Clinical Decision Support Positioning
Status: concept, vision — the deployment posture for the real-data phase.
PRISM is positioned as non-device clinical decision support: an information source that suggests established screening tests to licensed physicians, who retain complete authority over every clinical decision. Nothing PRISM emits can reach a patient except through a physician's independent judgment.
An information source, not a medical device
Regulatory frameworks for clinical decision support draw a line between software that supports clinical decisions and software that makes them. PRISM is built to stay on the supporting side of that line by construction, not by policy. Its audience is licensed healthcare professionals only. Every test it can suggest is an established, approved diagnostic with standard interpretation — the novelty is entirely in noticing when an existing test might be worth considering, never in the intervention itself. And the basis for each suggestion is independently reviewable: it rests on the patient's own claims history, which the physician can inspect, weigh, and disagree with.
The boundaries of what the system can do are architectural facts, not usage guidelines:
| action | PRISM |
|---|---|
| suggest an established screening test | yes — this is its entire output |
| order a test | no; the physician orders it, or doesn't |
| interpret results | no; PRISM never sees test results or lab values at all |
| diagnose a condition | no; a suggestion is a flag that a pattern warrants a look, nothing more |
| recommend or provide treatment | no; treatment is outside the output space entirely |
| deny, delay, or discourage care | no pathway exists — see constructive-only architecture |
The last row matters most. A system whose only possible output is "consider this test" cannot be repurposed into utilization review or care rationing, because the denial pathway was never built. Its silence is never a clearance — a patient no models flag is simply a patient the ensemble had nothing to say about.
Suggestions go to primary-care physicians only
Suggestions travel through the insurer's existing communication channels — the same routes already used for care-gap alerts, with no new integration — and they go to exactly one recipient: the patient's primary-care physician. Never to patients directly, never to specialists, never to anyone else.
The PCP is the right recipient because screening decisions belong inside a longitudinal relationship. The primary-care physician holds the whole-patient context — the presentation in the room, the testing done outside this insurer's claims, the preferences and circumstances that no billing code records — and coordinates care across everything else happening to the patient. PRISM, by contrast, sees only standardized billing codes in a six-column timeline. It does not know why a visit happened, only that it was billed. A single, clearly owned channel also prevents the failure modes of broadcast: specialists receiving flags outside their scope, patients receiving raw signals without clinical framing, or multiple providers acting on the same suggestion in conflicting ways.
A separate explanation model — a production design, not yet built — is intended to accompany each suggestion with a physician-readable narrative of the pattern behind it; that design is described in physician explanations.
Human-in-the-loop as the operating principle
Physician review is not a safety feature layered onto PRISM; it is the operating principle the whole deployment is shaped around. Without a physician choosing to act, the system is inert: no automated workflow proceeds, no alert escalates, no order fires by default. A suggestion nobody acts on does nothing at all.
This framing follows directly from what a suggestion is. A consensus flag — "N of M independent models surfaced this test" — is a flag, not a probability, and certainly not a diagnosis. Only a clinician can turn that flag into care, and only a clinician should. The human here is not a checkpoint that a future, more confident version of the system might optimize away for efficiency; the human is the mechanism by which pattern recognition becomes medicine. Remove the physician and PRISM does not become autonomous — it becomes useless.
No compliance tracking — because none is possible
There are no metrics on whether physicians follow suggestions, no dashboards scoring responsiveness, no penalties for dismissal, no rewards for uptake. This is not restraint; it is architecture. PRISM's only window into the world is claims data. It learns that a suggestion was acted on only if the suggested test later appears in claims carrying the PRISM-specific tracking code — the same mechanism that triggers results-based compensation. A physician who quietly dismisses every suggestion is, from the system's side, indistinguishable from one who never received them.
That invisibility is the point. A physician may have reasons no billing pattern can capture — recent equivalent testing elsewhere, a patient who has declined screening, a clinical picture that explains the pattern benignly. A system that cannot see whether it was followed cannot pressure anyone into following it.
Liability
The liability structure PRISM slots into is the existing one, unchanged. A physician who acts on a suggestion carries the same responsibility they carry for any decision informed by available data; PRISM supplies information for consideration, not medical advice, and makes no clinical decisions for which it could hold clinical responsibility. For the insurer, forwarding a suggestion is analogous to sharing a utilization report — claims-derived information, not a directive. Exposure is further narrowed by the constructive-only constraint: a system that can only propose additional established screening, and can never withhold, delay, or de-prioritize anything, has no pathway through which its output could deny a patient care.
None of this has been exercised yet. The 2026 synthetic prototype validated the screening method; this article describes the posture designed for deploying it against real claims, and it will be tested — legally and operationally — only when that phase arrives.
See also
- Constructive-only architecture — why no denial pathway exists to begin with
- Recall, not prediction — why silence is never reported as "no need"
- Results-based compensation — the tracking code as the only feedback channel
- Physician explanations — the planned narrative layer for suggestions