Article Text
PDF
Abstract
Purpose of study Initiatives to standardise hospital paper-based prescription charts are underway in various countries in an effort to reduce prescribing errors. The aim of this study was to investigate the extent to which prescribing error rates are influenced by prescription chart design and familiarity.
Study design In this prospective, randomised, cross-over study, Foundation Year 1 doctors working in five Scottish National Health Service (NHS) Boards participated in study sessions during which they were asked to prescribe lists of medications for five fictional patients using a different design of paper prescription chart for each patient. Each doctor was timed completing each set of prescriptions, and each chart was subsequently assessed against a predefined list of possible errors. A mixed modelling approach using three levels of variables (design of and familiarity with a chart, prescribing speed and individual prescriber) was employed.
Results A total of 72 Foundation Year 1 doctors participated in 10 data-collection sessions. Differences in prescription chart design were associated with significant variations in the rates of prescribing error. The charts from NHS Highland and NHS Grampian produced significantly higher error rates than the other three charts. Participants who took longer to complete their prescriptions made significantly fewer errors, but familiarity with a chart did not predict error rate.
Conclusions This study has important implications for prescription chart design and prescribing education. The inverse relationship between the time taken to complete a prescribing task and the rate of error emphasises the importance of attention to detail and workload as factors in error causation. Further work is required to identify the characteristics of prescription charts that are protective against errors.
- Clinical pharmacology
- Hospital medicine
- Human error
- Medication safety
Statistics from Altmetric.com
Introduction
In UK hospitals, prescribing errors are common. The General Medical Council commissioned EQUIP study (an indepth investigation into causes of prescribing errors by foundation trainees in relation to their medical education), published in 2009, detected a prescribing error rate of 8.4% among Foundation Year 1 doctors (FY1s).1 FY1s are newly qualified doctors who are undertaking their 1st year of postgraduate training; they are the most frequent prescribers in a hospital setting.1 Another recent UK study found 14.7% of medication orders to be erroneous, and concluded that drug chart design may influence certain types of prescribing errors.2 Recommendation 1a from the EQUIP report stated that ‘a standard drug chart should be introduced throughout the National Health Service (NHS)’,1 echoing the British Pharmacological Society's suggestion of a national prescription chart 3 years earlier.3 ,4
The assumption that standardisation of documentation will reduce errors has prompted several countries, including one within the UK, to implement national prescription charts. In 2004, the All Wales Medicines Strategy Group approved the introduction of a standardised inpatient medication administration chart throughout Wales.5 The chart was designed by a subgroup of the Welsh Pharmacists Committee and its implementation was accompanied by an e-learning package and nationally agreed prescription writing standards. To date, no data relating to the impact of the standardised chart on error rates have been made available.
In July 2003, the Australian Council for Safety and Quality in Health Care tasked the National Medication Chart Working Group with designing, piloting and implementing a national inpatient medication chart. By June 2006, all public hospitals in Australia were using a common prescription chart. Before and after prescribing audits undertaken in Queensland hospitals seem to indicate that standardisation was associated with a significant reduction in the frequency of prescribing errors, improved adverse drug reaction documentation and a decrease in the potential risks associated with warfarin management.6 However, a comparison of the national chart with the one previously used at the Royal Perth Hospital concluded that the new chart had ‘adverse design features’ and questioned the rationale for its introduction.7 Implementation of the standardised chart was accompanied by several other interventions including the provision of training sessions for medical, nursing and pharmacy staff, the placement of prescribing guidelines in every end-of-bed folder and a change in the administration time of warfarin. It is therefore impossible to attribute the reduction in prescribing errors found in Queensland hospitals solely to the process of standardisation.
In current UK practice there is a mixture of paper-based and electronic prescribing, with most hospitals still using the former. In Scotland, each NHS Board has designed and implemented its own unique paper-based prescription chart, two examples of which are shown as appendices (see online supplementary material). Doctors use the charts to prescribe drugs, pharmacists annotate them with additional information and nurses use them to record administration. Despite some recent critique of the initiative,8 the argument for standardisation of paper-based prescription charts throughout the NHS seems intuitively sensible. The potential for error should be reduced if prescribing documentation is familiar to prescribers who have also been trained to prescribe safely and effectively using the same agreed chart.5 Following this logic, the Academy of Medical Royal Colleges, in collaboration with the Royal Pharmaceutical Society and the Royal College of Nursing, recently produced a set of standards for the design of hospital drug charts9 that is intended to inform the design of future inpatient prescription charts.10
Although there is some evidence that the format and design of prescription charts may contribute to prescribing errors,11 ,12 this belief is yet to be supported by carefully controlled experimental data. We hypothesised that prescription chart design and familiarity are important variables that influence the likelihood of making a prescribing error. Therefore, the aims of this study were:
-
to investigate whether different prescription chart designs are associated with variation in the rate of prescribing error;
-
to use a mixed modelling approach to simultaneously explore the extent to which factors at the level of the documentation (prescription chart familiarity), speed of prescribing and the individual prescriber determine the frequency of prescribing transcription errors.
