Skip to main content

Blueprints Healthcare & Life Sciences Pharma · Diagnostics · Precision Medicine

Evidence Lens

The patient evidence pharma and care teams can act on. Clinical, claims, multi-omics, and wearable signal joined at patient level, reasoned through by agents.

Start a Sprint

Precision medicine has answered the molecular half of the question. The other half is what happens to the patient after the trial protocol ends. Real-world evidence is that other half — and it is now the currency. Five modalities join at patient level. An agentic layer reasons across them and returns patient-grade evidence to pharma, diagnostics, providers, and trial sponsors.

The context

Three facts.
The category direction is set.

The regulatory verdict

Real-world evidence was cited in 23 to 28% of FDA labeling-expansion approvals across 2022 to 2024, peaking at 27.7% in 2023. Oncology led at 43.6%. The 2018 RWE Framework has been followed by guidance on registries (final December 2023), digital health technologies for clinical investigations (final December 2023), EHR and claims data (final July 2024), external-control trials (draft February 2023, pending finalization), and the medical-device RWE update (December 2025). EMA's DARWIN EU is operational across 16 European countries, covering roughly 180 million patients on a common OMOP model.Deng, Girman & Ritchey, Therapeutic Innovation & Regulatory Science 2025 · FDA · EMA DARWIN EU

The commercial pull

Pharma is paying for linked real-world evidence at unit prices the market has not previously seen. AstraZeneca, Tempus, and Pathos announced a $200M agreement on April 17, 2025 to build a multimodal oncology foundation model on Tempus's de-identified patient records. GSK extended its Tempus partnership at $70M upfront in October 2022 for a three-year minimum, extendable to five. In January 2025, a coalition of US health systems launched the Truveta Genome Project, with the Regeneron Genetics Center as anchor sequencing partner in a $320M collaboration. CMS's Cell and Gene Therapy Access Model is operational across 32 states plus DC and Puerto Rico, with outcomes-based rebates tied to real-world results for Vertex's Casgevy and bluebird bio's Lyfgenia. The contract structure is policy now, not pilot.Tempus IR · GSK IR · Truveta · Regeneron IR · CMS CGT Access Model

The five-modality gap

Clinical, claims, genomic, multi-omic, and wearable data live in four parallel stacks and four standards. Even the leading players show the scarcity of the join at patient level. By the time Tempus unveiled its Fuses foundation-model program in May 2025, the multimodal library had grown past 40 million records, of which more than 1.5 million carry matched clinical and genomic data. Wearable signal is now regulator-grade in tightly scoped settings — SV95C became EMA's first wearable-derived primary endpoint in Duchenne muscular dystrophy in 2023. Outside those scoped settings, the wearable layer is rarely joined to the clinical record at scale.Tempus Fuses announcement, May 2025 · EMA SV95C Qualification Opinion, July 2023

Regulatory verdict

23–28 %

Share of FDA labeling-expansion approvals citing real-world evidence across 2022–2024. Oncology leads at 43.6%. EMA’s DARWIN EU is operational across 16 countries and roughly 180 million patients on OMOP. The RWE pull is regulatory policy, not pilot.

Deng, Girman & Ritchey, Therapeutic Innovation & Regulatory Science 2025 · EMA DARWIN EU

Commercial pull

$200M

AstraZeneca, Tempus, and Pathos announced a $200M agreement on April 17, 2025 to build a multimodal oncology foundation model on Tempus’s de-identified records. GSK extended Tempus at $70M upfront in October 2022. Linked multimodal RWE is paying at unit prices the market had not seen.

Tempus IR · GSK IR

The join is rare

1.5M / 40M

By Tempus’s May 2025 Fuses announcement, the multimodal library had grown past 40 million records. More than 1.5 million carry matched clinical and genomic data. The breadth is real. The depth at patient level is the scarce thing — and the substrate the next category step is being built on.

Tempus Fuses announcement, May 2025

The patient is the unit of evidence. Not the dataset, not the modality, not the cohort assembled study-by-study.

The math

The leak is in the join, not in any single modality.

What the silo costs

Between the EHR, the claims platform, the sequencer, and the wearable stream, evidence gets reconstructed cohort-by-cohort and study-by-study. Time-to-cohort runs in months. Causal signal is lost at the boundary between modalities. Shadow patient journeys outside the EHR — specialist visits, out-of-network labs, generic prescriptions, lifestyle change — go uncaptured. Every fragmented evidence base is paid for again on the next study.

