A previous study of prescribing in intensive care organised by the UK Intensive Care Society (ICS) showed up to 15% of all drug prescriptions written in ICU, including rewritten prescriptions, but excluding prescriptions for fluids, contained errors.1 These prescriptions are often written for drugs with a narrow therapeutic index given in multiple combinations to very sick patients, and 20% of prescription errors were reported as potentially serious or life-threatening.1 Results from other reviews of prescribing in critical care have shown a similar risk to patients.2 Our intensive care unit (ICU) had contributed to the ICS study;1 a repeat audit in 2005 also showed no reduction in prescribing errors, and we therefore decided to develop an education package to improve prescribing in critical care. The prescription error rate in our ICU was 8% in the ICS study of prescribing errors1 and was 13% when this was repeated in 2005.

The ICU is a 16-bed mixed general and neurosciences unit admitting over 700 adult patients a year. Twelve consultants provide clinical sessions, and trainee cover is provided at lower (Senior House Officer) and middle grade (Registrar or Clinical Fellow). Each rota has seven to eight trainees who work a full shift pattern, and 40 trainees passed through the unit between February and October 2006, for attachments of 3 or 6 months. It is outside unit policy for a visiting or referring doctor to prescribe directly on to an ICU drug chart. The unit is supported by a critical care pharmacist (25 h per week) who attends ward rounds, reviews prescriptions, ensures drug availability and provides drug information to all staff groups. Each bed area has a bed-end computer with prescribing information, including unit protocols and access to the Electronic Medicines Compendium, the British National Formulary and Uptodate (an online medical textbook with drug information). The unit has a two-sided A3 prescription sheet that has been repeatedly redesigned in response to previous critical incidents and is designed to be valid for only 72 h. Prescribing studies3^–5 suggest that electronic prescribing may reduce many errors; however, this was not going to be available in our ICU for some years. We therefore developed an educational approach to improve prescribing, which consisted of three parts: (1) setting specific standards for prescribing; (2) providing education both as a formal tutorial and as work-based learning; (3) developing an audit and feedback tool to allow trainees to learn from their own prescribing.

METHOD

The study was approved by the hospital audit committee, who judged the project to be a systems improvement and therefore did not require ethics approval. The prescribing standards were developed after extensive consultation with stakeholders in the hospital and followed guidance in the British National Formulary6 and were also made in response to specific critical incidents that had occurred in critical care areas in Greater Manchester.7 These standards, shown in box 1, were circulated to medical staff, and laminated copies were put on notice boards on the ICU and in the patients’ medical notes folders. The educational programme started with a tutorial, given by the pharmacist, which was designed to be delivered within 2 weeks of each new intake of trainees. The tutorial discussed problems of prescribing with real examples from critical care; it also reviewed the standards for prescribing, described where to find information about medications on the ICU and reviewed the process of the subsequent audit. The tutorial was then followed by consultant ward-round-based teaching during the next week’s ward rounds. This teaching aimed to demonstrate how to access drug information from the bed end computers; particular attention was given the prescription charts to demonstrate good practice or errors. The trainees were asked to give a general overview of one drug prescribed for each of the patients they had reviewed that morning. The final section of the educational process was audit and feedback. Prescribing was audited once prior to the educational sessions, once after the teaching and finally 6 weeks later. The trainees were aware the audits would be performed, although the exact dates of the audit were only known to those conducting the audit. We aimed to repeat this every 3 months to coincide with new intakes of trainees. The audit was conducted by one or two trainees under the supervision of the pharmacist, who reviewed prescription charts to identify the number of correct and incorrect prescriptions written by each trainee. Incorrect prescriptions were then classified according to the nature of the error using the previously described system of classification8 that had been used in the ICS study.1 The potential seriousness of errors was then classified as minor, significant (possibility of mild adverse effect), serious (reduction in probability of the treatment being timely or effective) or life-threatening.9 Any actual harm to patients was also noted (defined as temporary harm or increased length of stay or permanent harm or intervention required to sustain life or contribution to death). Although the pharmacist could have conducted the audit themselves, we felt the involvement of the trainees would be educational and improve acceptance of the system. Each audit reviewed identified prescriptions written over a 5-day period. The completed audit forms were then entered into a Microsoft Access database which we designed for the audit. This had a number of standard queries that allowed reports of the total number of incorrect prescriptions per trainee and a list of the different types of errors made by each trainee to be produced. Each consultant wrote a maximum of two prescriptions per audit cycle so they were all placed in a single group; the only other prescriptions not allocated to individual prescribers were those where the prescriber could not be identified, normally due to failure to sign or print the name legibly. All other prescriptions were allocated to specific doctors who were each identified by a letter code. The total numbers of correct prescriptions (clinically appropriate and accurately prescribed) per doctor were then added to the reports, and all information was then presented graphically. Each doctor then received graphs for the total number of correct and incorrect prescriptions written by all the ICU doctors (anonymised, and similar to that shown in fig 1) as well as a list of the specific prescription errors that were attributed to them with the number of error-free prescriptions they had written. These reports initially took 4 weeks to produce for the first two data sets in the first audit cycle but subsequently were available within 1 week of each audit. Although the feedback was given as written and graphical information without a discussion of the results, the trainee staff were already supported with educational supervision, and each trainee met regularly with their supervisor to be given feedback on their performance. This feedback did not include a discussion about their prescribing errors as measured by the audit unless the issue was raised by the trainee; this is because the supervisors were not aware of individual trainees’ results.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 Total number of prescriptions with no errors and number of prescriptions containing one or more errors written by prescribers during the first data collection of the second audit cycle in May 2006. A similar graph was given to each doctor with a list of the individual prescription errors they had made. They were given their own letter code but not the codes of other staff.

