The epidemiology of drug errors provides an eclectic range of results that differ according to the definition of key terms (Dean, 2003), method of detection, and elected measures. Yu et al. (2005) found the term “near miss” generated three functional meanings from 16 different definitions. The incidence of both errors and adverse drug events (ADEs) increases from voluntary self-reporting of errors at 0.15 per cent (Ross et al., 2000) through retrospective medical record review of ADE at 2 per cent (Wilson et al., 1995), and matched case control of ADEs at 2.43 per cent (Classen et al., 1997), to chart review of ADEs with stimulated self-report at 6.5 per cent (Leape et al., 1995). Observational methods have estimated error incidence at 19 per cent (Barker et al., 2002).

Errors, untoward incidents, and adverse events in healthcare, despite their sometimes negative consequences can be valuable opportunities for individual and organisational learning (Leape, 2002). The National Patient Safety Agency (NPSA) now has 97 per cent of NHS trusts linked up to the National Reporting and Learning System (www.npsa.nhs.uk accessed 27 February 2006). Nevertheless, the culture of reporting errors or adverse events in the British health service may remain less than positive. An independent inquiry into a series of infant deaths, clearly expressed the need for “a duty of candour” amongst health professionals, having recognised the traditional “blame” barrier to reporting (Kennedy, 2001, p. 107). In reality some practitioners may not engage in reporting due to this fear of blame (Vincent et al., 1999; Firth-Cozens, 2002) but also their subjectivity in choosing what is worth reporting (Shaw et al., 2005); time constraints (Waring, 2005); lack of feedback (Vincent et al., 1999; Evans et al., 2006); differences between professional cultures (Lawton and Parker, 2002); and the perpetual classification confusion (Tamuz et al., 2004). It also appears doctors are generally less likely to submit a report than their nurse colleagues in either electronic (Tuttle et al., 2004) or non-electronic systems (Lawton and Parker, 2002).

As well as barriers in the reporting of drug errors, the reported data is not representative of anything other than reported phenomena; furthermore valid, reliable conclusions about error incidence or baseline denominators cannot be drawn from incident reports (Battles and Lilford, 2003; Lilford, 2003; Wald and Shojania, 2003). Nevertheless, review of incident reports is unlikely to yield many false positives (Flynn et al., 2002), and there is documented optimism about the value of the qualitative data captured (Runciman, 2001; Kaplan and Barach, 2002).

Taxonomies can contribute to the original design of incident reporting systems (Zhang et al., 2004), and any post hoc analysis (Arah and Klazinga, 2004) to advance organisational learning. It appears the validity of existing taxonomies is, however, often untested; this type of study can offer an opportunity to do so which would in turn provide an evaluation of the reporting system. Content analysis could additionally evaluate the effectiveness of reporting specific incidents such as drug errors.

Within the trust studied, 800,000 drugs are dispensed and approximately 30 million drug doses are given annually. The trust operates a paper-based voluntary, non-confidential reporting system, which is linked to the National Learning and Reporting System at the NPSA. Following the entry of standard biographical data about the incident, the first of three free text boxes is completed to describe circumstances, a further free text is completed concerning causation, and a third text box is completed for action taken.

A 50 per cent random sample of all incident reports that were coded as drug errors (excluding blood transfusion errors) between 1999 and 2003 was analysed using a specific data extraction tool. Drug errors, near misses and adverse events were identified according to established definitions (NCCMERP, 1998, Department of Health, 2000); 10 per cent of the incident reports were subjected to co-rating.

Free text was analysed using a method of content analysis (Holsti, 1969), advocated by Scott (1990) for institutional documentary analysis. Ethical approval was given by the local research ethics committee and local research governance requirements were met.

The rate of clinical incident reporting compared to patient accident reporting in the study trust from 1996-2004 demonstrated a sharp rise from the summer of 2000, which coincided with a change in the reporting system and a publicity drive from the risk management department to encourage clinical reporting. Definitive drug errors from the reported drug incidents in the five-year period are shown in Figure 1, the total number of drug errors in the five-year period was 991. Adverse events resulted from ten (1 per cent) of the errors, and near misses 258 (26 per cent).

Figure 2 shows that the surgical department in the trust (including all theatre units) submitted most reports, followed by the medical and paediatric units. Mean rates of finished consultant episodes (FCEs) as a denominator for submitted reports can give a crude reporting rate, according to one measure of patient activity. Here it suggests a drug error is reported in 2.64 per cent of paediatric FCEs, 0.88 per cent of medical/elderly care FCEs, and 0.64 per cent of surgical FCEs.

