Move Pathology Add-On Tests Forward With Complete Information and Documented Authorization

It decides no clinical or payer question on its own. It creates an administrative evidence packet: the missing fact, recommendation or order evidence, configured requester authority, material status, payer path, current source, response state, confidence, and next owner. Qualified people decide clinical need, order acceptance, material use, payer action, patient communication, and billing.

No. A pathologist or other authorized clinician retains test-selection authority. The workflow can map an explicitly documented recommendation or order to the controlled catalog, show which evidence is present, and route the required administrative steps. It cannot expand a panel, infer a diagnosis, decide that a test is medically appropriate, or optimize documentation for coverage.

No replacement is assumed. The design reads approved fields and evidence from the LIS, EHR, document, material, payer, communication, and queue systems already in use, then writes back only agreed statuses and references. Interface availability, vendor effort, retries, downtime, and write-back ownership are scoped during implementation rather than advertised as universal integrations.

Each receives its own status and evidence. A pathologist's recommendation records clinical intent to consider an add-on. The organization's configured order path records who may authorize the exact test and what evidence is accepted. A payer authorization or determination records the plan's administrative response. A patient notice or financial choice is a separate human-led step. One never silently completes another.

No universal rule should be claimed. CMS says signatures are not always required for clinical diagnostic laboratory tests or physician pathology services when other evidence supports intent, while order and medical-necessity documentation still matter. Prior-authorization and coverage requirements depend on the current applicable service, source, jurisdiction, and payer path. Staff must verify the real case instead of applying a blanket rule.

It surfaces the available material category, location, recut or send-out state, outside ownership or return obligation, and competing requests before an administrative release. It does not calculate clinical sufficiency or direct a cut, depletion, shipment, return, or disposal. Pathology, histology, molecular, quality, and records staff make those decisions under local procedure.

Only through approved channels, templates, recipients, credentials, and scopes. A focused missing-information or authorization packet can be prepared and, where authorized, transmitted with delivery and response tracking. Sensitive outreach, clinical questions, attestations, patient-facing content, and ambiguous replies remain human reviewed. Portal credentials and response values are excluded from source code and marketing analytics.

The payer agent first verifies line of business and plan context, then loads only the applicable current source set. Original Medicare uses its order, coverage, and limited ABN framework. Medicare Advantage uses the plan's authorization, organization-determination, and notice process. Medicaid follows the state and delivery system, including third-party liability. Commercial review follows the exact plan, contract, network, state, and federal context.

The result remains unknown. The ledger records the failed source, retrieval time, affected decision, retry state, and manual fallback owner. It does not reuse a retired document, choose the nearest plan, or transform no response into approval. Qualified authorization or payer-policy staff verify the controlling source through an approved alternate channel before the case advances.

Run a shadow and parallel comparison on a representative, properly controlled sample while the established process remains authoritative. Review false clears, false holds, extraction errors, wrong source or payer matches, authority mistakes, missed material constraints, communication defects, and usability by role. Set risk-specific acceptance thresholds, sign-offs, fallback, and rollback instead of relying on one aggregate accuracy percentage.

The approved offer is $0 implementation fee and $0 customization charges for the agreed scope. Software subscription and usage charges are separate. Third-party fees, payer or interface vendor work, non-standard migration, reference-laboratory work, and services outside scope may be priced separately in the written order form. Buyers should review total cost, usage assumptions, renewal, and exit terms.

Use aggregate counts and durations by non-PHI reason category: missing operative report, missing site or laterality, order-authority review, provider response, payer response, policy no-match, material exception, queue age, override, send-out delay, rework, and downstream disposition. Marketing analytics should receive only route, specialty, workflow, content cluster, and CTA metadata—not case, patient, order, diagnosis, payer-response, or free-text values.

Build the library around the ways anatomic pathology work actually becomes ambiguous, not only around clean electronic requisitions. Intake examples should include multipart specimens, split cases, addended requisitions, handwritten abbreviations, inconsistent laterality, a renamed procedure, a missing operative note, conflicting specimen-part descriptions, duplicate submissions, cancelled requests, and outside consultations that arrive with a packing list but incomplete ordering context. Material examples should cover paraffin blocks, glass slides, unstained sections, serial levels, recuts, curls, punches, cytology cell blocks, decalcified tissue, a nearly exhausted block, material already loaned to another facility, a custodial laboratory's return deadline, and competing requests for consultation or molecular preparation. Add-on examples should span a single special stain, an immunohistochemistry request, a configured panel, in-situ hybridization, a molecular send-out, a post-sign-out addendum, a verbal request followed by written evidence, an amended or withdrawn order, and a test name that does not map cleanly to the local catalog. Source-failure examples should include a rotated scan, truncated fax, missing page, stale interface result, late document, duplicate attachment, unavailable archive, portal timeout, changed payer response, and provider reply that answers a different question. Every example needs an expected reason code, permitted next action, prohibited action, responsible role, evidence requirement, and recovery path. Clinical leaders should also identify work that stays outside the first release—such as urgent intraoperative decisions, disputed identity, forensic or legal-hold material, unusual chain-of-custody requests, research-only handling, or any case whose clinical risk exceeds the team's validated scope. The purpose is not to teach software how to make those decisions; it is to prove that the software recognizes the boundary, stops safely, and gives the correct specialist enough context to take over. The readout should also follow a material-centered trace from gross description and cassette map through processing, embedding, microtomy, staining, whole-slide imaging, archive, send-out packing, courier transfer, consultant receipt, and return. The pilot does not need to automate each bench step, but it should declare which event markers are authoritative and which are deliberately outside scope. Stress cases can include re-embedding, face trimming, deeper levels, ribbon loss, folds, chatter, floaters, contamination, cracked glass, failed controls, restaining, coverslip damage, scanner rescans, rejected images, missing thumbnails, and amended material counts. These are not decisions for the agent. They reveal whether an interface incorrectly hides a physical exception or announces readiness while technologists are still resolving it. For molecular handoffs, examine how the system carries—but never clinically judges—fixative, cold-ischemia interval, decalcifier exposure, block age, section thickness, unstained-slide count, tumor-content statement, macrodissection request, extraction state, nucleic-acid quality flag, germline comparator, leftover material, shipping-kit status, and reference-laboratory rejection. Use local terminology and have histotechnology, cytology, molecular diagnostics, and pathology sign off on the crosswalk. Sampling should reflect the laboratory's subspecialty mix without teaching the agent to practice pathology. A validation set might span dermatopathology shaves and punches, gastrointestinal biopsies with level requests, breast excisions with orientation sutures and inked margins, genitourinary core sequences, thoracic resections, gynecologic curettage, head-and-neck specimens, decalcified bone, soft-tissue masses, neuropathology frozen material, transplant consultations, placental examinations, and pediatric referrals. The administrative patterns differ: diagrams, margin keys, cassette legends, source-facility conventions, ancillary-study reservations, biorepository carve-outs, tumor-board deadlines, and return-loan terms. Subspecialty reviewers should decide which patterns enter scope and write the expected stop behavior for everything else.

