Diagnostic errors

INTRODUCTION — Diagnostic errors are common and important causes of preventable morbidity and mortality in a variety of medical settings. Diagnostic errors leading to harm are most likely to involve diagnoses of cancer, vascular events, or infection. This topic will focus on approaches to measurement, etiologies, and interventions to decrease diagnostic error.

DEFINITIONS

Diagnostic error — Diagnostic error is defined by the National Academies of Sciences, Engineering, and Medicine (NASEM) as the failure to either establish an accurate and timely explanation of the patient’s health problem(s) or communicate that explanation to the patient [1]. One challenge of the NASEM definition is that it can be difficult to ascertain if a diagnosis was “timely,” particularly in the emergency or inpatient setting where patients may evolve rapidly. In response, the diagnostic research community utilizes a variation on this definition. Diagnostic error is “a missed opportunity to make a timely or correct diagnosis, or take the next diagnostic action step, based on available evidence at the time” as a result of provider or system error [2]. Similar to other patient safety events, in which there may be an error that does not result in adverse event (eg, medication errors that do not result in an adverse drug event), both of these definitions recognize diagnostic error (missed, delayed, or wrong diagnosis) even in the absence of harm.

Because diagnosis is a process, it is important to note that not all changes in diagnosis are synonymous with error. Establishing the correct diagnosis may take time due to insufficient diagnostic information, atypical presentations, or clinical evolution (eg, a patient is eventually diagnosed with lupus but their presenting symptoms did not meet diagnostic criteria). A lengthy diagnostic process may be without process or outcome error.

Diagnostic adverse event — Diagnostic adverse event describes resultant patient harm from diagnostic error, typically due to a failure to treat or harm from treatment for a wrong diagnosis [3]. Malpractice studies indicate that diagnostic error is the most common cause of severe harm [4-6].

Diagnostic process error or “near miss” — There may also be diagnostic process failures that do not result in an ultimate diagnostic error (eg, procalcitonin misused but patient still diagnosed appropriately with viral pneumonia). These are analogous to a “near miss” and may still provide information about the safety of health care systems.

DIAGNOSTIC PROCESS — The National Academies of Sciences, Engineering, and Medicine (NASEM) representation of the diagnostic process is the most widely accepted model. The process begins with the patient's experience of a symptom or health condition and their engagement with the health care system, then enters a cyclical set of events in which clinicians gather, integrate, and interpret information to form a working diagnosis. The diagnostic process ends with the explanation of the health problem to the patient and management plan [1]. Whether patients respond as expected to a given treatment may inform the diagnostic process or lead to a return to diagnostic reasoning [7].

PREVALENCE — Diagnostic errors are common and may account for over 500,000 cases of preventable morbidity and mortality each year [8] based on error estimates and disease incidence [9]. Due to the fact that the majority of studies in diagnostic error document those that have led to adverse events (a subset of all diagnostic errors), this likely underestimates the true prevalence of all diagnostic error.

The errors often persist and/or propagate throughout multiple health care engagements, system settings, and interprofessional providers. Acknowledging measurement limitations, some population-based estimates of prevalence exist in the outpatient and inpatient setting [10].

Adult medicine

●Ambulatory practice – The prevalence of diagnostic error in the United States adult ambulatory setting has been estimated at nearly 12 million adult patients a year, or 5 percent of patients over the course of their engagement with the health care system [11]. This is based on extrapolation from three large observational studies of triggered case review [12-14].

●Hospital – Among hospitalized adults, previous adverse event studies estimated that between 5 to 18 percent of patients will experience an adverse event that may have been related to diagnostic error and that as many as 50 percent of adverse events leading to care escalation may contain diagnostic error [15,16]. A systematic review of 53 studies of autopsy-identified misdiagnosis spanning multiple decades from 1966 to 2002 informed a contemporary estimate that at least 8 percent of all hospital deaths may have a diagnostic error, with 4 to 6 percent of deaths attributable to that diagnostic error [17].

In a multicenter retrospective cohort study of 257 adults hospitalized under investigation for coronavirus disease 2019 (COVID-19) infection between February and July 2020, 36 (14 percent) had a diagnostic error, defined as a missed or delayed diagnosis [18]. Fourteen of these patients (35.9 percent of patients with errors and 5.4 percent overall) suffered harm or death due to the error. The most common diagnostic process faults were related to clinical assessment, including history taking; physical examination; test ordering, performance, or interpretation; and patient follow-up and monitoring. Diagnostic errors were not associated with clinical features of or risk factors for COVID-19 infection, or with patient factors such as age, comorbidities, language, or social determinants of health.

