Medication audit and feedback by a clinical pharmacist decrease medication errors at the PICU: An interrupted time series analysis

Abstract Objective Medication errors (MEs) are one of the most frequently occurring types of adverse events in hospitalized patients and potentially more harmful in children than in adults. To increase medication safety, we studied the effect of structured medication audit and feedback by a clinical pharmacist as part of the multidisciplinary team, on MEs in critically ill children. Method We performed an interrupted time series analysis with 6 preintervention and 6 postintervention data collection points, in a tertiary pediatric intensive care unit. We included intensive care patients admitted during July to December 2013 (preintervention) and July to December 2014 (postintervention). The primary endpoint was the prevalence of MEs per 100 prescriptions. We reviewed the clinical records of the patients and the incident reporting system for MEs. If an ME was suspected, a pediatrician‐intensivist and a clinical pharmacist determined causality and preventability. They classified MEs as harmful according to the National Coordinating Council for Medication Error Reporting and Prevention categories. Results We included 254 patients in the preintervention period and 230 patients in the postintervention period. We identified 153 MEs in the preintervention period, corresponding with 2.27 per 100 prescriptions, and 90 MEs in the postintervention period, corresponding with 1.71 per 100 prescriptions. Autoregressive integrated moving average analyses revealed a significant change in slopes between the preintervention and postintervention periods (β = −.21; 95% CI, −0.41 to −0.02; P = .04). We did not observe a significant decrease immediately after the start of the intervention (β = −.61; 95% CI, −1.31 to 0.08; P = .07). Conclusion The implementation of a structured medication audit, followed by feedback by a clinical pharmacist as part of the multidisciplinary team, resulted in a significant reduction of MEs in a tertiary pediatric intensive care unit.


| INTRODUCTION
Medication errors (MEs) are among the most frequently occurring types of adverse events in hospitalized patients, and 3% to 10% of MEs result in patient harm. [1][2][3] Medication errors are also associated with additional costs up to $8.500 per patient, as estimated for hospitals in the United States. 4 The reported prevalence of MEs varies from 5 to 24 per 100 prescriptions in pediatric inpatients. [5][6][7][8] A previous study suggested that MEs are potentially more harmful in children than in adults. 6 Children admitted to pediatric intensive care units (PICUs) are especially vulnerable to harmful MEs because of their dependence on multiple and life-supporting medications. 9 Because of the growing awareness of the complexity of the medication process and medication safety issues, it has been suggested that active involvement of a clinical pharmacist on pediatric wards might be of additional value. Three systematic reviews report a reduction of MEs after a pharmacist was employed on clinical wards, but the included studies do not provide a clear description of the interventions by the clinical pharmacist. [10][11][12] In addition, quality issues arise as most of the included publications involved observational studies, before-and-after designs were without a control group, or the MEs were self-reported by the intervening pharmacist. [10][11][12] A recent Cochrane systematic review 13 included only 1 high-quality, controlled before-after study that showed a significant reduction of serious MEs after the implementation of a multifaceted intervention by a full-time clinical pharmacist on a PICU. 14 Since available evidence is scarce, we decided to study the effect of a structured audit of prescribed medication, followed by feedback to the prescribing pediatrician-intensivist and bedside nurse by a clinical pharmacist as part of the multidisciplinary PICU team. We formulated the following research questions: • Do MEs and medication-related patient harm on a PICU decrease after the implementation of a structured medication audit, followed by feedback from a clinical pharmacist as part of a multidisciplinary team?
• What types of recommendations are made by the clinical pharmacist and to what extent are they accepted by the medical and nursing staff?

| METHODS
The Institutional Review Board of the Academic Medical Center in Amsterdam ascertained that medical ethical approval was not required.
All patients were informed about the fact that health-related data collected routinely could be used for quality improvement, evaluation of care, and scientific research. Patients were given the opportunity to refuse. All data were analyzed and reported anonymously. This is in line with the research code at the Amsterdam Medical Center, and it complies with Dutch Medical Ethics Law.

