Cancer Registry Abstraction Engine
Done-for-you, ODS/CTR-signed cancer registry abstracts — sold per completed, NAACCR-compliant case behind an accuracy-and-timeliness guarantee, not abstraction software the hospital operates

1.Title

Cancer Registry Abstraction Engine — an AI-native, registrar-supervised oncology-data service that delivers completed, audit-ready cancer registry abstracts (casefinding → abstraction → coding → entry → QC), not abstraction software the hospital drives. The customer receives finished abstracts inside their own registry system, each reviewed and signed off by an Oncology Data Specialist / Certified Tumor Registrar (ODS-CTR), ready for the state central cancer registry and the National Cancer Database (NCDB).

This is the cancer-registry data-abstraction business (clinical & regulatory data abstraction — catalog item 86 — in its highest-value, most-mandated vertical). It is explicitly distinct from the prior specialty-surgical-coding-engine (facility/professional billing coding for reimbursement), the risk-adjustment-integrity-engine (HCC/payer risk capture), and the clinical-research-coverage-analysis-engine runs. We do not bill payers; we produce the surveillance abstract that hospitals are legally required to report.

2.Final decision: Blueprint

3.Executive summary

Every hospital that diagnoses or treats cancer is legally required to report each case to its state central cancer registry — in nearly every state within six months (180 days) of diagnosis — and CoC-accredited programs must additionally submit to the NCDB Verified [13][14][4]. Producing each report is skilled work: a Certified Tumor Registrar reads the path report, operative notes, imaging, and clinical record, then codes the primary site and histology (ICD-O-3), stage (AJCC / Summary Stage), and dozens of site-specific data items, and enters a structured abstract. The work has historically depended on a scarce, aging, credentialed workforce — and that workforce is shrinking.

The work is already outsourced at scale: a stable industry of registry staffing/abstraction vendors (Savista — now the largest U.S. employer of Oncology Data Specialists after acquiring ONCO; Direct Difference; American Data Network; Clinical Registry Solutions; Carta Healthcare; In Record Time) sells abstraction to hospitals that cannot hire or retain CTRs Verified [3][6][7]. Pricing is already per-case; published cost-per-case for registry operations averages $60.77 (range $15.96–$233.48), with abstraction the single largest cost component Verified [5]. The incumbent unit cost is a registrar's hours — exactly the cost an AI-native casefinding-and-pre-abstraction engine collapses.

4.Thesis

Cancer-registry abstraction is an excellent AI-native service unit because the atomic object — one reportable case → one structured abstract — is a discrete, evidence-grounded, rule-bound task with a deterministic-heavy core and a small, namable judgment residual. Given the source documents, the engine: identifies whether the case is reportable (casefinding from path, cytology, and pharmacy/radiology feeds); consolidates the record; extracts and codes primary site, laterality, histology/behavior (ICD-O-3), grade, stage (AJCC TNM and Summary Stage), and the site-specific data items (SSDIs) and treatment fields per the NAACCR/SEER/STORE standards; and drafts the full abstract pre-populated in the registry software. An estimated ~70–85% of that work is casefinding, extraction, coding, and entry — precisely where frontier clinical NLP excels Inferred from the 93.9% exact-match result and commercial casefinding accuracy of 99% sensitivity/98% specificity [10][11]. The residual is genuine credentialed judgment — ambiguous primary site, multiple-primary/histology rules, stage assignment under edge cases, conflicting documents — concentrated at a chokepoint owned and signed by an ODS/CTR-credentialed registrar.

Because the work is already outsourced, because the buyer cares about the finished, NCDB-acceptable abstract (not the tool), and because the deliverable is reviewed and stood behind by a credentialed registrar, trust burden is low and budget already exists Verified [3][18]. Pricing is naturally per case (tiered by complexity / analytic vs. non-analytic) — never hourly. As models improve at clinical extraction, the autonomous share rises and cost-per-case falls while the registrar concentrates on a shrinking residual (Sam Altman test: strong pass). The durable asset is the abstraction OS: casefinding tuned per source system, the maintained standards/coding ruleset (which changes annually), QC that drives accuracy toward NAACCR thresholds, the growing private library of corrected abstracts, and the credentialed sign-off that makes each case acceptable to the state registry, the NCDB Call for Data, and a CoC survey.

5.Discovery rationale

This run independently scanned the AI-native services terrain — healthcare administration, clinical/regulatory data abstraction, insurance operations, legal/IP, and tax/financial compliance — screening against the six gates and the evidence threshold, and explicitly checking the 60 prior blueprints in the manifest for overlap. Candidate niches that were strong but adjacent to prior runs (medical-record review/chronology near the prior PI demand-package engine; CDI near the prior specialty-coding engine; Medicaid enrollment near the prior provider-enrollment engine) were down-weighted to preserve portfolio diversity and avoid duplication.

