds ratio of discharge with intravenous antibiotic error in the post-EDMRT era was 0.39 (0.18, 0.87) compared to the pre-EDMRT era. In the adjusted model, the total number of discharge medications was associated with increased OR of discharge error.

Originality/value -

To the authors’ knowledge, no other study has examined the impact of reconciliation on types and prevalence of medication errors at hospital discharge. The focus on intravenous antibiotics as a class of high-stakes medications with serious risks to patient safety during error events highlights the clinical importance of the findings. Electronic medication reconciliation may be an important tool in efforts to improve patient safety.

Introduction

In the USA, an estimated 1.5 million hospital patients experience medication errors each year, the majority of which occur during transitions of care (Institute of Medicine, 2007). In all, 44,000–98,000 people die each year at least in part due to medication errors (Institute of Medicine, 2000). Classen et al. (1997) and Bates et al. (1997) have estimated the cost of an adverse drug event (ADE) at more than $2,000 per event, and the cost of all ADEs occurring nationally in hospitals to be at $2 billion per year. The morbidity and mortality associated with medication errors have prompted medication management reform. Many healthcare systems have converted from paper-based processes to computerized physician order entry (CPOE) systems (Redley and Botti, 2013; Wang et al., 2007; Koppel et al., 2005; Menendez et al., 2012; Bates et al., 1999). While effective at reducing transcription errors, a CPOE system cannot detect an error if a physician omits medications the patient was taking at home (Agrawal, 2009). Medication reconciliation – the process of intentionally reviewing and reconciling outpatient and inpatient medications at transitions in healthcare – is one technique to decrease medication errors. In 2005, the Joint Commission on Accreditation of Health Care Organizations established medication reconciliation as a key patient safety goal (Institute of Medicine, 2007). Since then, a myriad of information technology systems have been implemented across the healthcare continuum to facilitate medication reconciliation (Agrawal, 2009). However, the effect of electronic medication reconciliation on reducing error rates is relatively unexplored (Agrawal and Wu, 2009). Our study seeks to address this knowledge gap by measuring the prevalence and types of discharge intravenous antibiotic medication errors before and after the addition of an electronic discharge medication reconciliation tool (EDMRT) to the inpatient discharge process.

Methods

Setting

Tufts Medical Center (Tufts MC) is a 415-bed, academic, tertiary care facility providing inpatient, outpatient and emergency services in Boston, Massachusetts. This research was approved by the Institutional Review Board of Tufts MC/Tufts University Health Sciences Campus.

Implementation of an electronic medication reconciliation tool

Tufts MC fully implemented an EDMRT into the electronic medical record in February 2011, after several months of training for staff and physicians (Soarian, Siemens Medical Systems copyright 2002–2014).

Prior to EDMRT, discharge medications were transcribed by hand from the inpatient electronic medication list, which did not highlight which medications were added during admission vs continued from home. With EDMRT, when physicians prescribe admission orders, they input home medications electronically, and have the option of easily selecting appropriate home medications for inpatient orders. The electronic list has mandatory fill-in boxes for strength, form, dose, route and time for every medication. The current inpatient list shows which medications were home medications and which are medications added during the inpatient stay. At time of hospital discharge, the prescribing physician creates an electronic list of medications from the current inpatient list. Inpatient and home medications that stay on the discharge list can be easily selected, and new medications for discharge can be added as required.

Medications and participants

Intravenous antibiotics prescribed at discharge were chosen for this study as they represent treatment for serious diseases, with high risk of patient harm if prescribed incorrectly. This study recruited subjects from a previously published database (Allison et al., 2013) of over 800 patients, aged 18 years or older, discharged and followed in the Tufts MC’s clinical Outpatient Parenteral Antimicrobial Therapy (OPAT) program from January 2009 to December 2011.

The clinical OPAT program began in 2009 and monitors more than 90 percent of patients discharged from Tufts MC with intravenous antibiotic therapy. The inpatient infectious disease (ID) consulting physician documents the final antibiotic recommendations in an electronic OPAT template form which is captured and used by the OPAT administrator to track patients through their outpatient treatment plan.

