Diagnostic error: the Improving Diagnosis report

You close the case. The blood pressure is documented, the electrocardiogram has been ordered, discharge seems safe. Hours later, another service finds what was missing: a different diagnosis, a clue that was in the chart and never entered the chain. The report Improving Diagnosis in Health Care, from the National Academies of Sciences, Engineering, and Medicine (IOM/NAM), starts from this type of failure and treats diagnostic error as a system problem, not just that of an isolated physician (Ball and Balogh, 2015; NAP, 2015).

NEXO is a daily medical diagnosis game: one clinical case a day, six progressive clues, and ICD-10 proximity feedback across five levels.

This text summarizes what the report put on the table, how it connects to the literature on incidence and clinical reasoning, what changes in study and preceptorship practice, and how to train without turning the topic into individual blame. It complements blog posts on hindsight bias, diagnostic pause, morbidity and mortality sessions and error disclosure: the angle here is the 2015 document and what it asked of the field.

Key points

• The report defines diagnosis as a collaborative process and defines diagnostic error as a failure to establish an accurate and timely explanation of the patient's health problem, or to communicate that explanation to the patient (NAP, 2015; Ball and Balogh, 2015).
• It places the topic at the same priority level as other patient safety failures and calls for teamwork, better use of technology, a just culture and continuous research (Ball and Balogh, 2015).
• Frequency estimates in outpatient settings and incidence reviews show that the problem is common enough to warrant system design, not just individual heroism (Singh et al., 2014; Graber, 2013).
• Training clinical reasoning, calibrating confidence and practicing error recovery fits the report's educational recommendations and the literature on error and reasoning (Norman and Eva, 2010; Eva, 2005).

What the report is

In 2015, the National Academies published Improving Diagnosis in Health Care, a report from the Committee on Diagnostic Error in Health Care. The full text is on the NAP portal; a summary for clinicians appeared in Annals of Internal Medicine (NAP, 2015; Ball and Balogh, 2015).

The document is not a checklist of diseases. It is a system map: it defines concepts, summarizes the evidence available at the time, points out measurement gaps and formulates recommendations for education, practice, patients, technology, payment and research. For the student and the resident, the useful reading is not memorizing the chapters: it is understanding the framing. Diagnostic error stops being "that physician made a mistake" and becomes "the process of reaching and communicating the explanation failed in time".

Why it came out in 2015

Decades of patient safety had advanced in infection, medication and surgery. Diagnosis was still handled in a fragmented way. Graber and others were already describing taxonomies and case series; Singh and colleagues were estimating frequency in outpatient settings; Norman and Eva linked error to failures of reasoning and knowledge (Graber et al., 2005; Singh et al., 2014; Norman and Eva, 2010). The NAP report gathered this field and gave managers, educators and clinicians a common language.

What it is not

It does not replace judgment at the bedside. It does not classify specialties as good or bad. It does not deliver a single universal error rate for all contexts. And it does not ask the student to "never err": it asks for systems and habits that detect, contain and learn from error. Anyone who has read about the second victim recognizes the care for the professional after a failure; the report also emphasizes the patient and the team.

How the report defines diagnostic error

The definition adopted in the report (and highlighted in the Annals summary) is practical: a diagnostic error occurs when there is a failure to obtain an accurate and timely explanation of the patient's health problem, or when that explanation is not communicated to the patient (Ball and Balogh, 2015; NAP, 2015).

Three ideas matter for training:

1. Process, not just the final label. Getting the name of the disease wrong is one form. Taking too long, omitting a critical hypothesis or not telling the patient what is known also count.
2. Collaboration. Diagnosis is described as teamwork: clinician, nursing, laboratory, imaging, patient and family. This changes the design of rounds and of M&M (morbidity and mortality sessions).
3. Communication as part of the diagnostic act. A correct label that does not reach the patient in time still fails the report's criterion.

Diagnosis as a journey

The report describes diagnosis as a continuous process: data collection, integration, hypothesis formulation, testing, review. This echoes the clinical cycle and the idea that closing too early is a risk, a theme linked to the diagnostic pause and to label overshadowing.

