This is the first systematic review of the cumulative medication errors of a country's healthcare system, linked to a discussion of the whole system. This approach allows the development of a system-wide approach to the improvement of quality in the use of medication. However, we recognize that there are limits to our methodology. We merged different types of studies and some areas of the map are more greatly populated than others. Weaknesses of the map are a reflection of the primary data. In the next section we reflect on the quality of the data which we found, as it affects our understanding of the whole system and also sets a research agenda of its own. We then discuss approaches to creating improvement.
Limitations of research reviewed
The limitations fall into four groups: (1) a dearth of studies, (2) the method of identifying and measuring the error, (3) the use of different units of measurement and (4) sampling limitations.
There is a dearth of evidence relating to some parts of the system, primarily related to prescribing. For example, studies which have investigated the accuracy of general practice medical records have focused on diagnosis rather than medication records [
48,
49]. From Figure
1 it can be seen how crucial the accuracy of the GP medication records are to the whole system and that further research in this area is urgently needed. In addition, research addressing the rates of the reviews of repeat prescriptions was conducted over 10 years ago and should now be updated [
28].
The methodology used to measure error has been problematic. Definitions of error have differed between studies which makes comparisons difficult. Some studies may not have been able to identify all the errors. For example, Shah
et al. [
21] conducted a study in which prescriptions were reviewed in pharmacies in order to determine the prescribing error rate: this is unlikely to detect all of the errors [
41]. Difficulties have also arisen when attempts have been made to determine the rates of errors in medication history taking when patients are admitted to hospital because there is no a gold standard with which to compare the medication history as general practice records have also been found to be inaccurate [
30]. Discrepancies rather than error rates have therefore been reported. Studies have classified discrepancies as intentional or unintentional when patients are issued repeat prescriptions after being discharged from hospital. However, in three of the studies GPs were not interviewed in order to check their intentions and therefore this classification may have been inaccurate [
36,
45,
46].
It is also difficult to compare error rates across the medicine management system as error rates have been measured using different units. Prescribing and dispensing errors have been measured using the number of items as the denominator in the majority of cases [
21,
24]. However, adherence has been measured per patient [
18,
19] and satisfaction with discharge communication has been measured per GP [
32,
33]. In future it may be best if error rates were expressed using more than one criterion, such as by act and by patient.
The limited sampling strategies employed in studies have led to some of the data collected being unrepresentative. Most studies conducted in the hospital setting have been carried out at a single site, sampled for convenience [
25,
27,
29,
37,
39,
40] and, in some cases, have been specific to a single patient group [
29,
37]. In addition, some studies have been carried out at a single GP surgery [
21,
22] or on a convenient sample of independent pharmacies [
24]. In some cases, there was no information given about the sampling strategy [
21,
26,
38] or the patients within a practice have not been randomly sampled [
22]. Lack of contextual information has meant that one cannot determine how representative the results have been. In addition, in other areas, such as adherence and preventable drug-related admissions, random sampling has been employed or there have been consistent results between studies leading to a cumulative validity.
Creating improvement
While we acknowledge the limitations of the literature, there is still sufficient information to indicate the ways in which the system could be made safer. There are several management techniques which have been designed to improve reliability and quality and to reduce waste - for example 'lean' aims to reduce waste by removing non-value added steps from a process and 'six sigma' aims to reduce the variation in order to produce a uniform process output. Given the opportunity for improvement based upon quantifiable error rates over periods of time, the six sigma approach to structured improvement is especially relevant to the improvement of quality of medication use. This approach uses a systematic methodology to define, measure, analyse, improve and control the situation. Such techniques have been successfully applied to healthcare [
10‐
13], including the reduction of medication error in secondary care [
14,
15]. However, there is little evidence of the successful application of these management strategies to the primary healthcare setting generally, or to the improvement in quality of medication use in primary care. Natarjan [
50] has identified a number of barriers for the application of improvement tools to primary care settings. They include: lack of awareness that problems exist; poor understanding of systems thinking; a traditional medical culture of individual responsibility; legal issues encouraging the concealment of error; poor information technology provision; poor data; and resource issues. In addition, unlike the situation seen in secondary care, patients have complete freedom of action and the healthcare may professionals come from several different organizations. Any solution must be able to address these challenges.
A systematic approach, based on the existing evidence, is required in order to identify how we should apply management strategies to the improvement of the quality of medication use in primary care. The approach requires a method of identifying priorities, a systematic measurement of error and the systematic design and testing of solutions. As problems of maintaining quality occur at every stage of the medicine management system in primary care, there is a need to prioritize the processes which need to be improved. We need to examine the impact of errors on the system as a whole and use that knowledge to develop an approach which will maximize its value to patients.
One method of choosing which processes need to be improved is to identify those that have both high error rates and which cause high levels of harm [
51]. Figure
1 shows the processes with the highest error rates. However, as data regarding harm is not available for all of the processes, this method is inappropriate. A more promising method would be to prioritize processes at the patient end of the system and gradually work backwards, thereby maximizing value to the patient [
17]. The aim of the system is to ensure that patients are taking medication successfully and that the medication is effective and not harmful. Improving these processes would be expected to lead to improvements in patient care. Medication adherence, effectiveness and lack of harm are therefore the stages on which we first should focus. These are the processes which are most important to the patient and in which there is high loss of quality. In order to improve these processes, we also a need to consider feedback loops within the system, another area in which there has been insufficient research. For example, it may not be possible to improve the NNT of a medication but effective feedback systems, such as medication reviews and monitoring, may allow prescribers and others to change ineffective or harmful medication and improve the quality of outcome.
Having identified the areas which should be given priority, the next stage is to measure and standardize processes at a local level. In order to assess the effects of the interventions, we need to establish standard methods for measuring system errors so that error rates can be monitored and compared. Statistical process monitoring is a management technique which can assist this process. It uses control charts to monitor processes [
52], allowing the measurement of changes in, and the predictability of, the mean error rate. It is necessary to know the predictability of error as management strategies cannot be applied to the reduction of error if the rates are unpredictable or chaotic; it is essential that error rates first be stabilized and that the adherence rates and clinical effects have been monitored as proxies for the desired clinical outcome. Adherence would probably need to be measured by techniques such as self report and dispensing records. The measurement of clinical effectiveness would depend on the drug being used, the condition being treated and number of admissions to hospital.
Once data has been collected, the analyses will indicate which solutions would be appropriate. If error rates are chaotic, the first stage would be to reduce variation. Control charts can identify whether the variations have a common or a specific cause. In order to reduce common causes of the variations we must improve the process, but to reduce specific causes we need to identify and act on factors which are extrinsic to the process [
52]. Once the process is stable, root cause analysis can be used as a tool to identify the causes of error. Once this knowledge is gained, appropriate solutions for reducing error rates can be identified and evaluated. It should be possible to extrapolate the information from representative and reproducible data collected at local levels and apply it at a national level.