Skip to content

AI uses for records

Laboratory administrative workflow dataset: grade the run-or-hold call

By SourceX Editorial · Reviewed by Noah Loul ·

9 sections

Short answer

A laboratory administrative workflow dataset should capture the call made when an order is clinically complete but something administrative is still open: a missing authorization, an unsigned requisition, an absent coverage notice or an unbooked collection. Label each case with one of six dispositions, from run and release to refer, and grade it against two clocks frozen at decision time: the specimen's stability limit and the usual wait for paperwork.

Call to transcript

Laboratory administrative

This page covers laboratory administrative workflow dataset: grade the run-or-hold call.Illustrative

Key takeaways

  • The decision is administrative: authorization, scheduling and documentation, never what a result means or whether the test was the right one.
  • Six dispositions include two deferrals, run now and hold the claim, and hold the specimen and ask, because waiting is often the right answer.
  • The specimen's stability limit against the usual paperwork wait decides most cases; freeze both at decision time.
  • In SENAITE's public workflow, no state marks a hold and 31 of 33 field-edit permissions stay open after receipt, so cases must be rebuilt from the audit trail.
  • Hard negatives run both ways: an unsigned Medicare requisition is neither an automatic cancel nor an automatic run.

What decision does a laboratory administrative workflow dataset capture?

It captures the call a laboratory's intake or billing staff make when an order is clinically complete, with tests, specimen and ordering provider all present, but something administrative is still open: an authorization not on file, a requisition without a signature, a coverage notice that was never signed, a collection that has not been booked. The label is what the laboratory should do next, and the outcome is whether that choice let the test run and the claim stand.

The answer is often conditional, partial or deferred. Run now and hold the claim, hold the specimen and ask for the missing item, request a new collection, or pass the question to the laboratory director: each can be right, and the specimen's own clock usually decides which. A blood sample that must be processed within hours cannot wait for a fax that takes days, while a stable sample can.

The dataset stays on the administrative side of the line. It records authorization, scheduling and documentation, and leaves out what a result means, whether the ordered test was the right one and anything a clinician would decide. A team building it needs dated cases with the order as it arrived, the open items, the specimen's stability limit, the call made and what followed, adjudicated by someone who knows laboratory billing and compliance, not by a model's guess.

Which records hold the run-or-hold call

Six record families hold the call, spread across the laboratory information system, the billing system and the communication logs. No single export contains a case.

The rules behind them are specific. Under the CLIA regulations, 42 CFR 493.1241, a laboratory must have a written or electronic request for patient testing from an authorized person; it may accept oral requests only if it solicits a written or electronic authorization within 30 days and keeps the authorization or a record of its efforts; and a patient's chart can serve as the requisition if it is available to the laboratory at the time of testing.

Medicare adds a documentation twist that shapes many labels. CMS's Medicare Learning Network says it does not require a physician's signature on orders for clinical diagnostic tests paid under the clinical laboratory fee schedule, provided the intent to order is documented in the medical record. Its laboratory documentation guidance adds that unsigned requisitions alone do not support that intent and that attestation statements are not accepted for them. The record families:

Which records hold the run-or-hold call
RecordWhere it usually livesFields the decision needsLeave out
Requisition or electronic orderOrder entry, interface messages, scanned paperTests ordered, channel (electronic, paper, phone), collection date and time, signature or authentication flag, ordering rolePatient identity, free-text clinical history
Status history and audit trailLaboratory information systemEvery state change with time and the actor's role, hold reasons, field edits with before and after valuesStaff names; keep roles
Specimen receiptReceiving log, specimen recordSpecimen type, received time, condition, stability limit for the test, storageResult values
Request and follow-up logHolds queue, fax, portal and phone notesWhat was requested, from which role, when, and when it arrivedMessage bodies carrying clinical detail
Coverage documentsBilling system, document storeCoverage notice status and its date against collection, authorization match by test and dates, eligibility check dateMember and policy numbers
Billing outcomeClaims and remittance recordsClaim status, denial reason codes, rebill or write-offAmounts, unless the task is about them

What one open-source LIMS records between order and result

A public laboratory system shows why a case has to be rebuilt from history rather than read from a final row. SENAITE, which describes itself as an open-source laboratory information management system for enterprise environments, publishes its sample workflow as a configuration file, and counting it at commit e85b04a (24 October 2026) gives three findings that carry over to any licensor.

First, the workflow has no place for an administrative hold. It defines 14 sample states and 27 transitions, 13 of the states reachable from Registered, and none of the 14 is named for a hold, a query or a pending item. Four states come before the sample is received on the forward path: Registered, To be sampled, Scheduled sampling and Sample due. An order waiting for paperwork simply sits in one of them, and the reason, if anyone writes it down, goes into a free-text Remarks field that stays editable through Verified.