Relation to bias and to knowledge

Norman and Eva argue that diagnostic error involves both knowledge and reasoning failures, and that dual process alone does not explain everything (Norman and Eva, 2010). The report does not choose a single cognitive model; it calls for education that trains reasoning, feedback and teamwork. For the student, the practical message is: studying content and training decisions under case conditions go together. The error notebook and confidence calibration are tools compatible with this framing.

What incidence and global burden show

The report works from the premise that diagnostic error is frequent and harmful enough to require action. The incidence literature does not deliver a single magic number for every setting, but it offers useful orders of magnitude.

Outpatient setting

Singh, Meyer and Thomas estimated the frequency of diagnostic errors in outpatient care from three large observational studies in the United States. The synthesis indicates that the phenomenon is common at the population level and that most adults may encounter at least one diagnostic error over a lifetime, with a fraction associated with serious harm (Singh et al., 2014). For the international reader, the point is not to copy the American figure as a local target: it is to accept that the outpatient setting is not a "safe zone" just because there is no ICU.

Series and taxonomies

Graber, Franklin and Gordon analyzed diagnostic errors in internal medicine and described combinations of system and cognition failures (Graber et al., 2005). In a later review, Graber synthesized the incidence of diagnostic error in medicine and reinforced the need for better measurement (Graber, 2013).

Primary care and global burden

Singh and colleagues reviewed the global burden of diagnostic errors in primary care and highlighted the difficulty of measurement and the impact in systems with limited resources (Singh et al., 2017). This matters for those studying outside tertiary centers: the NAP report is American in origin, but the problem is international.

Source	Focus	What helps the student remember
NAP / Ball and Balogh, 2015	Definition and system recommendations	Error = failure of timely explanation or of communication
Singh et al., 2014	Frequency in outpatient care (USA)	Outpatient care also errs; serious harm exists
Graber et al., 2005	Analysis in internal medicine	System + cognition usually combine
Graber, 2013	Incidence (synthesis)	Measuring is hard; ignoring is not an option
Singh et al., 2017	Burden in primary care	A global problem, not only of the university hospital
Norman and Eva, 2010	Error and reasoning	Knowledge and process; not just "bias"

When citing numbers from any study, use the link in the same sentence and state the context (country, setting). Avoid turning an estimate into a slogan.

Eight axes of recommendation

The Annals summary organizes the call to action into fronts that the report develops in chapters. For the blog, we translate the pedagogical spirit of each axis, without pretending to be the official text (Ball and Balogh, 2015; NAP, 2015).

1. Support teamwork in diagnosis

Rounds where only the most senior person speaks, a lab that cannot escalate a critical value, nursing that sees deterioration and has no channel: all of this is team design. Training clear presentation of hypotheses and of uncertainty (see how to communicate uncertainty) is part of the axis.

2. Involve patients and families

The report treats the patient as a partner in the diagnostic process. This connects with patient as partner and with narrative medicine: the story told by the patient is not "noise", it is data.

3. Education aimed at safe diagnosis

Curricula that only demand a list of diseases without training decision-making under uncertainty fail the spirit of the report. Case-based learning, rich feedback and deliberate practice of hypotheses fit here (case-based learning; feedback in cases).

4. Technology that supports, not just documents

An electronic record that hides the essential, alerts that tire, images without clinical context: technology can help or get in the way. The report calls for design centered on the diagnostic process.

5. Just culture and learning from error

Punishing the individual and hiding the event prevents improvement. The blog already covers just culture and reporting and hindsight bias: judging the error with the outcome in hand distorts learning.

6. Work environments that sustain diagnosis

Time, interruptions, fatigue and caseload affect the quality of the decision. There are sibling posts on interruptions and decision fatigue.

7. Measurement and research

Without metrics, the topic drops off the agenda. The report calls for better measurement of diagnostic error and research on interventions.

