Diagnostic Error in Pediatrics: A Narrative Review

The diagnostic process is complex, context dependent, iterative, and nonlinear, as highlighted in the NASEM overview (Fig 1). A diagnosis can evolve over minutes or weeks and involves patients, families, individual clinicians, and teams of health care providers. Thus, breakdowns in the diagnostic process are rarely clear and discrete events, which is in contrast to a wrong-site surgery, for example, where despite a variety of underlying causes, the error itself is easily defined. Diagnostic errors comprise a diverse group of medical errors that can, at times, be difficult to discern from natural progression of disease. However, several helpful definitions and models for operationalizing diagnostic error in clinical practice exist. The authors of the NASEM report define 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.”²  Although this definition includes several important concepts, most notably the focus on communicating diagnoses to patients and families, it is challenging to operationally define the concepts of timeliness and accuracy in research settings. Another definition of diagnostic errors proposed by Singh et al⁷  is the idea that diagnostic errors are “missed opportunities to make a correct or timely diagnosis based on the available evidence, regardless of patient harm.” This framing is more helpful in research settings, especially retrospective event review, because of its focus on diagnostic errors that are potentially preventable. In addition to these broad definitions is a variety of terminology used to categorize or further describe diagnostic errors (Table 1). In an effort to create a unified conceptual model for diagnostic errors, Newman-Toker⁸  noted that some of these terms can be helpful in further delineating subgroups or categories of diagnostic errors; however, an overlap often exists among these terms, which can make them difficult to consistently apply.⁹  Additionally, the authors of the NASEM report considered overdiagnosis, another growing concern in pediatrics, to be distinct from diagnostic errors in general.^2,10  Overdiagnosis is not a misdiagnosis, per se, but rather a correct diagnosis of a condition or disease that will not meaningfully affect a patient in their lifetime.¹¹  A frequently described potential overdiagnosis in pediatrics is the accurate identification of transient hypoxemia in an improving infant with bronchiolitis.¹²  Per the NASEM report, “a major reason overdiagnosis is not characterized as an error is because it is found primarily in population-based estimates; it is virtually impossible to assess whether overdiagnosis has occurred for an individual patient.”²  That said, it is clear that diagnoses (be they in individuals or populations) have consequences.

The precise incidence of diagnostic error in medicine is not known, but previous studies of predominantly adult populations estimated the diagnostic error rate to be 5% to 15% annually.^2,13,14  These estimates vary on the basis of both the clinical context of the study and the methodology used to identify diagnostic errors. However, few researchers of large studies have included or focused on diagnostic errors in pediatric patients. In a multisite survey study of both university-based and community pediatricians, more than half of the respondents reported making a diagnostic error 1 to 2 times per month, and approximately half reported making diagnostic errors that resulted in patient harm 1 to 2 times per year.¹⁵  Survey respondents identified viral illnesses being diagnosed as bacterial infections as the most common diagnostic error and medication side effects, psychiatric disorders, and appendicitis as other frequent diagnostic errors. In another survey study of general pediatricians from the American Academy of Pediatrics Quality Improvement Innovation Network, 35% of respondents reported making a diagnostic error at least monthly, and 33% reported making a diagnostic error resulting in an adverse event at least annually.¹⁶  Survey respondents identified missed diagnosis of hypertension, delayed diagnosis due to missed subspecialty referral, and delayed follow-up of abnormal laboratory results as targets for improvement. Although these survey studies revealed that diagnostic errors occur frequently in pediatrics, they did not provide data on the incidence and epidemiology of diagnostic errors. Many researchers have discussed condition-specific diagnostic errors (eg, delayed diagnosis of brain malignancy), the studies of which are summarized in Table 2. Although these studies contain valuable learnings, they are often single center and lack patient denominator data needed to determine the true incidence of diagnostic errors.

In addition to the limited data around the incidence of diagnostic errors in pediatric patients, there are even fewer data around which patients are most at risk for errors and which errors are most likely to result in significant patient harm. Furthermore, caution must be exercised when extrapolating estimates obtained from studies in adults to pediatric populations. The medication safety literature provides a useful precedent in that researchers have shown that the incidence of medical errors and adverse events are different for children and adults.¹⁸  For example, medication errors due to multiple formulation options (eg, various concentrations of liquid formulations) and weight-based dosing pose unique threats in pediatrics compared with adult populations.¹⁹  As for diagnostic errors, there may be some overlap in contributing factors, but diagnoses and high-risk populations likely differ significantly because of changes in disease prevalence and presentation by age. For example, researchers recently assessed a large database of adult medical malpractice and found that diagnostic errors of vascular events, infections, and cancer are most frequently associated with high-severity harm, including permanent disability and death.²⁰  However, given the much lower incidence of vascular events (eg, stroke, myocardial infarction) and malignancy (eg, lung, breast, colorectal, prostate cancer) in children, these data likely do not extrapolate well to pediatric populations.

