Medication Management: Reconciliation, Safety, and Clinical Data Workflows用药管理：连接用药史、核对、安全与临床数据工作流

Medication management can be viewed at two connected levels. At the individual level, it helps a person and their care team understand every prescription medicine, nonprescription product, vitamin, supplement, and relevant as-needed product; why it is used; how it is taken; whether it is helping; what monitoring is required; and what questions remain. At the system level, it covers the lifecycle through prescribing and ordering, order communication, dispensing, administration, documentation, monitoring, reconciliation, and reporting. A safe program must connect both levels because a technically correct order can still fail when the record is incomplete, the person cannot follow the regimen, a transition is poorly communicated, or follow-up never occurs.

The central deliverable is not merely a list. It is a current, source-attributed and time-aware account of medication use, plus a review trail. Each entry should distinguish an active prescription from a historical or discontinued item; an order from an administration; an intended schedule from reported behavior; and a suspected reaction from a verified allergy or adverse drug reaction. The record should also show who supplied the information, when it was observed, how conflicts were resolved, and which professional accepted the final interpretation.

Medication management is related to, but not identical with, medication therapy management, medicines optimization, adherence support, pharmacy inventory management, or medication administration. Those services may sit inside a broader cross-setting workflow for organizing evidence and professional review, while their exact scope varies by care setting and jurisdiction. Local law, professional scope, formulary rules, institutional policy, and the patient’s goals still control clinical decisions.

For focused prevention practices, high-risk situations, incident learning, and communication safeguards, continue with the medication safety guide用药安全指南.

Medication management is distributed work. The patient or caregiver contributes the lived record: what is actually taken, what was stopped, missed doses, practical barriers, affordability, swallowing difficulty, side effects, preferences, and use of products obtained outside the main health system. Prescribers define indications, goals, treatment intent, and follow-up. Pharmacists verify medication identity, formulation, supply, duplication, interaction information, administration practicality, and medication-related problems within their scope. Nurses and care workers may administer medicines, observe responses, document refusals or omissions, and escalate changes. Health information and data teams preserve identity, provenance, terminology, access, and longitudinal continuity.

No single source or role is automatically complete. An electronic health record may contain discontinued orders that look active. A dispensing feed may show supply without proving ingestion. An administration record may show that a dose was given but not why a long-term regimen changed. A patient-held list may be current but omit strength or prescriber. Claims can reveal fills across organizations but usually lack the clinical reason for change. The workflow should therefore treat each source as evidence with a defined meaning, not as an unquestionable truth.

The review owner must be explicit at each stage. Administrative or analytical staff can collect documents, normalize fields, flag contradictions, and prepare questions, but they should not resolve a clinical conflict outside their role. A physician or qualified prescriber may decide whether a medicine remains indicated. A pharmacist may conduct medication review or recommend an intervention according to local scope and collaboration arrangements. Urgent concerns must follow the organization’s escalation procedure rather than waiting for a routine data review.

Common settings include primary care, specialist clinics, hospital admission and discharge, emergency care, community pharmacy, home health, long-term care, behavioral health, and transitions between them. The same core record can support all settings, but the acceptable time window, review depth, and responsible professional differ. A discharge reconciliation may need completion before the patient leaves; a longitudinal population review may use a scheduled cycle; an emergency presentation may prioritize immediately relevant medicines, allergies, last doses, and high-risk conditions.

A medication history asks what the person has actually used over time, not simply what one system currently labels as active. Begin with all prescribed medicines, over-the-counter products, vitamins, supplements, herbal products, injections, inhalers, patches, creams, eye or ear preparations, and relevant intermittent or as-needed products. For each item, capture the generic and brand name when available, ingredient, strength, dose, route, formulation, frequency, indication, prescriber or source, start date, stop date, last dose, current status, adherence pattern, observed benefit, suspected adverse effects, and monitoring requirements. Record uncertainty rather than converting an estimate into a fact.

Use at least two sources when the consequences of error are important. Sources may include a patient or caregiver interview, medication containers, a patient-held list, an electronic medication list, dispensing history, pharmacy communication, discharge documentation, specialist letters, administration records, claims, and monitoring results. Reconcile identifiers before merging. A product name may represent different strengths or formulations; a changed prescription may coexist with an older refill; and a combination product may be duplicated by its individual ingredients.