8. Payment and policy (context)

There are recommendations on incentives and health policies. For the student, the hook is knowing that the problem goes beyond the classroom: when you become a manager or service leader, the report remains a reference.

Clinical reasoning and the report

Hypothesis, test and revision

The classic cycle (generating hypotheses, seeking discriminating data, revising) is the microcosm of the process the report describes at system scale. Eva summarizes what the teacher needs to know about clinical reasoning: it is not enough to "teach thinking"; practice with feedback and attention to how knowledge is organized are needed (Eva, 2005).

When the error is one of knowledge and when it is one of process

Confusing the two delays the remedy. A content gap calls for study and spaced retrieval (knowledge retention). A process failure calls for a pause, a second opinion, a mental checklist and a culture of voicing uncertainty. The post on knowledge or reasoning gap helps to separate them.

Biases without moralism

The report does not reduce error to a list of biases with English names. Still, knowing affective bias, implicit bias and omission bias helps to review one's own case without turning the debrief into a tribunal.

Metacognition as a habit

Connection with probability and tests

Diagnostic error is not just "the wrong name for the disease". It includes ordering the wrong test, interpreting without pretest probability or treating an incidental finding as if it resolved the picture. That is why the report connects well with blog texts on pretest probability and base rate: the diagnostic process includes calibrating the weight of each clue.

In study practice, after each case, write three lines: initial probability, the data point that moved the needle most, and what is still missing to close with confidence. This micro-habit mirrors the cycle the report describes at system scale.

Asking "what else could it be?" and "what would make me change my mind?" is cheap and aligned with the report. Metacognition in medicine and calibration of confidence are concrete training.

What changes in study and in preceptorship

For those who study alone

• Close cases with a line of "explanation to the patient": if you cannot say in clear language what you think and what is still uncertain, the report's communication criterion has still failed.
• In the error notebook, mark whether the failure was content, data collection, integration or premature closure.
• Train a short differential and a decisive test, not just long lists (differential diagnosis with clues).
• Alternate specialties and contexts (outpatient and emergency) so you do not build false expertise in a single setting.
• Review with retrieval questions before rereading: testing strengthens retention (Larsen et al., 2013; Raupach et al., 2016).

For those who precept

• Ask for the hypothesis and what would falsify it, not just the test plan.
• Normalize the phrase "I don't know yet" when it is true; this models safe communication.
• In M&M, separate what was visible at the time from what only appears with the outcome (hindsight bias).
• Include nursing, the lab and the patient's account when possible: the report insists on teamwork.
• Give specific feedback on reasoning, not just on the grade (feedback in cases).

Checklist aligned with the report

1. Definition: an explanation that is accurate, timely and communicated.
2. Process: data, hypotheses, test, review.
3. Team: who else needs to be in the loop.
4. Patient: what was said and what was left open.
5. System: what the environment made easier or blocked.
6. Learning: what changes tomorrow in the habit or in the service.
7. Care for the professional: avoid public humiliation (second victim).

Teaching example (fictional)

Fictional case with a real clinical presentation: a young adult with fever and sore throat; discharged with a diagnosis of viral infection. Two days later, the patient returns worse and another service identifies a bacterial condition that called for a different approach. In the debrief aligned with the report, the questions do not start with "how did you get it wrong?". They start with: what data was available; which hypotheses were generated; what would have changed the approach; how the uncertainty was communicated; which system barrier (time, queue, follow-up) contributed. The focus is recovery and redesign, not humiliation.

What the report changed in the discourse

Before 2015, many services treated diagnostic error as a matter of individual ethics or of "improving reasoning" in isolated classes. After the report, it became more legitimate to ask for indicators, channels for critical result feedback, protected time to discuss ambiguous cases and patient participation in reviewing the plan. This does not by itself solve the emergency department queue; it changes management's question: instead of "who got it wrong?", it becomes "where did the process let the explanation fail or arrive late?".