Why genomics alone is half an answer

AACR's Project GENIE Biopharma Collaborative is a ten-sponsor precompetitive collaboration targeting roughly 50,000 patients at clinico-genomic depth, with cohorts in NSCLC, CRC, prostate, pancreatic, and early-onset breast already released. The target is small for a reason. Linked clinico-molecular depth is the scarce thing, not raw genomic counts. A variant call without an outcome trajectory is a description, not evidence.AACR Project GENIE BPC

Why claims alone is half an answer

Komodo's Healthcare Map covers more than 325 million de-identified patients with longitudinal prescription, claims, EHR, payer, and lab signal. Truveta added linked closed claims for more than 200 million patients on top of its EHR foundation in February 2025. The breadth is real and useful. The limit is structural. Claims describe reimbursement, not clinical truth. Genotype is absent. Continuous physiologic signal is absent.Komodo Health 2024 · Truveta February 2025

Why wearables alone is half an answer

The qualified wearable endpoints to date work only inside tightly scoped patient populations. Continuous glucose, cardiac, sleep, and activity data are now normalized to open schemas — IEEE 1752 for mHealth metadata, sleep, and activity; Open mHealth JSON. What they lack on their own is the genotype and phenotype to interpret against. Continuous physiology without molecular and clinical context is signal without meaning.IEEE 1752.1-2021

The Blueprint

Four layers. One patient view.

Evidence Lens sits above the existing data estate — EHR, claims platform, sequencer pipeline, wearable ingestion. The systems of record stay in place. Four layers above them harmonize the modalities, store the patient evidence semantically, reason across it, and surface the work to four audiences.

Layer 04 · Surfaces

Persona surfaces

Bioinformatician

Research Workbench

Cohort spec to outcome trajectory by subgroup, with the audit trail attached.

Commercial Lead

Partner-Insight Surface

Cohort scoped, data-rights gated, deliverable priced against the partner's question.

Provider · EHR workflow

Patient-Journey Console

Patient labs, molecular profile, comorbidities, and matched-cohort RWE in one view.

Trial Sponsor

Cohort Explorer

Eligible-population picture across genotype, phenotype, catchment, and outcomes.

The surfaces are how the work shows up. The agents do the work.

Layer 03 · Reasoning

Agentic reasoning

Agent 01

Cohort Assembly

Resolves a population from a clinical, molecular, and longitudinal specification.

Agent 02

Phenotype Enrichment

Fills missing clinical features from notes, labs, and wearables to a Phenopackets-compatible profile.

Agent 03

Journey Reconstruction

Sequences each patient's clinical, claims, omics, and wearable record into a single timeline.

Agent 04

Evidence Synthesis

Answers research and care questions with cited spans pointing back to the underlying records.

Every agent action carries its retrieval trail.

Layer 02 · Vectors

Vector stores

Embeddings · Clinical

Clinical Vector Store

Encounters, notes, lab trajectories, medication histories, wearable-derived events as retrievable embeddings.

Embeddings · Genomics

Genomics Vector Store

Variant calls, expression signatures, proteomic profiles as retrievable embeddings.

The addressable patient evidence, ready for an agent or a query.

Layer 01 · Fabric

Data fabric

EHR · labs

Clinical

FHIR R4 / R5

closed + open

Claims

OMOP CDM v5.4

sequencing

Genomics

GA4GH VRS · Phenopackets v2 · Beacon v2

transcriptomic · proteomic · spatial

Multi-Omics

Ingested at the source

wearables · DHT

Wearables

IEEE 1752 · Open mHealth

Systems of record stay in place. Additive, not a replatform.

fig. 06 · four layers, one patient view

Trust posture

Five controls the medical, technology, and privacy office will ask about in the first working session.

Compliance

HIPAA BAA, SOC 2 Type II controls, HITRUST CSF alignment, GDPR Article 9 posture for European deployments. State-level frameworks (CCPA, TDPSA) addressed at deployment time.

Identity resolution

Patient-level join across modalities via tokenization (Datavant-class tokens or institutional MPI). Expert-determination de-identification for commercial use, full-identifier handling under BAA for clinical use. Re-identification risk is quantified per cohort, not assumed.

Deployment model

Customer-VPC or dedicated-tenant. AWS, Azure, and GCP supported. Data residency honored. Data does not leave the institution unless the use case requires it and consent permits it.