Discovery converged on cancer-registry abstraction for three reasons. First, demand is active, recurring, structurally rising, and legally compelled: 2.04M new U.S. cancer cases in 2025, every one reportable within ~6 months, against a shrinking, aging CTR workforce that has already produced 12-month backlogs Verified [1][2]. Second, it is an excellent AI-native fit: casefinding and field-level extraction from pathology and clinical text is exactly what clinical NLP and long-context LLMs do best, evidenced by a 2025 published 93.9% exact-match across 196 fields and commercial casefinding at 99% sensitivity Verified [10][11]. Third, the budget already exists and is per-case priced — a mature registry staffing/outsourcing industry exists precisely because hospitals cannot staff this internally — while the work itself is still done by scarce credentialed registrars. The scoping is deliberate: this is surveillance abstraction, distinct from billing coding, payer risk capture, and research coverage analysis already covered in the portfolio.

6.Candidate comparison

Six AI-native service candidates were generated and scored (1–5, higher better) across active-demand evidence, gross-margin potential, MVP narrowness, licensing/regulatory safety, and whitespace vs. existing tools/vendors and the 60 prior blueprints. None of the prior blueprints duplicated the winner.

The winner leads on the combination of legally-compelled active demand, MVP clarity, and licensing safety. Its only soft spots — a margin that starts thin (registrar review minutes are real) and an established outsourced category — are exactly the seams this blueprint is built around: an AI cost structure that drives review minutes down case-over-case, plus a credentialed-accuracy guarantee that legacy staffing pools and self-serve casefinding tools cannot match simultaneously.

7.CODE validation

C — Consumer / buyer trend

Three trends collide. (1) Rising case volume: U.S. cancer incidence keeps climbing (2.04M cases in 2025, ~5,600/day), so the abstraction backlog grows mechanically Verified [1]. (2) A shrinking, aging CTR workforce: more registrars are retiring than entering; the credential now requires an associate degree; large academic systems report 12-month backlogs and unfilled positions Verified [2]. (3) AI casefinding/abstraction has crossed the credibility threshold: CDC's NPCR runs NLP on pathology reports; NAACCR teaches AI/NLP in the registry field; commercial engines (Inspirata E-Path Plus) and a 2025 medRxiv model report high accuracy Verified [10][11][12][19]. More cases, fewer registrars, AI now trusted: the work flows to whoever can deliver compliant abstracts cheapest and fastest.

O — Opportunity

The underserved seam: mid-size and community CoC hospitals and their registry managers who must meet the 6-month deadline and CoC standards but cannot hire/retain CTRs, and who find pure staffing vendors expensive (a registrar's loaded hours, simply relocated) and self-serve casefinding tools incomplete (they still need a credentialed human to abstract, code, and sign). A service that fuses AI casefinding/pre-abstraction with credentialed review and a compliance guarantee owns the middle: it clears backlogs fast and keeps the program survey-ready.

D — Demand

Buyers are visibly spending and procuring: an established registry-staffing/outsourcing industry with published per-case service pages; vendor marketing explicitly framed around "filling the CTR gap" and clearing backlogs; the VA and other government programs procuring cancer-registry abstraction services on contract; and NCRA/NAACCR community discussion of the shortage and of AI tools Verified [3][6][7][18][22].

E — Economic sizing

Anchor figures: ~2.04M new U.S. cases/yr [1]; cost-per-case for registry operations averages ~$60.77 (abstraction the largest slice) [5]. A conservative serviceable wedge — ~1,500 CoC programs × ~1,000–4,000 analytic cases/yr × a ~$25–$45 per-abstract service fee ≈ $40–$270M of addressable abstraction spend in the accredited-hospital segment alone, before non-CoC hospitals, physician offices, pathology labs, and state/central registry contracts Inferred [4][5]. The cancer-registry software market (a proxy for adjacent spend) is ~$99M in 2025 growing ~11% CAGR to ~$168M by 2030 Verified (analyst est.) [16]. A small share of a legally-compelled, recurring market supports a meaningful, defensible business. Figures are illustrative, not a forecast.

8.Rubric scorecard (six gates)

Total: 28 / 30. Anti-commoditization check (Gate 6 corollary): even if a general model auto-codes a clean path report, the durable wins are casefinding tuned to each hospital's messy feeds, the ODS/CTR accountability and NCDB/CoC-grade QC, the maintained annually-changing standards ruleset (STORE, SEER, AJCC, SSDI), the private library of corrected abstracts, the timeliness-and-accuracy guarantee, and HIPAA-compliant managed operations — none of which a raw model or a self-serve casefinding tool provides.

9.Target buyer

10.Jobs-to-be-Done

• Functional: "Turn these reportable cases into complete, correctly-coded abstracts in our registry system, on time, so we meet the state deadline and NCDB submission."
• Risk/compliance: "Keep us survey-ready and out of a timeliness/completeness deficiency — don't let a backlog or a coding error threaten our CoC accreditation."
• Financial: "Get this done without paying to recruit, train, and retain a CTR I can't even find — at a predictable cost per case."
• Emotional: "Take the dread out of the growing queue and the looming survey — give me a credentialed partner I trust to close it."