For the international student, the report also serves as a lingua franca. In placements outside your own country, citing Improving Diagnosis helps align expectations with preceptors who already use the definition of error as a failure of timely explanation and communication. It is not jargon from a single health system.

Recovery: the error that is corrected in time

A useful reading of the report is its emphasis on recovery. Much harm does not come from the first wrong label; it comes from the delay in revising it when new data arrives. Training the habit of "reopening the case" when the course does not match the hypothesis is as important as getting it right the first time. In study practice, this means noting in your notebook not only the final diagnosis of the answer key, but the moment when a new clue should have overturned the previous hypothesis.

Communicating error and near miss

When the error reaches the patient, the conversation matters. The blog already covers how to communicate a diagnostic error. The report reinforces that communicating the explanation is part of the diagnostic act; communicating the failure, when it occurs, is part of safety and trust. Near misses also deserve recording in a just culture: they are gold for low-cost redesign.

Research still open

Zwaan, Schiff and Singh discussed the research agenda for reducing diagnostic error: better operational definitions, measurement methods and evaluation of interventions (Zwaan et al., 2013). Schiff and colleagues analyzed hundreds of errors reported by physicians and showed recurring patterns of failure (Schiff et al., 2009). Croskerry described the importance of cognitive errors and strategies to mitigate them, taking care not to turn the list of biases into moralism (Croskerry, 2003).

For those writing a thesis or starting undergraduate research, the report and these articles are a starting point: the field still needs well-designed intervention studies, not only descriptive series.

What to take to the next study shift

• Reread the report's definition in one sentence and apply it to a case from the day: was the explanation accurate, timely and communicated?
• Separate in your notebook what was a data, hypothesis, test or communication failure.
• Combine an active recall block (questions without looking) with a process debrief, not just a content debrief.
• If you study in a group, assign roles: one defends the hypothesis, another looks for what overturns it, a third writes the sentence to the patient.
• Link the case to a sibling blog post (diagnostic pause or confidence calibration) so you do not study the error only as guilt.

How NEXO helps you train

NEXO offers fictional cases with a real clinical presentation in which the task is to decide the next step with up to six clues. This trains the microprocess the report describes at a larger scale: generating hypotheses, asking for the discriminating data, revising when the new clue arrives. It does not replace M&M, the medical record or the conversation with the person receiving care. It complements study by making visible where the chain broke.

See the Economy. Those who coordinate a group or a league can get to know NEXO Closer. Researchers, reviewers and creators find NEXO Researcher, NEXO Curator and NEXO Creators.

Frequently asked questions

What is the Improving Diagnosis in Health Care report?

It is the 2015 report from the National Academies (IOM/NAM) on diagnostic error: definition, evidence, gaps and system recommendations (NAP, 2015; Ball and Balogh, 2015).

How does the report define diagnostic error?

As a failure to establish an accurate and timely explanation of the patient's health problem, or to communicate that explanation to the patient (Ball and Balogh, 2015).

Does the report blame the individual physician?

No. The framing is about process and system, with education, team, technology, culture and measurement (NAP, 2015).

What is the relation to cognitive biases?

Biases matter, but Norman and Eva show that knowledge and process also weigh; the report does not reduce everything to a list of biases (Norman and Eva, 2010).

Is diagnostic error rare in outpatient care?

Estimates in outpatient care in the US indicate that the phenomenon is common enough to deserve patient safety attention (Singh et al., 2014).

What can the student do in practice?

Train cases with feedback, mark the type of failure in the error notebook, practice communicating uncertainty and ask for the discriminating data before closing.

Does the report replace M&M sessions?

No. It gives language and priorities; M&M remains one of the places of collective learning (morbidity and mortality sessions).

How does the report relate to just culture?

It calls for learning without automatic humiliation of the individual; reporting and analyzing events requires psychological safety, a theme linked to just culture.

Where can I read the full text?

On the National Academies Press portal, catalog 21794 (NAP, 2015). The clinical summary is in Annals (Ball and Balogh, 2015).

Play the case of the day on NEXO and practice closing the chain with a clear, reviewable explanation.

