Week 11 update
Interim cumulative results for Weeks 1–11. These are not final pilot results. The earlier four-week report is preserved below and should not be read as Week 11 evidence.
The TheraCareAI pilot with Advocate LLC has onboarded 61 patients against a base target of 75 and stretch target of 100. This represents 81% of the base target and 97% of the Week 11 cumulative target.
481 discrepancies, tracked by source
| Source | Discrepancies |
|---|---|
| EHR | 131 |
| Exchange | 346 |
| Patient-reported | 4 |
| Total | 481 |
The mean is 7.9 discrepancies per onboarded patient, approximately eight. Counts are unevenly distributed: patients with longer medication lists can have many more, and some patients have none. These source-level observations do not establish how many distinct patients had discrepancies.
Discrepancies by type
| Type | Count |
|---|---|
| Outdated / inactive medication still listed | 259 |
| Missing medication | 97 |
| Incorrect dose | 65 |
| Incorrect medication | 20 |
| Duplicate entry | 15 |
| Other | 15 |
| Adherence | 10 |
| Total | 481 |
Approximately 54% of discrepancies involve outdated or inactive medications still listed, carried forward in both EHR and Exchange information.
82 high-significance observations
82 discrepancies, approximately 17% of all observations, were rated high significance, involving an active medication with the wrong drug, wrong dose, or a potentially harmful interaction.
One de-identified Week 11 example involved amlodipine 5 mg and metoprolol remaining listed for a participant no longer treated for hypertension. If administered, these medications could create a risk of significant hypotension. Another involved Vraylar listed at 4.5 mg while the patient reported taking 6 mg. These examples describe potential safety concerns, not demonstrated adverse events.
Platform progress
Five app issues were logged over 11 weeks: four fixed and one open. The Piedmont Healthcare FHIR connection remains open, logged since August 17. Exchange information is cross-checked for every pilot patient, and a photographed discharge medication list is now supported as a baseline.
What these findings do not establish
This is a 13-week directional pilot, not a clinical trial. Observations are counted separately by source, so an underlying issue may appear more than once. These findings do not show that harm occurred or was prevented, and do not establish clinical outcomes or population-level estimates.
The historical 92.6% figure belongs only to the first four weeks. The Week 11 update provides no equivalent affected-patient percentage. No exact Week 11 reporting date was provided with this update.
Historical report: first four weeks
The following original report describes data through August 28, 2026. Its participant rates and counts are historical, not current Week 11 totals.
Patient portals and electronic health records have made medication information easier to reach than ever. For someone managing several prescriptions across several clinicians, that access can make it feel like the problem has been solved: just log in and look at the list.
It hasn't been solved. A list pulled from a health record shows what has been prescribed or documented over time. It doesn't necessarily show what the patient is taking today.
That gap sits at the center of an ongoing pilot of TheraCareAI, a medication information management platform developed by Leap of Faith Technologies and run with our patient advocacy partner, Advocate LLC. The pilot began on August 1, 2026. The findings below come from its first four weeks. They are preliminary, but they already show why getting access to medication data is only the beginning of reconciliation.
What the pilot compares
The design is straightforward. Each participant's medication information is compared across three sources:
- The EHR. Participants install TheraCareAI and authorize access to their electronic health record, and medication and clinical information is retrieved from the connected system.
- An Exchange clinical summary, obtained separately and drawing on health information exchange (HIE) data.
- A verified active medication list. An Advocate clinician reviews the information with the participant and establishes what they report actually taking, verified by a healthcare professional.
It's worth being clear about what the review is not. It doesn't diagnose conditions or direct medication changes. Prescribing, stopping, or changing medications is outside the pilot's scope. The question is narrower: do the available records reflect the participant's current regimen?
The headline: almost everyone had a discrepancy
Of the 27 participants who had completed onboarding and clinical review as of August 28, 2026, 25 (92.6%) had at least one medication-related discrepancy.