11.Painful problem

The cancer abstract is the load-bearing record under cancer surveillance, accreditation, and research — and it is mandated, deadline-bound, and labor-scarce. Producing one means reading the full record and coding ~150–300 defined data items to controlled vocabularies under rules that change every year. Nearly every state requires reporting within ~6 months (180 days) of diagnosis, and CoC programs must additionally feed the NCDB Verified [13][14][4]. But the workforce that does this is shrinking and aging — retirements outpace new entrants, the credential now requires an associate degree, and the result is unfilled positions and 12-month abstraction backlogs at large academic centers Verified [2]. The consequences are concrete: a program that falls behind risks state timeliness/completeness deficiencies, a poor CoC survey, rejected or late NCDB submissions, and degraded surveillance and clinical-trial matching data. Hospitals already outsource to cope, but pure staffing simply relocates a registrar's expensive hours; AI now collapses the casefinding-and-extraction labor (a 2025 model coded 196 fields at 93.9% exact-match across ten cancers), making this the moment to industrialize the work as a per-case service Verified [11].

12.The outcome we sell

We do not sell casefinding software, a co-pilot, or a database the registrar operates. The client experiences an expert abstraction service; the AI is the internal production engine. The credentialed registrar is the customer-facing accountable interface, and the abstract lands where the hospital already works — its registry of record — not in a separate tool the client must learn.

13.First one-feature MVP wedge

14.Evidence summary

• Active, recurring, mandated demand — 2.04M new U.S. cases/yr, each reportable within ~6 months Verified [1][13].
• Worsening labor shortage — aging/retiring CTR workforce, unfilled roles, 12-month backlogs Verified [2].
• Already outsourced — mature registry staffing/abstraction vendor industry (Savista/ONCO, Direct Difference, ADN, CRS, Carta, In Record Time) Verified [3][6][7][22].
• Per-case pricing is the norm — cost-per-case ~$60.77 (range $15.96–$233.48), abstraction the largest component Verified [5].
• AI now credible & transformational — CDC NPCR NLP; commercial casefinding 99% sens/98% spec; 2025 model 93.9% exact-match over 196 fields Verified [10][11][12].
• Regulatory pull — CoC accreditation + state mandates + NCDB/NAACCR data standards Verified [4][13].
• Clean compliance boundary — abstraction is credentialed data work under HIPAA, not the practice of medicine Inferred (legal principle).
• Deterministic share ~70–85% — Inferred from casefinding/extraction accuracy results [10][11].
• Unit economics / conversion — per-case COGS, reviewer minutes, and pilot conversion are Inferred; carried as explicit pilot kill-criteria.

15.Claim table (Verified / Inferred / Unverified)

16.Source-claim matrix

17.Market and demand evidence

The opportunity sits on a legally-compelled, recurring base. Volume: 2,041,910 new U.S. cancer cases are projected in 2025 — about 5,600 per day — and incidence is rising Verified [1]. Mandate: nearly every state requires each case to be reported to its central registry within ~6 months of diagnosis, and ~1,400–1,500 CoC-accredited programs must additionally submit to the NCDB Verified [4][13][14]. Supply gap: the CTR/ODS workforce is shrinking and aging, leaving unfilled positions and 12-month backlogs Verified [2]. Spend: registry operations cost ~$60.77/case on average with abstraction the largest slice, and hospitals already pay outside vendors per case to cope Verified [5]. The adjacent cancer-registry software market (~$99M in 2025, ~11% CAGR) is a proxy for the surrounding budget and the direction of AI investment [16]. Demand is non-discretionary and event-driven: every diagnosis creates a deadline, every staffing gap creates a backlog, and every survey/Call-for-Data creates urgency.

18.Active buyer conversations

• Vendor service pages & pricing framing (Savista, Direct Difference, American Data Network, Clinical Registry Solutions, In Record Time) publish per-case/managed-abstraction offers — a category hospitals actively procure Verified [3][6][7][22].
• "Fill the CTR gap" / backlog content (Savista, Extract Systems, Medovent) written directly for registry managers weighing outsource-vs-hire — the buyer's own framing [2][3][20].
• Government procurement — VA/VISN solicitations for cancer-registry abstract services show institutional buyers contracting the work out [18].
• AI-tool roundups & NAACCR/CDC education — registrars actively learning and comparing AI/NLP abstraction, signaling intent to change how the work gets done [12][19][20].
• Job postings for CTR/ODS and "remote cancer registrar" roles at sustained volume — direct evidence of unmet demand and the labor cost being relocated [21].

19.Competitive landscape

Whitespace: the fused middle — AI-native casefinding/pre-abstraction speed and cost and ODS/CTR accountability with an accuracy-and-timeliness guarantee, delivered into the hospital's own registry — is held by neither the staffing vendors (no AI cost curve) nor the software tools (no done-for-you outcome, no sign-off-as-product).

20.Competitor & budget validation

Existing budget source: hospitals already pay registry-staffing/outsourcing vendors per case (or per managed FTE) and license registry software; operations cost ~$60.77/case with abstraction the largest slice [5]. Incumbent alternatives: staffing vendors, AI casefinding tools, registry software, in-house CTRs, offshore keying (above). Why current alternatives are insufficient: staffing relocates expensive scarce hours without an AI cost curve; AI tools shift the work and the accountability back onto the hospital's registrar; software is a system of record, not labor; offshore keying carries quality/HIPAA risk. Why we win: we deliver the AI cost/speed curve and a credentialed-signed, edit-clean, guaranteed abstract in the hospital's own system — the one combination none of the incumbents offer. Not a clone: we are neither a body-shop nor a tool vendor; we are an outcome service whose moat is the abstraction OS + credentialed accountability + guarantee.