Clinical decision-support stance

The agentic layer informs care. It does not decide care. Surfaces are designed under the FDA SaMD framework as decision-support, not autonomous diagnosis. The clinician retains authority. Every agent action carries its retrieval trail, citing the underlying records.

Bias and validation

Cohort representation is reported against benchmark populations per use case. Agent inferences are evaluated on golden sets per modality, with monitoring for drift. The evaluation dashboard is part of the working artifacts on the wall, not a black box.

The data foundation

Three tiers of evidence depth.
Tier 1 alone no longer answers.

Most real-world evidence platforms run on one tier. The shift in pharma, diagnostics, and provider use cases over 2024 and 2025 is that Tier 1 alone no longer answers the questions pharma, diagnostics, and provider teams are now asking.

Tier 01 · Baseline

Clinical + claims

Phenotype and utilization. FHIR for the clinical record. OMOP CDM v5.4 for population analytics. Adjudicated claims for journey continuity. The classic RWE substrate.

Useful, and where most platforms stop.

Tier 02 · Multi-omics

Joined at patient level

Genomic, transcriptomic, and proteomic data joined at the patient level. GA4GH Phenopackets v2 for clinical phenotype, VRS for variants, Beacon v2 for federated discovery. The patient population becomes addressable by molecular subtype, not only by ICD code.

The substrate for precision-medicine RWE.

Tier 03 · Wearables · DHT

Continuous signal

Continuous physiologic signal between visits. Step count, heart rhythm, glucose, sleep, activity-derived endpoints. IEEE 1752 for normalized mHealth metadata, sleep, and physical-activity measures.

The longitudinal layer that turns clinic visits into checkpoints in a continuous record.

North Star · The Reimagination

The patient is the unit of evidence.

Today, evidence is organized around the dataset. Tomorrow, it is organized around the patient. Drug development guided by living patient evidence in addition to the trial endpoint. Providers treating from RWE alongside guidelines, with the genomic, clinical, and physiologic picture in one place. Commercial teams shaping insight grounded in the joined patient, not the silo. Patients at the center of the evidence loop.

Two signals from the last twelve months point in this direction. In November 2024, Tempus's molecular profiling went natively available inside Flatiron's OncoEMR, putting molecular ordering inside the oncology EHR workflow across 800+ community cancer locations. In April 2025, MSK and City of Hope published InflaMix in Nature Medicine — a model integrating fourteen pre-infusion inflammation and end-organ markers to predict CAR-T response in non-Hodgkin lymphoma, validated across 688 patients. The pattern repeats in cardiology, metabolic disease, and rare disease. Commercial teams shape insight grounded in the joined patient, rather than the silo.Tempus + Flatiron, November 2024 · MSK / City of Hope, Nature Medicine April 2025

Personas

How the agents show up in each working surface.

The Bioinformatician

Research Workbench

A translational research scientist at a pharma sponsor is investigating early non-response to a targeted therapy in a rare oncogenic-driven subtype. The Workbench takes the molecular spec, the clinical phenotype, the regimen, and a candidate adherence signal from the wearable layer. Cohort Assembly returns the matched population across the five modalities. Phenotype Enrichment surfaces missing comorbidities. Evidence Synthesis returns the outcome trajectory by subgroup with cited spans pointing back to the underlying records.

The cohort that previously took a fellow weeks to assemble is in front of the scientist by mid-afternoon. Audit trail attached.

The Commercial Lead

Partner-Insight Surface

A pharma-data commercial lead is scoping a deliverable for a top-five sponsor. The Surface scopes a cohort, enforces the data-rights and consent gates, aggregates to the necessary tier, and prices the deliverable against the partner’s question. The differentiation in 2026 is the modality stack the partner buys.

Tier 3 wearable signal joined to Tiers 1 and 2 is the offer the market is moving toward — and the one most of the field has not yet built.

The Provider

Patient-Journey Console

An oncologist is choosing the next line for a patient who has progressed on a first-line targeted therapy. The Console reads the prior labs, the molecular profile, comorbidities, prior response, and a recent decline in step count from the wearable layer. The agent returns next-line options ranked against the patient's own profile, with the supporting RWE cohorts cited inline. NCCN and OncoKB remain the source of truth for guidelines. The RWE overlay shows how patients with this profile have actually responded.Tempus + OncoKB / NCCN integration 2024

The provider’s authority is unchanged. The picture supporting the decision is wider.

