Research and Publication
Status: vision — commitments that activate with real data.
When PRISM runs on real claims data, its findings — successes and failures alike — are committed to peer review, and the precursor patterns its models learn become research material in their own right.
Peer review, both disciplines, negative results included
PRISM sits at the intersection of two literatures, and commits to publishing in both. Computational venues are the right forum for the method itself: the pool-and-consensus architecture, the two training rounds, and the evaluation frame that never uses a hold-out set. Medical venues are the right forum for the claim that matters: whether flagged patients are actually detected earlier, and whether earlier detection changes their outcomes. Neither review alone is sufficient — a sound method with no clinical benefit is a curiosity, and a clinical result from an unexaminable method is not evidence.
The commitment explicitly includes negative results. The synthetic prototype proved the method works given a learnable precursor pattern; whether real conditions carry such patterns is the open question of the real-data phase, and for some conditions the honest answer will be no. A condition that resists this approach is a publishable finding, not a quiet omission — it maps the method's boundary and saves others the same detour. The project has practice at this posture: the prototype's own write-up treats a failed training approach as one of its most instructive results.
Learned patterns are hypotheses
The deeper research value is not validation of PRISM but what PRISM surfaces. When several independently trained models, sharing no data or weights, converge on the same precursor pattern for a condition, that agreement is evidence the condition leaves a consistent utilization phenotype — a recognizable trail in billing data before diagnosis. Individual models are wrong in tolerable, correctable ways; consistent cross-pool agreement about what precedes a diagnosis is something else: a hypothesis about disease progression, generated from population-scale observation rather than from theory.
That matters because many conditions sit in the screening gap precisely because their early course is scattered — a lab here, a specialist referral there, years apart, across providers who never see the whole. If models trained only to continue timelines keep firing on a particular constellation of events, that constellation deserves a clinician's attention, whether or not any textbook currently connects it to the diagnosis.
The caveat is structural: a utilization phenotype is a pattern in billing behavior, not in biology. It can reflect coding habits, referral customs, or benefit design as easily as pathology. Distinguishing signal from artifact requires clinical collaborators — which is why learned patterns are offered as hypotheses for medical research to test, never announced as discoveries.
Collaboration inside the privacy frame
Academic collaboration operates one level above even PRISM's own data posture. PRISM never holds identified patient data in the first place — anonymity is architectural, not procedural — and what crosses to research partners is a further abstraction: learned patterns, aggregate firing statistics, and model behavior under controlled prompts. No researcher ever sees a row of any patient's timeline. Within that frame, collaborators can propose hypotheses to test against ensemble behavior, probe which record types drive firing for a condition, and design prospective studies to validate a discovered phenotype clinically. The privacy boundary constrains the mechanics of collaboration, not its substance.
Open research, same principle as open collaboration
The publication commitment is the research-facing half of a single posture. The open-collaboration model shares trained models — never data — under licensing that mandates continuous contribution back to the collective; publication applies the same rule to knowledge. Methods are documented to the standard of replication, limits are stated as plainly as results, and what the models learn about disease progression is returned to the field that will ultimately judge it.