21.Pricing evidence & proposed pricing

Evidence: the market already prices per-case; published registry-operations cost-per-case averages ~$60.77 (range $15.96–$233.48), and outsourced vendors quote per-abstract/managed pricing that falls as volume rises Verified [5][22]. We mirror per-completed-abstract pricing (never hourly), positioned at or below incumbent fully-loaded cost while clearing backlogs faster.

Outcome/contingency note: pricing is per completed, QC-passing abstract (legal and standard). We avoid contingency tied to reimbursement or any clinical outcome — abstraction is surveillance data, not billing — which keeps incentives aligned with accuracy, not volume. The accuracy-and-timeliness guarantee (re-work/credit for an in-scope abstract failing NAACCR-standard QC or missing the agreed deadline) is a quality warranty, not a contingent fee.

22.Regulatory & compliance considerations

• Reporting mandates: state law requires reporting each case to the central registry within ~6 months of diagnosis; CoC accreditation requires a compliant registry and NCDB submission; NPCR (PL 102-515) underpins state registries Verified [13][15][4].
• Data standards: abstracts must conform to current NAACCR data standards, SEER coding rules, AJCC staging, and the CoC STORE manual; these change annually, so the engine's ruleset must be versioned and updated each cycle.
• HIPAA: we handle PHI as a Business Associate; a signed BAA, minimum-necessary access, encryption, audit logging, and breach procedures are mandatory. Sensitive matters stay on U.S.-controlled, access-restricted infrastructure.
• Not the practice of medicine: abstraction/coding is credentialed data work (ODS/CTR via NCRA), not clinical decision-making or a state-licensed medical activity; we never diagnose, stage clinically for treatment, or advise care Inferred (legal principle).
• Credentialing & accountability: the signing reviewer is ODS/CTR-credentialed; QC and a documented audit trail support state/NCDB acceptance and CoC survey readiness.
• Research/consent boundaries: registry data may feed IRB-governed research and clinical-trial matching; we operate strictly as the abstraction service and do not repurpose PHI.

23.Licensing boundary

Hard rule: every reportable-case abstract is signed by a credentialed ODS/CTR before it is finalized in the registry of record. We produce surveillance abstracts, not clinical advice or billing codes. Marketing never implies clinical judgment or reimbursement coding.

24.AI-native advantage

AI changes the economics and throughput of abstraction, not just typing speed. AI tasks: screen feeds to find reportable cases (casefinding); consolidate multi-document records; extract and pre-code primary site, laterality, histology/behavior (ICD-O-3), grade, stage (AJCC/Summary), SSDIs, and first-course treatment; flag conflicts and missing data; draft the abstract and run edits. Human tasks: reportability edge cases, ambiguous primary and multiple-primary/histology rules, stage judgment under conflicting documents, final QC and sign-off. Deterministic rules: standards/version mapping, edit-check logic, suspense/duplicate handling, deliverable templating into the registry system. QC: known-case recall tests, double-abstraction on high-stakes/ambiguous cases, reviewer checklist, accuracy sampling against NAACCR standards. Data inputs: path/cytology/operative/imaging/clinical text, suspense lists, the current standards manuals. Output artifacts: completed, edit-clean abstracts + QC trail. Never fully automated: the credentialed judgment that a case is reportable and correctly coded/staged — that stays with the ODS/CTR. The result: cost-per-case trends toward the cost of review + compute, while accuracy improves as models and the corrected-abstract corpus improve (the 93.9%-exact-match result is the leading indicator) [11].

25.Internal AI engine architecture

AIHuman chokepointDeterministic rulesQA / learning

26.AI-vs-human operations pipeline

Target automation share: ~65% at launch → ~78% by day 90 → ~85% within a year, with the registrar's review minutes-per-case falling as the QC library, casefinding tuning, and standards ruleset mature.

27.Dynasty translation layer

1. Buyer translation: the buyer is a cancer-program director / registry manager; their urgent problem is "I must report every case on time and stay survey-ready with a CTR I can't hire." The outcome they want is complete, accurate, on-time abstracts in their registry, signed by someone credentialed.
2. Service translation: done-for-you abstraction. Customer receives finished abstracts; AI handles casefinding/extraction/coding/entry; an ODS/CTR reviews and signs.
3. Workflow translation: intake (BAA, access, standards version) → casefind → consolidate → extract/code → edits → ODS/CTR review → QA → deliver into registry → feedback/renewal/audit.
4. Tooling translation: read access to the hospital's registry software + path/clinical feeds; clinical-NLP/LLM stack; standards ruleset; secure HIPAA infra + audit logging; lightweight workflow/CRM. Favor off-the-shelf data & models before custom build.
5. Sales translation: "We clear your cancer-registry backlog and keep you on time and survey-ready — AI-fast, ODS/CTR-signed, per case, with an accuracy-and-timeliness guarantee. Send us 25 backlog cases for a free pilot abstract set."
6. Delivery translation: first 3 hospitals delivered semi-manually with the founder-CTR in the loop; automate casefinding and extraction once site-specific patterns repeat.
7. Expansion translation: from one tumor site → all sites → ongoing managed abstraction → casefinding-only → QC audits → submission support → trauma/cardiac/other registries → state-registry contracts.