Using the NCCMERP definition of a drug error and the accompanying framework, it was possible to identify error types in all of the definitive drug errors (Figure 3). The leading error type was dosage error 267 (26.9 per cent) which when subdivided showed overdose to be the most common 136 (50.9 per cent). However, 26.1 per cent were classified as “other”; either being less than 2.5 per cent individually (e.g. wrong rate) and sometimes out with the taxonomy.

In 27 per cent of definitive drug errors, neither causes nor contributory factors could be identified from the incident report free text (Figure 4). Over the five-year period however, there was an overall fall in the percentage of drug error reports with no identifiable factors, from a peak of 37.0 per cent in 2000 to a mean of 25.4 per cent in the subsequent three years. Examination of the error types: wrong drug, and monitoring error allergy (where a patient received or was to potentially receive a drug to which they were known to be allergic) showed no identifiable cause or factor(s) in 30.9 per cent and 54.2 per cent of cases respectively. Where however contributory factors were identifiable in the free text, the two leading contributory factors were related to communication – whether this was written miscommunication, e.g. a senior house officer writing a prescription illegibly, or a communication system breakdown. Whilst Figure 4 may give the impression that the various factors exist independently, content analysis suggests many are not, for example:

Circumstances: discussed the incident with doctor, I asked to administer salbutamol 5 mgs via nebuliser, following which the patient developed a tremor and fast pulse. Prescribed medication then found to be inhaler and not nebuliser. I feel following this incident the demands placed upon all staff on this shift left us open to potential errors in practice” Cause: staff levels and work load. Action: [nil comment] (417/00).

Inter-rater reliabilities for key judgements were calculated on a 10 per cent random sample of all incident reports reviewed (n=1,253) using the Kappa (κ) statistic. Two independent reviewers identified 95 of the incidents as drug incidents. Agreement on what was/was not a definitive drug error was only fair (κ=0.39), on what was/was not a near miss was good (κ=0.71), and on error type was very good (κ=0.86). There were 25 options for contributory factors but Kappa was not used here as there were a considerable number of errors where multiple factors were identified. Alternatively contingency tables were constructed to assess agreement, there were notable differences in the attribution of “performance deficit” and “unknown” as factors (Tables I and II). Cross comparison of the two factors showed that 16 of the 22 primarily classified as “unknowns” were then classified as (individual) performance deficits by the independent reviewers.

Further content analysis demonstrated descriptions of “circumstances” that varied considerably in depth and clarity; “causation” was sometimes recorded as a summary iteration of the circumstances or simply that an error had occurred. “Action taken” sometimes inclined to comment on individual's deficiencies rather than on those of the system. Albeit sometimes difficult to identify causation, especially if the person discovering the error is in a different location to the perceived source, or has been otherwise unconnected, analysis of the causation of definitive errors in 2003 showed that 35 (10.9 per cent) made no comment on potential or actual systems factors, for example:

[…] about to give IV tazocin as prescribed when noticed that patient was allergic to penicillin. This was second dose, so one already given. Cause: patient allergy. Action: copy to all involved and discussion with nurse concerned (126/03).

Where there was an absence of detail or consideration of systems factors related to causation, the relationship between causation and action taken provided little detail, this being particularly evident in the following example:

[…] whilst giving patient's night time medications, I noticed that her evening dose of insulin had not been signed for, I felt it inappropriate to ask the patient if it had been given (had been confused), I decided as it was late, not to ring the nurse who had administered the drugs earlier. Blood sugar was 12.8, and I continued to monitor blood sugar levels overnight. I informed the nurse in charge next morning and realised I should have alerted the doctor. Cause: human error. Action: dealt with through disciplinary procedure (367/03).

Furthermore, as the reporting system studied was not confidential, ward staff and others outside of the unit were aware of the name of the person against whom disciplinary action was to be taken. Mention of disciplinary action against staff was also evident following incoherent and illogical statements concerning circumstances:

[…] informed by relatives that nurse had handed patient a syringe of oramorph and he had injected this via his venflon. When patient questioned, he said he had seen someone put it down there, which previously was pethidine for intercostals drain insertion. Spoke to nurse concerned who said she gave first dose orally. Doctor informed, and pharmacist. Check for skin reactions. Action: second drug related incident, so nurse seen in disciplinary meeting (298/03).