●Intensive care unit – In the adult intensive care unit (ICU), systematic review of autopsy studies from 1966 to 2011 revealed a diagnostic error rate of 28 percent, with an estimated 8 percent of these diagnostic errors events leading directly to the patient’s death [19]. This higher rate of error among critically ill patients may be related to higher patient acuity with more limited reserve, multitude of providers, patient instability influence on the diagnostic process, and selection bias of autopsy. A subsequent year-long retrospective cohort study of unplanned nonsurgical ICU admissions to a single site identified diagnostic error among 7 percent of cases, all causing harm and one-third causing significant harm [20].

●Emergency department – Most studies of diagnostic error within the emergency department setting have focused on a singular diagnosis and are heavily informed by malpractice studies. A systematic review of diagnostic error in the emergency department demonstrated an estimated error rate of 5.7 percent of all emergency department visits, a harm rate of 2 percent, serious harm 0.3 percent, and death 0.2 percent. Among the top clinical conditions associated with serious harm from misdiagnosis were stroke, myocardial infarction, aortic aneurysm and dissection, spinal cord compression and injury, venous thromboembolism, meningitis/encephalitis, and sepsis [21].

Obstetrics — Within obstetrics, maternal mortality was estimated to be 17.3 of 100,000 live births in 2017 within the United States [22]. In a retrospective review of 207 pregnancy-related deaths in California, 41 percent had a good to strong chance of preventability. The five most common causes of preventable death (n = 143) were cardiovascular disease, preeclampsia/eclampsia, obstetric hemorrhage, venous thromboembolism, and amniotic fluid embolism. While these are a heterogenous group of diagnoses, delayed recognition of clinical warning signs may have contributed in 60 to 80 percent of cases and misdiagnosis in 30 to 40 percent [23,24].

Pediatrics — A systematic review of observational studies with autopsy-confirmed diagnostic error in the pediatric ICU (PICU) and/or neonatal ICU (NICU) identified major diagnostic error in 19.6 percent of cases, nearly one-fourth of those potentially impacting survival [25]. A further systematic review focused on studies of diagnostic error among the PICU population noted a rate of 8 to 12 percent of diagnostic error [26]. Diagnostic error in the ambulatory and acute care setting has not been as extensively studied among pediatric populations.

Conditions with high rates of diagnostic error — Diagnostic error spans a wide array of diagnoses. For example, in one ambulatory study of diagnostic error, as many as 68 unique diagnoses were identified [27]. However, there is growing evidence that diagnostic error leading to harm is more likely to involve three categories of diseases: cancer, vascular events, and infection.

In a retrospective review of closed ambulatory malpractice claims with diagnostic error, malignancy was most heavily represented (59 percent, namely breast and colorectal cancer) followed by vascular diagnoses (12 percent) [4]. A subsequent review of 11,592 malpractice cases across care settings from the Controlled Risk Insurance Company (CRICO)’s Comparative Benchmarking System database (2006 to 2015) revealed 7379 cases with severe harm, of which 74 percent were either malignancy (38 percent), vascular events (23 percent), or infection (14 percent) [6]. Malignancy was more heavily represented in the ambulatory setting and vascular/infection in the emergency department and inpatient setting. Among pediatric malpractice cases, infection is the most significant category of misdiagnosis [25,28]. It is important to note that malpractice data reflect a very select population and may not be representative of diagnostic errors overall. However, in a survey of 310 physician-reported errors, vascular and malignancy etiologies remain among the most commonly reported missed diagnoses with the addition of drug reaction [29].

Racial, ethnic, and sex-based disparities exist for multiple common diagnoses, including dementia, schizophrenia, and acute myocardial infarction [30-32]. This likely represents a complex interplay between structural factors within health care and implicit bias by clinicians. Further evaluation and understanding of diagnostic error is essential for equity and improvement.

MEASUREMENT OF ERROR — The goal of measuring diagnostic errors is to improve the accuracy and timeliness of diagnosis by identifying system failures that increase the risk of diagnostic error for common presenting conditions. While individual clinician errors may contribute to diagnostic error, measurement should not be used for punitive reasons. As well, the goal of improving diagnosis must be balanced against the need to provide high-value care and conserve health care system resources.

The conceptual framework for measuring the quality of health care was developed by Avis Donabedian, who defined three categories of measurement [33]:

●Structure (how care is organized) – For example what systems were used for tracking critical lab results?

●Process (what care is delivered to the patient?) – For example, was there adherence to evidence-based diagnostic algorithms (eg, for pulmonary embolism)?

●Outcome (how did the patient ultimately fare?) – For example, were diagnostic errors identified at autopsy?