Study design

We conducted a retrospective medical chart review of 100 pre-EDMRT subjects and 100 post-EDMRT subjects selected by random number generation. All intravenous and oral antibiotics listed on the ID consultants’ final recommendations were compared to the antibiotics listed in the discharge medication orders. The ID consultants’ recommendations were considered the gold standard for comparative purposes. Patient data were abstracted from medical charts and ID recommendations into a secure electronic database using Research Electronic Data Capture (REDCap) (Harris et al., 2009, pp. 377–381). To ensure fidelity of outcome assessments, G.M.A. and B.W. independently evaluated the same 20 subjects from the pre-EDMRT era (selected by random number generation) and 20 subjects from the post-EDMRT era (selected by random number generation); inter-observer agreement was measured by the κ statistic (Viera and Garrett, 2005; Cohen, 1960).

Variables

Data collected for each subject: age (years), gender, date of discharge and intravenous antibiotic prescription (name, dose, route, frequency and duration), day of discharge (weekday vs weekend), type of inpatient service (medicine vs surgery), years of experience of prescribing physician and total number of discharge medications per patient. Medication discrepancies were categorized as: dose, frequency and route. When a drug was present on the gold standard list and absent on the discharge list, this was considered an antibiotic omission. Conversely, when a drug was absent on the gold standard list and present on the discharge list, this was considered an antibiotic addition. If a drug was prescribed that differed from the gold standard, this was considered two medication errors: an antibiotic omission and antibiotic addition. We summarized antibiotic error by type (omission, dose, route, frequency, addition), as well as the total errors in aggregate.

Statistical methods

With 100 subjects in each group, we have 80 percent power to detect a change of medication error rates from pre-EDMRT of 25 percent to post-EDMRT of 10 percent, with significance α = 0.05. We generated the primary outcome, crude prevalence of discharge intravenous antibiotic errors and analyzed the association of errors with the primary covariate of interest: pre- and post-EDMRT groups. Demographic data were compared using means and student’s t-test as appropriate. We performed logistic regression to adjust for potential confounders of the association between the primary outcome (presence of any antibiotic error) and primary covariate (pre- vs post-EDMRT implementation). Potential confounders analyzed were: day of discharge (weekend vs weekday), level of experience of prescribing resident physician (years of practice), type of inpatient service (medicine vs surgery) and total discharge medications per patient. Statistical analyses were performed using Stata for Mac version 12.0 (StataCorp; College Station, TX; 2011).

Results

Of the 257 subjects from the pre-EDMRT era and 127 subjects in the post-EDMRT era discharged with intravenous antibiotics and thus identified as potentially eligible for the study, 109 from each group were randomly selected and assessed for eligibility (Figure 1).

Figure 1.

Consort diagram

In the pre-EDMRT cohort, seven subjects were excluded because there was no ID recommendation, and two subjects were excluded because there was no electronic discharge medication list. In the post-EDMRT cohort, five subjects were excluded because there was no ID recommendation, and four subjects were excluded because there was no electronic discharge medication list. From there, 100 subjects from each era were analyzed. The κ coefficient was 0.90, indicating strong agreement between presence or absence of any medication error as assigned by B.W. and G.M.A (Viera and Garrett, 2005).

Demographic, antibiotic and clinical characteristics are shown in Table I.Patients were discharged with approximately 12 medications in the pre- and post-EDMRT periods, and the most common intravenous antibiotic used was vancomycin in both periods.

Table I.

Study population characteristics

	Pre-EDMRT (95% confidence interval)	Post-EDMRT (95% confidence interval)	Confidence intervals of the difference	p-Value for the difference
Number of subjects	100	100
Average age, years	57 (53.9, 60.2)	62 (58.6, 65.0)	(−9.28, −0.43)	0.03
Gender ratio	64% male (54.6%, 73.4%)	56% male (46.2%, 65.7%)	(−0.06, 0.22)	0.25
Average no. of medications per subject	12.2 (11.0, 13.4)	12.8 (11.7, 13.9)	(−2.28, 1.00)	0.44
Four most common intravenous antibiotics (percentage of subjects on each medication)	38% vancomycin 19% ertapenem 11% ceftriaxone 10% daptomycin	34% vancomycin 16% oxacillin 14% ceftriaxone 14% ertapenem
Weekend vs weekday discharge	12% weekend (5.6%, 18.4%)	12% weekend (5.6%, 18.4%)	(−0.09, 0.09)	1.00
Average years of experience by prescribing physician	3.9 (2.3, 5.5)	2.1 (1.3, 2.9)	(0.07, 3.59)	0.04
Type of inpatient service (medicine vs surgery)	63% medicine (53.5%, 72.5%)	55% medicine (45.2%, 64.8%)	(−0.06, 0.22)	0.25

In the 200 subjects recruited, 34 patients had at least one antibiotic error (Table II). The crude odds ratio of discharge with at least one antibiotic error was 0.41 (95 percent CI: 0.19, 0.90; p = 0.027) lower for the post-EDMRT era compared to the pre-EDMRT era.