Methods
Setting and population
The study was undertaken in the five largest NHS Boards in Scotland, based on the sizes of the populations that they serve: NHS Lothian, NHS Greater Glasgow and Clyde, NHS Tayside, NHS Grampian and NHS Highland. Although some aspects of prescribing practice differ slightly between regions, guidelines are extremely similar and abbreviations such as ‘mcg’ for micrograms and ‘PO’ for oral are not permitted in any of the five NHS Boards. A hospital-based pharmacist or consultant physician from each area was recruited to facilitate room bookings and assist with local recruitment of FY1s. Two 1-h session times were arranged in each area, usually at different hospitals and outwith normal working hours to maximise attendance. Volunteer FY1s were sought via email. The nature and purpose of the study was explained, along with information about the location and times of sessions. A £20 book token was offered as an incentive to each eligible FY1 who participated. All data collection sessions took place during October and November 2011; FY1s were eligible provided that they were completing their first 4-month attachment and had not undertaken paid employment as a doctor prior to August 2011.
Design
In this prospective, randomised, cross-over study, lists of medications for five fictional hospital inpatients were devised by three clinicians: a Consultant Pharmacologist (SRJM), a Specialty Registrar (VRT) and a Foundation Doctor (RLH). Each fictional patient was given a name, date of birth, Community Health Index number, address, weight and single drug allergy, with the nature of the allergy specified. Each medication list contained a total of eight commonly-prescribed items, and details regarding route, dose and frequency were provided. Six of the medications required regular prescriptions: three were once daily preparations and three were required regularly between twice daily and four times daily. One of the regular medications in each list was an antibiotic for which the course duration was specified. One medication in each list was to be prescribed ‘as required’, with a maximum frequency and indication provided. The final medication in each list was a ‘once only’ prescription, to be administered immediately.
Study packs containing medication lists for the five fictional patients (labelled as Ann, David, Doris, Henry and Lorraine) along with a single, authentic prescription chart from each of the five NHS Boards were compiled. Five ‘types’ of study packs, labelled A to E, were developed. The pairings of prescription chart to medication lists varied in each of the five pack types to ensure that overall the medication list for each fictional patient was prescribed onto each of the five prescription charts. To reduce order and fatigue effects, the sequence in which the FY1s received each drug chart varied between pack types. Details of the medication list and prescription chart pairings, along with the order of each pairing within each pack type, are shown in table 1.
Details of study packs
Data collection
Ten data collection sessions were undertaken in total, two in each NHS Board. Each session lasted approximately 1 h, was facilitated by either VRT or RLH, and took place outwith clinical areas, usually in a hospital education facility. To avoid interruption, FY1s were asked to attend bleep-free, and at a time when they had no clinical duties. The nature and purpose of the study was explained at the beginning of each session, and the participating FY1s were given the opportunity to ask questions and clarify instructions. Each FY1 was provided with a digital stopwatch with lap-split time memory facility along with written and verbal instructions. They were then asked to prescribe the five lists of medications (with appropriate patient details) onto the five prescription charts in the order in which they had been presented. To minimise confusion, each drug list was printed on a separate sheet of paper and the five sheets were stapled in the appropriate order. Furthermore, the prescription charts for each FY1 were prenumbered in accordance with the order in table 1. They were asked to start their stopwatch at the beginning of the task, and press ‘split’ to record a lap-time at the point when they had finished prescribing each of the medication lists. A table for noting lap times was provided and participants were asked to write down each time after pressing the ‘split’ button. The FY1s were informed that their prescriptions would be assessed for accuracy, completeness and time taken. They were asked to transcribe the medication lists exactly, and were reassured that there were no inaccurate prescriptions or tricks. Finally, they were asked to prescribe as though it was a typically busy day on the wards and their prescriptions would be administered directly to patients without any additional checks. Following transcription of all five medication lists, FY1s were requested to remain at their desks to avoid disturbing others. Following the session, all split times were retrieved from the stopwatches by the study facilitators and checked against the times recorded by the FY1s.
Data extraction
Informed by the error classifications used in previous studies,1 ,13 VRT and RLH devised two lists of possible errors: one relating to patient information errors (such as date of birth) and one relating to each prescription type (regular, as required or once only). As no drug choices had been made by the participants, the lists were restricted to possible transcription errors, including omissions. The lists of possible errors are shown in tables 2 and 3.
