Pattern-Plus-Exception Recognition¶

Why It Matters¶

Pure pattern recognition says "this looks like X, therefore treat for X." But most dangerous misdiagnoses occur when common patterns have uncommon underlying causes. A "flu-like illness" is usually viral, but can be early sepsis or meningitis. Pattern-plus-exception corrects the systematic error of stopping at pattern recognition without checking if this instance is typical.

The Key Move¶

When you recognize a pattern, immediately ask: "What would be different if this were actually something else masquerading as this pattern?" Generate specific exceptions to look for. Check them explicitly. If you find exceptions, weight them heavily - they indicate this instance deviates from the typical case, which dramatically changes probability estimates.

Classic Application¶

"Typical chest pain" recognition. Crushing substernal chest pain radiating to the jaw is the classic heart attack pattern. But 30% of actual heart attacks present atypically. So pattern-plus-exception: recognize the pattern, then check exceptions: Is the patient very young? Female? Diabetic? (Heart attacks present atypically in these groups.) Is the pain sharp rather than pressure? Is it reproducible with palpation? These exceptions override the initial pattern match.

Surprising Application¶

Fraud detection in financial systems. A transaction matches the pattern of legitimate customer behavior (location, amount, merchant type), so it passes. Pattern-plus-exception adds: "What would be different if this were fraud mimicking legitimate behavior?" Check exceptions: Is the velocity unusual (three similar transactions in an hour)? Is the device fingerprint new? Is the amount just below reporting thresholds? The exceptions catch sophisticated fraud that mimics patterns.

Failure Modes¶

Seeing exceptions everywhere creates false alarms - every case has some atypical features. The tool requires calibration about which exceptions matter. It fails when you don't know what exceptions to look for (requires domain knowledge of how conditions present atypically). Overweighting rare exceptions above common patterns is also dangerous - hoofbeats are usually horses, not zebras, even if the horse has an unusual marking. The tool works best when exceptions have high predictive value, not just because they're surprising.

Go Deeper¶

Norman, G. (2005). "Research in Clinical Reasoning: Past History and Current Trends." Medical Education, 39(4), 418-427; Pearl, J. (2000). Causality: Models, Reasoning, and Inference, on how exceptions provide diagnostic leverage.