Although the diversity, complexity, and heterogeneity of diagnostic errors make the systematic measurement of diagnostic errors challenging, various strategies have been used, predominantly in adult populations, to identify diagnostic errors. We review the available literature on the incidence of diagnostic errors and assess the strengths and weaknesses of the various methods most commonly used in diagnostic error research. We also highlight successful applications of these strategies in pediatric populations.

Medical record review is a well-established method for identifying a broad range of medical errors and adverse events. A multistep retrospective chart review of pediatric patients admitted with an acute-onset illness to a community hospital over a 90-day period revealed a diagnostic error rate of 5%.²¹  In a study of hospitals across Canada that used a multistep chart review process adapted from the Harvard Medical Practice Study protocol, the researchers found that 34% of adverse events were due to diagnostic errors.²²  Although both these studies help with estimating the incidence of diagnostic errors and associated harm in hospital settings, the researchers did not identify events in the outpatient setting, and the results are susceptible to hindsight bias. Targeted or electronically triggered medical record review is another promising method for identifying diagnostic errors.⁴⁰  However, widely available tools, such as the Institute for Health Care Improvement’s Global Trigger Tool and the Global Assessment of Pediatric Patient Safety, currently have limitations in assessing for diagnostic errors.^41,42  Electronic triggers have been successfully used to identify diagnostic error in the adult ambulatory setting, including an unplanned hospitalization or follow-up visit within 14 days of an index primary care visit.¹⁴  Electronic triggers for diagnostic error are also being explored in pediatrics. In a single-center study of PICU patients, researchers were successful in identifying errors in 26 of 214 high-risk patient records, which included patients who received an autopsy, were seen in the outpatient setting within 2 weeks before ICU admission, were transferred emergently to the ICU, and were transferred to the ICU after a rapid response team evaluation but did not experience any clinical decompensation.⁴³  Finally, Rinke et al⁴⁴  had primary care practices perform retrospective chart reviews to identify high-frequency, but subacute errors in diagnosing adolescent depression, missed elevated blood pressure, and delayed abnormal laboratory tests. In adolescent patients presenting for health supervision visits, 62% did not receive appropriate screening or evaluation for depression. For patients with elevated blood pressure documented during a clinical encounter, 54% did not receive appropriate response either through rechecking the blood pressure during the visit or a documented follow-up plan. Among patients with abnormal laboratory results, 9% had no documentation reflecting recognition of the abnormal laboratory value or an appropriate action documented without delay.

Medical malpractice claims alleging negligent misdiagnosis are the most prevalent type of claim in the United States,⁴⁵  including pediatrics. An assessment of pediatric-specific data from 1985 to 2005 revealed that errors in diagnosis comprised 31.9% of medical misadventures or alleged departures from accepted medical practice.⁴⁶  Pediatric medical malpractice claims have most frequently involved care within the hospital (49.0%), followed by the outpatient setting (43.4%). With respect to diagnostic error, the most frequently cited conditions in pediatric malpractice claims data, in descending order, are meningitis, appendicitis, pneumonia, nonteratogenic anomalies (eg, developmental dysplasia of the hips, spina bifida occulta), and brain damage in infants.⁴⁷  There are several important limitations of medical malpractice claims data, including that only a very small proportion of medical errors result in a malpractice claim, and some claims may be made in the absence of an error. Claims are skewed toward events that cause significant morbidity and mortality. Additionally, medical malpractice claims data are likely not generalizable to populations, as lower socioeconomic status and uninsured patients are significantly less likely to file a malpractice claim.⁴⁸ 

Researchers have shown that 25% of autopsies reveal major diagnostic errors, with 10% of those errors likely contributing to the patient’s death or would have resulted in a change in management (ie, Goldman Classification System class I errors).²  A challenge to learning from autopsies is that rates of diagnostic errors discovered on autopsy have declined over time, as noted in a systematic review of both adult and pediatric studies.⁴⁹  Another systematic review, revealed class I errors in 6.4% and 3.7% of autopsies of PICU and NICU patients, respectively.⁵⁰  For both PICU and NICU patients, the most common misdiagnosed conditions included infections and vascular events (eg, hemorrhage, thrombosis, ischemic bowel).