Second, administrative facts stay editable after the decision they would explain. While a sample is Received, 31 of the 33 field-edit permissions in the file are open (acquired or granted to at least one role), including client order number, contact, priority, sampling date, date sampled and invoice exclusion; at Verified only 4 remain, at Published 1, and in the five closed states none. A dataset built from current field values can therefore show a value entered after the call it is meant to teach. SENAITE keeps an audit log for that history, but its interface warns when the global audit log has been disabled in setup, so a buyer has to ask whether it was on for the whole period.

Third, the gates the system enforces are thin. The guard on the receive transition checks only that a sampling date is set; sampling needs a sampling date and a sampler, or a user holding the sampler role; rejection is available only if the rejection workflow is enabled. The role map limits receiving a sample to four roles, and lets the record's owner, as well as laboratory roles, cancel it. Everything else a laboratory treats as a prerequisite, authorization and coverage included, lives outside these checks. The count by state group:

What one open-source LIMS records between order and result
State groupStatesField-edit permissions open, of 33What it means for the dataset
Before receiptRegistered, To be sampled, Scheduled sampling, Sample due29 to 30Holds leave no state of their own; reasons sit in Remarks or outside the system
Received and awaiting verificationReceived, To be verified31Order number, contact, priority and dates can still change: rebuild from the audit trail
VerifiedVerified4Remarks and interpretation fields stay open; administrative fields close
PublishedPublished1Only the publication specification can change
ClosedRejected, Invalid, Cancelled, Dispatched, Disposed0Frozen: the reason must already be on file
Defined but unreachableTo be preservedNo transition leads to itA state in the configuration need not appear in any record
A state diagram of SENAITE's sample workflow: four states before receipt and four in the lab joined by arrows, five closed states in a side column and an unreachable To be preserved state, each reachable state badged with its open field-edit permissions, 29 to 31 before verification, 4 at Verified, 1 at Published and 0 when closed.
Count of the SENAITE sample workflow at commit e85b04a (2026): open field-edit permissions, of 33, by state.

Six dispositions, and how each outcome becomes known

Six dispositions cover the administrative calls; each is graded against what was knowable at decision time and read off records that appear later. The label is the disposition, and the outcome is evidence about whether it was the right one.

Two clocks decide most cases. The specimen clock is the stability limit for the test from collection; the paperwork clock is the time a missing item usually takes to arrive, which a laboratory can estimate from its own request logs. When the paperwork clock is shorter, holding is safe; when it is longer, the choice is between running now and holding the claim, or asking for a new collection. Freeze both clocks in the case as they stood at the decision, so a later arrival time does not leak into the label.

A model trained on these labels is being taught when to wait, not only what to do, so the schema gives deferral its own classes rather than folding it into a failure. The six:

Six dispositions, and how each outcome becomes known
DispositionCorrect when, at decision timeEvidence in the recordOutcome and how it is known
Run and releaseEvery prerequisite the laboratory's policy requires is on fileAuthenticated order, coverage documents present or not neededResult released and claim accepted, from status history and remittance
Run now, hold the claimTesting is authorized, a billing document is missing and the specimen cannot waitStability limit shorter than the usual wait; request loggedDocument arrives and the claim goes out, or the policy's fallback applies, from billing history
Hold the specimen and askThe missing item normally arrives inside the stability limitRequest time, owner and stability limitReleased and run inside the limit, or expired, from hold release time against the limit
Request a new collectionThe specimen cannot serve the order as written, or the order must changeReceipt condition, collection time, the change neededA linked recollection accession, or none
Cancel with noticeNo authorized request can be obtained, or the ordering office withdrawsRequest attempts and the responseCancelled status and a logged notice to the ordering office
Refer to the laboratory directorThe question is clinical or beyond administrative authorityReferral reason category onlyThe administrative label ends at the referral; the director's decision is out of scope

Hard negatives: orders that look ready and are not, and the reverse

Hard negatives are cases where the obvious answer is wrong. They are what separates a dataset that teaches judgment from one that teaches a checklist, and a laboratory's history contains them in both directions. Seven worth sampling deliberately:

  • The unsigned requisition that should not be cancelled. For Medicare clinical laboratory tests a missing signature is not fatal if intent is documented; Medicare's Program Integrity Manual accepts an unsigned requisition listing the tests together with an authenticated medical record supporting the intent to order. The right call is to request that documentation, not to cancel.
  • The unsigned requisition that cannot simply be run and billed. The reverse case: CMS says an unsigned requisition alone does not support intent and that attestation statements are not accepted, so running and releasing without the record request is the wrong label.
  • The phone order that ran cleanly. The test was performed and resulted, so the case looks closed, but no written or electronic authorization was solicited within 30 days of the oral request; the label should flag the missing follow-up.
  • The frequency-limited test with no coverage notice. CMS's instructions for the Advance Beneficiary Notice of Noncoverage use reasons such as Medicare not paying for a test as often as ordered; a case where that applied and no notice was obtained is not a clean run and release.
  • The prudent hold that destroyed the order. Staff held a short-stability specimen for a document that arrived after the limit; the hold looked careful and cost a recollection. Run now and hold the claim, or recollect at once, was the better call.
  • The authorization that matches the wrong thing. An authorization is on file, but for a different test or date range; reading only the presence flag produces a false run and release.
  • The reflex test outside the notice. A follow-on test was added after a coverage notice that did not mention further testing; CMS's form instructions let the notifier note that further testing is possible, and a case where nobody did is a different label from one where they did.

The case packet and the domain review checklist

Each case should travel as one packet frozen at decision time, and each label should pass a domain review before it enters training or evaluation. The packet holds the order as received, the open items, both clocks, the actor's role, the disposition and the later outcome records, with the decision timestamp marking which fields the model may see.

The review is done by someone who knows laboratory billing and compliance for the laboratory's payers and period, with a second reviewer on a sample of cases and on every hard negative. The checklist:

Checklist · 8 items

  1. Confirm the case is administrative: no result value, interpretation or test-selection question is needed to label it; otherwise label it Refer and stop.
  2. Confirm the policy version: the laboratory's own hold and billing policy, and the payer rules, in force on the decision date.
  3. Rebuild the order's state at the decision time from the audit trail, not from current field values.
  4. Check both clocks: the stability limit for the specific test and specimen type, and the wait time the laboratory's logs support.
  5. Check authorization against the test and dates it covers, not just its presence.
  6. Check coverage notice timing against collection, and whether added tests were covered by it.
  7. Assign one of the six dispositions and record the evidence pointer for each open item.
  8. Record disagreement between reviewers and resolve it before the case is used; keep unresolved cases out of evaluation sets.

0 of 8 ticked. Tick each item as you go.

What a laboratory's records can and cannot show a buyer

A laboratory's records can show the order as it arrived, every status change with its time and role, the requests it sent and when replies came, how long specimens waited and how claims ended. They cannot show what sits in the ordering office's chart, calls nobody logged, or the clinical reasons behind an order, and those gaps limit which labels can be adjudicated at all.

Rights come before any of this. The laboratory holds these records, but they describe patients and ordering providers, and some may fall under agreements with client practices, hospitals or payers; whether the laboratory can license them, and under which terms, is a question for its counsel on its own facts. Privacy rules shape the content too. HHS describes two ways to de-identify health information under HIPAA, expert determination and the safe harbor method, and the safe harbor removes date elements tied to an individual except the year. A dataset whose labels depend on hours between collection and receipt therefore needs relative times inside each case, or an expert determination that supports more date detail.

System configuration is the last limit. If the audit log was off for part of the period, or holds were tracked in a spreadsheet beside the system, the decision-time state cannot be rebuilt for those cases, and they belong outside the training set rather than inside it with guessed timestamps.

Requesting a laboratory administrative decision set

A buyer asks for this data by describing the task, the cases and the test it will use to accept them. SourceX sources datasets when a buyer asks rather than keeping them in stock, and its request form collects the domain, source systems, volume, format, geography, language, historical timeframe, timeline, licensing needs and de-identification requirements.

For a laboratory administrative decision set, state the dispositions you need and the share of hard negatives, the laboratory type (hospital outreach, reference or independent clinical laboratory), the systems the cases must span, whether payer outcome records are required, the decision-time rule for what a model may see, and the acceptance test: for example, that a second domain reviewer agrees with the recorded disposition on a stated share of a blind sample. Say plainly that clinical interpretation is out of scope, so the supplier's privacy and rights review can exclude it from the start.

Nothing in this article means such a set exists today at any laboratory or through SourceX. It describes what a usable one contains, so a request can be specific enough to source, review and test.

Methods note: how the SENAITE workflow was counted

The count was made on 2026-10-28 from a shallow clone of senaite/senaite.core, branch 2.x, at commit e85b04afc8e134ec22448c5f4f3cdf8990ecf77f, by a short script (xs0817_senaite_count.py) that parses two files, the sample workflow definition and the role map, as XML; it runs no SENAITE code, uses no randomness and strips a UTF-8 byte-order mark if one is present (neither file has one). Its output holds state, transition, field and role names and counts, and no personal data.