Table II.

Prevalence and odds ratio of medication errors

	Pre-EDMRT (n = 100)	Post-EDMRT (n = 100)	Odds ratio (95% confidence interval)
Subject with at least one error	23	11	0.41 (0.19, 0.90)

Table III presents descriptive statistics of types and prevalence of medication errors. Duration errors in both the pre- and post-EDMRT eras were excluded because physicians lacked a consistent way of recording medication duration information in the medical record. A multivariable model was generated to account for potentially confounding variables (Table IV). The following variables met our a priori inclusion criteria of significance level p < 0.25 and therefore were included in the multivariable model: day of discharge (weekday vs weekend), and total number of discharge medications. Prescriber experience did not meet criteria and was not included. The adjusted odds ratio of discharge with an intravenous antibiotic error in the post-EDMRT error was 0.39 (95 percent CI: 0.18, 0.87; p = 0.021) compared to the pre-EDMRT era. The adjusted odds ratio (0.39) and crude odds ratio models (0.41) were extremely close, showing that adjusting for these potential confounders did not change the exposure-outcome odds ratio association. We also found in the adjusted model that for every five additional discharge medications, the odds ratio increased by 1.33 (p = 0.065) holding all other variables constant. Day of discharge (weekday vs weekend) was not associated with higher prevalence of antibiotic errors.

Table III.

Descriptive statistics of error categorization

Type of error	Pre-EDMRT (n = 100)	Post-EDMRT(n = 100)
Omission	4	4
Dose	13	5
Route	2	0
Frequency	5	5
Addition	6	1
Total	30	15

Note: A total of 45 errors were recorded; one subject may have more than one type of error

Table IV.

Multivariable model

Variable	Odds ratio	95 % confidence interval
Group (post-EDMRT vs pre-EDMRT)	0.39	(0.18, 0.87)
Day of discharge (weekend vs weekday)	2.69	(0.59,12.29)
Total number of discharge medications	1.06	(1.00,1.13)

Discussion

Our findings indicate that EDMRT implementation was associated with a significant decrease in overall intravenous antibiotic errors at the time of hospital discharge. We also found that antibiotic errors were positively associated with total number of discharge medications. Interestingly, we did not find a correlation of errors with years of prescriber experience or day of week of prescription.

Given our result of positive association of total number of medications with antibiotic errors, we hypothesize that one underlying mechanism of antibiotic errors at discharge may be “distracted prescribing” (Nichols et al., 2008; McDowell et al., 2009). The process redesign and implementation of EDMRT may address potential distracted prescribing in the following ways. First, EDMRT lifts the burden of handwriting discharge medication lists because admission medications are already electronically entered, and subsequent medications are ordered via CPOE. Second, home vs inpatient medications are clearly identified by icons, which act as visual cues, and free physicians from sifting through a written chart. Finally, by making use of EDMRT mandatory, the process redesign involved comprehensive physician training and support so that reconciliation becomes an accepted part of work flow and work culture. Schnipper et al. (2011) and Bails et al. (2008) have previously shown optional reconciliation has low usage in actual clinical settings. Greater facility with the mechanics of EDMRT may sharpen focus on the medical decision aspect of prescribing with fewer distractions from the mechanics of prescribing.

