AI in Medication Reconciliation

Reconciliation compares what a patient was taking before a transition with what is ordered afterward. Every difference has to be either justified or corrected. The starting point is a best possible medication history (BPMH), a term from patient safety programs such as the WHO High 5s project. A BPMH is built from at least two sources, one of which is usually a structured patient interview. In the US, the Joint Commission's National Patient Safety Goals require hospitals to maintain and communicate accurate medication information. Each source has blind spots, and that is where software helps. Pharmacy fill data, often delivered through medication history networks, shows what was dispensed, not what is taken. It misses cash purchases, samples, over-the-counter drugs and supplements. The EHR list may be years out of date. Patients forget drugs, especially inhalers, injectables and drops. Notes contain decisions like "hold warfarin until INR checked" that never reach a structured list. AI systems do four main jobs. They normalize drug names and strengths to standard concepts so brand and generic match. They parse free-text directions. They infer whether a medicine is probably still active from fill dates and days supplied. They extract stop, hold or change intentions from notes. They then list discrepancies for a pharmacist or clinician, often sorted by type: omission, commission, wrong dose, wrong frequency, or duplication. Pharmacists usually separate unintentional discrepancies from intentional ones. Intentional discrepancies are documented or undocumented changes the prescriber meant to make. Only unintentional discrepancies are errors, although undocumented intentional ones still create confusion. A common misconception is that the software does the reconciliation. It prepares evidence, and a clinician still confirms it with the patient and makes the decision. Another misconception is that fill data is ground truth. A regular fill proves pickup, not use.