Diagnostic Atypia

Diagnostic Atypia 🦓 🐎

Atypical presentations are described as “a shortage of prototypical features that are most frequently encountered in patients with the disease, features encountered in advanced presentations of the disease, or simply features of the disease commonly listed in medical textbooks.”

Atypical presentations can make diagnosis more challenging and distract from the diagnostic process. Atypical presentations have been increasingly recognized for their impact on diagnostic errors. For example, they are associated with a higher prevalence of diagnostic errors compared to typical presentations, and the major contributing factor to diagnostic errors. Previous studies suggested that if physicians were aware of some specific patterns of atypical presentations, the prevalence of diagnostic errors in patients with atypical presentations could be lower.

To counter this, Harada et al. propose that each presenting feature of a disease can be classified into 1 of 4 feature types (”PSUC”):

• Primary features
• Suggestive features
• Uncommon features
• Chameleon features

Primary Features 💡

The textbook-classic findings. The ones that activate immediate pattern recognition. Chest pain in MI, sudden focal deficit in stroke, RLQ pain in appendicitis.

Suggestive Features 🔙

Not classic for the disease, but when present, they should nudge you toward it. Think: acute pulmonary edema or cardiogenic shock as a presentation of MI. These aren't in the "chief complaint" box of your illness script, but they mechanistically make sense — an infarcting ventricle can't pump, so the downstream hemodynamic consequences become the presenting syndrome. The pathophysiology connects them; you just have to follow the chain.

Uncommon Features ❓

Real features of the disease that are genuinely low-frequency. Fatigue and balance problems in MS, or dizziness as a stroke presentation. These are part of the disease's pathophysiological repertoire, but they sit at the tail of the distribution. You won't see them in many textbook tables. The danger is that their rarity makes them invisible to pattern recognition — you've likely never built a strong association between them and the diagnosis.

Chameleon Features 🦎

The most dangerous category. These are features that are not classical for the clinical unknown but are primary features of a different disease entirely. For example, acute psychiatric symptoms as the presentation of stroke, throat discomfort as the presentation of ACS. The patient's presentation perfectly matches a different illness script, so you get anchored to the wrong diagnosis with high confidence. The chameleon feature actively misdirects you.

“PSUC” ⊕ Diagnostic Errors 💭

The following DDx shares 4 patterns of PSUC features that are strongly associated with diagnostic errors.

Pattern 1: Suggestive ❓

What's happening: lacks Primary disease features (i.e. features always written in textbooks), but has Suggestive features (i.e. can still stimulate consideration of specific disease).

Example: isolated acute pulmonary edema as the presentation of ACS (i.e. no chest pain), isolated AMS as UTI in the elderly (i.e. no dysuria, no frequency), isolated neutrophilic leukocytosis as cholecystitis (i.e. no RUQ pain, no colic)

Why you miss it: your illness script demands the Primary feature as a gatekeeper, and the Suggestive feature gets attributed to a "more obvious" cause or a “proximal” diagnosis.

Cognitive workaround:

• Reverse the causal chain. Work backward through the pathophysiology. Don't start from the diagnosis and look for confirming features. Start from the physiological derangement and generate a list of upstream causes.
• Pursuit of “endpoint diagnosis.” Seek the underlying causative explanation for a patient’s signs, symptoms, and laboratory and radiographic data, exhausting additional relevant diagnostic evaluation.

Pattern 2: Suggestive ⊕ Uncommon ❓

What's happening: both Suggestive and Uncommon features are present, but Primary features are absent (i.e. the patient has real features of the disease, but they're the low-frequency ones that don't trigger recognition).

Example: lung cancer presenting as new neurological symptoms in a patient with known COPD. The patient already carries a diagnosis of COPD, so chronic cough and dyspnea get attributed to COPD exacerbation (i.e. the pre-existing condition absorbs the signal). Meanwhile, the cancer has metastasized to the brain, producing focal neurological deficits — an Uncommon presenting feature of primary lung cancer.

Why you miss it: uncommon features live in the long tail of your illness scripts. You've maybe seen them mentioned in a review article but never personally encountered them. Worse, the co-occurring Suggestive features may get attributed to a pre-existing condition (e.g. new neuro symptoms in a COPD patient get worked up as a separate problem, and nobody connects them to an occult thoracic malignancy).

Cognitive workaround:

• Force a “unifying diagnosis.” When a patient has ≥2 complaints you're tempted to explain with separate diagnoses, stop and ask: "Is there one disease that could produce all of this?" This is especially important when the patient has a known chronic condition that's absorbing signal. For example, the COPD patient with weight loss, new hemoptysis, and now a seizure — COPD explains the cough, but it doesn't explain the weight loss and seizure. Meanwhile, one diagnosis (lung cancer with brain mets) can unify all 3.
• Deliberately expand your illness scripts for high-stakes diagnoses. For the diseases where missing the diagnosis causes serious harm (ACS, stroke, PE, aortic dissection, cancer), actively learn the uncommon presentations. These are the ones that show up in M&M conferences.

Pattern 3: Uncommon ❓

What's happening: the patient presents with only Uncommon features (i.e. the presentation is a known-but-rare manifestation of the disease in isolation).