Rules as the script applies them. A state is a state element and a transition a transition element. Reachable states come from a breadth-first search from the initial state, Registered, over each state's exit transitions to the transition's new state; transitions with an empty new state add no edge. The forward path to receipt starts at Registered and follows only the To be sampled, No sampling workflow, Schedule sampling and Sample transitions, stopping at states that list Receive among their exits. A state counts as hold-named if its identifier, with underscores read as spaces, or its title matches, as a whole word and ignoring case, hold, held, pending, pend, query, queried, await, awaiting, wait, waiting, incomplete, missing or info; none does. A field-edit permission is a permission map whose name contains Field: Edit, which leaves out maps such as Edit Field Results that govern field analyses rather than sample fields; it counts as open in a state when the map is acquired or names at least one role. The script also reports which of ten administrative fields it names are open in Received and To be verified (all ten are), the states where Remarks is open, and the roles listed for five transition permissions in the role map, without following acquisition.

Read by hand rather than counted: the guard functions in guards.py, the README and the audit log notice. Not measured: any laboratory's records, how often holds occur, how many SENAITE installations enable the sampling, rejection or audit log features, and how other laboratory systems model the same steps. The label schema, hard negatives and checklist were drawn up for this article on 2026-10-28 and describe no real laboratory's cases.

Frequently asked questions

What is a laboratory administrative workflow dataset?

A set of dated laboratory cases in which the order was clinically complete but an administrative item, such as authorization, a signature, a coverage notice or a booked collection, was open. Each case records the open items, the specimen's stability limit, the call staff made and the records that show how it turned out.

Does a lab need a physician's signature on a requisition?

For Medicare clinical diagnostic tests paid under the clinical laboratory fee schedule, CMS says a signature is not required if the intent to order is documented in the medical record, although an unsigned requisition alone does not show that intent. Other payers and state rules can differ, so a dataset should record which rule applied to each case.

Why can't the dataset be built from a LIMS export of final values?

Because administrative fields often stay editable after the decision. In the SENAITE workflow counted here, 31 of 33 field-edit permissions remain open after a sample is received, so a final value may postdate the call. The audit trail is the source for decision-time state.

What should be excluded from a lab administrative dataset?

Result values, interpretations, test-selection questions, free-text clinical history and patient identifiers. Coded diagnosis detail can often be replaced by a flag saying whether the coverage rule was met; cases that need clinical judgment are labeled as referrals and go no further.

How should a model be scored on these cases?

Against the adjudicated disposition, with deferral counted as a correct answer when the clocks and records support it. Weight hard negatives heavily, report errors separately for holds that should have been runs and runs that should have been holds, and keep cases with unresolved reviewer disagreement out of the evaluation set.

Sources

  • 42 CFR 493.1241 requires a written or electronic test request from an authorized person, allows oral requests only if written or electronic authorization is solicited within 30 days and kept or the effort documented, and lets the chart serve as the requisition if available at testing. Source
  • CMS's Medicare Learning Network says a signature is not required on orders for clinical diagnostic tests paid under the clinical laboratory fee schedule, but the intent to order must be documented in the medical record. Source
  • CMS's laboratory documentation guidance says unsigned requisitions alone do not support physician intent to order and attestation statements are not accepted for unsigned orders or requisitions. Source
  • Medicare's Program Integrity Manual, chapter 6, treats order requirements as met by a signed order or requisition, or by an unsigned order or requisition listing the tests together with an authenticated medical record supporting the intent to order. Source
  • CMS's ABN form instructions include, as an example reason, that Medicare does not pay for a test as often as ordered, and allow the notifier to indicate that further testing may be needed. Source
  • HHS describes expert determination and safe harbor as the two methods of de-identifying protected health information, with the safe harbor removing date elements related to an individual except the year. Source
  • SENAITE describes SENAITE.CORE as an open-source laboratory information management system for enterprise environments. Source
  • SENAITE's sample workflow definition at commit e85b04a defines 14 states and 27 transitions with per-state field-edit permissions; counted for this article. Source
  • SENAITE's sample guards allow receive only when a sampling date is set, sample when a sampling date and sampler are set or the user holds the sampler role, and reject only when the rejection workflow is enabled. Source
  • SENAITE's role map grants the Receive Sample transition to four laboratory roles and the Cancel transition to laboratory roles and the record's owner. Source
  • SENAITE's interface shows a warning when the global audit log is disabled and points to setup to enable it again. Source
  • SourceX's request page says datasets are sourced on request, not held in stock, and its form collects domain, source systems, volume, format, geography, language, timeframe, timeline, licensing and de-identification requirements. Source

Tell us what data you need

Share scope, volume, format, timing and licensing needs. Datasets are sourced on request.

See if your company qualifies