Outpatient Synapse — Imaging/results & Physical exams

Redesign grounded in the 7/15 Outpatient Working Session transcript. Where the meeting decided, I built it. Where it didn't, there are options.

This version is grounded — the last one wasn't

I pulled the full 3-hour transcript. My first pass was for acute tele-stroke and was wrong on the premise; this one is built against what you, Melanie, Prachi, Neha, Smriti and Dr. Peeyoosh actually said. The time rail is parked for the Synapse 3.0 acute conversation.

Settled in the meeting

Not re-litigating these

Nine calls the room actually made. They're baked into the mockups below rather than offered as choices.

✓
Yes/No dies. It manufactures junk documentation — "in my note, it says, was that scan done? No. Why the hell did I even click that?"
✓
Pills, never dropdowns. "Please, please not a drop down, please not a drop down, please… I hate drop down[s]. I'm trying to get rid of that click."
✓
Three groups: diagnostic imaging · diagnostic testing/procedures · labs. Labs are free text only — "labs can be really overwhelming and can be organised differently."
✓
Pill → subtype → free text + date. Type, paste, or dictate. No "Add study" button — the pill is the add.
✓
This tab is history, not orders. "This is not ordering. This is all about recording." That's what resolves Prachi's data-mining concern — the mineable structure lives on Recommendations.
✓
New fields: EEG as its own discrete trackable field ("EEG is not under MRI. EEG is its own thing") and spine imaging across CT/MRI — cervical, thoracic, lumbar.
✓
Exam timestamp removed. "The timestamp is not relevant any longer… that's irrelevant."
✓
Exam pill order: free text → clickable → scales. "I would have free text first and then the score pills afterwards."
✓
Nothing mandatory. You: "I personally wouldn't make anything mandatory." Melanie: "I despise them." Scales surface by AI nudge, never by requirement.

Tab 1

Imaging / results

The entry model is settled. What's open is the half you flagged as missing entirely — "In an actual EHR, in Epic, I go to the diagnostics tab and I can see everything that they've had forever. We don't have that."

The entry model — built to spec

History
Imaging/results
Physical exams
Recommendation
New this visit
Diagnostic imaging
Optional
CT
MRI
Vascular imaging
PET
Other imaging
CT — select study
CT head
CT cervical spine NEW
CT thoracic spine NEW
CT lumbar spine NEW
CT head
01/16/2026
No acute intracranial abnormality. Chronic small-vessel ischemic change, stable from prior.
🎤 ⚡ ✨
↳ You picked a study — what did it show?
Diagnostic testing & procedures
Optional
EEG NEW
EMG / NCS
Lumbar puncture
Sleep study
EEG — routine
12/02/2025
Normal awake and drowsy EEG. No epileptiform discharges.
🎤 ⚡ ✨
Labs
Optional FREE TEXT ONLY
Type, paste, or dictate any labs worth keeping…
🎤 ⚡ ✨
🔖 Flag a result to carry forward to every future visit CONCEPT
Pill → subtype → free text + date. No yes/no. No dropdowns. Nothing required.
One concept I drew that you parked without a mechanism

The 🔖 flag on labs is your "save this, this is important… so it always is always flagged visit to visit" — the genetic test, the "crazy antibody that I found five years ago." No mechanism was decided in the meeting, so treat this as a strawman for one: a per-result pin that survives into every future encounter's header, separate from the note body. It's a small build with outsized clinical value, and it's the only idea in the session that had no owner when the meeting ended.

The open half — displaying longitudinal data

Smriti asked for exactly this: "We should discuss this with the UI UX team… we should maybe suggest a couple of the ways." So — three ways. Your constraint on all of them: "That doesn't need to go into my note from 2015, but I still want to remember that they had it in 2010. So that's the history versus the documentation."