Require a reproducible acceptance matrix, not a polished demonstration or one overall accuracy claim. Stratify the readout by intake channel, document type, specimen complexity, add-on category, payer path, material state, source quality, and risk tier so a strong result on clean interfaces cannot hide failures on faxed operative reports, outside blocks, or payer exceptions. Show extraction errors, false matches, false clears, false holds, duplicate outreach, missed conflicts, wrong-recipient attempts, stale-source use, premature write-backs, and material-release stops separately. For every defect, show the original evidence, expected behavior, observed behavior, severity, owner, correction, regression test, and whether the manual fallback worked. Ask operations to report first-touch resolution, reopen rate, response completeness, handoff count, backlog-age distribution, escalation timeliness, and time spent searching, contacting, checking status, and reconstructing history. Ask pathology and quality leaders to review whether recommendations stayed distinct from orders, whether clinical interpretation remained human, whether specimen and material uncertainty stopped correctly, and whether an addendum, amendment, cancellation, or competing-use signal remained visible. Ask revenue-cycle and authorization leaders to verify plan matching, effective dates, jurisdiction, source provenance, request contents, response classification, and separation of authorization from payment. Ask security and privacy teams for the interface inventory, field-level data flow, role permissions, least-privilege design, credential vault boundary, audit export, retention behavior, break-glass procedure, downtime queue, retry controls, and restoration test. Finally, require finance to replace the illustrative cases, minutes, and loaded rate with measured inputs and subtract subscription, usage, interface, training, governance, and internal support costs. The organization—not the vendor—sets risk-specific thresholds and decides whether the pilot is ready, needs repair, should remain in shadow mode, or should stop. Document testing should include TIFF, PDF/A, JPEG, word-processing exports, scanned color images, duplex blanks, rotated pages, skew, bleed-through, stamps, handwriting, fax noise, clipped margins, corrupted files, password-protected attachments, mixed-language notes, and duplicate packets in which only one page changed. Confirm whether optical character recognition retains page coordinates, distinguishes printed headers from handwritten additions, preserves strikethroughs and checkboxes, and allows a reviewer to open the original. Catalog testing should include client mnemonics, local aliases, legacy names, retired entries, bundled versus standalone services, reference-lab synonyms, hyphenation changes, and mappings that differ across facilities. None of those variations should be silently normalized when meaning is uncertain. Interface resilience deserves its own drill. Simulate an interface-feed outage, negative acknowledgment, duplicate event, delayed replay, out-of-order message, clock drift, time-zone conversion, daylight-saving change, schema revision, absent field, explicit null, truncated narrative, invalid character, code-set update, retry storm, rate throttling, dead-letter entry, poison message, partial write, and downstream rejection. Require idempotent write behavior so a replay cannot create a second task, plus a reconciliation report that identifies what arrived, what was accepted, what was refused, and what needs human repair. Confirm that a rollback checkpoint does not erase reviewer history or material custody. Measure recovery-point and recovery-time behavior in the organization's own environment rather than accepting a generic uptime statement. Security evidence should identify single sign-on, multifactor authentication, user provisioning, role removal, service accounts, credential rotation, secret vaulting, transport encryption, storage encryption, tenant isolation, field masking, attachment malware scanning, export controls, session timeout, break-glass access, and administrative audit review. Ask who can view an operative report, download an attachment, change a mapping, approve a release rule, replay a message, or export a ledger. Examine separation of duties between configuration, clinical approval, production support, and audit. Procurement should request the applicable architecture diagram, data-flow inventory, subcontractor list, incident-response path, vulnerability-management process, backup restoration evidence, and deletion verification without assuming any certification that has not been produced and independently evaluated. Finally, review human factors at the gross room, histology bench, pathologist worklist, molecular desk, provider-relations queue, authorization desk, patient-access team, archive, loading dock, and client-services function. Test whether labels are understandable without hovering, whether the reason and next action appear together, whether color is never the only warning, whether long specimen descriptions wrap, whether keyboard users can reach disclosures, and whether urgent and routine work are distinguishable without sounding an alarm on every case. Ask frontline staff to narrate what they believe happened, what they would do next, and which person has authority. A successful readout shows shared understanding and safe recovery, not merely fast clicking.