Many sites described in the literature utilize a multidisciplinary system of physicians, nurses and pharmacists to perform medication reconciliation (Murphy et al., 2009; Varkey et al., 2007; Boockvar et al., 2006; Bartick and Baron, 2006; Kramer et al., 2007). Although these multidisciplinary teams (usually led by pharmacists) are generally effective at reducing medication discrepancies, the reconciliation process without electronic aid remains costly and time-consuming for healthcare providers (Kramer et al., 2007). Schnipper et al. (2011), Bails et al. (2008) and Poon et al. (2006) describe the development and implementation of an electronic medication reconciliation tool in the ambulatory medical record, but do not report its effectiveness in decreasing errors. Agrawal and Wu (2009) demonstrate that the development of a “MedRecon system” in the electronic health record decreased medication errors by 93 percent at hospital admission. However, unlike our study, these authors were unable to evaluate medication discrepancies at discharge, which is a particularly high-risk time for medication errors because patients are less strictly monitored. Schnipper et al. (2009) performed a randomized control trial to determine how electronic medication reconciliation and process redesign affects potential adverse drug events at hospital discharge. These authors found a significant reduction in errors at one of two hospital sites, and the study was not powered to compare the two sites. The importance of an accurate discharge medication list cannot be overstated. Makaryus and Friedman (2005) found that only about 30 percent of patients remember their medication-related information (i.e. name, and purpose of the drug) relayed to them by their prescribing physician. Both patients and healthcare providers rely on an accurate discharge medication list for information about whether to stop or change outpatient medications, and which new medications to add when they are discharged from the hospital.

We acknowledge that our study has some limitations. First, although considered the gold standard, ID physicians may have made errors entering medications in their forms. However, forms were reviewed at the time they were created by multiple physicians engaged in patient care, thus reducing likelihood of these errors. While this study reports fewer errors per patient compared to other published studies (Wong et al., 2008; Cornish et al., 2005; Pippins et al., 2008; Vira et al., 2006), our study focussed only on antibiotic errors rather than all medications at discharge. Another limitation is the time lag between ID physicians final recommendations and discharge medication prescribing was not assessed in this study, although our clinical practice is that most final recommendations occur on the day of discharge to account for patients’ changing conditions during hospital admission. Blinding researchers to subjects’ group (pre-EDMRT vs post-EDMRT) was not feasible because medication lists appear differently in the electronic record. To discourage expectancy bias, we created a standardized data capture system that only assigned an error to a subject after data from both the discharge list and gold standard were recorded. Additionally, the κ statistic analysis implies inter-observer agreement and reproducible results. Lastly, this is a single center study; however, Tufts MC uses a commercially available electronic medical record so the results should be applicable to other healthcare centers and hospitals utilizing similar electronic medical records.

The strengths of our study include capture of initial data in real time, thus avoiding recall bias that can afflict retrospective studies. In addition, subjects were drawn from a well-defined subject cohort (Allison et al., 2013). The focus on intravenous antibiotics is clinically relevant because intravenous antibiotics are high-stakes medications for complex infections.

In summary, our study suggests that using an electronic medication reconciliation tool may decrease the prevalence of antibiotic errors at the time of hospital discharge. This study highlights the need to continue developing information technology systems designed to decrease medication errors. Furthermore, by uncovering a positive relationship between antibiotic errors and total number of medications, our study suggests that an underlying mechanism of errors may be distraction at the time of prescribing. Prospective studies examining this mechanism are urgently needed in order to improve interventions and save lives.

About the authors

Dr Genève M. Allison is the Director of the OPAT program at the Tufts Medical Center, Boston, MA and an Assistant Professor of Medicine at the Tufts University School of Medicine. She was awarded an NIH-funded Career-Development Grant supporting a Masters in Clinical-Translational Science at the Tufts University studying OPAT predictive modeling and graduated in June 2013. She graduated from the Harvard College (BA, Biology), University of Massachusetts (MD), Internal Medicine Residency and Chief Residency at the Alameda County Medical Center (Oakland, CA) and Infectious Diseases Fellowship at the Tufts-NEMC (Boston, MA, 2007). Her research interests include OPAT outcomes, guideline development, complex care systems and biofilms on PICC lines.

Bernard Weigel graduated summa cum laude from Tufts University in 2014 with a concentration in economics and pre-medical program studies. In June 2013, he was awarded a Nathan Gantcher Student Summer Scholars Grant to conduct a clinical research project under the mentorship of Dr. Geneve Allison in the Division of Geographic Medicine and Infectious Diseases. Bernard’s research interests include OPAT patient outcomes, quality improvement, and cardiovascular health in the setting of HIV infection.

Christina Holcroft joined the National Fire Protection Association in July 2014 as director of the organization’s Fire Analysis and Research Division. Dr. Holcroft was previously an assistant professor of medicine at the Sackler School of Graduate Biomedical Sciences at Tufts University in Boston, where her research examined methods by which data analysis can contribute to advances in healthcare. She holds a B.S. from the Massachusetts Institute of Technology, M.S. and ScD degrees in biostatistics from the Harvard School of Public Health, and received postdoctoral training at the University of Massachusetts–Lowell.