A broadly used strategy is a voluntary incident reporting system for identifying all types of medical errors, adverse events, and near-miss events. Such reporting systems have been shown to be an effective method for identifying and learning from diagnostic errors in adult emergency department and inpatient populations.^51,52  In a multisite study, researchers asked internal medicine physicians to describe up to 3 diagnostic errors along with the perceived causes, seriousness, and frequency and identified 583 events across >20 clinical sites.⁵³  Although different event analysis procedures were used in all 3 of these studies, all revealed the potential to identify frequent diagnoses associated with diagnostic errors as well as potential targets for systems improvement. However, voluntary incident reporting also has several important limitations. Physicians, in particular, have been shown to be low utilizers of incident reporting systems, and it has been well established that incident reporting only captures a small subset of safety threats and medical errors.⁵⁴  Several factors are known to increase reporting in general (eg, clearly defining events of interest, nonpunitive and systems-oriented review processes, provider feedback) that can be used as guiding principles for developing diagnostic error reporting systems.⁵⁵  In a recently published quality improvement project by Marshall et al,⁵⁶  pediatric hospital medicine attending physician reporting of suspected diagnostic errors, strategically called diagnostic learning opportunities, was successfully increased. Marshall et al used several of the aforementioned guiding principles, as well as a focus on psychological safety, to increase and sustain the filing of reports from 0 to 1.6 per 100 patient admissions.

In addition to methods for identifying potential errors, given the nuance and complexity of diagnostic errors, a need exists for reliable tools to confirm the presence of a diagnostic error as well as to understand where and why the diagnostic process is failing so that more-effective strategies can be developed to reduce diagnostic errors. Several helpful tools can be adapted from the adult literature.

The most well-established diagnostic error screening tool is the Safer Dx Instrument, which was initially designed and tested in the adult ambulatory care setting.⁵⁷  The Safer Dx Instrument is based on the conceptualization of diagnostic errors as missed opportunities and is a Likert scale survey tool. The original tool was recently revised to make it easier to use and apply in a diverse range of clinical settings.⁵⁸  The diagnostic error and evaluation research taxonomy is a process map that was developed from a review of the literature and that breaks the diagnostic process down into 7 overarching stages, including health care assessment, history taking, physical examination, diagnostic testing, diagnostic data assessment, referrals and consultations, and patient follow-up.⁵⁹  Another promising strategy for holistically evaluating diagnostic errors is the use of a modified fishbone diagram, which aids in the detection of both systems and cognitive factors that may contribute to these events.⁶⁰  Finally, the identification of undesirable diagnostic events, defined as “specific, measurable, and actionable clinical situations likely to denote the presence of a diagnostic error,” may allow rapid identification and review of specific diagnostic errors.⁶¹  Undesirable diagnostic events were conceptualized as a way to standardize identification and reporting of diagnostic errors for high-risk conditions (eg, bacterial meningitis) and to measure the impact of interventions aimed at improving diagnosis of these conditions.

Given the complexity of the diagnostic process and the multifactorial nature of diagnostic errors, it is important to develop and test strategies that target both the individual or cognitive factors and the systems and environmental factors that contribute to these errors. There are several tested and proposed strategies from the literature but a paucity of pediatric-specific interventions, as highlighted by a recent systematic review on strategies to reduce diagnostic errors, which contained only 2 studies that included children.⁶² 

Diagnosis is a challenging and context-dependent task that occurs in high-stress, often chaotic, conditions, and the fallibility of human cognition is exacerbated by the system in which health care occurs. Individual- or clinician-level factors that lead to diagnostic error often stem from cognitive errors or breakdowns in the decision-making process. Although there has been much enthusiasm for focusing on dual-process theory and the potential efficacy of avoiding diagnostic errors by avoiding heuristics and cognitive bias, these approaches have led to very limited, if any, success.^63–66  Cognitive bias is a helpful framework to normalize diagnostic errors, but caution should be used in solely emphasizing on them. Another framework to help identify the causes of diagnostic error is that of situated cognition, which describes the interaction between people and their environment.⁶⁷ 

Several strategies have been used to augment or support clinical reasoning both in real time and as part of reflective practice. Diagnostic checklists, trialed in adult primary care and emergency department settings, increased the consideration of alternate diagnoses but did not reduce the rate of diagnostic errors.^68,69  Cognitive autopsies, which prompt reflection on diagnostic errors, are well received by learners and have been proposed as a way to augment learning in morbidity and mortality conferences.^70,71  Several application- or Web-based tools have also been developed to support clinical reasoning. With the use of differential diagnosis generators, such as Isabel, researchers seek to translate symptoms and patient demographics into a differential diagnosis, often with particular attention paid to “red flag” or “can’t miss” diagnoses, and have shown improvement in accuracy over time.⁷²  Bayesian reasoning uses base rate probabilities and clinical information to assess the probability of a diagnosis, with phone-based applications offering a promising means of teaching and incorporating this approach in clinical settings.⁷³  However, additional research is needed to better integrate these tools into the care system and to reveal their real-world effectiveness.

