Principal findings
Descriptive and thematic analysis of a large number of nationally reported patient safety incidents of diagnostic error showed that a third related to errors in clinical assessment, a quarter to inappropriate response to diagnostic imaging/investigations and one in 12 to failing to order diagnostic imaging/investigations. Staff human factors, including mistakes, were common. This was consistent for both delayed and wrong diagnoses and across most diagnoses.
Key diagnoses implicated in reports of diagnostic error included hip and cervical spine fractures, myocardial infarctions and intracranial bleeds. Most of these reports detailed incidents of misinterpretation of radiographs, failure to order correct investigations and a lack of sufficient assessment of the patient. Common related contributory factors with these reports concerned inadequate skill and clinician mistakes.
Strengths and limitations
Underreporting is an established methodological problem in patient safety incident studies but also in similar analyses of data in other high-risk industries [
14,
40]. The true incidence of diagnostic error in emergency departments will be higher than we have found. Focussing our analysis on reports where the incident type was ‘diagnostic error’ relies on a reporter suspecting that a diagnostic error has occurred; many diagnostic errors will not be reported as clinicians may be unaware a diagnostic error has occurred. Thus there may be other reports contained within other NRLS categories, such as reports concerning treatment error, which would have not been included in our analysis. No missed diagnoses were coded in the sample. However, for the definition we have used, these may be more likely to present in primary care than the emergency department and such cases may be less likely to be coded as a safety incident by hospital staff [
41,
42]. The NRLS is known to have limitations, with incident reporting often influenced by campaigns and alerts that raise awareness of certain incidents and disease, and its reports criticised for having poor data standardisation [
43]. Development of the Patient Safety Incident Management System (DPSIMS) is currently in progress to replace the NRLS and address these limitations [
44].
The reasons for submitting reports are also complex, meaning there will be a degree of selection bias that it is impossible to quantify [
45]. Several reports were excluded (57%) as they contained insufficient detail or were irrelevant to the subject of diagnostic error. Only a limited number of reports could be evaluated for harm severity. There is a risk of detection bias in the selection and subsequent coding of reports, as this depends on the application of the Primary Care Patient Safety (PISA) taxonomy by report raters. We attempted to counteract this with 10% of the reports double-coded, showing a kappa score of 0.868. Scores higher than 0.700 have been accepted in similar research studies [
34,
46] and our methods and training have mirrored these previous research studies.
Though we were able to ascertain the frequency of the types of diagnoses mentioned in reports, we do not know what are the commonest conditions that present to emergency departments. It is difficult to determine whether diagnostic errors are reported with the conditions frequently mentioned because these conditions are more prone to diagnostic error or because these conditions are common presentations in the acute care setting.
The number of reports and their breadth across the UK is informative and potentially transferable for looking at common diagnostic errors nationally. Consistent patterns and inferences, particularly for important conditions or contributory factors, enable the identification of interventions that could be applied to all emergency departments. We could find no previous studies of this size that have analysed patient safety incident reports of diagnostic error occurring in emergency departments.
Comparison with the literature
The high levels of insufficient assessment reports across a number of diagnoses suggest that there are common sources of these types of errors. These include cognitive and system errors [
47]. Cognitive errors are recognised in most cases [
17], and are often related to clinician expertise and experience [
48]. These human mistakes can be worsened in the emergency department by time constraints on staff for patient assessment and investigation [
49].
Several interventions have been suggested to reduce the occurrence of diagnostic error cases. Few of these suggestions have been tested in clinical trials [
50,
51]. System-based modifications that optimise clinician skills and use processes for mitigating errors have been shown to reduce the rates of adverse events significantly [
52]. Simple programmes, including a whole systems examination intervention [
53], assigned training in electrocardiograms (ECG) interpretation [
54] and diagnostic checklists [
50] can be effective in localised settings. Alongside formal emergency department staff teamwork training [
55], these could help mitigate contributory factors, such as limitations in knowledge and cognitive mistakes, and reduce rates of diagnostic error.
Imaging errors, encompassing failure to image appropriately and errors in interpretation, featured prominently in our analysis. Measures that support junior colleagues to more accurately interpret investigations could reduce the number of diagnostic errors [
56]. Potential strategies have been cited [
57], with recognition that interventions should focus on adapting both educational and system approaches. Changes in both these elements are needed to successfully reduce diagnostic investigation interpretation errors [
58]. Radiology interventions can be non-technological, such as structured reporting [
59] or double reading [
60] of imaging results, or technological, such as perceptual feedback or attentional guidance [
61]. Though these interventions show promise, it is unlikely the majority of emergency department clinicians will reach the same technical standard as radiologists [
62]. Thus more prompt secondary reviews of radiographs are also needed to reduce the impact of missed fractures [
63]. Few of these interventions have been tested [
64] but some have shown promise including radiological checklists [
64] and computer-aided detection [
65].
Other identified errors in our analysis included failure to correctly interpret and follow up other investigations including laboratory results and ECGs. Both continuous education feedback strategies [
66] and standardised forms to drive follow-up of investigations are effective interventions [
67]. Specific diagnoses, such as abdominal aortic aneurysms, require specific interventions that address challenges in their diagnostic pathway. For example, a low threshold for immediate CT scanning and greater involvement of emergency department clinicians in ultrasound examinations may help reduce missed abdominal aortic aneurysms cases [
68,
69]. Similar thresholds or decision tools are applicable to detection of high-risk fractures such as hip and cervical spine fractures. Increasing the utility of these tools and awareness of them could improve emergency department diagnosis for these patients [
70,
71].
Diagnostic error is a challenging field to act upon [
72‐
74] but opportunities for improvement can be addressed using a Plan-Do-Study-Act model and through system quality improvement [
75]. Small adaptations, across the drivers of Fig.
2, that add up to an overall system modification could help address the multiple causes of diagnostic error and improve emergency department diagnosis. “Blame and shame” approaches do not contribute to learning and system improvement [
76]. Instead, future research should be directed towards implementing suggested interventions with a system-oriented direction. These are needed alongside cultural shifts and organisational restructure to be sustainable [
77].