Skip to main content

Offline Clinic Work and ABDM: Reconcile Before Sending

Offline clinic work and external health-data exchange are separate capabilities. After connectivity returns, identify each queued item, confirm its patient and current version, recheck the relevant permission and reconcile any earlier delivery before sending. A locally saved note is not evidence that ABDM or a laboratory received it. An uncertain acknowledgement is not a reason to blindly replay every queued action.

By Bractis · Updated · 8 min read

Indian clinics evaluating local work, delayed uploads and ABDM or other external connections during intermittent connectivity.

Ask what actually works offline

'Offline-first' can describe several things: viewing a cached page, editing a local draft, saving a signed record or preparing an external request for later transmission. These have different evidence and recovery needs. Ask the vendor to demonstrate the exact capability being offered.

A clinic can continue some authorised local documentation while an external service is unreachable, but that does not make the external exchange offline. The receiver still needs an actual transmission and its own processing or acknowledgement when connectivity is available.

MoHFW's official ABDM explanation states that health-data exchange between intended stakeholders requires patient consent. A local queue entry does not establish that the relevant permission remains valid when an exchange is attempted later.

Distinguish consent to an external data exchange from authorisation of a clinical order and from general application login. One does not automatically supply the others. The exact controls depend on the connected workflow and current implementation.

Source: MoHFW: Consent and security in ABDM, 5 August 2025

Give every pending item enough identity to reconcile

A useful queue record identifies patient, encounter, object type, local version, intended recipient and the action proposed. It also records creation time, any attempted send and the last confirmed outcome. A list of filenames labelled pending is rarely enough.

Separate a document upload from a laboratory order, medicine request or payment action. Repeating an upload may create duplicate documents; repeating an order can initiate real work. The consequence determines the recovery decision.

Keep the source's actual clinical time separate from the later synchronisation time. A visit documented at 10:00 and transmitted at 14:00 should not look as if the examination occurred at 14:00 merely because that was the network event.

Make pending work visible to the responsible role without exposing every patient's information to every user. The doctor may need to resolve a clinical version conflict; an authorised administrator may only need to know that a technical delivery remains unresolved.

Preserve a safe reference for an external operation when the receiver provides one. It can help distinguish an uncertain response from a genuinely new request, but it should not become a bearer link that grants uncontrolled record access.

Work through a small queue before switching on bulk replay

Consider six fictional queued items after connectivity returns. One is a current document ready for its authorised exchange. One has already reached the receiver despite a missing local acknowledgement. One has been superseded by a corrected version. One has a patient-identity mismatch. One needs current permission resolved. One is still a draft.

Only the first is ready for a new send on those facts. The previously received item needs acknowledgement reconciliation; the other four need their respective hold or review decisions. One of six, or 16.7%, being send-ready is a description of this queue, not a system reliability score.

The table makes the action explicit. Do not convert every row to green by deleting unresolved work or treating an attempted connection as delivery. Keep a responsible owner and a visible next decision.

After an actual send, confirm the outcome supported by the receiving service. Technical acceptance may still be different from a clinician reviewing a result or a laboratory accepting an order for collection.

On a narrow screen, scroll the table horizontally to read all columns.

Fictional six-item offline queue: different states need different actions
ItemObserved stateNext decision
ACurrent version and permission confirmedSend through the supported route
BReceiver already has the originalReconcile acknowledgement; avoid duplicate
CSource corrected after queuingReview and select the current version
DPatient association conflictsHold until identity is resolved
ERelevant permission unresolvedRecheck before exchange
FLocal content remains a draftKeep draft; do not imply authorisation

Resolve duplicates and conflicts without guessing from timestamps

A lost response can leave the sender uncertain even after the receiver acted. RFC 9110 explains why repeating a non-idempotent operation needs knowledge that repetition is safe or that the original action was not applied. A clinic should ask how the integration resolves that uncertainty.

An operation identifier is useful only if the relevant systems use it consistently. Do not assume that adding a retry number or resending the same PDF filename guarantees deduplication. Request a demonstrated recovery case in the offered integration.

Conflicting versions require more than choosing whichever timestamp is later. In a fictional example, a note is written at 10:00, corrected at 10:15 and the older queued version attempts to send at 10:20. The later send time does not make the older clinical content current.

Keep the correction history and intended recipient clear. If an obsolete version was already exchanged, the clinic may need an appropriate correction communication rather than simply sending another unexplained document.

Cancellation has its own uncertainty. Removing an item from the local queue does not prove that a receiver cancelled work already accepted. Confirm the supported remote action and its acknowledgement where applicable.

