An electronic health record may document a medication order without recording what happened afterward. Did the patient fill it, take it as intended, experience side effects or stop because of cost? Linking these events to the condition being treated is part of understanding the course of care.
Medication and problem lists do not always preserve the links between a treatment, its purpose and its outcome.
The first paper in our LIPI-KG series, presented at AMIA IPC in August 2026, uses medication reconciliation to connect Lawrence Weed's problem-oriented record with Warner Slack's patient–computer dialogue.
Two pioneers, two halves of the answer
Lawrence Weed gave us the problem-oriented medical record. His insight was that a record organized by source, service, or date is the wrong shape for clinical reasoning. The patient's problem should be the organizing unit, and observations, assessments, plans, and follow-up should all hang from it. For Weed, the record was not a passive archive; it was an instrument that should guide care and teach.
Today's charts preserve the form of that vision (we still write SOAP notes) but often lose its logic. Problems are copied forward without re-evaluation. Plans are disconnected from what happened next. Medication orders pile up without a durable link to their indication or the reason they were changed.
Warner Slack showed, as early as the 1960s, that a computer could conduct a medical history directly with a patient. He treated the patient as an active source of structured clinical information rather than a passive subject of documentation, and he later extended that work to histories patients completed at home before a visit. Crucially, he saw patient–computer dialogue as a way to strengthen the patient–clinician relationship, not replace it.
Weed supplied the organizing unit: the problem. Slack supplied the reach: the patient, including at home. Our work argues that you need both.
Why medication reconciliation is the right test
Medication information exists in several states that are easy to confuse: prescribed, ordered, dispensed, available in the home, reported by the patient, and actually taken. Treat them as interchangeable and you get false certainty.
The research literature makes the gap concrete. In one study of older home-care clients, 83% had discrepancies between in-home interviews and their electronic records, and 40% had discrepancies judged capable of compromising treatment. Earlier work found that clinic-derived lists captured a complete list of regular medications for only about half of patients. None of this means the patient is always right; it means no single list tells the whole truth.
Consider a patient with hypertension who takes their antihypertensive only some days because the morning dose leaves them dizzy. A conventional reconciliation flags "not taking as prescribed" and updates the list. A problem-oriented, longitudinal account would go further. It would connect the dizziness to timing and dose, relate the intermittent use to blood-pressure readings, keep the patient's explanation, record the clinician's response, and follow whether the revised plan brought both the blood pressure and the symptom under control.
That is the difference between correcting a list and understanding a course of care.
Introducing Orientation
We use the term Orientation for the explicit act of determining where a patient currently stands within the course of a specific problem. Orientation asks a handful of questions before anything else:
What is the current state? What evidence supports it? What has already been tried? What did the patient actually do? What happened as a result? What is still unresolved?
Placing Orientation before the familiar History–Exam–Assessment–Plan sequence gives us O-HEAP. This is more than reordering a note. It changes the frame of inquiry, so that everything gathered afterward is interpreted in light of where the problem stands. For medication reconciliation, Orientation turns a discrepancy into the start of reasoning instead of the end of documentation.
Looking ahead, an AI-generated patient summary could propose Orientations for each relevant problem, and each Orientation could then pull together the longitudinal evidence behind it: how the problem was recognized, what was tried, what the patient did at home, and what followed. The aim is not to have AI summarize the chart or predict the next event. It is to show how a problem has been approached over time, what worked, and what didn't.
The foundation underneath
This approach needs an information model that outlives any single encounter. Our work builds on a long lineage:
The Longitudinal Electronic Medical Record (LEMR) provides the metadata foundation for connecting concepts, observations, decisions, interventions, and outcomes over time. The Intelligent Electronic Medical Record (iEMR) offers an early implementation of problem-oriented reasoning. O-HEAP makes the orientation step explicit. Together they support LIPI, the Longitudinal Intelligent Patient-Centric Phenotype, and its graph form, LIPI-KG, the Patient Orientation Knowledge Graph.
Five principles for a patient-oriented medication record
The paper closes its argument with five design principles that we think any next-generation record should honor:
- Link every medication to purpose and problem. Keep why a therapy was started, continued, changed, or stopped, and keep the many-to-many relationships between drugs and problems explicit.
- Distinguish intended therapy from actual use. Represent prescribed, dispensed, available, reported, and taken as separate states with their sources, and preserve uncertainty when they conflict.
- Treat the patient's home experience as clinical evidence. Routines, symptoms, cost, beliefs, caregivers, and access often decide whether a plan succeeds.
- Preserve reasoning and outcome across time. A corrected list is not a finished clinical story.
- Use computation to strengthen human partnership. Reduce the burden of reconstructing the story, make the evidence visible, and prepare patients and clinicians for a better conversation, never substitute an opaque output for their judgment.
What we are and aren't claiming
Directed patient graphs, temporal knowledge graphs, medication lists, and patient portals are not new on their own. Our contribution is their problem-oriented synthesis: Orientation as the entry point, LIPI as the evolving phenotype, and a directed longitudinal representation of problem solving. These claims still need technical validation, testing on longitudinal data, and careful attention to provenance, privacy, bias, workflow burden, and patient control.
Medication reconciliation is the first use case, not the last. The same model applies to any problem whose management depends on what happens between visits.
What's next
The proposed record links medication use to the patient's problems, earlier decisions and outcomes. Patient-reported experience at home supplies information that may be missing from the clinical chart.
In Part 2, we move from concept to data model: what exactly is a Problem Orientation, and how do you represent it so a machine can reason with it?
This post is adapted from "From the Problem-Oriented Medical Record to Patient Orientation: Medication reconciliation as a bridge between the visions of Lawrence Weed and Warner Slack," Paper One of the LIPI-KG series, presented at AMIA IPC, August 20, 2026.
Key references: Weed LL, N Engl J Med 1968;278:593–600 · Slack WV et al., N Engl J Med 1966;274:194–198 · Slack WV et al., JAMIA 2012;19:545–548 · Tiihonen M et al., Pharmacoepidemiol Drug Saf 2016;25:100–105 · Yang JC et al., J Gen Intern Med 2001;16:112–115.