This framework can be used to measure diagnostic safety within health care organizations. However, each category has certain limitations. There is limited evidence that structural elements reliably impact diagnostic accuracy, diagnostic algorithms do not exist for many diseases, and measurement of patient outcomes is labor-intensive and can only occur after the event.

Organizations should seek to evaluate the quality of each part of the diagnostic process:

●Patient/clinician encounter

●Performance and interpretation of diagnostic tests

●Tracking of information over time (including test result follow-up)

●Referrals/consults

Organizations will need to use multiple methods to identify diagnostic errors, as no single data source will provide data on the different phases of the diagnostic process [34,35]. Routinely collected safety data (such as incident reports, Morbidity and Mortality conferences, or malpractice data) may provide detailed examples of errors that are clinically significant, but are necessarily retrospective, and may miss low-harm or “near miss” events [36]. Solicited data (from patients or clinicians) may capture a complementary spectrum of diagnostic errors [37]. Both of these sources can help identify specific issues and targets for improvement, but they cannot provide estimates of the overall frequency or scope of diagnostic errors within an organization (or compared to other organizations). There has been considerable research around developing trigger tools for screening medical records or administrative data for diagnostic errors. For example, in the outpatient setting, electronic triggers have been developed that identify patients with abnormal test results (such as iron deficiency anemia) who have not received the recommended follow-up test (colonoscopy) within a reasonable period of time [12]. This may be indicative of a diagnostic process breakdown. Symptom-disease Pair Analysis of Diagnostic Error (SPADE) has been proposed as a means of screening large administrative data sets to quantify and monitor diagnostic errors over time for acute conditions with high risk of morbidity or mortality, using either prospective or retrospective methods to identify cases in which diagnostic errors occurred despite the presence of warning symptoms [38].

Although candidate measures of diagnostic safety have been proposed, at this time, few organizations are routinely measuring diagnostic safety, in part because diagnostic safety metrics are not among those required by payors or accrediting agencies.

CAUSES OF DIAGNOSTIC ERRORS

Elements of the diagnostic process — Diagnostic adverse events, as with other patient safety events, often involve multiple interdependent underlying causes that include patient, provider, and systems factors. It is rare for a single individual error to directly result in a diagnostic adverse event. Rather, diagnostic errors cause patient harm as a result of health care system flaws that allow errors made by individuals to reach the patient. Because of this complexity, it may be beneficial to evaluate clustering of errors or patterns of error as opposed to a single provider or system’s error when reviewing diagnosis.

The diagnostic process can be used as a framework for considering underlying causes of error [1,29,34].

The process includes the following domains:

●Patient-clinician encounter (access, history, and physical)

●Performance and interpretation of diagnostic tests

●Team-based care (including consultation and referrals)

●Monitoring of patients over time

Examples of error in each of these domains are included in the table (table 1).

Provider decision-making — Provider assessment and clinical decision-making can be influenced by cognitive bias, experience, implicit bias, emotional state, and cognitive load.

Cognitive biases, or systematic cognitive shortcuts, have received the most attention [39,40]. These commonly include framing effect, anchoring/premature closure, availability bias, base-rate neglect, and confirmation bias. Examples are provided in the table (table 2).

There is growing recognition that implicit bias on the part of clinicians plays a role in diagnostic error, influenced by the persistence of racism, sexism, prejudice towards LGBTQ+ people, ageism, and many other discriminatory aspects of society [30,41].

INTERVENTIONS — Interventions to improve diagnosis should incorporate both systems and individual clinician approaches. Measurement challenges have limited the availability of high-quality, unbiased study of diagnostic interventions in both of these domains, often limiting analysis to surrogate process outcomes rather than patient-centered outcomes such as morbidity and mortality. Much of the literature that does exist focuses on the perceptive diagnostic processes (electrocardiography, radiology, pathology) as opposed to other clinical diagnostic processes, due in part to easier gold-standard measurement.

For the purposes of this section, we will be focusing on the general clinical diagnostic process, which, of course, integrates these perceptive diagnosis fields [42-47]. We will also be focusing on general health care delivery interventions as opposed to disease-specific interventions. Within the diagnosis community, there is debate about taking a disease-oriented versus disease-agnostic approach to improvement; the ideal state is likely a balance of both with measurement of specific diagnosis outcomes as well as broader systems change impacting universal diagnostic workflows [48].

System action — System actions should be informed by underlying root causes of error within the diagnostic process (NASEM).

