While the true rate of misdiagnosis in modern medicine remains uncertain, our best estimates suggest that the rate of diagnostic error has remained constant over time. When misdiagnosis results in obvious patient harm, prolonged hospital stay, multiple diagnostic tests, or overutilization of health care resources, then we are prompted to detect the cause and to attribute responsibility for diagnostic error appropriately. When a patient complains, or when a diagnosis results in a course of expensive treatment and/or testing, then the diagnosis in question seems to receive additional scrutiny by default. It is likely that many cases of diagnostic error do not result in permanent patient harm, and certainly many patients get better despite their physicians’ lack of diagnostic skill. In such cases, it is likely that misdiagnosis will remain undetected. However, it is quite possible that many diagnostic errors remain undetected even when such errors contribute to real patient harm. This may happen, for example, when a patient is lost to follow-up. In the worst case scenario, diagnostic error will remain permanently concealed when a patient’s death is explained by the wrong diagnosis.

It is difficult for any given physician to estimate his or her own rate of diagnostic error. When diagnostic error occurs, the physician in question is, by definition, unaware of it. So, it should come as no surprise that physicians are poor judges of their own performance. In fact, physicians’ certainty regarding their own diagnoses has a poor correlation with diagnostic accuracy. Friedman et al. performed a study comparing the concordance between diagnostic certainty and diagnostic accuracy among faculty internists, senior residents, and medical students [1]. All participants generated a list of differential diagnoses from a standard set of clinical case scenarios, and they were questioned about their certainty regarding each diagnosis. Students tended to have a higher rate of agreement between confidence and accuracy, reflecting their general tendency for underconfidence to match their higher rate of diagnostic error. On the other hand, faculty internists demonstrated the highest rate of correct diagnoses, but the alignment between confidence and accuracy was relatively poor: 64 % [1].

Davis et al. performed a comprehensive meta-analysis of studies evaluating the accuracy of physician self-assessment [2]. While there is wide variation in study design and quality, most studies demonstrate that physicians have an inaccurate understanding of their own performance. In general, there tends to be a poor correlation between physician self-assessment and external objective measures of performance. Furthermore, those physicians who demonstrate poor performance scores tend to have the least accurate assessment of their own ability. It may come as no surprise that those physicians most in need of improvement are the least likely to recognize their own errors.

The baseline rate of diagnostic error tends to vary according to specialty. The rate of error for the so-called “perceptive fields” of radiology and pathology can be measured directly with over-reads of initial diagnoses and with chart review. With these specialties, it is easier to isolate the diagnostic episode (interpretation of an xray, reading of a histologic specimen, etc.) and to determine whether the original diagnosis was correct. The rate of diagnostic error in these specialties tends to be around 5 % [3]. Clinical specialties, such as internal medicine, family practice, or emergency medicine tend to demonstrate a higher rate of diagnostic error. In these specialties, the process of diagnosis is not a self-contained discrete episode. Rather, the process of diagnosis runs concurrently with other processes of patient care, including data acquisition, ordering of tests, communication with patients, and provision of treatment. The rate of delayed diagnosis and misdiagnosis in clinical specialties as measured by chart review is approximately 10–15 % [3]. It is possible that a number of misdiagnoses remain undetected, such as when a patient is lost to follow-up or when an illness resolves spontaneously despite the wrong diagnosis.

In the worst case, when a diagnostic error results in a patient’s death, the ultimate measure of this is autopsy. In a meta-analysis of multiple autopsy series spanning four decades, Shojania et al. demonstrate that the diagnostic error rate as reported in the English literature has tended to decrease over time [4]. This may reflect improvement in diagnostic accuracy, or it may reflect selection bias. In the United States, the autopsy rate has decreased considerably over the last half the twentieth century, from 40 % in the 1960s to less than 6 % in the 1990s [4]. So, it is possible that a number of lethal misdiagnoses remain undetected. Even so, modern autopsy series demonstrate a significant rate of diagnostic error. Shojania et al. estimate that major diagnoses remain undetected at least 8.4 % of the time, and nearly half of these misdiagnoses may represent class I error [4]. Class I error is defined as a major diagnostic error that results in or contributes to a preventable patient death.

Diagnostic error has obvious practical consequences for patients, and may result in irrevocable harm. Two recent reviews of the United States National Practitioner Data Bank demonstrate that diagnostic error is responsible for a significant proportion of medical malpractice claims [5, 6]. Compared to claims associated with other allegation categories, claims alleging errors in diagnosis are most common, and these claims account for the highest proportion of payments [6]. While claims originating in the outpatient setting are more frequent, claims originating in the inpatient setting are most likely to be related to more serious harm [6]. As diagnostic error tends to occur upstream in the delivery of healthcare, prior to any surgical or medical intervention, it tends to represent a common root cause of patient harm. Clearly, diagnostic error can be dangerous and expensive. To the degree that it can be prevented, diagnostic error it serves as an appropriate focus for those who seek to improve patient safety.

