Background

Diagnostic errors represent a source of iatrogenic harms whose adverse impact on patients may exceed the total morbidity and mortality from all other medical errors combined.¹ The number of diagnostic errors each year in the United States is estimated to be at least 12 million but could be an order of magnitude higher.¹ Estimates for the number of Americans seriously harmed each year as a result of diagnostic error (so-called “misdiagnosis-related harms”²) range from approximately 40,000 to 10 million,¹ but the most plausible values are likely in the hundreds of thousands.³ Misdiagnosis-related harms range from none to serious (i.e., permanent disability or death). Costs to the U.S. healthcare system may exceed $100 billion per year.⁴

Despite their toll on patients and society, diagnostic errors remain largely invisible. This is mostly because diagnostic errors are rarely evident at the time when they occur and only surface later, often when they are discovered by another clinician or after misdiagnosis-related harms have occurred. Furthermore, diagnostic errors are variably defined, difficult to measure, and not routinely tracked as part of patient safety or quality improvement initiatives. The National Academy of Medicine (NAM) defines diagnostic error as “the failure to (a) establish an accurate and timely explanation of the patient’s health problem(s) or (b) communicate that explanation to the patient.”⁵ Notably, this definition (which is used in this report) does not require a care process failure (e.g., a specific clinical reasoning “mistake” on the part of an individual clinician) and is agnostic with respect to any resulting harms or their preventability. Furthermore, it does not elaborate on the words “accurate” or “timely,” nor does it draw distinctions between false negative and false positive errors or specify how management differences might be used inferentially in assessing the “correctness” of diagnostic decision-making. There is no clear consensus on how to define “diagnostic error” at this deeper level,⁶ but some authors focused on emergency department (ED) diagnosis have made important attempts to do so. For example, a Swiss group examining diagnostic errors among admitted ED patients divided differences between ED and final hospital discharge diagnoses into those that were deemed, in their view, not to represent ED diagnostic errors (ED diagnosis was somewhat underspecified or a complication not present at the time of the ED visit later became the primary inpatient diagnosis) and those that were considered diagnostic errors (ED missed a second, more important diagnosis or ED diagnosis was qualitatively incorrect).⁷ There is even less certainty about how best to capture communication failures between ED clinician and patient, and very few studies have sought to address this aspect of diagnostic error definitions.⁸ Whenever possible, we relied on the NAM definition of diagnostic error (e.g., to differentiate diagnostic errors from diagnostic errors with adverse events or harms), but we also relied, as necessary, on individual study-based operational definitions, including more granular determinations of error, harms, and preventable harms that were used in the included studies.

Diagnostic errors occur across all medical settings. Overall diagnostic error rates are not known, but a commonly cited figure is that 10-15 percent of medical diagnoses are wrong.⁹ Hospital autopsy studies tend to corroborate this figure, with estimated major error rates of 8 percent to 24 percent, even after accounting for selection bias in what deaths undergo autopsy.¹⁰ On sheer volume alone, most diagnostic errors happen in ambulatory clinic settings, where diagnostic errors are estimated to affect 6.3 percent of primary care encounters, translating to more than 12 million Americans suffering errors each year.¹¹ Because this figure is based largely on studies that searched retrospectively for adverse events (e.g., unplanned revisits), the true error rate is likely higher, but the rate of misdiagnosis-related harms in primary care is likely lower and closer to 0.1% (calculated as 187/212,165 from Singh et al., 2013).¹² Diagnostic adverse event rates (defined similarly to misdiagnosis-related harms) are estimated to occur in 0.7% of hospitalizations³; that this adverse event rate would be higher than in primary care makes sense, given the higher intensity of illness present among hospitalized patients.