References

1. National Academies of Sciences, Engineering, and Medicine. Improving Diagnosis in Health Care. Washington (DC): National Academies Press; 2015. Available at: [https://nap.nationalacademies.org/catalog/21794/improving-diagnosis-in-health-care](https://nap.nationalacademies.org/catalog/21794/improving-diagnosis-in-health-care)
2. Ball JR, Balogh E. Improving Diagnosis in Health Care: Highlights of a Report From the National Academies of Sciences, Engineering, and Medicine. Ann Intern Med. 2015. PMID: [26414299](https://pubmed.ncbi.nlm.nih.gov/26414299/)
3. Graber ML, Franklin N, Gordon R. Diagnostic error in internal medicine. Arch Intern Med. 2005. PMID: [16009864](https://pubmed.ncbi.nlm.nih.gov/16009864/)
4. Graber ML. The incidence of diagnostic error in medicine. BMJ Qual Saf. 2013. PMID: [23771902](https://pubmed.ncbi.nlm.nih.gov/23771902/)
5. Singh H, Meyer AN, Thomas EJ. The frequency of diagnostic errors in outpatient care: estimations from three large observational studies involving US adult populations. BMJ Qual Saf. 2014. PMID: [24742777](https://pubmed.ncbi.nlm.nih.gov/24742777/)
6. Singh H, Schiff GD, Graber ML, Onakpoya I, Thompson MJ. The global burden of diagnostic errors in primary care. BMJ Qual Saf. 2017. PMID: [27530239](https://pubmed.ncbi.nlm.nih.gov/27530239/)
7. Norman GR, Eva KW. Diagnostic error and clinical reasoning. Med Educ. 2010. PMID: [20078760](https://pubmed.ncbi.nlm.nih.gov/20078760/)
8. Eva KW. What every teacher needs to know about clinical reasoning. Med Educ. 2005. PMID: [15612906](https://pubmed.ncbi.nlm.nih.gov/15612906/)
9. Larsen DP, Butler AC, Roediger HL 3rd. Comparative effects of test-enhanced learning and self-explanation on long-term retention. Med Educ. 2013. PMID: [23746156](https://pubmed.ncbi.nlm.nih.gov/23746156/)
10. Raupach T, Andresen JC, Meyer K, et al. Test-enhanced learning of clinical reasoning: a crossover randomised trial. Med Educ. 2016. PMID: [27295475](https://pubmed.ncbi.nlm.nih.gov/27295475/)
11. Schiff GD, Hasan O, Kim S, et al. Diagnostic error in medicine: analysis of 583 physician-reported errors. Arch Intern Med. 2009. PMID: [19901140](https://pubmed.ncbi.nlm.nih.gov/19901140/)
12. Zwaan L, Schiff GD, Singh H. Advancing the research agenda for diagnostic error reduction. BMJ Qual Saf. 2013. PMID: [23942182](https://pubmed.ncbi.nlm.nih.gov/23942182/)
13. Croskerry P. The importance of cognitive errors in diagnosis and strategies to minimize them. Acad Med. 2003. PMID: [12915363](https://pubmed.ncbi.nlm.nih.gov/12915363/)
14. Miller GE. The assessment of clinical skills/competence/performance. Acad Med. 1990. PMID: [2400509](https://pubmed.ncbi.nlm.nih.gov/2400509/)

<!-- revisao_clinica_thiago:

• Definition of diagnostic error: check wording against NAP 2015 / Ball Balogh 2015 (accurate and timely explanation + communication to the patient).
• Singh 2014: avoid citing a percentage without the paper open; in the text I used qualitative language ("common", "fraction with serious harm").
• Fictional case of pharyngitis/viral: no prescriptive conduct; only pedagogical debrief.
• Relation to biases: Norman/Eva 2010 as a brake on reductionism.
• PMIDs Schiff 19901140 and Zwaan 23942182: confirm titles in the review (auxiliary searches).
• No phone, no /programas, no NEXO efficacy.

-->