This is a 13-week directional pilot, not a clinical trial, and 27 people is far too small and too early a sample to support a population-level estimate. What it does is make the underlying problem visible. Retrieving medication data doesn't automatically answer the practical question every clinician and caregiver actually needs answered: what is this patient taking right now?
Not just spelling errors
It would be easy to assume most discrepancies are cosmetic: a brand name versus a generic, a formatting difference. The early reviews found something more substantive:
- medications still listed after the participant had stopped taking them
- medications the participant reported taking that were missing from the record
- incorrect dosages
- duplicate entries
- differences between the medication documented in a source and the one the participant reported taking
172 observations, tracked by source
Through August 28, the team logged 172 observations, tracked separately by where they came from:
| Source | Observations |
|---|---|
| EHR medication information | 56 |
| Exchange clinical summaries | 115 |
| Patient-reported adherence | 1 |
An important caveat: these are source-level observations, not 172 distinct medication problems. A single underlying issue can show up in more than one source. Tracking each source separately is deliberate, because it shows exactly where the EHR record, the HIE-based summary, and the clinically reviewed list diverge, rather than blending them into a single "error rate."
Some differences matter more than others
Each observation was classified using the pilot's preliminary significance criteria:
| Significance | Count | What it means |
|---|---|---|
| High | 40 | An active medication with the wrong drug, wrong dose, or a potentially harmful interaction that could immediately affect a clinical decision |
| Medium | 69 | Missing or duplicate medications, or an incorrect dose history, without an immediate safety risk |
| Low | 63 | Outdated, inactive, or administrative information that doesn't affect current care |
Being careful about the claim
These classifications do not show that an adverse clinical event occurred, and they don't show that TheraCareAI prevented one. What they indicate is that a meaningful share of these differences go beyond record cleanup.
The early reviews reinforce one critical distinction: medication information retrieved from an existing record may not reflect what the patient is actually taking at the time of review.
Why a verified baseline is valuable
TheraCareAI is meant to support the work that sits between retrieving data and using it well. Its value isn't displaying yet another medication list. It's bringing the available information together, making it accessible to the patient and the people supporting their care, and supporting a structured review against what the patient actually takes.
That verified baseline is most useful exactly where records tend to drift: when a patient sees multiple clinicians or health systems, when prescriptions have changed over time, or when inactive medications linger in the chart.
The pilot is also following participants over time to identify medications added or discontinued after later provider visits, and to understand when another review may be warranted. A verified list is a snapshot; keeping it true is the longer problem.
How this connects to our research
These early observations line up closely with the argument in our LIPI-KG research series. There, we described medication information as existing in several distinct states: prescribed, dispensed, available at home, reported by the patient, and actually taken. Treating those states as interchangeable creates false certainty, and the patient is often the only person who can explain the difference between the intended regimen and actual use.
TheraCareAI puts that idea into practice. Each source is kept separate, disagreements are recorded rather than overwritten, and the patient's own account, reviewed with a clinician, becomes the reference point. It's a practical first step toward a record that reflects lived therapy, not just recorded therapy.
What this pilot doesn't (yet) answer
The current pilot does not directly measure whether the workflow saves time, reduces hospital readmissions, improves clinical outcomes, or produces a financial return. Those claims need larger samples, defined outcome measures, and further evaluation.
The pilot continues through October 2026, aiming for at least 75 fully onboarded participants with the option to expand to 100. At its conclusion, Advocate LLC is expected to report on comparative accuracy, workflow, and usability.
The first four weeks can't answer the big questions. They do show why those questions matter.
A medication list is useful. A medication list that accurately reflects what the patient is taking is considerably more valuable.
TheraCareAI is developed by Leap of Faith Technologies. The pilot is conducted by Advocate LLC in partnership with Leap of Faith Technologies, using EHR and Exchange information. Data reported here are as of August 28, 2026, and are preliminary. Learn more at advocate-patient.com.