Time is essential. “Current” should have an as-of date. A start or stop date may be exact, approximate, inferred, or unknown, and the record should say which. Separate order date, dispense date, administration date, reported-use period, and verification date. These events answer different questions. If a person reports taking half a tablet while the latest order says one tablet, preserve both statements, mark the discrepancy, and route it for professional confirmation instead of silently choosing one.

High-quality history taking also asks about behavior and context without blame. Explore difficulty opening packaging, remembering schedules, obtaining refills, paying for medicines, swallowing, using devices, understanding instructions, and fitting treatment into meals, work, sleep, or cultural practices. These observations may explain gaps, but they do not justify changing the plan without review. The detailed field-level method is covered in the related medication history guide.

Continue with the dedicated medication history workflow用药史整理方法 when the task is to construct or repair the longitudinal list itself.

Medication reconciliation is a defined component of medication management. It compares the best available pre-transition medication history with new orders or the planned post-transition regimen, identifies every difference, and determines whether each difference is intentional, documented, clinically appropriate, and communicated. It is especially important at admission, internal transfer, discharge, referral, change of prescriber, and return to community care. A copied list is not reconciliation; the work is the comparison, explanation, resolution, and accountable handoff.

Use a side-by-side view that preserves both states. For every medicine, compare ingredient, strength, dose, route, formulation, frequency, indication, status, and timing. Classify a difference as an intentional start, stop, substitution, dose change, route change, temporary hold, formulary substitution, duplicate, omission, unexplained continuation, or uncertain item. Avoid classifying a difference solely from text similarity. A generic substitution may be equivalent; a similar-looking product may not be; and a route or formulation change may materially alter use.

Intent must be traceable. Record the clinician or source supporting the change, the date, the reason when available, any monitoring or follow-up, and how the patient or next care team was informed. If the reason is not available, label it unresolved and assign follow-up. Do not infer that a missing discharge item was intentionally stopped, or that a medication appearing on an old list should automatically continue. The professional reviewer must decide after considering diagnoses, current observations, laboratory results, goals, allergies, procedures, and other relevant context.

Completion should be measured by resolved discrepancies, not by whether a form was opened. Useful process checks include the percentage of transitions with a documented source history, time from transition to reconciliation, unresolved high-priority discrepancies, communication to the next setting, patient understanding, and correction of the longitudinal record. Metrics require clearly defined denominators and should not be interpreted as proof of clinical benefit without appropriate evaluation.

The focused medication reconciliation process用药核对流程 provides a detailed transition checklist and discrepancy framework.

Interaction checking is a screening step, not an independent treatment decision. A result may describe a potential drug–drug, drug–disease, drug–food, drug–laboratory, or duplicate-therapy concern. Its significance depends on the exact ingredients, formulation, dose, route, timing, duration, indication, organ function, laboratory context, age, pregnancy status where relevant, comorbidities, previous response, and monitoring plan. A database severity label summarizes a rule or evidence classification; it does not fully represent the person’s situation.

Prepare a complete and current list before screening. Include prescribed and nonprescription products, supplements, recently stopped medicines with lingering effects, allergies or reactions, relevant laboratory results, diagnoses, and the actual use pattern. Normalize ingredients and combination products so the same active substance is not counted twice under different names. Preserve the source and version of the interaction knowledge used, because databases differ and recommendations change.

Review findings in a queue that separates evidence from disposition. Show the interacting items, type of concern, evidence source, mechanism or practical explanation when available, patient-specific modifiers, existing safeguards, unanswered questions, and responsible reviewer. The reviewer may determine that a finding is not applicable, already monitored, clinically accepted for a documented reason, or requires follow-up. The data workflow should record that disposition without converting it into a generic instruction for other patients.

Never use a checker result alone to stop a medicine, skip a dose, substitute a product, or change timing or dose. Some abrupt changes can themselves cause harm, and the risk of untreated disease may exceed the interaction concern. Urgent symptoms or suspected serious reactions should follow local emergency and clinical escalation procedures. The drug interaction checker safety guide药物相互作用检查器安全指南 explains how to prepare a list and discuss a result with a doctor or pharmacist.

A concise record summary helps a reviewer see why a regimen exists and how it changed. Begin with the active problems and treatment goals relevant to medication use, then align key medication events with encounters, procedures, laboratory and imaging results, physiological observations, allergies or reactions, and documented decisions. Preserve the difference between event time and documentation time. A late note may explain an earlier change; displaying it only by note date can make the sequence misleading.