●Patient-provider encounter – Efforts may include increasing access to timely care, enhancing electronic health record (EHR) transparency, ensuring flow of information between patients and health professionals, optimizing communication strategies for the patients level of health literacy, and using standardized diagnostic instruments [49,50].

•In a trial of patients presenting to the emergency department with nonspecific symptoms potentially associated with occult psychiatric disease, the use of a validated, self-administered, psychiatric interview (Primary Care Evaluation of Mental Disorders [PRIME-MD]) improved the diagnosis of a mental health condition compared with physician evaluation alone (42 versus 5 percent) [50].

●Performance and interpretation of diagnostic tests – Efforts may include building pathways to allow for access to appropriate advanced diagnostics, testing new diagnostic capabilities, ensuring workflow for reliable result communication, utilizing artificial intelligence to enhance diagnosis, developing clinical decision support or algorithms, or improving timely access to specialist interpretation of tests [51-53].

•In a multicenter trial of 1502 patients presenting to the emergency department with traumatic musculoskeletal injuries, immediate reporting of trained radiologist interpretation of imaging was compared with delayed reporting. Immediate reporting reduced emergency department provider interpretive errors and patient recall [53].

●Team-based care (consultation, referrals, communication) – Efforts may include enhancing intra- and interprofessional collaboration in the diagnostic process, bolstering formal second-opinion programs, tracking referral completion, establishing highly structured handoffs, and remote nurse monitoring [54-56]. (See "Patient handoffs".)

•A study of real-time nurse remote monitoring of EHRs of patients at high risk for clinical deterioration, with results of monitoring communicated to rapid-response teams at hospitals, demonstrated lower mortality within 30 days in the intervention cohort than the comparison cohort [56]. The intervention was also associated with shorter length of stay and lower incidence of intensive care unit (ICU) admission.

●Follow-up and monitoring – Efforts may include enhancing health IT interoperability, improving data display for clinicians, virtually monitoring inpatient trajectory, and ensuring EHR methodology to track abnormal results [57,58].

•In a trial among primary care providers of electronic triggers identifying patients with possible delayed evaluation for lung, colorectal, or prostate cancer, triggers improved the proportion of all patients who received follow-up (73.4 versus 52.2 percent) and reduced the time to diagnostic evaluation in those with colorectal (median 104 versus 200 days) and prostate cancer triggers (144 versus 192 days) compared with usual care [57].

There are a variety of patient factors that may also influence the diagnostic process including medical complexity, psychiatric comorbidity, limited language proficiency, and other social determinants of health. While patient factors may not be modifiable, systems change will need to recognize the interplay between patient, provider, and system factors. Consideration should be given to the ways interventions may be tailored to vulnerable or high-risk populations. For example, older adult patients are at greater risk of diagnostic error due to differences in disease presentation, communication challenges, medical complexity, and ageism [59]. Interventions targeting improving diagnosis for older adult patients will need to recognize the potential harms that could occur through attempts to obtain a definitive diagnosis (such as the well-documented harms of hospitalization for older adult patients.)

What clinicians can do to reduce diagnostic errors — Existing evidence on improving clinician reasoning is generally based on simulation studies primarily in learners, and evidence of impact on diagnostic errors or adverse events is limited. However, the existing evidence base combined with adult learning theory and expert recommendations does identify a number of promising strategies that individual clinicians can use to minimize the risk of diagnostic error in their practice [60]. A 2022 article outlined five strategies for clinicians to use to improve their diagnostic skills, several of these designed to be applied real-time at the bedside [41].

Prior to employing these strategies, clinicians should identify clinical scenarios that may be high risk for diagnostic error in their practice environment, based on clinical course or predisposing factors. Signs of diagnostic error may include atypical clinical trajectory (eg, presumed community-acquired pneumonia that is not improving with appropriate antibiotic therapy, which may indicate a noninfectious diagnosis such as drug-induced pulmonary toxicity) or frequent healthcare interactions (eg, repeat presentation to the emergency department for neck pain, which may indicate epidural abscess). In terms of predisposing factors, one may consider the patient, the provider, and the process. Patient factors that may heighten provider awareness of bias or error include the patient belonging to a historically marginalized or stigmatized population (gender, race, housing insecurity, substance use) or having multiple comorbidities or communication barriers (encephalopathy, mental health, mistrust). Provider factors that may increase risk of error include inexperience with given complaint/diagnosis or high degree of uncertainty. Systems factors may include high census or frequent handoffs.

