A site query can only return what routine care recorded.
Recruitment does not begin inside your protocol. It begins in ordinary clinical records, before a single protocol-mandated test has been drawn — where the urine albumin result is frequently absent and the creatinine looks unremarkable.
Many site queries are restricted to documented diagnoses and the kidney measurements already in the record. We read the routine panels that were already drawn, and surface candidates such a query cannot return.
The Problem
A creatinine threshold reads only one of the two core measurements in CKD evaluation
KDIGO evaluates CKD on filtration and on albuminuria. A rule built on creatinine or eGFR reads the first measurement and cannot read the second.
The measurement gap
Urine albumin is an excellent early marker — a patient with a normal eGFR and a raised albumin-to-creatinine ratio already meets the definition of CKD. The problem is not the marker. The problem is that for most at-risk patients, nobody ordered the test, so that half of the definition is simply absent from the record.
SourcesApplications
One platform, five evidence paths
The same routine-panel inputs, deployment architecture and IP support five applications for sponsors, across recruitment, progression, safety and on-treatment decisions. Each needs its own model and its own evidence, so each carries the stage it has actually reached.
- STRATASurfacing the candidate population for kidney trialsOpen for sponsor collaboration
- CADENCESerial panels as a trajectory — validation study designedValidation study designed
- SENTINELRenal risk characterization in post-approval populationsIn development
- CALIBERRenal trajectory across treatment cyclesIn development
- BRIDGEWhether the routine-panel signature crosses speciesEarly research
STRATA
Open for sponsor collaborationCohort construction for kidney and cardio-renal programs
In kidney and cardio-renal trials, much of the recruitable population is invisible before you can rank anything: most adults with CKD do not know they have it, so they are absent from the cohorts a conventional site query returns. STRATA reads the routine panels a health system already holds and surfaces a candidate population for confirmatory review, including patients with no CKD diagnosis code and no urine albumin result on file.
Site feasibility and cohort sizing
Estimate the candidate population a site actually holds before you select it, rather than after.
Recruitment yield
Bring the undiagnosed and untested population into the funnel, instead of recruiting mainly from patients who already carry a diagnosis.
Screen-failure research
Albuminuria entry criteria are a major source of screen failure in kidney outcome trials — the test is often not on file, and varies between measurements in the same person. Whether routine-panel prescreening changes that is a study outcome we want to test with a sponsor, not a result we are claiming.
Where this stops: STRATA is not an eligibility determination and not a general trial-query platform. It is kidney-specific probabilistic prescreening for the case where diagnosis codes and urine testing are incomplete. Protocol-mandated confirmatory testing still governs enrolment.
Clinical operations · Feasibility · Clinical development · Medical affairs
CADENCE
Validation study designedLongitudinal renal trajectory
Serial panels across time carry information that no single draw does. CADENCE is the progression-model validation program — a separate model from STRATA, to be trained against a locked longitudinal outcome. The validation study is designed and the analysis plan is defined; it has not run. Until it runs, CADENCE cannot support claims about event rate or sample size.
Where this stops: Not offered as a trial endpoint. Regulatory acceptance of a novel renal endpoint is a multi-year, consortium-level process, and no sponsor should put a seed-stage company's proprietary endpoint into a registrational trial.
Biostatistics · Epidemiology · Clinical development
SENTINEL
In developmentReal-world renal safety on a marketed asset
After approval, patients on a nephrotoxic or cardio-renal asset live in ordinary care, where labs are sparse and nobody is ordering a urine albumin test. SENTINEL will evaluate whether serial routine panels can surface renal risk patterns in those populations, to support risk characterization on a marketed compound.
What has to be true first: A retrospective exposure cohort with serial routine panels and adjudicated renal outcomes, on a marketed compound — sponsor data or a governed health-system extract. The analysis is already specified; obtaining that cohort is the remaining requirement.
Scope: Scoped to post-approval populations in routine care, not to Phase 1/2 dose-escalation support — and not a replacement for the regulator-qualified kidney safety biomarkers that exist for early-phase drug-induced injury. Those assays report tubular injury directly; this reads routine chemistry.
Drug safety · Pharmacovigilance · Real-world evidence · Medical affairs
CALIBER
In developmentRenal function for dose and eligibility decisions
Platinum-based chemotherapy and several antibody-drug conjugates gate eligibility and dosing on creatinine clearance. In oncology patients a single serum creatinine is a volatile basis for that decision: sarcopenia, hydration shifts and intercurrent acute kidney injury all move it, so a threshold decision can rest on a value that does not reflect durable renal change. CALIBER would ask, retrospectively, whether the routine panels already drawn each cycle give a steadier basis for that decision than the single value does.
What has to be true first: A de-identified retrospective dataset with serial per-cycle chemistry and the recorded dose or eligibility decisions — completed trial lab data or an oncology health-system extract.
Scope: Not a dosing recommendation and not a substitute for creatinine clearance, which is what protocols and labels specify. Any output would inform clinical judgment about whether a single value reflects a durable change, inside the existing decision rather than around it. Distinct from SENTINEL: patient-level and on-treatment, on dense per-cycle chemistry.
