💊 AI Allergy Cross-Reactivity Alert Simulator

A Documented Allergy Sits in the Chart

One past reaction becomes a lifelong safety flag.

• Penicillin: Allergen on file (beta-lactam class)
• Variable: Reaction severity (mild rash to anaphylaxis)
• ~10%: US patients allergy-labeled (report penicillin allergy)
• <10%: Labels confirmed true allergy (on formal testing)

Where the record lives

EHR allergy fields store drug, reaction, and severity.

Why it matters downstream

Every future order gets silently checked against this list.

A single label, many drugs

One allergy can implicate an entire chemical family.

A Related Drug Gets Prescribed

A clinician orders a drug from a chemically adjacent class.

• Cephalexin: New order example (cephalosporin class)
• Beta-lactam: Shared ring system (core reactive structure)
• Every one: Orders needing allergy check (point-of-prescribing screen)
• Common: Missed interaction alerts (without structural AI)

Why the new class was chosen

It treats the infection but shares a chemical backbone.

The naive check fails

Exact drug-name matching misses cross-class chemistry.

Where AI steps in

Structure-aware screening catches what name matching cannot.

Comparing Chemical Scaffolds Atom by Atom

The AI overlays molecular graphs to find shared substructure.

• Graph overlay: Comparison method (ring and side-chain matching)
• Beta-lactam ring: Shared core in penicillins/cephalosporins (primary allergenic trigger)
• ~1–3%: Typical reported cross-reactivity (penicillin to cephalosporin)
• 0–100%: Similarity score range (overlap of shared regions)

Encoding molecules as graphs

Atoms become nodes, bonds become edges for comparison.

Finding shared substructure

Subgraph matching isolates the reactive ring in common.

Scoring the overlap

A similarity percentage summarizes structural closeness.

Turning Similarity Into a Risk Probability

Structural overlap and reaction history combine into one score.

• 2 factors: Model inputs (similarity + severity history)
• Highest: Anaphylaxis history weight (raises risk multiplier)
• 0–100%: Risk output (cross-reactivity probability)
• ~50%: Alert threshold (typical) (triggers clinician warning)

Combining the two signals

Risk scales with structural overlap and past severity.

Why severity history counts

Anaphylaxis history raises caution more than a mild rash.

Calibrating the threshold

A risk cutoff decides when to interrupt the clinician.

The Alert Reaches the Prescriber in Real Time

A risk-scored banner interrupts ordering before the drug is given.

• Order entry: Alert delivery point (before signing the prescription)
• Risk %: Alert content (plus shared structure note)
• Yes: Clinician override allowed (with documented reasoning)
• Prevent reaction: Goal (without blocking needed care)

What the clinician sees

A percentage risk and the shared chemical structure.

Deciding to override or switch

Clinician can pick an unrelated alternative drug class.

Closing the loop

The decision is logged back into the patient record.