28.Anti-duplication analysis

• What similar things exist? Registry staffing/abstraction vendors, AI casefinding software, registry software with AI add-ons, in-house registrars, offshore keying (see §19).
• Why not a copy? Staffing vendors sell relocated registrar hours with no AI cost curve and no guarantee-as-product; software vendors sell tools the hospital operates. We sell a guaranteed, credentialed-signed, edit-clean abstract delivered into the hospital's registry, built on an AI engine — a different product and economic model.
• Narrow wedge: backlog clearance for one tumor site at one CoC hospital — a fast, low-risk, measurable entry no enterprise software sale matches.
• Under-served segment: community / mid-size CoC hospitals that can't hire CTRs and can't justify enterprise tooling.
• Unsolved manual pain: the credentialed review, NCDB/CoC-grade QC, and HIPAA-compliant delivery that tools leave on the hospital's plate.
• Differentiator: the abstraction OS + corrected-abstract corpus + credentialed accountability + timeliness guarantee — none of which a tool or a body-shop provides.
• Portfolio check: distinct from prior runs (specialty/surgical coding = billing; risk-adjustment = payer HCC; clinical-research coverage = trial billing) — this is mandated surveillance abstraction.

29.Anti-commoditization analysis

If a future general model auto-codes a clean pathology report end-to-end, does this business evaporate? No — for six reasons. (1) Casefinding from messy, hospital-specific feeds (multiple EHRs, scanned PDFs, free text, lab interfaces) is integration and tuning work a raw model doesn't do. (2) Credentialed accountability: the state registry, NCDB, and CoC surveyors expect a responsible ODS/CTR behind the data — a model can't be the accountable signer. (3) Annually-changing standards (STORE/SEER/AJCC/SSDI) demand a maintained, versioned ruleset and edit logic. (4) NCDB/CoC-grade QC and a defensible audit trail are an operating system, not a prompt. (5) HIPAA-compliant managed operations (BAA, access control, breach posture) are table stakes hospitals won't hand to a self-serve tool. (6) The buyer doesn't want to operate software — they want finished abstracts in their registry. As models improve, our cost-per-case falls and our margin and speed advantage widen; the human residual shrinks but the accountability, QC, integration, and guarantee layers remain the product. Sam Altman test: strong pass — better models make us cheaper and faster, not obsolete.

30.Service delivery workflow

1. Onboard: sign BAA; obtain scoped access to registry software + source feeds; confirm standards version, suspense process, and accuracy/timeliness SLA.
2. Scope: pull the backlog/suspense list for the pilot tumor site; size case count and complexity.
3. Casefind & consolidate: engine flags reportable cases and assembles each record.
4. Pre-abstract: engine extracts/pre-codes all data items and runs edit checks; drafts the abstract.
5. Review & sign: ODS/CTR reviews flagged judgment items, corrects, and signs off.
6. QA: accuracy sampling + edit-clean verification before finalize.
7. Deliver: finalize abstracts in the hospital's registry on the agreed cadence; report burn-down/timeliness.
8. Close-loop: feed corrections back into the engine; renew into managed abstraction or next site.

31.Operations as product

The operation is the product: variance elimination turns credentialed abstraction into a repeatable production line.

• SOPs per tumor site and per registry-software platform; structured intake checklists (access, feeds, standards version, SLA).
• Required-evidence lists per case (path, operative, imaging, clinical) and automated completeness checks.
• Exception queues for ambiguous primary/multiple-primary/stage; reviewer-assignment logic by site expertise.
• Confidence scoring on each extracted field; low-confidence fields routed to mandatory human review.
• Audit trails and version control of standards rules and every abstract change; gold-standard examples per site.
• Red-team checks on high-stakes cases (rare histologies, multiple primaries); customer-ready output = edit-clean abstract in their system.
• Root-cause analysis on any QC failure or rejected/late case; a postmortem loop converting each miss into an SOP/rule/prompt/QC update.

32.No-holes quality engine

• Casefinding recall test: seeded known reportable cases must be caught; track false-negative rate against feeds.
• Field-level accuracy sampling against NAACCR standards; site-specific thresholds (e.g., ~95–98%).
• Double-abstraction on ambiguous/high-stakes cases; reconcile discrepancies before sign-off.
• Edit-check gate: no abstract is delivered unless it passes the registry/NAACCR edit set.
• Reviewer checklist per case; confidence-scored fields force review where the engine is unsure.
• Inter-rater monitoring across reviewers; periodic recoding audits to catch drift.
• Standards-change control: annual STORE/SEER/AJCC/SSDI updates tested before go-live each cycle.

33.What the human expert actually does

Indicative minutes (Inferred). At launch ~28 min credentialed time per complex analytic case; target ~15 by day 90 as automation and QC mature — the core lever of the unit-economics thesis. Non-analytic/simple cases run materially lower.