Nurses were the most prolific reporters of definitive drug errors (80.2 per cent). Doctors formed 6.97 per cent of reporters and of these 56.5 per cent were senior registrars or consultants. Pharmacists appeared to form only 5.05 per cent, but within this trust, the majority of pharmacists use a regional reporting system, sometimes additional to the trust system but also instead of the trust system. Expressing the figures however as a percentage of their respective professional populations, demonstrates that 54.1 per cent of the nursing workforce reported, 12.9 per cent of doctors, and 151 per cent of pharmacists (each pharmacist reported at least one error). The population of protagonists in a definitive drug error, mindful of the above percentages of reporters, was dominated by nurses but by a much smaller margin (40.8 per cent); also the number of protagonists who were doctors was greater (12.7 per cent) clearly demonstrating that reporters can be vicarious or “second hand”. In 37 reports, nurses were subject to disciplinary action and or counselling following an error, compared to five pharmacy staff, and one doctor.

There is potential value in voluntary self-reporting systems, such as the revealing of latent conditions (Thomas and Peterson, 2003). Yet this is dependent on the reporting process allowing simplicity of completion, effective feedback, and a non-punitive approach (Leape, 2002), as well as the ability of staff to adequately complete an incident report. This study does not of course provide an error rate but simply a reporting rate for drug incidents and errors. The increasing rate of reporting across the trust may be suggestive of a greater awareness of the value of reporting, and is reflective of the British national trend (National Audit Office, 2005). The proportion of errors reported that were near misses (26 per cent) is somewhat higher than the 20 per cent estimated by the National Audit Office (2005) for acute trusts in the UK which is also encouraging, knowing that near miss reporting is typically uncommon despite its value in uncovering error rescue (Kaplan and Fastman, 2003; Evans et al., 2006). A reporting system should not, however, be evaluated on the basis of reports received but on the quality of information, and its ability in “identifying vulnerabilities” (Bagian et al., 2001, p. 524).

The different reporting rates across specialities is nevertheless a safety alert in itself – the measure applied here, albeit crude may demonstrate a rate of reporting that is three times higher in paediatrics (including neonates) than medical and elderly care. Questions might be raised as to why such a speciality differential exists and how much is due to complexity of pharmacological interventions and dosage calculations in for example, paediatrics (Simpson et al., 2004; Kaushal et al., 2001); or tangential to this a raised staff awareness of the lower threshold for harm after a drug error. The differential may also be attributable to a reduced tolerance of patient safety violations (Shaw et al., 2005), allied to a more open and supportive culture (Giles et al., 2006).

Defining a drug error and the error type should be straightforward, although the inter-rater scores do illuminate problems of interpretation even at the level of definition. That the most common reported error type identified was dosing error (26.94 per cent), is reflective of the findings of Selbst et al. (1999), and Ross et al. (2000), who also drew on incident report analyses but in paediatric settings. The leading identifiable contributory factors were concerned with communication. Dean et al. (2002) found that poor communication of treatments was seen by junior doctors as a key factor in the genesis of prescribing errors. More generally, but again with junior doctors, Sutcliffe et al. (2004) confirmed communication difficulties as one of the leading factors in medical error. Establishing a knowledge of any relationship between contributory factor and error type was difficult due to the superficial nature of the report data, importantly this may be partly attributable to a lack of structured prompts for reporters (Malpass et al., 1999). Such prompts may also give reporters the opportunity to draw contributory factors from a comprehensive list and if several are selected – grade their influence.

While the associated Kappa scores on error type were very good, the confusion that can arise between what is an error and what is a contributory factor is a substantial challenge to both reporters and risk managers. For example, is a written miscommunication such as an illegible prescription the cause of an error or the effect of a systems problem such as reduced staffing and high workload? The inter-rater process clarified the potential problem in using related terminology. Furthermore, the primary rater (GA) was more aware of the superficial nature of the incident report text having reviewed the whole sample and thus less inclined to apply the value laden term “performance deficit” as a contributory factor than the secondary raters – indeed more inclined to classify most of the same cases as having “unknown” factors. This perhaps highlights a weakness in the taxonomy applied here. Error causation appears predicated on external manifestations but neglecting the cognitive processes of those present (Kostopoulou, 2006); this suggests the absence of a systematic theoretical foundation that may impede an understanding of the mechanisms inherent in error (Zhang et al., 2004; Kostopoulou, 2006), and inadvertently assign blame.