Use layers rather than compressing everything into one paragraph. The first layer can show the current medication state, urgent unresolved questions, allergies, recent major changes, and required monitoring. A second layer can provide the longitudinal timeline. A third can link each claim to source records. This design helps a professional scan quickly while retaining the ability to verify. Every summary statement should be traceable to one or more sources or explicitly labeled as patient-reported, inferred, or pending confirmation.

Do not hide disagreement in a polished narrative. If two records conflict on dose or stop date, state the conflict, show both sources, and identify who must resolve it. Do not turn absence of data into evidence that an event did not occur. Do not copy every historic diagnosis into the medication context unless it helps explain an indication, contraindication, risk modifier, or monitoring need. A concise summary is selective, but its selection rules must be visible.

Useful quality checks include source-link completeness, coverage of the requested time window, correct medication status, preservation of units and reference ranges, unresolved discrepancies, reviewer confirmation, and whether a second qualified reader reaches the same understanding. The dedicated medical record summary guide病历摘要指南 provides a source-linked timeline method.

Analytics can support medication management at individual, service, and population levels, but the question and unit of analysis must be explicit. At the individual level, a timeline may help a reviewer relate medication changes to laboratory values, symptoms, encounters, adherence observations, or outcomes. At the service level, teams may examine reconciliation completion, unresolved discrepancies, monitoring follow-up, refill processes, or alert burden. At the population level, governed analysis may identify groups needing record review or outreach. None of these outputs proves a medication caused an outcome without an appropriate design and professional interpretation.

Clinical analytics focuses on clinically meaningful care processes and outcomes: whether monitoring occurred, how treatment pathways vary, or where safety-relevant gaps appear. Patient analytics focuses on patient-level attributes, utilization, needs, and outcomes for a defined population. Patient journey analytics examines sequences across touchpoints, such as referral, consultation, pharmacy, discharge, and follow-up, to locate delay, dropout, duplication, or handoff failure. These lenses can use overlapping data, but they answer different questions and should not be collapsed into one undifferentiated dashboard.

Before analysis, define the population, index event, observation window, exposure logic, outcome, exclusions, and data cutoff. Separate prescribed, dispensed, administered, and reported use. Control immortal-time and look-ahead errors when time matters. Explain missingness, coding change, incomplete outside-system data, and whether a refill proxy can reasonably support the intended question. When comparing groups, show counts and denominators, baseline differences, and sensitivity checks. When reporting trends, state whether changes may reflect documentation or workflow changes rather than clinical change.

Use the clinical analytics guide临床分析指南 for care-process and outcome questions, the patient analytics framework患者数据分析框架 for patient-level cohort work, and the patient journey analytics method患者旅程分析方法 for cross-touchpoint sequences. Each page preserves its own audience, data model, and decision boundary.

Medication management rarely lives in one system. Relevant evidence may be distributed across EHR medication orders, medication administration records, community and hospital pharmacy feeds, dispensing or claims data, laboratory systems, clinical notes, discharge documents, patient portals, device records, and external organizations. Integration must preserve identity, event time, status, terminology, units, provenance, and access controls. Moving fields into one warehouse does not make them clinically equivalent.

Start with a source contract. For each feed, document what event it represents, when the event occurred, when the record was created and received, how corrections appear, which identifiers are used, how deleted or entered-in-error items are represented, what code systems and units apply, and which periods or facilities are covered. Medication data needs ingredient, product, strength, form, route, dose, frequency, status, reason, and linkage to orders or administrations where available. Normalize only after retaining the original value and source.

Identity matching deserves separate governance. A duplicate patient can split one history; an incorrect merge can attach another person’s medicine. Use approved matching rules, confidence levels, manual review for ambiguous cases, and an audit trail for merges and unmerges. Do not expose direct identifiers in an analytical workspace unless they are necessary, authorized, and protected. Access should follow role, purpose, minimum necessary use, and applicable privacy and retention requirements.

Healthcare data integration addresses the broader enterprise architecture spanning clinical, operational, claims, registry, device, financial, and other sources. Clinical data integration concentrates on making clinical concepts—including medications, allergies, conditions, observations, procedures, and notes—semantically consistent and usable in a particular care workflow. The difference matters because reliable transport alone does not guarantee correct clinical meaning.

For enterprise transport, governance, and interoperability, use the healthcare data integration guide医疗数据集成指南. For clinical terminology, status, negation, time, and provenance in a care workflow, use the clinical data integration guide临床数据集成指南.