34.Minimum viable offer

35.Fulfillment process

First 3 customers, semi-manually: the founder-CTR (or a contracted ODS/CTR) runs the chokepoint personally while the engine does casefinding, consolidation, and pre-abstraction. Tools on day one: scoped access to the hospital's registry software; a clinical-NLP/LLM extraction pipeline; the current standards manuals encoded as a ruleset; secure HIPAA infrastructure with audit logging; a simple case-tracking board. Automate later: casefinding tuning per feed, confidence-scored field routing, edit-check automation, and the corrected-abstract learning loop — only after patterns repeat. Don't automate first: reportability edge cases, primary/histology/stage judgment, and sign-off. First paid offer: the Backlog Clear-Out. Team: one founder + 1–2 contract ODS/CTR reviewers + an engineer; scale reviewers sublinearly to volume as automation share rises.

36.Tools and systems

• Source/registry access: hospital registry software (Onco/CNExT, METRIQ, ClinicalPath) + path/clinical feeds, via scoped, audited access.
• AI engine: clinical-NLP + long-context LLM extraction with a provider-abstraction layer (swap models as frontier improves).
• Standards ruleset: versioned STORE/SEER/NAACCR/AJCC/SSDI rules + edit-check logic.
• Security/compliance: U.S.-controlled HIPAA infrastructure, encryption, access control, audit logging, BAA management.
• Ops: case-tracking/workflow board, QC sampling tooling, lightweight CRM for pilots/renewals.

37.Human-in-the-loop quality control

Every reportable-case abstract passes a credentialed chokepoint before finalize. Low-confidence extracted fields are force-routed to human review; ambiguous/high-stakes cases get double-abstraction; nothing ships that fails the edit set. Accuracy is sampled against NAACCR standards, inter-rater agreement is monitored, and periodic recoding audits catch drift. The signing ODS/CTR is accountable to the hospital, the state registry, and the NCDB/CoC — the trust interface a tool cannot replace.

38.Nonlinear scaling and unit economics

All unit economics are Inferred and carried as explicit pilot kill-criteria. Other targets: revenue-per-FTE rising as automation share grows (the thesis is decoupling revenue from registrar headcount); throughput per reviewer per day climbing as minutes-per-case fall; rework rate <3%; QC failure rate <2%; escalation/exception rate trending down; CAC payback <6 months on per-case margin; lead-magnet (free pilot set) → paid conversion target ~30–40%; pilot → ongoing-managed conversion target ~40–55%; net case-volume retention >100% as hospitals expand from one site to all sites.

COGS tracked from day one: model inference, document processing/storage, hosting, registry-system access, credentialed review labor, QC, casefinding compute, support, standards-update engineering, rework, and sales follow-up. Labor is never hidden inside "operations."

39.Distribution proof table

40.Sales and outreach plan

Lead with a diagnosis, not a demo: offer a free Backlog & Timeliness Risk Review (how many cases are open, how close to the 6-month wall, what a CoC survey would find) and a free 25-case pilot abstract set. Convert via a paid Backlog Clear-Out, then expand to all sites and ongoing managed abstraction. Three motions: founder/expert-led content (build registrar trust), warm outreach to associations/diagnostic users, and targeted outbound to CoC programs with an approaching survey or known staffing gap.

41.Founder-led content plan

Publish as a credentialed registry expert, teaching managers to understand and de-risk their own program: the real cost of a timeliness deficiency; how the CTR shortage actually hits a community program; what CoC surveyors look for; how AI casefinding changes the math; common coding/staging error patterns; and how to evaluate an abstraction partner. High-performing organic pieces become paid-ad creative later.

42.First 30 days of content

10 educational posts

• "The 6-month clock: what your state actually requires (and what happens if you miss it)."
• "Why you can't hire a CTR — and what the shortage means for community programs."
• "What a CoC surveyor checks in your registry (and how backlogs show up)."
• "Casefinding is where backlogs are born: the feeds you're probably missing."
• "ICD-O-3 vs. billing codes: why your biller can't do your abstracts."
• "Multiple primaries & histology rules: the judgment AI still can't own."
• "AJCC vs. Summary Stage: a plain-English refresher for program leaders."
• "What 93.9% field accuracy from AI does — and doesn't — mean for your registry."
• "NCDB Call for Data: a survival checklist for a thin team."
• "Outsource vs. hire vs. AI: the real cost-per-case comparison."

3 diagnostic teardown formats

• "Backlog & Timeliness Risk Review" — open cases vs. the 6-month wall, with a burn-down plan.
• "Casefinding Gap Scan" — sample feeds, estimate missed reportable cases.
• "QC Recode Spotlight" — re-abstract a small sample, show field-accuracy & edit failures.

2 lead-magnet angles

• Free 25-case pilot abstract set (paid-at-cost) delivered into the hospital's registry.
• "Survey-Readiness Scorecard" — a self-assessment of timeliness, completeness, and QC.

1 webinar / live review

• "Clearing a cancer-registry backlog before your CoC survey — live walk-through with a CTR."