| Setting and study population
We performed our study in the tertiary PICU of Emma Children's Hospital/Academic Medical Center, Amsterdam, The Netherlands. This mixed PICU has a capacity of 12 beds and provides care to approximately 600 intensive care patients and 300 high-care patients annually, ranging in age from newborns to 18 years.
At the time of the study, all medications were prescribed or altered during the morning round, using a stand-alone patient data management system (PDMS). This PDMS is a generic ordering system and is not equipped with a medication safety monitoring or decision support system. At the start of every nursing shift, an electronically generated sign-off medication list was printed for every patient separately.
Electronic alterations could be made to the medication list by the attending resident, fellow, or staff member. After a mandatory double check, the prescribed medications were administered to the patient, and both nurses signed off the medications on the list. Guidelines of all medications were available on the ward in a hospital formulary.
We included all intensive care patients with at least 1 medication prescription and with an expected length of stay in the PICU of more than 24 hours. We excluded high-care patients from our study.

| Study design and endpoints
We performed an interrupted time series (ITS) with 6 preintervention and 6 postintervention data collection points. We considered 1-month intervals between data collection points as adequate to identify trends in the occurrence rate of MEs. For accurate comparison of the preintervention and postintervention data, the data collection took place during the same calendar months of 2 consecutive years to rule out seasonal effect. The primary endpoint was the prevalence of MEs per 100 prescriptions. Secondary outcomes were medication-related patient harm per 100 prescriptions, the types of the recommendations by the clinical pharmacist, and their acceptance by the clinicians.
We used the definitions and categories for error and harm as described by the National Coordinating Council for Medication Error Reporting and Prevention 15 (Appendix S1). High-alert medications were recorded according to the list for pediatric patients. 16 A prescription was defined as a recipe written by the pediatrician-intensivist to start or change medication, including change of dose.

| Interventions by the clinical pharmacist
The study intervention was the expansion of the PICU team with a clinical pharmacist. The clinical pharmacists received mandatory training before the implementation period on the PICU started. During their training, they familiarized themselves with prevailing medication protocols and guidelines and with data collection from the electronic hospital systems, including the PDMS.
The clinical pharmacist was present on the PICU for a maximum of followed by feedback and recommendations to the attending pediatrician-intensivist and nurse during the ward round later the same morning. Administration of medication was discussed with the bedside nurse, eg, compatibility of medication administration, and infusion pump rates. A structured form was used for the medication audit and bedside evaluation (Appendix S2).

| Data collection
Data on MEs and patient harm were collected for all included patients, ie, both the patients who were audited by the clinical pharmacist and the nonaudited patients. To establish the prevalence of MEs and patient harm, we used a 3-step approach that was validated in a previous study. 17 During the first step, the clinical records of discharged patients were retrospectively reviewed by one of the investigators (J. M. or M. T.). Potential MEs were identified by reviewing all medication summaries, check-off lists, medical and nursing daily notes, symptom registration, and postoperative notes. We systematically compared the potential MEs with the local protocols and the Dutch pediatric formulary. 18 In addition, the hospital incident reporting system was reviewed for reported MEs during the study period.
During the second step, we presented the identified potential MEs to a blinded pediatrician-intensivist and a clinical pharmacist, who deemed the identification of potential MEs to be true or false. In the third step, they classified the MEs as harmful according to the National Coordinating Council for Medication Error Reporting and Prevention categories. The process of data collection is visualized in Figure 1.  and discrepancies were discussed until consensus was reached. During the other study data collection periods, the investigators performed double checks on the patient files that were considered complex by discretion of the researchers.

| Power calculation and statistical analyses
We estimated a prevalence of 10 Table 1.

| Patient harm
Of the 153 MEs that had occurred in the preintervention period, we The experts classified the observed harm as temporary and requiring intervention in 23 harmful MEs (79%) and temporary with prolonged PICU hospitalization in 6 harmful MEs (21%).  Table 4.