The significance of the reduction in the number of errors was assessed using simple logistic regression, cycles 2 and 3 being compared with cycle 1 using indicator variables. The overall effect of training was tested using a Poisson regression model, with the standard errors and variance–covariance matrix adjusted for lack of independence within doctors’ error rates using the Huber–White sandwich estimator. The incidence rate ratio (IRR) between pre- and post-training was then assessed, and the overall effect of training was assessed using a chi-squared test.

RESULTS

The audit was repeated over three 3-month audit cycles starting in February 2006. The final audit cycle from August to October 2006 missed the initial pretutorial audit due to the authors’ annual leave. There were therefore a total of eight sets of data collection, two initial sets, three sets after the teaching sessions and three final sets. There were 40 trainees who worked on the unit during this time, with a range of 10 to 16 (median of 15) in each audit cycle. Thirty-four worked through one audit cycle, and six, who were on 6-month attachments, worked through two cycles. Twenty-two trainees were on the first on-call (senior house officer rota), 10 of these were anaesthetic trainees, the others were from a range of other disciplines, and all but two had had previous posts since registration. Thirteen of the 18 trainees on the second on-call rota were anaesthetic trainees with the remainder from medical specialties. There were a total of 3952 individual drug prescriptions audited, 363 of these prescriptions contained one or more errors, 284 contained a single error, 69 had two errors, and 10 had three errors. Types of error that occurred in more than five individual prescriptions are shown in fig 2. Forty-five of 363 prescription errors were considered serious, significant or life-threatening, 33 significant, 11 serious (including underdosing of antibiotics twice and combining non-steroidal anti-inflammatories with anticoagulants twice) and one life-threatening (under dosing vancomycin in a patient with septicaemia). We did not identify any direct patient harm as a result of a prescription error during these audit periods. The percentages of prescriptions with errors and the total numbers of prescriptions/prescriptions with errors of for the different audit periods are shown in fig 3. The total numbers of prescriptions written (total number of prescriptions with errors) for each audit were as follows:

• Download figure
• Open in new tab
• Download powerpoint

Figure 2 Frequency of different prescribing errors that occurred more than five times. There were a total of 452 errors in 363 error containing prescriptions in a total of 3952 prescriptions audited.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3 Error rates expressed as percentage of prescriptions that contained one or more errors for individual data collections. The data collection was repeated in three audit cycles, each designed to have three separate episodes of data collection, before training was provided, after training and again 6 weeks later. Each cycle lasted 3 months, from February to April, May to July and August to October 2006. The initial audit in August was not conducted due to authors’ annual leave. There was a significant reduction in errors (p<0.005) in each audit cycle. The number of prescriptions with errors/total prescriptions is shown under the x axis for each individual audit.

• pretraining audits 373 (97), 318 (59)—with one missing data point;

• post-training audits 513 (116), 231 (14), 230 (25);

• 6-week final audits 374 (26), 171 (7), 477 (24).