Patient information errors
Specific prescription errors
All charts were reviewed by a single assessor, a recently-retired chief pharmacist who had not been involved in either study design or data collection. For every participant, each of the five completed prescription charts was reviewed for the specific errors detailed in tables 2 and 3. Errors were indexed using Excel (Microsoft Office 2007) to facilitate cross-referencing with information relating to prescribing speed, prescription chart design and chart familiarity.
Analysis
In order to explore whether the five medication lists were comparable in terms of complexity, variation in error rates between the five lists (labelled as Ann, David, Doris, Henry and Lorraine) was examined. Multilevel models are increasingly being employed to examine sources of variation at different levels within the health service structure.14 ,15 The variables included in the analysis relating to documentation were familiarity with a prescription chart (indicated by prior usage in their role as a doctor) and design of the prescription chart as defined by the source NHS Board. The documentation, prescribing speed and individual prescriber variables were chosen on the basis of a theoretical justification or evidence from previous literature indicating that they may influence the incidence of prescribing errors. Participants, prescribing speed and familiarity were modelled as between-subject effects, and speed and familiarity were modelled as within-subject effects. Between-subject effects describe how one study participant differed from another; for example, the question ‘person A was faster than person B, but did they make more errors?’ is exploring between-subject effects in relation to timing. Within-subject effects describe how an individual participant's performance was influenced by chart design and other factors; for example, the question ‘did person A make more errors on the prescription charts on which they prescribed more quickly?’ is exploring within-subject effects in relation to timing.
Results
A total of 72 FY1s participated in 10 data collection sessions: 14 working in NHS Lothian, 19 working in NHS Greater Glasgow and Clyde, 16 working in NHS Tayside, 10 working in NHS Grampian and 13 working in NHS Highland. Incomplete sets of ‘lap’ times resulted in the data from 11 participants being excluded from the final analysis (two from NHS Lothian, one from NHS Greater Glasgow and Clyde, four from NHS Tayside, one from NHS Grampian and three from NHS Highland). The maximum number of errors that could be made by each FY1 was 64 per prescription chart (six patient information errors and 58 specific prescription errors). Each FY1 could therefore make up to 320 possible errors across the five prescription charts included in the study. The proportion of errors made by each participant was expressed as a value between 0 (no errors were made) and 1 (all possible errors were made). The mean total error rate was 0.17 (SD=0.22), indicating that, on average, just over a sixth of all the errors listed in tables 2 and 3 were made by each FY1. Time for prescription chart completion was recorded in seconds (mean=406.63, SD=128.72 s).
A one-way analysis of variance (ANOVA) performed to explore comparability of the five medication lists showed no significant difference in error rate according to medication list (F4,355=0.62, p=0.65) and no significant difference in speed of prescribing according to medication list (F4,344=0.68, p=0.61). This indicated that participants did not find any of the five patients (Ann, David, Doris, Henry or Lorraine) easier or harder to prescribe than any others; they were thus comparable in terms of prescribing complexity. An analysis of error rate in relation to prescription chart design was also performed. A one-way ANOVA examining errors for each prescription chart showed a significant difference (F4,359=11.96, p=0.001), indicating that participants more frequently made errors on some chart designs compared with others. Post hoc Bonferroni comparisons suggested that NHS Highland and NHS Grampian charts produced significantly higher error rates than the other charts. The remaining three charts (NHS Lothian, NHS Greater Glasgow and Clyde, and NHS Tayside) performed similarly.
In addition to analysing whether different prescription chart designs produced differing error rates, the study aimed to examine the influences of individual familiarity with a particular prescription chart and speed of prescribing. Prescribing speed significantly predicted the error rate between participants; those who took longer to complete their prescriptions made fewer errors. However, as a within-subject effect, prescribing speed was marginally non-significant. In other words, an individual's prescribing speed on each of the five prescription charts did not influence their error rate between charts, although a slightly larger sample size may have produced a result in keeping with the between-subject finding. Familiarity with a chart did not predict error rate. Effect sizes were small or lower as indicated via η2 (η2=0.01). Full model statistics can be found in table 4.
Summary of output from mixed model predicting the proportion of errors caused by prescribing speed and chart familiarity
Discussion
This study is the first to explore the associations between prescription chart design, familiarity, prescribing speed and error. It has demonstrated that different prescription chart designs are associated with a significant variation in the rate of prescribing error. However, the key hypothesis in support of prescription chart standardisation (that significant variation in prescribing error rate occurred as a result of familiarity, in addition to any variation seen at the levels of prescribing speed and individual prescriber) was disproved. It is, however, possible that any such association was disguised by the removal of individual prescriber and timing effects; for example, some participants slowed down when prescribing on unfamiliar documentation, suggesting that they took extra care and perhaps double-checked their prescriptions.
