Medication-Order Reconciliation — Christopher Lyvers

Overview

This simulated heart-failure assignment tested whether a model could reconcile home, discharge and current medication records while renal function and potassium changed across three laboratory time points. The work centered on record identity, source conflict, calculation verification and consistent order classification.

Challenge

A reassuring scanned lab belonged to a different Margaret Donnelly. Meanwhile, the matched records disagreed about medications, and the newest laboratory values were materially worse than the admission values. Correct reasoning depended on excluding the mismatched record and anchoring every calculation to the latest valid data.

Source material

Current inpatient medication orders
Prior discharge summary and home medication list
Serial basic metabolic panel spreadsheet
Vital-sign flowsheet and nursing note
Scanned outpatient lab report belonging to a different patient

My work

Cross-checked name, middle initial, date of birth and MRN across all seven files before combining data.
Excluded the mismatched outpatient lab and documented why its normal values could not be used.
Reconciled medications as continued, changed, added or dropped across home, discharge and current records.
Trended potassium, creatinine and eGFR across three dates and calculated an estimated creatinine clearance from age, sex, weight and current creatinine.
Applied the assignment's threshold and interaction rules to all 11 active orders and produced one disposition per order.
Separated factual record conflict from domain recommendations and kept the final decision with the simulated licensed reviewer.

Analytical approach

Gate on identityDo not blend data until patient identifiers agree.

Choose the current anchorUse the most recent matched laboratory result for renal and electrolyte logic.

Reconcile across settingsCompare what the patient reported, what the prior discharge intended and what is active now.

Check derived valuesRecompute clearance and route/dose conversions rather than copying a draft conclusion.

Classify every orderProvide a consistent, auditable disposition and evidence trail for all 11 orders.

Key findings

Specific examples from the completed work.

Wrong-patient record excluded

Six files matched Margaret A. Donnelly. The scanned report named Margaret R. Donnelly, used a transposed MRN and a different birth date. Its potassium 4.1, creatinine 0.90 and eGFR 72 were excluded because they would have masked the matched inpatient trend.

Deteriorating trend identified

Across 03/16–03/18, potassium rose 4.4 → 4.9 → 5.4, creatinine rose 1.10 → 1.60 → 2.10, and eGFR fell 51 → 33 → 26. The assignment's Cockcroft–Gault check produced approximately 22 mL/min using age 76, female sex, 62 kg and creatinine 2.10.

Medication-history contradiction

Lisinopril and metformin had been explicitly discontinued at the prior discharge but reappeared through an unreliable home list and became active again in the current orders.

Interaction and duplication logic

The active list combined lisinopril with sacubitril/valsartan despite the assignment's 36-hour washout rule, and stacked four potassium-raising agents while potassium was rising.

Route, formulary and dose checks

The review also identified IV-to-oral conversion questions, a non-formulary statin switch, a renal dose ceiling issue and omissions from the prior regimen.

Deliverable

A structured reconciliation log with identity status, source-by-source medication comparison, current laboratory anchors, 11 order classifications, calculation notes, discrepancies, and a concise reviewer message.

Skills demonstrated

Record identity validation
Medication-list reconciliation
Multi-source QA
Trend analysis
Derived-value verification
Rule-based classification
Contradiction detection
Structured decision logging

Why it matters

The transferable strength is not clinical authority; it is disciplined reconciliation. The work shows how I prevent record contamination, choose the correct evidence anchor, verify calculations, and turn conflicting files into a complete classification set.