There are approximately 130 million ED visits each year¹³ at the roughly 5,000 EDs in the United States.¹⁴ The ED is believed to be a high-risk site for diagnostic error,^15–20 but error rates are highly variable by disease¹ and symptom presentation.^{21, 22} Diagnostic error rates for myocardial infarction, for example, are impressively low at about 1 to 2 percent¹ (in part due to the availability of electrocardiograms and a reliable lab test [i.e., troponin assays]), but delays in diagnosing spinal abscess occur in up to 75 percent of such encounters in the ED.²³ Diagnostic error rates for stroke, a condition that tends to be more reliant on bedside diagnosis, are approximately 4 percent for those presenting with weakness but approximately 40 percent for those presenting with dizziness or vertigo.²¹ Most misdiagnosed patients do not suffer significant adverse events. However, even if preventable misdiagnosis-related harm rates in the ED were as low as approximately 0.1 percent (n=13 of 13,495 ED visits),²⁴ with 130 million ED visits per year in the United States,¹³ this would translate to approximately 125,000 ED patients harmed each year. Perhaps most importantly, there is evidence that error and harm rates are highly variable across individual EDs—for example, one epidemiologic study found that rates of missed myocardial infarction varied from 0 to 29 percent across hospitals²⁵ while another found that rates of missed appendicitis varied from 1 to 16 percent, with higher error rates correlating strongly with higher harm rates.²⁶

There are many reasons why the ED may be the most challenging clinical setting for diagnosis. The breadth of clinical presentations is extremely wide, with dozens of common symptoms caused by thousands of possible diseases. Patients seen in the ED frequently have dangerous underlying conditions causing their symptoms, but the manifestations may be vague or atypical. Patients are often seen early in the natural history of their disease presentation (sometimes just minutes or hours after the onset of symptoms), so “obvious” features may not yet have developed, and diagnoses may be highly uncertain. ED physicians usually have no prior relationship with their patients, and information at the time of evaluation is often incomplete (e.g., patients may not provide accurate information on past medical history or medications; the history of present illness may not be clear, as with altered mental status). Many EDs have limited access to specialty consultants or advanced diagnostic tests, such as magnetic resonance imaging (MRI). The ED environment is often fast-paced, chaotic, and distraction-rich, increasing cognitive load for clinicians.²⁷ Finally, the goal of ED care is often not to make a firm diagnosis, but instead to make a diagnosis-informed management decision about immediate on-site testing, treatment, or hospitalization, and generally to do so in a matter of hours. The asymmetry of false negative versus false positive diagnostic errors often factors heavily in such considerations—an “undercall” (discharging a patient with a dangerous disease) is generally viewed as less desirable than an “overcall” (admitting a patient without a dangerous disease). As a result, some forms of diagnostic “error” (e.g., symptom-only diagnoses or false positive overcalls among admitted patients) may simply reflect goals and context of care.

Understanding more about the frequency, contexts, and causes of ED diagnostic error offers a springboard for both research and operational quality improvement efforts aimed at reducing or eliminating preventable misdiagnosis-related harms. For example, if 15 specific diseases account for two-thirds of the serious harms from ED diagnostic error, then these diseases can become top targets for solutions both at the institutional level and nationally. If stroke is diagnosed well among patients with unilateral weakness but poorly among patients with dizziness, local clinical pathways and national guidelines for diagnosing dizziness could be developed. If local hospital diagnostic performance benchmarks are better than the overall average ED performance nationally but lag for aortic dissection, then that disease can be made the focus of a new local quality improvement program. Thus, this report offers decision-makers at every level (individual clinicians, ED directors, hospital safety officers, national policy makers, etc.) critical insights about diagnostic failures. It provides a detailed view of what is known, as well as what is not known, about ED diagnostic errors and misdiagnosis-related harms.

Purpose and Scope of the Systematic Review

Although diagnostic errors are known to occur across all clinical settings, the scope of this evidence review, commissioned by the Agency for Healthcare Research and Quality, was limited to the ED. The key decisional dilemma for this evidence review is “What are the most common and significant medical diagnostic failures in the ED, and why do they happen?” We conducted a systematic review to determine the most frequent diseases and clinical presentations associated with diagnostic errors (and resulting harms) in the ED, measure error and harm frequency, as well as assess causal factors and identify risk predictors for these errors.