A working diagnosis is useful because it gives us a framework for deciding what to do next.
But there is a dangerous moment in clinical medicine, when the diagnosis starts becoming more important than the evidence.
The initial impression may have been reasonable. The treatment may even have been appropriate at the time. The problem begins when new information arrives and we continue trying to make that information fit the original diagnosis.
Modern diagnostic-safety frameworks emphasize that diagnosis is an iterative process: clinicians gather information, interpret it, form a working diagnosis, test that hypothesis and revise it as new information becomes available. 
That means changing your mind is not diagnostic failure.
Sometimes, changing your mind is the correct diagnostic action.
Here are five clinical clues that should trigger a deliberate reassessment.

1. One important finding does not fit
This is probably the most useful clue.
You have a working diagnosis that explains most of the presentation—but one clinically important feature refuses to fit.
It may be:

• the severity of the symptoms
• the time course
• an examination finding
• a laboratory abnormality
• an imaging finding
• an unexpected demographic/contextual feature
  or a symptom that the diagnosis simply cannot explain satisfactorily.

The temptation is to call it an “atypical presentation.”
Sometimes it is.
But sometimes the atypical feature is the clue.
A practical question
Instead of asking:
“Can this diagnosis cause this finding?”
ask:
“How well does my diagnosis explain the entire picture?”
Those are different questions.
Almost any diagnosis can be made to accommodate an isolated finding if enough exceptions are allowed.
The more exceptions you need, the weaker the explanatory power of the diagnosis becomes.
The National Academies’ framework specifically describes diagnosis as an ongoing process in which clinicians should continue gathering and interpreting information when the available information is not consistent with the working diagnosis. 
Clinical example
A patient presents with fever, cough and tachycardia.
Pneumonia seems straightforward.
But the patient has:
profound hypotension,
severe abdominal pain,
a relatively unimpressive chest examination,
and a rapidly worsening lactate.
It is possible to have severe pneumonia with an unusual presentation.
But before repeatedly explaining away the abdominal findings, ask:
What else could produce this combination?
The important clue may not be the fever.
It may be the finding that pneumonia doesn’t adequately explain.

2. The patient’s trajectory is wrong
A diagnosis is not just expected to explain where the patient is now.
It should also make reasonable predictions about where the patient is going.
This is why the patient’s clinical trajectory is diagnostically valuable.
Ask:
“If my diagnosis is correct, what should happen next?”
Then compare that prediction with reality.
Reconsider when:

• the patient deteriorates unexpectedly
• symptoms progress despite appropriate initial treatment
• new organ dysfunction appears
• the clinical course is substantially more severe than expected
• the time course does not resemble the presumed disease
• or the patient repeatedly returns with the “same” problem without behaving like someone with that diagnosis.

Treatment itself can provide diagnostic information. The National Academies’ model explicitly describes treatment response as part of the feedback loop that can refine a working diagnosis. 
Example
A patient is diagnosed with uncomplicated cellulitis and started on appropriate antibiotics.
Twenty-four hours later:
pain is dramatically increasing,
the patient develops systemic toxicity,
the skin findings are progressing rapidly,
and the degree of pain seems disproportionate to the visible changes.
The question is no longer simply:
“Should the antibiotic be changed?”
It should also be:
“Was cellulitis the correct diagnosis in the first place?”
The worsening trajectory is new diagnostic evidence.

3. A new result directly contradicts your working diagnosis
This is where clinicians can fall into confirmation bias.
A test result that supports the diagnosis is readily incorporated.
A result that conflicts with it may be:
dismissed as unreliable,
attributed to laboratory error,
labelled incidental,
or interpreted in a way that preserves the original diagnosis.
Sometimes that is completely appropriate.
Tests have false positives, false negatives and limitations.
But a strongly discordant result deserves active reconciliation, not automatic dismissal.
Diagnostic reasoning literature recognizes anchoring and confirmation bias among important contributors to diagnostic error. 
Try this instead
When an important result conflicts with your diagnosis, ask three questions:
1. Could the test be wrong?
2. Could I be interpreting the result incorrectly?
3. If the result is correct, what diagnosis would make the whole picture make more sense?
The third question is often the one that gets skipped.

4. You are accumulating explanations instead of evidence
This is a subtle warning sign.
Suppose your working diagnosis requires you to say:

• “That symptom is unusual, but possible.”
• “That laboratory result is probably incidental.”
• “The age is atypical, but it can happen.”
• “The treatment isn’t working yet.”
• “The imaging finding is probably unrelated.”

Any individual explanation might be reasonable.
But when several are required simultaneously, pause.
You may be protecting the diagnosis rather than testing it.
A useful mental experiment is:
“If I encountered this patient for the first time today, without knowing the previous diagnosis, would I reach the same conclusion?”
If the answer is no, diagnostic momentum may already be influencing your thinking.
This is closely related to anchoring and premature closure—continuing with an initial diagnostic impression despite subsequent information that should prompt reconsideration. 