1 outbound diagnosis template

• "[Program], your CoC survey window is approaching and CTR hiring is hard. We can return edit-clean, ODS/CTR-signed abstracts in your system at $X/case — want a free 25-case pilot?"

43.Lead magnet and waitlist plan

Lead magnet: the free Backlog & Timeliness Risk Review + a 25-case paid-at-cost pilot abstract set — high-value, trust-building, and a direct read on the buyer's pain (backlog size, deadline proximity). Waitlist: a "Backlog Clear-Out — limited pilot cohort" signup that captures program name, accreditation status, registry software, open-case count, and next survey/Call-for-Data date. What the buyer receives before paying: a concrete picture of their timeliness risk and a sample of finished, signed abstracts in their own system. Qualification: CoC/reporting hospital + a real backlog or staffing gap + an owner of the registry budget = sales-ready.

44.Warm GTM plan

Work association networks (NCRA/NAACCR, state registrar groups), former colleagues of the founder-CTR, and every diagnostic/scorecard user. Offer a free risk review and a scoped pilot. Registry-staffing vendors with overflow backlogs are warm partners for white-label pre-abstraction. Software vendors are warm referral partners (we fill the system they sell).

45.Targeted outbound plan

Build a perfect-fit list from the public CoC accreditation directory [4], prioritized by signals of pain: recent CTR job postings (staffing gap), an approaching survey cycle, or a state timeliness notice. Personalize each outreach with a backlog/timeliness observation and lead with the free pilot — never a generic demo ask.

46.Answer-engine / search visibility plan

Own the questions registry managers ask ChatGPT/Perplexity/Google: "cancer registry backlog outsourcing," "how to meet the 6-month cancer reporting deadline," "CTR shortage solutions," "AI cancer abstraction accuracy," "CoC survey registry readiness." Publish authoritative, citable guides and structured FAQs so the engine surfaces us when buyers research the problem.

47.Pilot design and early-demand trap mitigation

First cohort: 3–5 CoC community hospitals with active backlogs, capped tightly. Early-access incentive: pilot abstracts at cost + a locked per-case rate for the first managed term. Trap mitigation: the pilot is one tumor site and one outcome (edit-clean, signed abstracts on cadence) — not unlimited custom registry consulting. Anything outside abstraction (program redesign, software migration) is explicitly out of scope. Feedback is captured daily; corrections become engine/rule/QC improvements, not bespoke side-work.

48.Early-access feedback flywheel

Each reviewer correction is logged and triaged: is it a product signal (a recurring extraction/coding miss → fix the prompt/rule/QC check) or custom work (a one-off hospital quirk → SOP, not engine change)? Recurring misses convert into extraction tuning, standards-rule updates, gold-standard examples, and new edit checks. Daily pilot stand-ups review accuracy samples and exception queues. The corrected-abstract corpus is the compounding asset.

49.Build-before-scale checkpoints

• After 5 hospitals: harden intake (access/feeds/standards version), casefinding recall, and the edit-check gate.
• After 10 hospitals: harden SOPs per tumor site & per registry platform, exception queues, reviewer checklists, and delivery templates.
• After 20 hospitals: pause new pilots until COGS, reviewer-minutes-per-case, rework rate, escalation rate, and timeliness are measured and trending to target. Acceptable temporary workarounds: manual casefinding on a new feed, manual standards mapping for a rare site. Unacceptable (signals non-scalable): per-hospital bespoke logic that can't be templated, or review minutes not falling cohort-over-cohort.

50.7-day launch plan

• Days 1–2: lock the MVP (one tumor site, backlog clearance); encode current standards ruleset + edit checks; stand up HIPAA infra + BAA template.
• Days 3–4: wire the casefinding/extraction pipeline against sample de-identified path reports; build the reviewer checklist + QC sampling.
• Day 5: publish the Backlog & Timeliness Risk Review lead magnet + pilot waitlist page.
• Days 6–7: outbound to 25 CoC programs with backlog/survey signals; line up the founder-CTR reviewer; book 3 pilot calls.

51.30-day launch plan

• Sign 2–3 pilot hospitals; complete BAAs and scoped access.
• Deliver first 25-case pilot sets; measure field accuracy, edit-clean rate, reviewer minutes/case.
• Publish 10 educational posts + run 1 webinar; collect diagnostic signups.
• Instrument COGS and reviewer-minutes dashboards from case one.

52.90-day launch plan

• Convert pilots to paid Backlog Clear-Out, then to ongoing managed abstraction for ≥2 hospitals.
• Expand from one tumor site to all sites for the first converted hospital.
• Drive automation share to ~78% and review minutes/complex case toward ~15.
• Harden SOPs (5-hospital checkpoint); sign one staffing/software partner; stand up the corrected-abstract learning loop.

53.Metrics and KPIs

• Quality: field-level accuracy vs. NAACCR standard; casefinding recall; edit-clean rate; rework <3%; QC failure <2%.
• Timeliness: % cases delivered inside the agreed window; backlog burn-down rate.
• Economics: COGS/case; reviewer minutes/case; gross margin; revenue-per-FTE; CAC payback.
• Automation: automation share; low-confidence routing rate; escalation/exception rate.
• Growth: pilot→paid and paid→managed conversion; one-site→all-sites expansion; net case-volume retention.