When diagnostic errors occur, the correct diagnosis is missed or delayed, or a decision is made to accept an incorrect diagnosis. When diagnostic errors are discovered, it is the responsibility of the physician not only to disclose the error to the patient and the patient’s family, but the physician should make every effort to examine what went wrong. By reviewing each case, it may be possible to identify certain root causes that contribute to a diagnostic delay and misdiagnosis, and to prevent similar errors in the future. It is one thing to take responsibility for an error, but it is another thing to learn from an error and to identify regular lapses in the diagnostic process that can be corrected. This is the goal of regular morbidity and mortality conferences and the aim of every quality assurance committee. Each case of diagnostic error may represent an opportunity for improvement, but this improvement is possible only if we take the time to dissect our errors.

In a landmark study, Graber et al. evaluated a series of 100 diagnostic errors collected from five tertiary medical centers [7]. They proposed a regular taxonomy whereby each case of diagnostic error could be described as belonging to one of three categories: no-fault errors, system related errors, or cognitive errors. No fault errors are errors that cannot be corrected, and these include cases of masked or atypical presentation of disease, or cases in which the patient is uncooperative or deliberately deceptive. System related errors are either technical or organizational. An example of a technical error would be a faulty laboratory reading due to faulty instrument calibration. An example of an organizational error would be an error in communication such that consultations are not completed as expected or laboratory results or not provided in a timely fashion. Cognitive errors are divided into three types: faulty knowledge, faulty data gathering, or faulty synthesis. Using this taxonomy, diagnostic errors can be classified according to etiology, and common or recurring problems can be identified.

From this study, it appears that most cases of diagnostic error have a multifactorial etiology, including a combination of systemic and cognitive errors. While system-related error occurs in a majority of cases, cognitive error is most often present. When a system error is identified, it is most often a form of organizational error, such as poor communication between providers or poor communication of test results. Cognitive error was identified in 74 % of cases, often combined with system errors, and sometimes found in isolation. Further analysis of cognitive errors demonstrates that most errors do not result from poor data gathering or knowledge deficits. Rather, most cognitive error is described as an error in synthesis of data, or an error in clinical reasoning. The most common flaw identified is described as “premature closure,” or the decision to accept a diagnosis too early, without giving sufficient attention to other competing diagnoses [7].

Another study evaluating the occurrence of diagnostic error provides slightly different results. Schiff et al. evaluated the causes of diagnostic error for a series of 669 self-reported cases [8]. These authors used a different taxonomy system, classifying errors in a somewhat chronological fashion, according to the time they occur in the diagnostic process. Errors can occur at any point along the diagnostic timeline: presentation, history, physical exam, testing (laboratory or radiological), assessment, referral/consultation, and follow-up. Using this methodology, Schiff et al. conclude that most errors (44 %) occur during the testing phase of the diagnostic work-up, while errors in clinician assessment (32 %) are the second most-common cause of diagnostic error [8]. These results tend to downplay the role of cognitive error, such as failure to synthesize data, or failure to consider the correct diagnosis. But it is clear that much of the testing phase requires appropriate physician assessment skills. Indeed, of all the failures in the testing phase recorded by Schiff et al., many are attributable to some cognitive error on the part of the physician: from failure to order the correct tests to failure to interpret tests correctly. From this, it is clear that testing and assessment errors are difficult to separate. In clinical practice, testing and assessment occur simultaneously, with each process informed by the other. So, it is likely that an error in one process will contribute to an error in the other.

In a study of adverse events resulting from misdiagnosis, Zwaan et al. evaluated the causes of diagnostic error according to the Eindhoven classification: human, organizational, technical, patient related, or other. In nearly all cases (96.3 %), human error was identified as the chief cause of the adverse event [9]. Specifically, lack of knowledge was identified as the predominant cause of diagnostic error, and several knowledge-based lapses were identified. Physicians either lacked appropriate knowledge to form the right diagnosis, or they failed to use their knowledge correctly.

In a separate study, the same authors evaluated the rate of cognitive error as determined by so-called “suboptimal cognitive acts” or SCAs, using a retrospective chart review [10]. They found that not all cognitive errors result in an adverse event, but errors do occur at a surprising rate. SCAs were present in 66 % of all patient records, with an average of 2.3 SCAs per record. Not all error results in patient harm, but each “suboptimal cognitive act” appears to have a cumulative effect. Not surprisingly, adverse events can be predicted by a higher SCA count [10]. Like Schiff et al., Zwaan et al. found that most cognitive lapses tend to occur during the data gathering stage of the diagnostic process. This stage includes history taking and the ordering of laboratory and radiological studies. As data gathering occurs early in the diagnostic process, errors at this stage may delay the correct diagnosis. If they remain undetected, errors in data gathering may be compounded by other errors of clinical reasoning, and misdiagnosis will be the result.