5. A dangerous alternative diagnosis still explains the case
Not every alternative deserves equal attention.
But some alternatives deserve attention because the consequences of missing them are substantial.
Examples include presentations in which you must consider:

• acute coronary syndrome
• pulmonary embolism
• aortic catastrophe
• meningitis
• intracranial hemorrhage
• bowel ischemia
• perforation
• sepsis
• ectopic pregnancy
• necrotizing soft-tissue infection
• serious drug toxicity

The point is not to order every test for every dangerous diagnosis.
It is to recognize when the current diagnosis has left a high-consequence alternative insufficiently addressed.
Ask:
“What is the most dangerous plausible diagnosis that I have not adequately excluded?”
Then ask:
“What finding would I expect if that diagnosis were true?”
And finally:
“Do I have enough evidence to safely move away from it?”
This approach is particularly important because diagnostic error is not limited to making the wrong label. It includes missed and delayed diagnoses, and the diagnostic process continues through follow-up and reassessment. 

The five-clue diagnostic reset
When something feels wrong about the diagnosis, don’t simply generate a longer differential.
Try this short reset:
1. Fit
What important finding does my diagnosis fail to explain?
2. Trajectory
Is the patient behaving the way I would expect with this diagnosis?
3. Contradiction
What new information argues against my diagnosis?
4. Exceptions
How many special explanations am I using to preserve it?
5. Consequence
What dangerous alternative remains plausible if I am wrong?

If several answers point in the same direction, the threshold for reassessment should be low.
A diagnosis is a hypothesis, not a verdict.
This is perhaps the most important distinction.
A working diagnosis does not need to be perfectly certain before treatment begins. Clinical decisions often have to be made under uncertainty. The goal of diagnostic reasoning is to reduce uncertainty enough to make a reasonable clinical decision—not to achieve absolute certainty. 
But once new evidence appears, the probability of the diagnosis should be allowed to change.
That means:
Initial impression → new evidence → updated probability → revised plan
—not—
Initial impression → new evidence → explanation for why the initial impression must still be right.
Diagnostic reasoning is inherently iterative. 

What doctors commonly miss
The most dangerous diagnostic error is not necessarily choosing the wrong diagnosis at the first encounter.
Sometimes the initial diagnosis was entirely reasonable.
The problem is failing to recognize when the evidence has changed.
A patient can start with a plausible diagnosis and later develop evidence that makes another diagnosis more likely.
Therefore, one of the safest habits in clinical medicine is to deliberately ask:
“What would make me change my mind?”
And when that finding appears:
Actually change your mind.

MedRounds Clinical Takeaway
Don’t ask only whether your diagnosis is possible.
Ask whether it is still the best explanation of the patient’s current evidence.
The strongest clinicians are not those who never revise a diagnosis.
They are the ones who recognize when revision is necessary—and do it early enough to matter.

References
1. National Academies of Sciences, Engineering, and Medicine. Improving Diagnosis in Health Care. Washington, DC: The National Academies Press; 2015. doi:10.17226/21794.
Web: National Academies — Improving Diagnosis in Health Care⁠
2. World Health Organization. Global Patient Safety Report 2024. Geneva: World Health Organization; 2024. ISBN: 978-92-4-009545-8.
Web: WHO — Global Patient Safety Report 2024⁠
3. Agency for Healthcare Research and Quality. Diagnostic Errors. Patient Safety Network.
Web: AHRQ PSNet — Diagnostic Errors⁠
4. Elstein AS, Schwarz A. Clinical problem solving and diagnostic decision making: selective review of the cognitive literature. BMJ. 2002;324:729–732. doi:10.1136/bmj.324.7339.729.
Web: BMJ — Clinical problem solving and diagnostic decision making⁠
5. Graber ML, Kissam S, Payne VL, et al. Cognitive interventions to reduce diagnostic error: a narrative review. BMJ Qual Saf. 2012;21(7):535–557. doi:10.1136/bmjqs-2011-000149.
Web: PubMed — Cognitive interventions to reduce diagnostic error⁠
6. Norman GR, Eva KW. Diagnostic error and clinical reasoning. Med Educ. 2010;44(1):94–100. doi:10.1111/j.1365-2923.2009.03507.x.
Web: PubMed — Diagnostic error and clinical reasoning⁠
7. Etchells E. Anchoring Bias With Critical Implications. Patient Safety Network. Rockville, MD: Agency for Healthcare Research and Quality; 2015.
Web: AHRQ PSNet — Anchoring Bias With Critical Implications⁠

Editorial note for MedRounds
Author: MedRounds Editorial Team
Evidence reviewed: 25 August 2026
Next review: Update when major diagnostic-safety or clinical-reasoning evidence materially changes the recommendations.

Disclaimer: This article is intended for healthcare professionals and medical learners for educational purposes. It is not a substitute for clinical judgment, local protocols, specialist consultation, or patient-specific medical advice.