Source: IETF RFC 9110: HTTP semantics, section 9.2.2

Recheck permission and protect the surviving local work

An outage can outlast an exchange permission, change the intended recipient or coincide with a patient's withdrawal or correction request. The clinic should have a defined decision point before delayed transmission, not assume that an earlier screen selection permits every later action.

This is an evaluation requirement for the workflow, not a claim that every system can operate a consent manager offline. If the relevant external permission cannot be established, use the appropriate hold and care-continuity path.

Local persistence must also be demonstrated. Browser storage is not uniformly permanent; MDN describes best-effort and persistent storage and private-session limitations. A queue that exists only on one device needs a documented recovery boundary.

Do not clear browser storage, change user accounts or replace a device as casual troubleshooting while unresolved local work remains. First identify what is safely acknowledged elsewhere and what depends on the current device.

Use fictional data to test loss of connectivity, refresh, account change and return to service. The result should explain what survives, what requires intervention and how the clinic knows it has not sent the wrong version.

Source: MDN: Browser storage quotas and eviction

Close the clinical task separately from the network task

A synchronisation dashboard can show no pending network messages while clinical work remains open. A result may have arrived but not been reviewed; an order may have been transmitted but not accepted by the provider. Keep these next responsibilities visible.

In the six-item example, resolving all technical queue states does not establish six completed episodes of care. Document which items were sent, acknowledged, superseded, held or retained as drafts, and which patient-facing action still needs an owner.

Prepare a shift handover that identifies unresolved items by safe reference and consequence. 'Internet is back' is not a sufficient handover when the previous team does not know which external requests were accepted.

During a Bractis demonstration, ask for one already-received item, one corrected document and one unresolved permission in a fictional queue. Inspect the reconciliation outcome in the actual offered connection rather than relying on a generic offline claim.

The receiver has the document, but the clinic never saw the acknowledgement

In a fictional outage, a document reaches an external service just before the connection drops. The local queue still shows an uncertain outcome. On reconnection, the team uses the supported reference/status route and confirms receipt before retrying. A second item has been corrected while offline and is reviewed separately. The queue is reconciled item by item; neither a bulk replay nor the newest network timestamp determines the correct clinical version.

Offline exchange reconciliation ledger

Use these prompts in your clinic’s own notes. Record the result, responsible person and next action for each item.

Offline scope
Which exact viewing, drafting, saving or queuing actions are supported without connectivity?
Identity and version
What binds each queued action to its patient, encounter, clinical time and current source version?
Permission and prior action
What must be rechecked before exchange, and how is uncertain prior receipt resolved?
Completion
Which technical outcome is confirmed and which clinical review, communication or service action remains open?

Questions doctors ask

Does an offline EMR exchange ABDM records while disconnected?

Local work and external exchange are separate. A queued action still needs connectivity, the relevant permission and successful receiver interaction through the supported workflow.

Should every pending item be sent automatically when the internet returns?

Not blindly. Some items may already be received, superseded, still drafts or blocked by identity or permission questions. Reconcile their states before deciding what can be sent.

Is the latest timestamp always the correct record version?

No. Clinical event time, edit time and transmission time mean different things. Review the actual source version and correction history rather than choosing the last network event.

Does deleting a local queue item cancel a remote order?

Not necessarily. If the receiver already acted, a separate supported cancellation or correction process may be required. Check its acknowledgement rather than inferring cancellation from the local list.

Does successful synchronisation mean the patient's care is complete?

No. Receipt, clinical review, patient communication and service completion are separate states. Assign the remaining work even when the network queue is clear.

Sources and scope

Updated on . The queue and status model are original evaluation aids. They do not claim a particular ABDM implementation, offline consent capability or automatic reconciliation feature. No independent clinical or legal review is claimed.

  • MoHFW: Consent and security in ABDM, 5 August 2025

    Official government explanation accessed 25 October 2026. Supports consent before ABDM health-data exchange; the offline queue scenarios are original evaluation examples, not a claim that ABDM exchanges work offline.

  • IETF RFC 9110: HTTP semantics, section 9.2.2

    Internet standard published June 2022, relevant retry semantics accessed 25 October 2026. Transport retries do not independently prove one clinical action or one billable processing job.

  • MDN: Browser storage quotas and eviction

    Primary browser documentation accessed 25 October 2026. Explains best-effort and persistent storage and private-session limits. It does not verify any vendor's recovery or backup implementation.

Try the workflow
with Bractis.

Use a fictional six-item queue to review Bractis's actual offline and external-connection scope, including an uncertain acknowledgement, correction and unresolved permission.

Request a clinic demo