| DISCUSSION
Our study shows that the implementation of structured medication audit, followed by timely feedback by a clinical pharmacist as part of the multidisciplinary team, resulted in a significant reduction of MEs in a tertiary PICU. We observed a nonsignificant decrease in medication-related patient harm. The proactive role of the clinical pharmacist resulted in recommendations with a high acceptance rate.
We identified only 1 previous high-quality study that investigated interventions by a clinical pharmacist on a PICU. 14    Relative effect directly after intervention 23% Abbreviation: ME, medication error. β1 estimates the preintervention slope. β2 estimates the difference between the observed level just after the intervention started and that predicted by the preintervention slope. β3 estimates the difference in slopes between the preintervention and postintervention periods.
approximately 3 h/day on the PICU. Our study demonstrates that a comparable decrease in the incidence of MEs after the introduction of a clinical pharmacist can be achieved also with a more cost-effective protocol. Other studies that have investigated the effect of the presence of a clinical pharmacist on a PICU involved single-arm designs without a control group and focused on the recommendations and their acceptance by doctors and nurses rather than on the reduction of MEs. [19][20][21][22][23] Our finding that most recommendations of the clinical pharmacist concerned dosages is in accordance with the aforementioned studies, but the acceptance rates of the recommendations of 95% and 98% were higher than the 63% acceptance rate in our study. 19,20,23 In our study, the clinical pharmacist was actively involved in the medication process of 1 to 2 patients per day, who were considered most at risk for MEs. We performed a post hoc analysis to explore differences in the prevalence of MEs between patients whose medications were audited and discussed in the PICU team and patients without the medication audit. This analysis showed a significant difference between the 2 groups (mean difference = −1.71; 95% CI, −3.13 to −0.28; P = .03), meaning the prevalence of MEs per 100 prescriptions is significantly lower in patients with medication audit than those without. This result suggests that the intervention has no effect (or a delayed effect) in the nonaudited patients, but this hypothesis must be investigated in future research.
We found no significant effect of the interventions of the clinical pharmacist on patient harm. This might be explained by the low baseline rate of harmful MEs, and our study may have been underpowered to detect a difference. Although the low number of harm incidents is consistent with previous studies (6.9), these results may be underestimated as we studied patient harm during the stay on the PICU only, and we did not perform a follow-up after transfer or discharge.
It can be expected that in the future computerized physician order entry systems will increasingly support the medication prescription process, possibly marginalizing the role of the clinical pharmacist.

Consequences for patients
No consequences, n (%) 130 (85) 84 (93) .08 Temporary harm, requiring intervention, n (%) 17 (11) 6 (7) .25 Temporary harm, prolonged PICU stay, n (%) 6 (4) 0 .14 Permanent harm, life threatening or fatal 0 0 1.00 Abbreviations: ME, medication error; PICU, pediatric intensive care units. Several risk factors have been studied, such as age, severity of illness, and surgery, but the existing studies are limited and report nonconclusive results. 8,9,29,30 Only the number of prescriptions seems to be an independent risk factor for MEs. 9,31 Our study was designed as a single-center study. In such a setting and anticipating that the study intervention would influence behavior of the professionals and the organization of care, an ITS design is the recommended approach. 32 The optimal number of data collection points is still under debate, with recommendations that vary from 3 to 12 points. [33][34][35] We collected data at 6 points before and 6 points after the intervention, which is in line with the Cochrane Collaboration guidelines. 35 An ITS does not provide protection against the effect of other events occurring at the same time as the study intervention. A comparable patient group that could be used as a control group was not available at our hospital. To increase the confidence in the study results, we studied the rate of safety incidents during the study periods as a control variable. This analysis shows no significant differences between the preintervention and postintervention periods (β2 = .16; 95% CI, −0.03 to 0.36; P = .11 and β3 = .03; 95% CI, −0.02 to 0.08; P = .18). Also, the capacity of both nursing and medical staffing, a known risk factor for adverse events, was stable. 36,37 We recognize several limitations in our study. First, because of limited resources, the clinical pharmacist was present from Monday to Friday. We are aware that patients on a PICU may be instable and