Clinical decision support can place medication information at a useful point in workflow: a reconciliation task at admission, an allergy or interaction warning during ordering, a monitoring reminder after a relevant result, or a follow-up queue after discharge. A useful tool presents the right evidence, for the right person, in the right format, through the right channel, at the right time. It should show provenance, logic, uncertainty, and the action owner. More alerts do not necessarily create safer care; low-specificity or poorly timed alerts can add burden and be ignored.

Early warning scores and predictive models may use physiological, laboratory, utilization, medication, or other variables to estimate deterioration or future events. They answer different questions from medication reconciliation. Before use, define the prediction time, outcome, horizon, eligible population, exclusions, threshold, calibration, subgroup performance, missing-data behavior, and response pathway. Prevent leakage from information that would not have been available at the decision time. Validate the model in the intended setting and monitor changes in data and workflow after deployment.

A model estimate is not a diagnosis, and an alert is not an order. Display what data contributed, when it was measured, whether it may be stale, and which limitations apply. Give professionals a way to inspect source information, document disagreement, record disposition, and escalate. If no one owns the action or the service cannot respond within the required time, adding a risk signal may create false reassurance or unmanageable queues.

Evaluate tools with representative cases, including missing values, conflicting lists, unusual formulations, boundary values, duplicate records, late results, and cases that should not trigger. Measure sensitivity and specificity where appropriate, but also measure alert burden, response time, overrides, reasons for disagreement, unresolved items, workflow delay, and downstream corrections. Use the clinical decision support tools evaluation临床决策支持工具评估, early warning score guide早期预警评分指南, and predictive analytics validation guide医疗预测分析验证指南 for their materially different methods.

1. Define the purpose and review owner.

   State the care setting, patient or population, decision to support, time available, and qualified professional responsible for final review. Identify what would make the workflow urgent, routine, or out of scope.

2. Collect the best available evidence.

   Gather patient or caregiver reports, medication containers, orders, dispensing and administration records, discharge documents, laboratory context, allergies, diagnoses, and relevant notes. Record source, event time, update time, and confidence.

3. Normalize without erasing the source.

   Align ingredients, products, strength, dose, units, route, formulation, frequency, status, and terminology. Keep original values beside normalized values and version the mappings.

4. Build the longitudinal medication history.

   Separate current, historical, stopped, held, planned, and uncertain items. Distinguish orders, supplies, administrations, and reported use. Place changes beside the relevant clinical events and monitoring data.

5. Reconcile at the decision point.

   Compare the prior history with the intended new regimen. Classify every difference, identify its source and reason, and assign unresolved discrepancies for professional review.

6. Screen for reviewable concerns.

   Surface possible interactions, duplicate ingredients, allergy conflicts, missing monitoring, unusual changes, and inconsistent instructions. Present evidence and patient-specific context; do not convert flags into autonomous recommendations.

7. Conduct shared professional review.

   The qualified clinician or pharmacist evaluates indication, effectiveness, safety, practicality, goals, preferences, and the complete clinical context. The patient or caregiver contributes experience and questions. Record the conclusion and responsible person.

8. Communicate an accountable plan.

   Update the official record, explain what changed and what remains unresolved, provide an understandable current list, define monitoring and follow-up, and send the information to the next care setting through approved channels.

9. Verify, monitor, and correct.

   Check that required data and actions were completed, monitor new results and reported experience, capture corrections, review overrides and delays, and repeat the cycle after relevant transitions or changes.

Use the table as a review aid, not a scoring instrument. The importance of each domain depends on the decision and the possible consequence of error. Define acceptance criteria before applying automation. For safety-relevant fields, sample the underlying records and have an appropriate professional confirm clinical meaning. For data transformations, reproduce results from versioned inputs and mappings. For workflow acceptance, verify that the right person receives the result in time and can inspect its evidence.

Hypothetical example only: an adult is discharged after a hospital stay and will be followed by primary care and a community pharmacy. The hospital order list, administration record, discharge summary, community dispensing history, patient-held list, and recent laboratory results do not perfectly agree. One product appears under a brand name in the patient list and as two ingredients in the EHR. A medicine present before admission is absent from the draft discharge list. Another has a different frequency in the dispensing history. These are data observations, not proof that any prescription is wrong.