The Chief Medical Officer

Cohort Explorer

A health-system CMO is asked by the board whether the institution should sign a value-based oncology arrangement on a new CAR-T product. The Console and the Explorer give the CMO the picture across the attributed population — who is eligible by genotype and phenotype, who sits in the catchment, what the matched-population outcomes look like in published RWE, what the cost-of-care trajectory looks like across the modality stack.

Existing Tempus, Flatiron, Komodo, or Truveta relationships continue to populate the modalities they already serve. Evidence Lens joins them at patient level.

Clinical trials

One platform. Five trial use cases.

Clinical-trial sponsors are a distinct audience with a distinct budget and a distinct buying cycle. Evidence Lens applies to five recurring use cases across phase 2, phase 3, and post-marketing.

TC-01

Cohort identification

Find the patients who match a rare phenotype-by-genotype combination, with the journey continuity needed to recruit them. AACR Project GENIE has shown the model at registry scale. The agentic layer brings it to query-time across the five-modality fabric.

TC-02

Synthetic control arms

Build externally controlled comparators from matched real-world cohorts. Across 2022 to 2024, externally controlled trials were a notable subset of the RWE-supported labeling-expansion approvals. FDA's 2023 guidance set the bar high. Producing a defensible cohort against that bar is what trial sponsors are looking for.FDA, Externally Controlled Trials, 2023

TC-03

Decentralized-trial enablement

Wearable-derived endpoints have moved from supportive to qualified primary — SV95C in Duchenne muscular dystrophy being the proof point. Continuous glucose in metabolic disease and cardiac-patch ECG in cardiology research are next. The Tier 3 layer is the operational substrate that makes these endpoints assemblable across trials, not bespoke per trial.EMA SV95C qualification 2023 · Diabetes, Obesity & Metabolism 2024

TC-04

Site selection

Where patients with a given phenotype-by-genotype combination actually receive care. Five-modality coverage at population scale answers this question with patient-grade precision, not zip-code aggregates.

TC-05

Post-marketing surveillance

The same platform that supported the approval can run the outcomes-based contract that follows. The CMS Cell and Gene Therapy Access Model and the Casgevy and Lyfgenia structures depend on durable outcomes capture at patient level. The substrate for the approval is the substrate for the post-marketing commitment.CMS CGT Access Model 2025

Key use cases

Three places the work shows up today.

UC-01

Pharma · Label expansion via RWE

A sponsor builds the dossier for a labeling expansion using a matched-cohort analysis or an external-control arm. The Phenopackets-compatible cohort, the molecular subset, and the outcome trajectory are assembled and audit-trailed on Evidence Lens. The scale is set by the share of FDA labeling-expansion approvals already citing RWE.

UC-02

Diagnostics · Clinical-utility evidence

A diagnostics business generating clinical-utility evidence for a novel test pairs the molecular result with downstream care decisions, outcomes, and resource utilization across the matched cohort. The submission goes in with linked evidence in place of narrative inference.

UC-03

Provider network · Patient-journey optimization

A provider network deploys the Patient-Journey Console at the point of care. Treatment choice is supported by the patient’s own full record and the matched-population RWE. The pattern is now visible in published response models from leading cancer centers, with the same shape repeating in cardiology, metabolic disease, and rare disease.

The bet

The patient-level join is the asset. The agentic layer is what makes it usable.

Regulators, payers, and pharma have moved in the same direction. The five-modality join at patient level is what each of them is asking for, and the vendor stack today does not yet hold it at scale.

The reading

Regulators, payers, and pharma have moved in the same direction over the last twenty-four months. The patient-level join across modalities is what each of them is asking for, and the vendor stack today does not yet hold it at scale. The category is converging on a single shape.

The integration posture

Evidence Lens does not replace the EHR, the claims platform, the sequencer pipeline, or the wearable ingestion. The integration cost is reading them, not migrating them. The deployment is additive to the data estate the institution already runs.

The compounding return

Each new patient cohort tunes the agentic layer. Each new sponsor question expands the synthesis library. Each new wearable stream extends the Tier 3 layer. The Clinical and Genomics Vector Stores compound across modalities as the patient population grows. The next question is answered faster than the last.

The window

The regulatory and commercial pulls are public. The five-modality join is still rare in 2026. The platform that holds it across pharma, diagnostics, provider, and trial use cases owns the next category step.

