The 70% No One Demos
Every clinical AI demo opens with a perfect post-op patient: two weeks of longitudinal data, a clean wound, a compliant exercise log. The post-op case is tidy because the raw material is already sitting in the system — the AI just summarizes a trend.
That is not the problem most clinics face. A typical orthopedic outpatient day is dominated by first visits: new back pain, a twisted knee, a shoulder that has been aching for months. The patient has no history in the system. Roughly 60–70% of these cases will never require surgery. Almost no published clinical AI shows what a good summary looks like for this group.
This post is about designing for that majority. iRehab calls the one-line summary that lands on the physician's screen before a first-visit consult the Brief. A companion piece, Intake Compression, Not Form Consolidation, covers the post-op follow-up case.
Briefing a Single Moment, Not Weeks of History
For a post-op patient the system has material to work with — a VAS (Visual Analog Scale, 0–10 pain score) trend, wound photos across days, exercise completion rates. The AI distills a pattern out of a rich time series.
A first-visit patient has none of that. What the system has is whatever the patient wrote on a pre-consult form ten minutes before walking in, plus maybe a tap on a body figure and a verbal complaint once they sit down. The Brief's job is different: extract specialty-relevant signal from a single structured moment.
The underlying principle is the same — four fields, five-second read, diagnostic hypothesis assembles in the physician's head. But the input shape changes everything about how the form is designed and what the AI is allowed to infer.
Patient-Answerable: The Constraint That Shapes Everything
Walk up to a patient in a waiting room and ask them which tissue is involved. The answers are either a shrug, a polite guess, or the wrong tissue named with confidence. Patients are not trained to distinguish tendon from ligament from nerve, and they should not be. A question that requires clinical training to answer reliably is not an intake question — it is a diagnostic hypothesis dressed up as data collection.
iRehab's pre-visit form treats this as a hard design constraint: every patient-facing question must be answerable by someone without medical training. That single rule forces dozens of downstream decisions.
- "Is this a tendon, ligament, or nerve issue?" — removed. Patients cannot answer this. The AI infers it from questions the patient can answer (next section).
- "Where does it hurt?" — becomes a body figure the patient taps directly, no anatomy labels. The patient points to a spot on a human silhouette.
- "What does it feel like?" — becomes a short list: pain / numbness / weakness / multiple / other. Weakness sits alongside pain because neurological weakness is a red flag that pain alone will not surface.
- "When did it start?" — becomes "what were you doing when it started?": impact / lifting / repetitive motion / woke up with it / crept up gradually / don't remember. Every option is a verb a layperson uses.
- "How long has it been going on?" — becomes five bucketed choices from today to more than a month. Precise durations are notoriously unreliable from patient recall; buckets are easier to answer and more faithful to reality.
Every option passes the same test: can a family member with no medical background answer this without asking what it means? If the answer is no, the option does not ship.
What a First-Visit Brief Actually Looks Like
What arrives on the physician's screen, after all that structured input is compressed, is a one-line telegraphic summary in the format the clinic has used for decades:
45F. CC: R't knee pain going up/down stairs x2wk. VAS 6. Dur <1wk. Onset: lifting. Pattern: intermittent. Sensation: pain only. Wound: no. Zones: R't knee (front). Pt: "The numbness kicks in after I walk for a bit."
Five seconds to read. Clinical handwriting conventions are preserved — R't / L't / B/L for laterality, enum codes left uncompressed so the physician reads them as signals rather than as sentences, a verbatim patient quote in quotation marks to keep temporal nuance the structured fields cannot capture.
The four fields that drive the decision are all there — site, severity, onset type, duration — with two optional fields (wound status, patient quote) layered in when present. A prose version is also generated for documentation export. The shorthand Brief is what loads first because bedside reading speed is the design target.
Tissue Type: Refusing to Ask the Patient
An earlier version of the form asked the patient to classify their problem into one of six tissue categories: joint, muscle, bone, nerve, skin, trauma. Completion rates were decent and many of the answers were usable. The question was removed anyway.
The reason is not that patients answer it badly. The reason is that every wrong-but-confident answer quietly contaminates downstream reasoning. A patient labelling their sciatica as "muscle" because the gluteal muscle hurts has given the physician a cognitive nudge in the wrong direction before the physician has even read the chief complaint.
Tissue type is now inferred by AI from the fields the patient can answer. The inferred category sits on the physician's screen as a small purple badge, collapsed by default:
AI inference: nerve Confidence: medium Based on: numbness and pain together, gradually worsening, VAS 6, right cervical plus right upper limb. [AI-assisted suggestion. Clinical judgment governs.]
The inference lives under hard rules, none of them negotiable:
- It cannot affect red-flag evaluation. Red-flag rules read only the fields the patient filled; AI inference never feeds back into them.
- It cannot block any patient flow. Not triage, not urgent routing, not queue position.
- If the inference fails, the patient's submit still succeeds. A back-end error must not bounce the form back to the patient.
- It is never treated as ground truth. The badge literally says "AI inference" — a suggestion, not a conclusion.
- A physician can override it at any time, for any reason. No justification, no approval, no delay.
A physician who wants to ignore the badge sees, for the rest of their career, two words. A physician who wants the reasoning taps once. The AI is present but never in the way.
The Brief Extends Past the Consult
Most first-visit outpatients do not need surgery, daily exercise regimens, or weekly PROM (Patient-Reported Outcome Measure) questionnaires. They need a follow-up in two weeks, and a lightweight check-in before that follow-up so the physician does not walk in cold.
iRehab exposes this as a single button on the pre-consult detail page: one-click conservative follow-up. One tap promotes the patient onto the conservative-treatment pathway:
- No daily tasks.
- One PROM push three days before the expected visit: a VAS (0–10) and a GROC (Global Rating of Change, 7-point "better / worse / same compared to last time").
- No red-flag re-evaluation cron — the acute window is past.
- Upgrade path preserved — if symptoms worsen, the physician re-promotes the episode in place onto a surgical or generic-rehab pathway; the original episode is kept, not replaced.
The design intent is simple: the Brief should not stop at the door of the consult room. A first visit that ends with a conservative plan should still benefit from pre-visit structure for the next appointment, without adding surveillance burden on a patient whose problem is mild.
Bottom Line
Most clinical AI demos feature the tidy case — the post-op patient with a rich longitudinal record. The clinic-reality majority are first visits without any history in the system. The Brief principle transfers; the implementation shifts.
A post-op Brief summarizes over time. A first-visit Brief performs structural capture in a single moment, constrained by what a patient without medical training can honestly answer. Both end at the same target: the physician reads four fields in five seconds and forms a diagnostic hypothesis before the patient finishes sitting down.
The four fields are not glamorous. They are the durable core of outpatient orthopedics. If a Brief lands on the physician's screen before the visit starts — without ever asking the patient a question they could not answer — the fifteen-minute consult stops being a typing contest and becomes a conversation.