The Insurance Vantage Point
Status: concept — why claims data is the substrate.
Exactly one party in the healthcare system sees a patient's complete journey across every provider, every specialty, and every year of coverage: the payer. That view exists as a byproduct of billing, arrives pre-standardized, and is the entire reason PRISM is built on insurance claims.
The only complete view
A patient sees a primary-care physician on Monday, a specialist on Wednesday, fills a prescription on Friday, and has lab work drawn the next week. Each provider sees their own fragment. The insurer sees all of it, because every one of those encounters generated a claim.
This visibility has two dimensions, and both matter. It is cross-provider: the payer sees claims from every physician, pharmacy, laboratory, and facility the patient touches, regardless of which health system employs them or whether they have ever heard of each other. And it is longitudinal: coverage persists across years, so the payer's record captures the slow arc of a developing condition — the escalating medications, the widening circle of specialists, the intervals between symptom and workup — rather than the disconnected snapshots any single provider works from.
The conditions PRISM targets are precisely the ones that hide in this fragmentation. A utilization phenotype that spans cardiology, nephrology, and a string of vague symptom visits is invisible to each specialist individually and legible only in the assembled record. The payer holds the only assembled record there is.
Completeness is a financial byproduct
Nobody designed claims data as a research asset. Its completeness follows from one blunt fact: providers who want to be paid must submit claims. Every billable service — routine refills, lab panels, imaging, ER visits, equipment — produces a record automatically, with no reliance on voluntary reporting, patient recall, or inter-office communication. The mundane entries accumulate alongside the significant ones, and for pattern recognition the mundane entries are the context.
The same financial machinery enforces standardization. Payment requires that every provider describe every service in the same shared code systems — procedure, diagnosis, place of service, provider taxonomy, medication (the code systems in detail). A rule written to make adjudication tractable accidentally produced something close to an ideal substrate for machine learning: a finite, documented vocabulary in which the same condition leaves similar traces regardless of who treated it or where. PRISM's six-column timeline format is a thin, direct rendering of what the claims stream already contains — the format adds structure, not information.
This is also why PRISM can run without touching anyone's systems: the substrate already sits inside the insurer, fully formed, which is the basis of the zero-integration posture.
Why not hospitals, health systems, or EHRs
Provider-side data sources are the obvious alternatives, and each breaks in a structural way that no diligence can fix.
| candidate | where its view breaks |
|---|---|
| A single provider | Sees only their own encounters. No visibility into other physicians' prescriptions, other facilities' tests, or patterns spanning specialties. |
| A hospital or health system | Sees only patients who choose its facilities — and loses them to outside specialists, retail pharmacies, out-of-network emergencies, and every move between cities. |
| EHR networks | Competing systems treat records as competitive assets. Decades of interoperability mandates have produced partial, inconsistent sharing; a record that arrives as an unsearchable document is not data. |
These are not failures of effort. They are consequences of a delivery system in which patients move freely, institutions compete, and records live where care happened. The payer relationship is the one thread that follows the patient through all of it: coverage does not care which building the claim came from.
What claims data is not
The honest limit: claims describe utilization, not clinical truth. A claim records that a test was billed, not what it found; that a diagnosis code was attached, not how confident the physician was. There are no lab values, no vitals, no notes, and coding is shaped by reimbursement incentives as much as by medicine. PRISM's method is built around this constraint rather than against it — it learns patterns of utilization and suggests one constructive action, a screening test, which is a claim about what the billing record resembles, never a diagnosis.
See also
- The screening gap — the class of condition this vantage point can catch.
- The patient timeline format — how claims render into model input.
- Anonymous by architecture — comprehensive visibility without identity.
- Zero integration — the operational consequence of data that already exists.