Despite this negative result, the study findings have important implications for prescription chart design and prescribing education. The novel finding that some designs of prescription chart are more likely to provoke FY1 prescribing errors than others warrants attention by those seeking to maximise the safety of hospitalised patients. Future work should investigate the characteristics of charts which appear to make errors more or less likely, and how findings gleaned from this study and others can be applied to increasingly popular electronic prescribing systems. Due to the chronology of events, not all of the charts used in this study meet the recent Academy of Medical Royal Colleges standards for the design of hospital drug charts. Any standardised prescription chart should fulfil such criteria, as well as incorporate the features found to be protective against error.
The finding that participants who took longer to complete their prescriptions made fewer errors has implications for prescribing education. Excessive workload and distractions have long since been recognised as factors in error causation, but this study demonstrates a high residual error rate when FY1s were prescribing familiar drugs in a non-clinical environment with minimal distractions. Medical students and FY1s should frequently be reminded about the need for careful prescribing, and educated about common sources and types of error.
This study has developed a novel methodology for investigating prescribing error which could be used to explore error rates in other groups of prescribers. Furthermore, the model could easily be adapted to examine other prescribing-related issues such as the utility of prescription chart design for dyslexic prescribers. It might also be used as a method of piloting future prescription charts, particularly any proposed national chart.
Limitations
This study employed an innovative methodology in conjunction with a mixed modelling approach to explore the complex but endemic problem of prescribing error. It is, however, limited by the relatively small sample size and incorporation of participants from only five of Scotland's 14 NHS Boards. As with any study that recruits volunteers, it is possible that the FY1 participants were not representative of the entire population and disproportionally included those with an interest in prescribing initiatives or patient safety. FY1s who had moved region following graduation from medical school may have had some degree of familiarity with the chart from the region in which they completed their undergraduate training, but would not have actually used it to write real patient prescriptions. Reliance on the participants to record their own ‘lap’ times for each patient prescription unfortunately resulted in several incomplete datasets which had to be excluded from the analysis.
The environment in which the study took place was undoubtedly artificial. The instruction to ‘prescribe as though it was a typically busy day on the wards and the prescriptions would be administered directly to patients without any additional checks’ may have been insufficient to mimic prescribing in the workplace. The study environment provided fewer distractions and time pressures than a ward setting. Furthermore, this study only assessed transcription errors; FY1s were not asked to make decisions regarding drug type, dose, route or suitability. It is possible that in a more realistic environment, with the need to make complex prescription decisions in the context of time pressure and frequent distractions, opportunities for double-checking may be reduced and familiarity with documentation may influence error rates. Given the effect of timing (participants slowed down when prescribing on unfamiliar documentation), it would be interesting to see if error rates changed when the prescriber was not permitted to have as much time as they required. Finally, while the NHS Highland and NHS Grampian charts appeared to produce significantly higher error rates, this study was not powered to detect associations between specific errors (eg, drug dosage errors) and chart design.
Conclusions
This study has shown that some prescription chart designs are more likely to provoke errors than others. It also demonstrated that FY1s who took longer to complete their prescriptions made fewer errors. Familiarity with a particular prescription chart did not appear to influence error rate, although this may be an artefact of the study design and statistical tests employed. If prescribing errors are to be reduced, prescribing education must emphasise the importance of due care, and the workload of FY1s must allow adequate time for prescription chart completion.
Acknowledgments
With thanks to the 72 FY1s who gave their time and enthusiasm. Also thanks to Mr James Wallace for his assistance with data extraction, Mr David Hope for the statistical analysis, and the clinicians who coordinated local recruitment of FY1s: Dr Stewart Lambie and Dr Grant Franklin in NHS Highland, Dr Gerard McKay and Dr Alistair Dorward in NHS Greater Glasgow and Clyde, Dr Gordon Christie, Dr Sarah Ross and Dr Divya Raviraj in NHS Grampian and Dr Kerri Baker in NHS Tayside.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online supplement
- Data supplement 2 - Online supplement
Footnotes
-
Contributors VRT and RLH jointly designed the study, collected the data, performed the data analysis and drafted the paper. NGD helped to refine the research questions, advised on study design and critically revised the manuscript. SRJM advised on all stages of the study and was involved in the overall design, decisions regarding the error coding framework and critical revision of the manuscript. All authors approved the final manuscript for publication.
-
Funding This work was supported by a grant from the Association of the British Pharmaceutical Industry but they had no involvement in study design, data collection or analysis, writing of the report or the decision to submit for publication.
-
Competing interests VRT, RLH, NGD and SRJM are members of the Single Prescription and Administration Record for Scotland (SPARS) group which has the remit of driving forward evidence-based design, implementation and evaluation of a unified prescription chart in Scotland.
-
Ethics approval Ethical approval for this study was waived by the South East Scotland Research Ethics Service under the terms of the Governance Arrangements for Research Ethics Committees in the UK.
-
Provenance and peer review Not commissioned; externally peer reviewed.