OPTION ARecommended
Two zones
Literally what you described: what's new sits on top, everything ever done sits below it in one scrollable record.
New this visit
CT head
01/16/26
No acute abnormality…
Prior record
01/16/26CT head — no acute
12/02/25EEG — normal
08/14/24MRI brain — 2 T2 lesions
03/02/21MRI c-spine — mild sten.
06/11/18MRI brain — unremarkable
09/30/10LP — OCB positive
+Your own words, drawn. Lowest risk of missing the intent.
+Cleanly separates history from documentation — the top zone is what enters the note, the bottom never does.
+Scrolls back to 2010 without a mode switch.
−Long list gets unwieldy for a 15-year patient with 40 studies.
−Comparing two MRIs four years apart means scanning, not reading side by side.
OPTION B
Flowsheet grid
Modality down, time across — the Epic diagnostics view you said Sevaro doesn't have. Tap any cell for the report.
Study'10'18'21'24'25'26
MRI brain
MRI c-spine
CT head
EEG
LP
EMG/NCS
Normal Abnormal Pending
+Fifteen years fits on one screen. Density no list can match.
+Gaps are as visible as data — "never had an MRI" reads instantly.
+Closest thing to the Epic view you said you miss.
−Needs a normal/abnormal flag per study, which means one structured field you don't have today and said you didn't want.
−Every cell is a click to read anything. Density costs depth.
−Sparse early years waste most of the grid.
OPTION C
Stack with inline priors
Group by study type; the current entry shows, with every prior of the same type nested dimmed underneath. Already shipped in OPS.
MRI brain
08/14/24
Two new T2 hyperintense lesions, periventricular.
Prior · 06/11/18
Unremarkable. No demyelinating…
Prior · 09/30/10
Normal study…
+Already built and shipped in OPS. Cheapest path by far.
+Best in the set for the actual clinical question — "is this MRI different from the last one?" Priors sit right under the current text.
+No new structured fields needed. Pure free text, as you want.
−No cross-modality view. Can't see the whole patient at once.
−Answers "what changed" but not "what have they ever had" — which is the thing you said is missing.
My read

A as the page, C as the behavior inside it. They're complementary rather than rival: two zones structure the page, and expanding any row in the prior record gives you C's inline-prior comparison. That covers both questions — what have they ever had (the list) and is this one different (the nesting) — with no new structured fields.

I'd hold B. It's the most impressive view here and the one that most looks like Epic, but it only works if every study carries a normal/abnormal flag, and that's a structured field you explicitly said you didn't want to force. B is worth revisiting only if the AI can infer that flag from the free text reliably — which is testable, and worth testing before anyone commits to the grid.


Tab 2

Physical exams

The pill order is settled. The clickable exam is the thing you've built repeatedly and never liked — so that gets the real work.

The shell — built to spec

History
Imaging/results
Physical exams
Recommendation
✨
Seen for stroke 60 days ago. Add the NIH Stroke Scale?
Add NIHSS
✕
Examination
Optional
Free text exam
Clickable exam
+ NIHSS
+ EDSS
+ UPDRS
+ Search scales…
Alert and oriented to person, place, time and situation. Language fluent without paraphasic error. Cranial nerves II–XII intact. Motor 5/5 throughout with no pronator drift. Sensation intact to light touch…
🎤 ⚡ ✨
↳ Free text first, exactly as specified — clickable and scales are opt-in
NIHSS
AI SUGGESTED LONGITUDINAL
NIH Stroke Scale — trend
3
11/17
12
12/09
7
01/02
4
Today
3
↳ "I want to be able to see the old values and then… enter the new values and see those trend lines over time."
Vital signs
NEVER DISCUSSED
Glucose · BP · O₂ sat · Pulse · Febrile/Afebrile — these survive from the current build untouched. They didn't come up once in three hours.
Timestamp gone. Free text first. Scales opt-in via nudge, never required.

The hard one — why your clickable exams keep disappointing you

A theory, offered for you to shoot down

You said it twice: "I've built this several times and I'm never really satisfied… it's never very efficient for some reason, and I struggle with that" and "my most recent AI build was a million different click boxes… ugly as sin. And I hate it."

Here's my read on the "some reason": every clickable exam is built around the wrong ratio. A normal neuro exam is roughly forty assertions. An abnormal one is those same forty, minus three. So the information you're actually adding is three items — but every design makes you handle all forty to express them. Your own instinct already names the fix — "save as their favorite… and then they can just pull up their favourite and then unselect the items that are abnormal" — but a saved template still renders forty rows, so the screen stays as noisy as before even though the clicking got cheaper. The clicks went away; the cognitive load didn't.

The efficient design makes the exception the only thing that occupies space. All three options below are variations on that, differing in how much of the normal stays visible — which matters, because what you're attesting to has to stay legible.