The coordinator first verifies identity and the time window, then preserves each source rather than overwriting it. The analyst normalizes ingredients, strength, form, route, and frequency, while retaining original text. The system groups possible duplicates and displays before/after lists. It labels the absent medicine as an unresolved omission rather than “stopped,” because no documented reason is available. It labels the frequency difference as conflicting evidence. A potential interaction appears, along with the source knowledge version and relevant laboratory context.

The pharmacist and prescribing clinician review indication, documented decisions, actual administration, patient report, current laboratory information, and planned monitoring. They determine the final status within their professional roles, document reasons, and define any follow-up. The patient receives an understandable current list and knows whom to contact with questions. The primary care team receives the reconciled record, unresolved items, monitoring plan, and source links. The data team records which mappings and rules were used.

Validation asks more than whether the final list exists. A second reviewer should be able to trace each material statement to a source, see why differences were classified, identify who accepted the final state, and confirm that handoff and follow-up occurred. If later evidence corrects a source or decision, the record should preserve the previous version, show the correction, and identify any downstream summary or analysis that needs revision.

Begin implementation with one bounded use case rather than a universal medication platform. Define the care setting, target population, decision point, source systems, review owner, expected response time, and consequences of error. Document what the workflow will and will not do. A discharge reconciliation queue, for example, has different data latency and escalation needs from a retrospective population review. Pilot with representative records before expanding scope.

Create a data dictionary and source register. Define every critical field, allowed values, units, terminology, event-time meaning, status transitions, provenance, update behavior, and known limitations. Version mapping tables and calculation rules. Include test cases for duplicates, combination products, entered-in-error orders, late corrections, missing strengths, ambiguous routes, patient-reported medicines, temporary holds, conflicting sources, and records outside the expected time window. Retain expected results and the professional explanation for each safety-relevant test.

Separate technical, semantic, clinical, and workflow validation. Technical validation confirms files, messages, transformations, and queries run. Semantic validation confirms codes, units, status, and time mean what the workflow assumes. Clinical validation asks whether the information is appropriate for the decision and whether exceptions are visible. Workflow validation checks timeliness, responsibility, communication, burden, and whether action can be completed. A pipeline can pass technical tests while failing the other three.

Govern access and changes. Use role-based authorization, minimum necessary data, secure transfer and storage, retention rules, audit logs, correction procedures, and review of downstream impact. A terminology update, source-system change, new formulary, altered documentation workflow, or interface revision can change results without changing analytical code. Monitor source volume, missingness, mapping failures, identity conflicts, data latency, rule triggers, manual overrides, unresolved reviews, response time, correction rate, and user-reported problems.

Measure outcomes against the original purpose. Process measures may include completion, time to reconciliation, unresolved discrepancies, follow-up completion, and source-link coverage. Safety or clinical outcome evaluation requires an appropriate design and should not be inferred from process improvement alone. Review performance by relevant setting and subgroup, investigate unexpected differences, and document limitations. Avoid targets that encourage superficial completion, such as closing a task without resolving the evidence conflict.

Plan for failure and rollback. If a source feed stops, a mapping becomes unreliable, or the review queue exceeds capacity, define how results are blocked, labeled, or routed. Maintain a change log and the ability to identify affected outputs. Notify responsible teams when prior summaries or decisions may need re-review. Test downtime and manual procedures. The safest workflow is not the one that never reports problems; it is the one that makes problems visible, contains their impact, and supports accountable correction.

Common failures include an incomplete history, unverified patient identity, copied-forward inactive medicines, duplicate ingredients hidden by brand names, order status confused with actual use, missing start or stop dates, event time confused with documentation time, units or routes lost during normalization, outside-system medicines omitted, patient-reported information presented as verified, and discrepancies closed without a documented reason. Analytical failures include ambiguous denominators, look-ahead leakage, treating refill data as ingestion, treating association as causation, and hiding missingness behind a single score.

Workflow failures are equally important: no named reviewer, alerts sent to someone who cannot act, unclear escalation, excessive queue burden, inaccessible source records, corrections that do not propagate, and handoffs that update one system but not the patient or next team. Automation can repeat an error at scale. A well-designed workflow therefore favors traceability, human review, exceptions, and rollback over a polished but opaque answer.

Implementation should use current organizational policy, local law, professional scope, and primary sources. The references below support the broad process and safety principles; they do not replace patient-specific review.

• AHRQ medication management strategy for patient and family engagement
• WHO Medication Without Harm initiative and medication-safety resources
• NHS England structured medication reviews and medicines optimisation
• Medicare medication therapy management and safety-management information
• 医数智析 official gallery description