Clinical pharmacology · Drug safety · Oncology clinical development · Medical affairs
BRIDGE
Early researchPreclinical translation
Whether a routine-panel signature has any translational analogue in preclinical species, and could contribute to a translational safety screen. This is the earliest thing on our roadmap and the one furthest from evidence; we list it because it is a question we want to work on with the right collaborator.
What has to be true first: A collaborator with paired preclinical and clinical chemistry panels, and a species-comparison design that a translational safety group would accept. That is a research partnership, not a service engagement.
Scope: No study is running and nothing here is offered as a preclinical assay.
Preclinical safety · Research collaborators
Study Concept
What a first study looks like
Not a pilot and not a procurement. A retrospective comparison on records an institution already holds, specified before anything is run.
The question
Whether model-assisted review of routine panels would surface candidates for confirmatory kidney evaluation that a standard EHR query does not return — and at what review cost.
Minimum dataset
Routine adult panels with basic demographics and history flags, read retrospectively. The records stay where they are: the dataset is defined, held and governed entirely by the institution. No new draw and no prospective collection; consent or waiver is determined by the institution and its IRB.
Comparator
A standard EHR query — diagnosis codes plus whatever lab thresholds are on file — run against model-assisted discovery on the same records. The arms differ only in how the record is read.
What it would report
- Incremental candidates surfaced over the standard query
- Confirmatory yield among reviewed candidates
- Number needed to review
- Subgroup consistency
- Workflow burden on the reviewing team
Review burden: Review effort would be bounded by protocol: a defined chart-review sample, fixed before the analysis is run, rather than the full surfaced list. Everything above is study design. Nothing in it is a result, and none of these measures has a value we can quote yet.
Model-to-data by design
The shape of a governed retrospective study
This is the form a STRATA engagement would take. The analysis travels to the data, institutional approvals define what it may read, and only aggregate outputs come back.
- 01Retrospective
Sponsor research question
The engagement opens on a question about data that already exists. The question and its analysis plan are specified before any record is touched.
How many candidates does routine data hold that a site query misses?
- 02Approvals
Institution-approved dataset
Health-system data never becomes available to a sponsor without institution-specific IRB, data use agreement, privacy and governance approval. Those approvals define the dataset before the analysis is written.
- IRB
- Data use agreement
- Governance sign-off
- 03Architecture
Model-to-data analysis
The analysis goes to the data rather than the data to the sponsor. Inference would run inside the institution's environment as a query-only container: patient-level data does not leave it, and the model is not handed over as weights.
Stated as the deployment architecture — the design an engagement is built to, not an installation running today.
- 04Deliverables
Aggregate outputs only
What returns to the sponsor is aggregate. No patient-level data is transferred, and anything leaving the environment passes the institution's own release gate.
- Aggregate report
- Attestation
- Gated export
Existing risk equations answer: given a patient already known to have kidney disease, how fast will it progress?
We begin upstream: in a patient nobody has flagged, whose creatinine reads normal and who never had a urine test, should this person be prioritized for confirmatory kidney evaluation at all?
Different question, different population. We are not arguing that our model is better than a progression equation — we are pointing at the patient who never reaches one.
One Platform
One platform, whichever question you arrived with
Health systems and trial sponsors ask different questions. Each application requires its own model and its own validation, but all of them are built on one platform — the same routine-panel inputs, the same deployment architecture, the same IP. These are its properties in both settings.
Routine inputs, already on file
The model reads values already present in a routine adult workup — drawn and paid for as part of ordinary care. No new assay, no new visit, no new hardware.
Model-to-data by architecture
Institutional deployments are built so inference runs inside your environment as a query-only container — patient-level data does not leave it, and the model is not transferred to you as weights.
Patent-pending technology. Provisional patent filed March 2026.
Evidence
Where the evidence stands
Stated plainly, because it is the first question a safety or biostatistics group will ask.
Retrospective feasibility demonstrated on a US adult clinical cohort. Independent external validation is in progress and is not yet complete. Detailed performance characteristics, cohort provenance, subgroup analyses, calibration and decision-curve results are available under a mutual confidentiality agreement.
What we do not claim
We are not claiming to read creatinine better than creatinine, and we are not claiming to predict progression to a renal endpoint — that is a separate model and a separate validation program. What we are testing is narrower: whether a routine panel can surface risk that a creatinine threshold cannot return, in patients who have never had a urine test. Our studies are designed to examine that specific question, and until they read out we will describe it as a hypothesis under test.
Getting Started
Three shapes a first engagement can take
All three are research, not procurement. We would rather be useful on a real asset than general about a category.
Retrospective look-back
We run the model against a de-identified cohort from a completed or ongoing study, and report what the routine-panel signal would have surfaced.
What it settles: Whether the signal exists in your data, on your compound, before anything is committed.
Design consultation
We work with your clinical-development or feasibility group on specifying routine-panel prescreening for an upcoming protocol.
What it settles: Whether it is implementable inside your operating model.
Research collaboration
A jointly designed, co-authored study, structured so the method stays with us and the data stays with you.
What it settles: Both — and it is designed for co-authorship and publication.
If there is a program where the kidney question is live, that is the conversation we want.
We are early, we are specific about what is and is not yet validated, and we would rather start with a retrospective look-back on your data than a slide about a category.