The risk

What we do not know.
What we will not do.

What we don't know

We don’t yet know how the agentic layer will perform on your specific cohort until we run it against yours. Cohort representation against benchmark populations, phenotype-enrichment accuracy on your notes and labs, evidence-synthesis citation fidelity on your records. Sprint returns a confidence range. Enable turns it into evidence on your data, or it doesn’t.

What we will not do

The agentic layer informs care. It does not decide care. The blueprint does not replace the EHR, the claims platform, the sequencer pipeline, or the wearable ingestion. Every clinical call stays with the clinician. Every commercial deliverable stays with the data-rights gate. Regulatory accountability stays with your team. The line does not move to us.

What we phase carefully

Tier 2 multi-omics and Tier 3 wearable joins ship after the data-rights, consent, and de-identification posture is on file per cohort. The re-identification risk is quantified per cohort, not assumed. We do not include identified or contested data in any external collateral until the legal pass is in place.

The engagement model

Prove the join first. Ship one use case next.

Three phases, starting with a defined Sprint slice. Sprint stands up the data fabric and the Clinical Vector Store on a sample of your records and proves cohort assembly against a published RWE benchmark. Provectus operators then join your RWE/HEOR or translational research team to ship one use case end-to-end. We scale only once the comparison favours the agentic loop on the locked scorecard.

01

Sprint

Weeks 1–2

Prove the join on your own patients.

  • Ingest a defined slice of your clinical (FHIR) and claims (OMOP) records under BAA. Genomic and wearable layers added where data rights and consent permit.
  • Stand up the Clinical Vector Store on the slice. Run Cohort Assembly against a published RWE benchmark question on your cohort.
  • Produce the back-test artifact: cohort assembly time, phenotype completeness, evidence-synthesis citation accuracy against ground truth.
  • Lock the shared scorecard for Enable: cohort delivery time, phenotype coverage, audit-trail completeness, persona-surface adoption.
  • Pick one use case for Enable — pharma label expansion, diagnostics clinical-utility, point-of-care, or trial cohort.

02

Enable

One use case

Ship one use case end-to-end.

  • Provectus operators sit on the RWE/HEOR or translational research crew through the cycle.
  • Tier 2 multi-omics layer joined at patient level for the chosen use case. Tier 3 wearable layer added where the use case requires it.
  • Persona surface activated for the audience — Research Workbench, Partner-Insight Surface, Patient-Journey Console, or Cohort Explorer.
  • Head-to-head against the prior workflow on the locked scorecard. Joint accountability for the outcomes, not just the software.

03

Realize

Cycle over cycle

Scale across use cases. Own the loop.

  • Expand to remaining use cases once Enable proves out on the scorecard.
  • Synthesis library compounds as each cohort adds to the patient evidence base.
  • Business outcomes tracked and reported: label-expansion submissions, clinical-utility dossiers, value-based-contract outcomes, trial-cohort delivery time.

Commitments

What we sign up for.

Bounded confidence

One use case in Enable. Head-to-head on cohort delivery time, phenotype coverage, evidence-synthesis citation accuracy, and persona-surface adoption. We scale only when the comparison favours the closed loop.

Confidence range, not promise

Sprint returns a measured range on what Enable would deliver on your data — cohort accuracy ceilings, time-to-cohort deltas, citation-fidelity bounds. The range is based on the patients we see. We commit to the range, not to numbers we cannot yet justify.

Designed for the loop

Every patient cohort tunes the agentic layer. Every sponsor question expands the synthesis library. Every wearable stream extends the Tier 3 layer. The Clinical and Genomics Vector Stores compound across modalities as the patient population grows. We design for that contract on day one.

Named trade-off

The blueprint sits above the systems of record — EHR, claims platform, sequencer pipeline, wearable ingestion — not inside them. We accept that constraint. It keeps the regulatory audit story clean and the adoption path short. We will not chase problems that belong inside the EHR or the sequencer.

Next Step

Let's start with one use case.

If this looks like the right shape for your evidence question, the next step is a Baseline Assessment. One to two weeks. Read-only against a defined slice of your clinical and claims records under BAA. We measure cohort assembly time, phenotype completeness, and evidence-synthesis citation fidelity against ground truth. Then we decide together whether to proceed. If the back-test does not support a strong business case, we say so openly.

Schedule a working session
Stepan Pushkarev CEO, Provectus