One fundamental factor leading to, and reinforcing, diagnostic errors is that clinicians receive little feedback about the outcomes of their diagnostic decision-making (eg, an open-loop system). Thus, clinicians are led to believe that they are accurate and make few errors, despite much evidence to the contrary, because they are rarely informed of their own errors. There are published approaches to start to close these loops and improve diagnostic calibration, although much work remains to be done.^74–76  For example, a study evaluating the impact of structured peer feedback to residents on overnight admissions, including diagnostic changes noted in 43.7% of admissions, revealed this approach to be promising for improving diagnostic calibration.⁷⁷  Diagnostic feedback is particularly challenging when care is fragmented, such as when patients and families seek follow-up care elsewhere. In addition to structured and regular feedback on diagnostic performance, education interventions, such as structured assessment of clinical reasoning, dedicated diagnostic error curricula for learners, and virtual patient cases, are promising strategies that warrant additional research.^78,79 

Identifying systems-based factors and adopting systems solutions are also promising strategies for reducing diagnostic errors. Within pediatrics, a collaborative quality improvement study group, Project RedDE (Reducing Diagnostic Errors in Pediatric Primary Care), sought to reduce diagnostic errors, or missed opportunities for diagnosis, by systematically identifying and targeting opportunities for improvement in the ambulatory pediatric setting by using quality improvement collaborative methodologies to accelerate change and learning.^80–82  By using quality improvement collaborative methodology, this group reduced blood pressure and adolescent depression errors in the primary analysis and abnormal laboratory errors in analyses during longer-term follow-up.

Among the top recommendations in the NASEM report was to promote more effective teamwork in the diagnostic process.²  Graber et al⁸³  elaborated on this topic by proposing a “new diagnostic team” that focuses on the ambulatory care setting; places the patient at the center of the diagnostic team; highlights the roles of the primary care physician, subspecialists, nurses, and allied health professionals in the diagnostic process; and highlights the supportive role of the electronic health record. An ethnographic study of inpatient, internal medicine teaching teams, including medical students, residents, and attending physicians, revealed among its key themes that diagnosis is a social phenomenon, also highlighting the need to promote effective teamwork.⁸⁴  Although improving diagnostic teamwork is a promising area for future research, we were unable to find additional literature on studies that tested specific team-based interventions to prevent diagnostic errors.

Finally, a critical need exists to develop and implement effective health informatics–based support systems that better facilitate the diagnostic process and provide timely and effective clinical decision support.⁸⁵  A systematic review of communication and audit-based strategies for reducing diagnostic error, including both the previously mentioned application- and Web- based support systems and clinical decision support and alert systems embedded in the electronic health record, revealed that few systems currently are used to improve diagnostic error rates in real clinical settings.⁶² 

Diagnostic error in pediatrics remains an area in need of both rigorous research and quality improvement efforts to apply learnings from our growing evidence base. We have highlighted several key research questions (Table 3) critical to advancing the field of pediatric diagnostic error research.

Given what is already known about the harm associated with diagnostic errors from the general literature, it is imperative that we as researchers focus on improving our understanding of the epidemiology of diagnostic errors in pediatrics. To meaningfully reduce diagnostic errors in pediatrics, we need to identify the high-risk clinical scenarios, patient populations, and groups of diagnoses most prone to diagnostic error and associated harm. Additionally, given the increasing complexity of health care systems, including the size of the health care team and fragmentation of care, effective team-based strategies recommended by the NASEM report are a particularly high priority area for research into preventing diagnostic errors.

This young field of pediatric diagnostic errors is growing, and research findings will help to inform additional research and quality improvement efforts. Perry et al⁸⁶  developed a pediatric diagnostic error index by using a combination of existing data sources, including data from a traditional safety event reporting system, institutional root cause analyses, autopsy results, and morbidity and mortality conferences, as well as an electronic trigger tool for abdominal pain to track diagnostic errors on a monthly basis. Although not comprehensive, the diagnostic error index is a helpful model for how institutions can use largely existing data to further local understanding of the incidence of diagnostic error. The aim of a recent quality improvement study was to improve shared situational awareness of hospitalized children with uncertain diagnoses because lack of communication around diagnostic uncertainty had previously been identified as a contributing factor to safety events.^87,88  The findings led researchers to develop a novel uncertain diagnosis label in the electronic health record that clearly identifies patients with uncertain diagnoses and facilitates open discussion and contingency planning when uncertainty is present. Next steps will be to examine the association of this work and diagnostic outcomes, but as this research occurs, institutions and researchers can begin to improve diagnosis in pediatrics by exploring innovative ways to improve aspects of the diagnostic process.

Diagnostic errors are highly prevalent in medicine and result in significant patient harm, yet critical gaps in knowledge remain about the incidence and epidemiology of diagnostic errors and their associated harm in pediatric patients. To improve diagnosis in pediatrics, it is imperative that pediatric researchers work collaboratively to more reliably identify and evaluate pediatric diagnostic errors.

NASEM

National Academies of Science, Engineering, and Medicine