54.Risks and mitigations

The exhaustive register follows; headline risks: (1) AI extraction errors causing reportable-data inaccuracy → credentialed sign-off + QC sampling + edit-check gate; (2) HIPAA/PHI breach → U.S.-controlled, access-controlled, audited infra + BAA; (3) standards changing annually → versioned ruleset + change control; (4) incumbents/EHRs bundling AI abstraction → out-execute on the corpus/guarantee/accountability layer; (5) thin launch margin → drive review minutes down fast or stop.

55.Exhaustive risk register

56.What could kill this

• Margin never bends: if credentialed review minutes/case don't fall with automation, the model stays a low-margin body-shop — the single most important metric to watch.
• A trust failure: a HIPAA breach or a high-profile mis-reported case would be existential for a data-accountability business.
• Incumbent EHR/registry-software bundling a "good enough" signed-abstraction service at zero marginal price to existing customers.
• Recall skepticism: if registry managers don't trust AI casefinding, the free pilot must consistently match/beat in-house and incumbent quality.

57.Go/no-go reasoning

GO. The candidate clears every evidence-threshold criterion: a clear buyer (cancer-program/registry managers + staffing/software partners), a painful, specific, legally-mandated recurring problem (report every case within ~6 months against a shrinking CTR workforce), proof the problem exists and is already paid for (mature outsourcing market + per-case cost data), active demand (vendor "fill-the-gap" content, government procurement, sustained job postings), competitor/budget validation, a credible win reason (AI cost curve + credentialed accountability + guarantee), a narrow MVP wedge (one-site backlog clearance), a service-first path needing no large platform, no unresolved fatal regulatory blocker (clean HIPAA/credentialed-data boundary; not the practice of medicine), a credible 50–65%+ margin path, and a believable distribution path. The main watch-item — launch margin thinness — is explicitly gated by build-before-scale checkpoints.

58.Final recommendation

59.Source list

1. American Cancer Society — Cancer Facts & Figures 2025; Siegel et al., Cancer statistics, 2025 (2,041,910 new U.S. cases projected; ~5,600/day)
2. Extract Systems — The Cancer Registrar Shortage and the Impact of CoC's CTR Standard 5.1 (aging workforce; 12-month backlogs; associate-degree requirement)
3. Savista — Cancer & Clinical Registry Management Services; How To Face The Certified Tumor Registrar Shortage (largest U.S. employer of ODS after ONCO acquisition)
4. American College of Surgeons — Commission on Cancer Accreditation (~1,400–1,500 accredited programs; NCDB submission required)
5. Subramanian et al. — The Cost of Cancer Registry Operations: Impact of Volume on Cost per Case ($60.77 avg; $15.96–$233.48 range; abstraction largest cost component)
6. Direct Difference — Cancer Registry Data Abstraction Services
7. American Data Network — Clinical Data Abstraction Services (98.4% accuracy across measures/registries)
8. Clinical Registry Solutions — Remote Clinical Data Abstraction Services (>98% accuracy)
9. Carta Healthcare — Expert Clinical Data Abstraction Services (AI-assisted abstraction)
10. Inspirata — E-Path & E-Path Plus: Automate Cancer Registry (99% sensitivity / 98% specificity casefinding; 124+ data elements)
11. medRxiv (2025) — Automating cancer registry abstraction with an autonomous, resource-efficient AI for multi-cancer pathology reports (93.9% mean exact-match across 196 fields, 10 cancers)
12. CDC — Natural Language Processing for Cancer Surveillance (NPCR)
13. New Jersey Department of Health — NJ State Cancer Registry — Reporting (case reports within six months of first contact)
14. New York State Department of Health — Cancer Reporting (notice within 180 days of every cancer case)
15. CDC — National Program of Cancer Registries (NPCR) (Cancer Registries Amendment Act of 1992, PL 102-515)
16. Mordor Intelligence — Cancer Registry Software Market ($99.06M 2025 → $167.59M 2030, 11.17% CAGR)
17. Grand View Research — Cancer Registry Software Market Size & Share Report
18. GovernmentContracts.us — Q701 — Cancer Registry Abstract Services (VA / VISN)
19. NAACCR — Artificial Intelligence and Natural Language Processing in the Cancer Registry Field
20. John Snow Labs — Automating Cancer Registries: Overcoming Data Bottlenecks with AI-Powered Abstraction
21. ZipRecruiter — Certified Tumor Registrar Jobs (wages ~$20–$45/hr; sustained demand)
22. In Record Time, Inc. — Cancer Registry Outsourcing & Oncology Data Management
23. Medovent Solutions — The Cost Benefits of Using a Cancer Registry Staffing Company

Market-size figures are third-party analyst estimates and vary by source; cost/pricing ranges are illustrative of published norms and vary by case complexity and volume. The cost-per-case study reflects registry operations and is used as a directional benchmark. Inferred unit economics and conversion rates are carried as explicit pilot kill-criteria. Not medical or legal advice. Deliverables are surveillance abstracts produced under HIPAA and signed by an ODS/CTR-credentialed registrar — not clinical decisions, diagnoses, or billing codes.