OPTION A
Exception-only
The entire normal exam collapses to one green line. You add only what's abnormal. Forty rows become one plus your exceptions.
✓
Normal general neuro exam
View 41 findings
Motor — right arm
4+/5 with pronator drift
✕
Gait
Mildly wide-based, unsteady tandem
✕
+ Add abnormal finding
+Fixes the ratio outright. Two exceptions cost two interactions and two rows.
+The screen finally matches the clinical reality — normal is the boring default, abnormal is the story.
+Scales to a 41-item exam without ever looking like 41 items.
−The real objection: one click asserts 41 findings you never looked at. That's a lot of documentation behind a checkbox, and it's a genuine medico-legal exposure — hence "View 41 findings" being permanently on screen rather than buried.
−Needs a good finding-search or "add abnormal" becomes its own dropdown hell.
OPTION B
Module chips
Your seven modules as chips, green by default from your favorite. Open one at a time. You never see more than one module's boxes.
Mental status
Cranial nerves
Motor
Coordination
Reflexes
Sensory
Gait
Motor
✓Bulk & tone normal
5/5 throughout
No pronator drift
✓No fasciculations
Right arm 4+/5 with drift…
+Preserves the mental model you dictated — the seven modules, in your order.
+Bounded pixels: one module open means four to six boxes, never forty.
+Chip color makes an abnormal module findable from across the page. Reflexes can just stay grey — "reflexes on video are like worthless."
+Per-module free text keeps you "relatively broad" and lets granularity go in prose.
−Still seven interactions to review a normal exam, even if each is cheap.
−The closest to what you've built before — which is the honest warning, since that's the thing that keeps disappointing you.
OPTION CRecommended
Live prose mirror
Exception-only on the left; the actual note text renders live on the right, with your changes highlighted. You attest to prose, not checkboxes.
Normal exam
Motor — R arm
4+/5, drift
Gait
Wide-based
+ Add finding
📄 Note preview
Alert, oriented ×4. Language fluent. CN II–XII intact. Motor: right arm 4+/5 with pronator drift, otherwise 5/5. Sensation intact. Coordination intact. Gait mildly wide-based with unsteady tandem.
+Solves A's trust problem without paying B's pixel cost. You see every word you're signing, so the collapsed normal stops being a leap of faith.
+The output is the interface. No mental translation from checkbox to note text — which I suspect is the other half of why clickable exams feel wrong.
+Highlighting makes your three exceptions pop out of the prose instantly.
+The prose stays editable — free text and structure stop being a either/or.
−Needs the full page width. Fine for outpatient; wouldn't survive a narrow panel.
−Most build effort of the three: a real template→prose renderer that has to sound like a neurologist wrote it, not a form.
My read on the clickable exam

C. Your stated preference is free text, and every clickable exam you've built has fought that preference instead of serving it. C doesn't fight it — the structure is just a fast way to produce prose you can still edit, and you always see the prose. If my theory about the ratio is right, C is the one that fixes both halves of it: the exception is the only thing you touch, and you never have to imagine what the checkboxes will read like.

If C is too much build, B is the safe fallback — but I'd say plainly that B is closest to what you've built before, so if the theory is wrong, B is where you'd find out by being disappointed a fourth time.


Still open

What the meeting didn't settle

1. Where do UPDRS, EDSS and MoCA live? — you own this

You: "I'm thinking like the UPDRS — is that an exam or is that a scale? … similar with the EDSS for MS. Where do we put those kind of things? And I'll have to actually kind of think about that a little bit."

Your working answer was NIHSS stays exam-embedded, everything else goes to Scales under History. The mockups above assume exactly that. It's the one structural assumption I'd most like you to confirm or break, since it changes both tabs.

2. Vital signs and the fever chips — a real gap

Glucose, BP, O₂ sat, Pulse, and Febrile/Afebrile are in the current Figma and were never mentioned once in three hours. I left them untouched and greyed rather than redesign something nobody discussed. Worth a direct question to Neha and Smriti — in a tele-outpatient visit where an MA is assisting, it's not obvious these should be typed at all.

3. The "flag forever" mechanism had no owner

The genetic-test / five-year-old-antibody problem was raised, agreed to be valuable, and then the conversation moved on without a mechanism or an owner. My 🔖 pin above is a strawman, not a decision.

4. The tension worth naming out loud

Prachi: "This needs to be positioned in a way where we can mine data off of it quickly… it can't be all these free text."

Your history-vs-orders split resolves this cleanly on paper, and I've built to it. But note what it costs: Option B for diagnostics (the flowsheet) dies under that split, because a grid needs a structured normal/abnormal flag that pure free text can't give it. That's the one place where "free text everywhere" has a visible price, and it's worth deciding with eyes open rather than discovering later.

One thing I'd flag about the Scrum Launch plan

You framed the ambition deliberately: "can we push it further and faster and harder than we would our own internal people?" Prachi's counterweight was equally deliberate: "we still have to maintain this internally. So we need to be reasonable."

Both are right, and the split isn't 50/50 across this document. The diagnostics tab is mostly cheap — pills and free text are ordinary work, and Option A is a list. The exam tab is where the real money is, and Option C's prose renderer is the single most expensive thing here. If something gets handed to an external team, C is the candidate that actually justifies it — it's self-contained, spec-able, and it's the piece your internal team would most likely descope under pressure. The pills you could ship yourselves next sprint.