The IRR between pre- and post-training was 0.69 (95% CI 0.48 to 0.98). The IRR between pre- and 6 weeks final audits was 0.25 (95% CI 0.19 to 0.33) (p<0.005). The wide variation in the number of prescriptions written by trainees and the difference in error rates between trainees are shown in fig 1, which summarises the first audit in cycle 2 (May 2006); this shows a range of prescriptions written by prescribers varying from 1 to 61 (median 18) with error rates of 0% to 100% (median 15%), and similar results were found in all other cycles. While investigating the reasons for the variation in the number of prescriptions written by different prescribers, it became apparent that the night staff rewrote a disproportionate number of prescriptions, a practice we have subsequently tried to discourage to ensure even distribution of work between staff. The profile of error types also varied between trainees—for example, a trainee who was good at transcription may have been weak at another aspect of prescribing. In the first audit cycle, we were unable to identify the prescriber in 48 prescriptions that had an error rate of 63%. We subsequently collected the initials as well as the signatures for all new staff so that the number of prescriptions with no identified prescriber did not exceed 10 in any subsequent audit. Some consultant staff resented the imposition on the ward rounds, which are conducted over 2–3 h between the availability of investigations and the start of visiting time. We did not formally assess the views of trainees, but informally the trainees welcomed the feedback provided about their prescribing.

DISCUSSION

Our study showed that an educational programme starting with the setting of standards, using tutorials and ward-round teaching and repeated feedback reduces prescribing errors in ICU. We also showed there was a very wide range in the number of prescriptions written by doctors and in their rates and types of prescribing error. Fortunately, most errors were minor, and of those errors that were serious, significant or life-threatening, none resulted in actual harm to patients. The fact that no patient was obviously harmed when a number of life-threatening errors were made was in part because each audit only collected audit data over a few days, so the pretraining audits with the highest error rates represent only 10 days’ prescription writing. Many errors (such as omission of signature or omission of a drug on rewriting the chart) would appear to have been avoidable had electronic prescribing been available.3^–5 The prescribing error rate, types of errors and their significance were similar to those previously described in UK ICUs.1 This is higher than rates observed on general wards10 but is comparable with rates found in ICUs in other countries.2 Opportunities for improving prescribing include improving the prescribing culture to ensure the importance of correct prescribing is recognised, the involvement of a pharmacist, provision of high-quality and accessible drug information, feedback to prescribers and electronic prescribing.11 When prescribing errors do take place, it is important that prescribers12 and the team13 learn from these errors. The importance of feedback in medical education is well recognised.1415

Although some formal education about prescribing was important to provide factual information, the provision of feedback was key to improvements in prescribing. This feedback was helped by providing specific factual information about both correct and incorrect prescribing that was repeated, anonymised and timely as well as being given with the context of the performance of other prescribers. Although the feedback was a one-way provision of information, it was given in a supportive education environment with active educational supervision. Evidence of the relative importance of the feedback can be seen by the very small improvement between the pre- and post-training audits in cycle 1 (where technical problems with data analysis delayed the provision of feedback until after the post-training audit) when compared with cycle 2, where feedback was provided between the pre- and post-training audits. As the trainees have been in clinical practice for at least 12 months or more, it seems unlikely their prescribing would have improved spontaneously in the few weeks of their ICU attachments.

The identification of prescribing errors such as legibility of handwriting is partly subjective on the part of the data collector, and the wide range of prescription error rates reported in different ICUs in the ICS study was partly due to differences in interpretation of what constituted a prescribing error.1 We tried to limit this problem by the use of specific error categories and direct supervision of the audit by same pharmacist. The unit pharmacist has more than 5 years’ experience in critical care, but any competent qualified pharmacist could fulfil this role in other clinical areas. The audit therefore requires a critical care pharmacist as well as secretarial support to produce the audit reports and clinical leadership to introduce the process. The established process now takes about 3 h of trainee time, 1 h of pharmacist time and 1 h of secretarial time a month to run the audit and 1 h every 3 months to deliver the tutorial. Due to changes in medical staffing, we no longer have groups of trainees starting at set times, therefore a CD training package has been developed and trainees receive feedback every month of their ICU attachment.

In summary, our study shows that ICU trainees vary widely in the number of prescriptions they write and in the error rate of these prescriptions. An educational programme using the setting of standards, formal education and repeated feedback reduces these prescribing errors.