Example: chest pain in duodenal perforation. Extraocular complaints (e.g. headache) in acute angle-closure glaucoma.

Why you miss it: there's no classic feature and no downstream clue to redirect you. The isolated Uncommon feature gets diagnosed as whatever it most obviously resembles in isolation. Using the examples above, chest pain becomes "ACS," the headache in angle-closure becomes "tension headache."

Cognitive workaround:

• Learn how “horses” can present as “zebras.” For every high-volume chief complaint you see, have a short mental list of the dangerous diagnoses that can present with only that symptom (e.g. headache → subarachnoid hemorrhage, angle-closure glaucoma, temporal arteritis; dizziness → posterior-circulation stroke). The key is that this list is short, rehearsed, and triggered automatically. You're not generating a full differential; you're running a 10-second safety screen.
• Use age-sex-comorbidity mismatch as a trigger. An isolated Uncommon feature becomes far more concerning in a patient with risk factors for the serious diagnosis. For example, abdominal discomfort in an elder with hypertension, diabetes, and CKD deserves an ECG and troponin.
• Management “safety net.” Explicitly set a timeline for expected improvement and document it. Treatment failure is a signal to reconsider whether your initial diagnosis is correct.
• Pay attention to the "naive observer." Sometimes nurses, family members, or the patients themselves notice something that doesn't fit your working diagnosis. The family says "she seems different.” These observations from people who aren't constrained by illness scripts can be the subtle discordance that flags a Pattern 3 presentation. Take them seriously.

Pattern 4: Uncommon ⊕ Chameleon ❓

What's happening: both Uncommon and Chameleon features are present, and the patient has features that are primary for a different disease (i.e. the presentation actively mimics something else).

Example: acute psychiatric symptoms (confusion, agitation, psychosis) as the presentation of frontal lobe stroke (i.e. these are primary features of a primary psychiatric disorder). Nausea, vomiting, and epigastric pain as the presentation of inferior MI.

Why you miss it: chameleon features satisfy your pattern recognition for the wrong illness script, so you stop searching. Premature closure is nearly guaranteed because the presentation "makes sense.”

Cognitive workaround:

• Calibrate your confidence against the base rate of mimicry. For certain diagnostic pairs (common vs. mimicker), chameleon presentations are common enough that you should expect them. For example, certain antibody-mediated encephalitis diseases mimic psychiatric illness in the majority of cases early in the course. When your working diagnosis is one of the "commonly mimicked" conditions (primary psychiatric illness), your confidence should be automatically reduced by the known base rate of this chameleon presentation.
• Flag "first episode" presentations of diseases that usually have a history. First episode of psychosis in someone over 40? First seizure? First-time "panic attack" in someone without a psychiatric history? These deserve extra scrutiny because the chameleon diseases (e.g. stroke, encephalitis) are mimicking conditions that usually have a longitudinal history.
• Look for subtle discordances. Chameleon presentations are rarely perfect copies. There are usually small findings that are inconsistent with the mimicked disease. The "psychiatric" patient with subtle asymmetric reflexes. The "vestibular neuritis" patient with subtle skew deviation on HINTS. The "gastritis" patient with diaphoresis. The "migraine" patient with personality change. These discordances are the breadcrumbs back to the true diagnosis, but you have to be actively looking for them. The cognitive move is to ask: "If my working diagnosis is correct, what findings would be unexpected?" Then specifically examine for those unexpected findings.

1. Atypicality is not random — it's predictable by population.

Women, elderly patients, and patients with multiple comorbidities are consistently overrepresented in atypical presentations across diseases. When you're seeing a patient in one of these demographic groups, actively calibrate your index of suspicion upward for atypical presentations of serious disease.

2. The "atypicality gap" is subjective and trainable.

Harada et al. point out that atypicality is measured by the gap between the physician's mental image of the disease and the patient's actual presentation. This means you can shrink the gap by deliberately expanding your illness scripts. Every M&M case, every case report of a missed diagnosis — these are opportunities to update your prototypes. The more presentations you encode as "this is also what Disease X can look like," the fewer patients will fall through.

3. Use time as a diagnostic tool.

For Patterns 1–3 especially, re-evaluation over time can reveal the diagnosis. Scheduled reassessment, clear return precautions, and explicit documentation of diagnostic uncertainty ("atypical presentation, cannot exclude X, plan to reassess in Y hours") are safety nets that catch what initial pattern recognition misses.

4. When your gut says something is wrong but you can't name the diagnosis — trust the gut and test broadly.

The subjective feeling of diagnostic uncertainty is itself a signal. If a presentation makes you uncomfortable, that discomfort is data. Widen the net.

5. Build "if-then" triggers into your practice.

• If acute pulmonary edema with no clear precipitant → get an ECG and troponin
• If "benign" chief complaint + cardiovascular risk factors → rule out vascular emergency
• If diagnosis doesn't respond to treatment as expected → reconsider the diagnosis entirely

• Atypical Presentations at Risk for Diagnostic Errors in Internal Medicine: A Scoping Review

• Pursuit of “endpoint diagnoses” as a cognitive forcing strategy to avoid premature diagnostic closure