In the paper-based system studied here, which utilised the three stage model of circumstances, cause, and action taken, it was clear that even when practitioners were able to elucidate an adequate description of circumstances, the contributory factors were not consistently identifiable, and the action taken did not consistently acknowledge the influence of active failures and latent conditions, which are seen to be instrumental in the aetiology of errors (Reason, 1990a, b). The illogical relationship sometimes seen between circumstances and cause may have been exacerbated by the process of allocating staff separate and senior to the protagonist in order to identify causation. Using the term “human error” in isolation provides little information but may crucially suggest staff require specific guidance on what to detail.

Understanding the context of error from incident report analysis may then be difficult but an analysis of the context of reporting less so. The free text concerning the circumstances, suggested causation and action taken is also in part, a social product of the organisation and its culture and subcultures, and thus it may provide valuable information about individual, local and organisational responses to error. Variations in reporting are thought to be more closely related to incentives and local culture than quality of health care (Cullen et al., 1995). As suggested, the lack of a clear relationship between causation and action taken may imply difficulty with the principles of risk management, but also a belief that allocating individual blame may be necessary. As Runciman et al. (2003, p. 976) has argued, “outcome is a poor index of blameworthiness.”

Assessing an organisational culture is however complicated by professional differences and the subcultures therein. Doctors may remain reluctant reporters. Daniels and Marlow (2005) have documented that even in relation to self-injury, whilst surgeons suffer the most needle stick injuries they were least likely amongst health professionals to report them. Familiar reasons for this were put forward including time constraints, lack of knowledge about reporting and concerns about confidentiality. The rates of reporting by profession bear a symmetry with the first report from the NPSA's (2005) National Reporting and Learning System which showed that 73 per cent of their received reports were from nurses and 7 per cent from doctors (Vere-Jones, 2005); and other studies (Uribe et al., 2002; Lawton and Parker, 2002). Here again nurses have reported far more frequently than doctors, but Leape et al. (1995) have also suggested nurses to be the principle interceptors of doctor's prescribing errors, although pharmacists might dispute this.

A key limitation lay in the process and structure of incident reporting: they are voluntarily submitted, biased by a number of known psychological phenomena, not necessarily representative of the organisation other than those who report, and are limited by their degree of detail, objectivity and clarity. This study was carried out in one health care organisation in one city and as such the quantitative findings may not be generalisable. However in respect of the generalisability of the qualitative data, it is claimed that the organisation and its risk management processes were theoretically representative of many acute hospital trusts across the country. It would then be useful to establish any similarity between the findings here and in qualitative studies of other trust's error reporting behaviours. Measuring the degree of agreement between researchers in assigning phenomena to discrete categories is designed to assure the process was not solely based on the subjective judgements of an individual researcher (Pope et al., 2000) but here only 10 per cent of the sample was subjected to the measure.

Finally, the most recent data here are now over two years old but do provide a longitudinal assessment of the way in which a healthcare organisation has addressed reporting over a period of time when there was an increasing emphasis on risk and reporting. Data such as these can provide a useful evaluation of the reporting systems.

It is the organisational response to reporting that is more likely to improve patient care than the reporting process (WHO, 2004). Yet if the process is not best equipped to draw out the right detail and the practitioners involved lack an understanding of the system's objectives, the response will be less than effective. Reporting health care errors is challenging to health care professionals for various reasons, but ambiguity in terminology is a key practical consideration (Dineen and Walshe, 1999) – whether in taxonomies, or the reports themselves. The problem is particularly evident in medication safety (Yu et al., 2005). A single taxonomy was applied here to define drug error, error type, and contributory factors; this was very helpful but not problem free.

While practitioners sometimes complete incident reports under strained circumstances without the option of protected time for risk management activities, their task should be made easier by giving them sufficient options concerning both type and contributory factors with additional cues to comment on human (active failures) and system factors (latent conditions), and perhaps a specific incident report for drug incidents. Identifying vulnerabilities may then be easier for all concerned. These developments, which are in place in some organisations would be easily transferable to electronic or telephone reporting systems.

In summary, an incident can provide a “window” on the system (Vincent et al., 2004) but further research is required, ideally a multi-centred study for an appropriate assessment of reporting standards mindful that best practice in incident reporting remains unknown (Nash, 2003).