Once a clinician recognizes the risk of diagnostic error, one or more of the following strategies may help avert error [41]:

●Foster critical thinking – Clinicians should actively use strategies like diagnostic pauses to reflect on their own diagnostic process and consciously identify alternate diagnoses. A diagnostic time out will generally include the use of de-biasing questions (“What doesn’t fit? Why couldn’t this be something else? What is the must not miss diagnosis? What more information do I need?”), aids to support differential diagnosis (checklists or web-based differential diagnosis generators) and diagnostic algorithms/protocols (eg, low-risk chest pain pathway) [61]. Critical thinking will also require reducing cognitive load, as the diagnostic process is highly influenced by patient volume, patient acuity, and interruptions.

●Make diagnosis a team sport – Clinicians should seek input from colleagues, including the interdisciplinary team involved in the patient’s care, to maximize the likelihood of obtaining the correct diagnosis. This may mean pursuing a colleague’s second opinion (shown to have benefit in the radiology, pathology fields), further specialty consultation or engagement of structured response teams (eg, trauma teams, rapid response teams) [55,56]. It is important to remember the patient and family role on this team as well.

●Consider biases – Since cognitive biases and implicit bias are powerful contributors to diagnostic errors, clinicians should recognize these biases and the situations in which they occur, as well as seek data on disparities in diagnosis. While these steps are important, debiasing interventions alone have been difficult to implement and have had limited effect. A study utilizing a knowledge-oriented intervention found a short-term mitigation of availability bias [62]. This may mean that strategies to impact knowledge, as above, may have more efficacy in bias immunization than those to enhance provider awareness of bias or metacognition.

The following strategies are likely to help clinicians improve their overall diagnostic accuracy outside of individual clinical scenarios:

●Seek feedback on diagnostic decisions – Most clinicians do not have a systematic method of following up on their diagnostic decisions. This is especially true for clinicians who do not have a longitudinal relationship with a patient. Maintaining patient lists and seeking feedback from colleagues can help with self-directed feedback; ultimately, EHRs could be configured to directly feedback information to clinicians on changes in diagnoses or clinical status. Diagnostic calibration, the correlation between perceived and actual diagnostic accuracy, can be important to a diagnostician’s bias recognition and metacognition development. While done retrospectively as opposed to in real-time, diagnostic feedback is likely to inform prospective performance [63].

●From a health care system standpoint, the model of the “Diagnostic Learning Cycle” describes the development of consistent and standardized programs that allow diagnostic feedback [64]. These programs would include provider diagnosis, reasoning, and confidence as well as patient outcome in both the short and long term. While these individual patient stories are crucial to provider learning, they can also add value to the diagnostic learning health system [65,66].

●Improving content-specific knowledge – This strategy refers to efforts to improve disease-specific knowledge through training programs and simulation. These tend to provide the most consistent (albeit small) benefit among cognitive interventions in improving diagnostic accuracy.

COMMUNICATING ERROR TO PATIENTS AND OTHER PROVIDERS — Disclosure of diagnostic error has steps that are parallel to other disclosure practices: disclosure of medical error, explanation of why error occurred, plan to minimize the error’s impact on the patient, and system efforts to prevent recurrence [67]. (See "Disclosure of errors in surgical procedures".)

One complicated aspect of disclosure that may be more germane to diagnostic error is the breadth of the teams and settings that may be involved in the diagnostic process. Because of handoffs, consultant models and transfers, a provider may find themselves in a position in which they are disclosing what feels like another provider or system’s error. The first step in this circumstance is to communicate with the involved provider as well as each individual/site’s risk management to perform analysis and provide joint disclosure if a preventable event is confirmed [68]. Of note, this should take place when there is a high degree of certainty that error led to the delay or wrong diagnosis.

SUMMARY AND RECOMMENDATIONS

●Definition – A diagnostic error is “a missed opportunity to make a timely or correct diagnosis, or take the next diagnostic action step, based on available evidence at the time” as a result of provider or system error. (See 'Definitions' above.)

●Prevalence – Diagnostic errors are common and likely account for thousands of cases of preventable morbidity and mortality each year. The errors often persist and/or propagate throughout multiple health care engagements, system settings, and interprofessional providers. (See 'Prevalence' above.)

•Certain conditions tend to have higher rates of error. (See 'Conditions with high rates of diagnostic error' above.)

●Causes – Diagnostic errors often involve multiple interdependent underlying causes that include patient, provider, and systems factors, and domains of the diagnostic process (table 1). (See 'Causes of diagnostic errors' above.)

•Provider assessment and clinical decision-making can be influenced by cognitive bias, experience, implicit bias, emotional state, and cognitive load (table 2). (See 'Provider decision-making' above.)

●Interventions – The most effective improvement work will incorporate a mix of both individual clinician and system-level interventions. (See 'Interventions' above.)