See Chart w/GPT in Action
Whether you’re using Chart w/GPT, or you’re queued for activation, we wanted to provide you with a candid build update using feedback from different user groups.
We’ll keep sending these every few weeks so you can see what’s improving, what still needs work, and how your feedback is shaping the product you’ll use.
Below is a synthesis of recent sessions with primary care physicians focused on first-visit prep, preventive care, medication reality vs. EMR lore, and “first-pass resolution” for quality and risk.
TL;DR (Primary Care Workflow We’re Designing Around)
- Start with meds → infer conditions and spot mismatches.
- Preventive status → show what’s truly due vs. already done (with documents).
- Risk stratify fast → utilization at a glance + stability signals (A1C, COPD bursts, HF weight trends).
- Act now → produce a one-page, source-cited snapshot you can file in the EMR and use to drive today’s visit.
See a Chart w/GPT Demo on YouTube: https://youtu.be/bFaGdAPE8Ws
What We Tested (and Learned)
A. Medications as the fastest clinical picture
- Reality check vs. EMR lists. Active meds are frequently wrong or stale; lists “accumulate” unless someone cleans them. Fill history (claims/Surescripts) is the most trusted proxy for adherence.
- Clinical inference. Providers naturally infer conditions from meds (e.g., nifedipine → consider chronic angina; anastrozole → active—not historical—breast cancer). Chart w/GPT should both list meds and surface likely conditions/HCCs with evidence.
- What builds trust. A visible, one-click “Why we believe this” line with citations to the exact source record (or fill data when available).
- “This line is very valuable… I’d want an asterisk that says how we know this is true.” — Dr. Brackman
B. Preventive screenings the way clinical teams measure them
- Guideline alignment matters. Recommendations must match USPSTF/CDC cutoffs (e.g., colon cancer screening intervals, Pap stop at 65, pneumococcal PCV20 sufficiency after 65). Inconsistency erodes trust.
- Documents or it didn’t happen. To close HEDIS/Stars gaps, teams often must download the actual report (e.g., mammogram, LDCT) and file it in the EMR. Patient- or provider-reported alone won’t earn credit in many cases.
- “Find the report” beats “five portals.” If Chart w/GPT can locate the imaging note and hand you a link/download, you win back 30–60 minutes of portal chase.
- “You don’t get credit unless you have the document… Links to direct documents will be incredibly handy.” — Josh
C. Risk stratification in two moves
- Utilization headline first. Open with a bold line like:
“6 ED/admits in last 6 months” → instant care-management trigger (nurse CM, SW, tighter cadence, etc.).
- Stability signal next. Flag out-of-range A1C, frequent COPD “burst” scripts (steroids/abx), CHF weight volatility—condition-specific “unstable” tells.
- Why it matters. Earlier identification = earlier resources = fewer avoidable admissions.
- “Give utilization up top so I know to deploy resources now.” — primary care physician
D. “First-pass resolution” as a product principle
- Definition. Can a provider accept/close a prompt on first review because the evidence is already attached (e.g., CKD3A with two qualifying GFRs 3+ months apart; dementia with MoCA score + ADL/IADL deficits; a specialist note that nails the diagnosis)?
- Implication. Every summary we return should have evidence expandable in one click—reducing dismissals and rework.
- “If the evidence is there, I accept it and move on.” — primary care physician
What Landed Well
- Prompts with curated logic outperformed ad-hoc free-text (“the juice was worth the squeeze”).
- Preventive tab that pairs “status” with clear next-due logic and rationale got strong marks—when guidelines were correct.
- Hospitalization/ED summary that reads like a case manager brief (“why they went in, patterns, what to do next”).
- The vision of a one-page, source-cited face sheet for new patients was called a “game changer.”
- “Basically, this is wonderful. I spent two minutes and have a great understanding of this person.” — primary care physician
Gaps & Requests (We’re On It)
- Source truth everywhere. Citations/links on every critical line (meds, screening dates, labs, imaging). Download when possible.
- Guideline fidelity. Lock to USPSTF/CDC recs by age/sex/hx (e.g., Pap stop at 65; FIT-DNA every 3 years; PCV20 rules).
- Utilization topline. A bold metric at the top of summaries (e.g., “3+ ED/admits in 90 days”) with quick route to details.
- Stability cues by condition. Diabetes (A1C), COPD (burst fills), HF (weights), etc.
- Medication reality. Show last fill dates when we have them; clearly label confidence when we don’t.
- Fewer hops. Reduce “prompt-hunting.” One New-Patient Snapshot prompt should assemble meds → preventive → risk (utilization + stability) with evidence.
- Face-sheet export. One-page PDF/HTML you can drop in the EMR and reuse every visit.
- Performance consistency. Reduce response lag; make curated prompts and free-text produce comparable depth.
Changes Shipping Next (In Progress)
New-Patient Snapshot (beta): One prompt that returns:
-
- Active meds (with fill evidence when available) and likely conditions
- Preventive status with USPSTF/CDC-aligned next-due and doc links
- Utilization headline (ER/inpatient counts with dates)
- Stability indicators for diabetes/COPD/HF
- Actions to take today (orders, tests, referrals)
- Citations + Download for each fact
- Guideline engine hardening: Explicit rulesets for age cutoffs/intervals; unit tests to prevent drift.
- Citations everywhere: Inline “Why we believe this” with hover → source and “Open document” (download when allowed).
- Face-sheet export (1–2 pages): Timestamped, source-cited, EMR-friendly.
- Utilization badge: Prominent line at the top of visit history summaries.
- Medication confidence labels: “Active per record,” “Recently filled,” or “Active—fill unknown” (with rationale).
Open Questions We’re Bringing Back to You
- Medication reconciliation workflow: If we show “probable duplicates” or “likely discontinued,” do you want bulk resolve inside Chart w/GPT, or just clear flags to fix in the EMR?
- AWV & SDOH signals: Which Annual Wellness Visit elements (PHQ-9, GAD-7, MoCA, ADLs/IADLs, POA/DNR) are highest value to surface first?
- Download vs. deep link: When is a link sufficient vs. a download mandatory for your quality team? (We’ll default to download where permitted.)
If You’re Waiting to Go Live on Chart w/GPT...
We’re finishing verification → scheduler feed → record pulls for new practices.
Typical timeline: up to 10 days.
As soon as your first patients are mapped, you’ll get access with the features above (and we’ll keep shipping improvements behind the scenes).
In the meantime, if you want your New-Patient Snapshot tuned to your clinic’s preferences (e.g., which stability markers to surface first), contact us with your top five “must-see” items.
For a complete rundown of our platform, please watch this video.