Treatment intelligence

Treatment intelligence for rheumatology.

TENSOR Med turns each patient's encounters, labs, treatments and outcome scores into one continuously updated report, with a reasoned, evidence-grounded next step that reflects everything in their record and refreshes as new results and letters arrive.

axSpARAPsAGoutSLEGCAPMR
Advisory. It never blocks a clinical decision.
S. Chen 34F · axSpA · HLA-B27+
Living report
3.4ASDAS-CRP
High activity
Two NSAIDs failed at max dose. Above the treatment target.
TB screen negative. FBC and LFTs current.
AI read · next step
Start a TNF inhibitor
3 of 4 analysts agree · confidence moderate
One dissent: confirm NSAID adherence first.
ASAS/EULAR 2022ABILITY-1ATLAS
Record · behind the summary
12 FebASDAS-CRP3.4
08 NovASDAS-CRP3.2
04 NovNaproxen 500 mg BDongoing
18 JanQuantiFERON (TB)negative
18 JanFBC, LFTsnormal
Source · letter vs extract
Clinic letter
Extracted
Drug Naproxen
Dose 500 mg BD
ASDAS-CRP 3.4
TB screen Negative
Why it exists

The information is already there. Reconstructing it is the work.

Nearly half the working day now goes to the record and desk work, and much of that is retrieval, not documentation: pulling together results, letters and prior decisions that already exist but sit in different systems. TENSOR Med assembles them into one place and keeps it current.

Time-and-motion data: Sinsky et al., Annals of Internal Medicine, 2016.
~2:1
Record to patient time. About two hours on the record and desk work for every hour of direct patient care.
~49%
Of the working day spent in the record and on desk work, rather than with patients.
How it works
The living report

Each patient is a single report with four levels of depth.

From a quick read to the original source. Most visits live in the first level; the rest is one click away when you need to check the working.

1Dashboard
Current disease activity and how it has moved, position on the guideline pathway, monitoring due or overdue, and active safety flags. This is what most visits need.
ASDAS-CRP3.4
At target: no2 NSAIDs failedReview due
2AI read
Four analysts review the whole record and return a next-step recommendation, with the reasoning and sources behind it, refreshed as new data arrives.
Start a TNF inhibitor
3 of 4 agreeModerateASAS/EULAR
3Record
Every encounter, lab, treatment and score behind the summary, so nothing is taken on faith.
12 FebASDAS-CRP3.4
04 NovNaproxen 500 BDongoing
18 JanQuantiFERONnegative
4Source documents
The original letters, PDFs and imported files, shown beside exactly what was extracted from them.
Clinic letter
Extracted
Drug Naproxen
Dose 500 BD
ASDAS 3.4
How the AI works

Watch it reason about a patient.

The same case, run end to end. Four analysts examine a de-identified record through different clinical lenses, respond to each other, and a moderator reconciles them into one recommendation.

1De-identify 2Analyse 3Discuss 4Recommend
Patient record
PatientSarah Chen · MRN 4471982[name] · [id]
DiagnosisaxSpA · HLA-B27 positive · 34F
Current therapyNaproxen 500 mg BD, 6 months
Prior NSAIDIbuprofen 600 mg TDS (failed)
ASDAS-CRP3.1 → 3.2 → 3.4
ScreeningTB neg (18 Jan)[date] · bloods current
De-identified Names, identifiers and exact dates removed before anything reaches the model.
Four analysts
Clinical
Treatment response
analysing
Naproxen 6 months. ASDAS-CRP 3.1 to 3.4, above target, not improving.
Guideline
Pathway position
analysing
Two NSAIDs failed, ASDAS ≥ 2.1. Meets ASAS/EULAR criteria to start a biologic.
Evidence
Trial relevance
analysing
Bio-naive axSpA on a TNFi: ASAS40 ~40 to 60% vs ~25% placebo (ABILITY-1, ATLAS).
Safety
Monitoring & risk
analysing
TB negative, bloods current. No contraindication. Confirm NSAID adherence first.
Discussion, round two
SafetyBefore a biologic, worth confirming the NSAIDs were actually taken as prescribed.
GuidelineBoth were at maximum dose for over four weeks, so the escalation criteria are met either way.
ClinicalSix months with no fall in ASDAS supports moving now, with adherence checked in parallel.
Moderator reconciles
Recommendation
Start a TNF inhibitor
3 of 4 analysts agree
One dissent: the safety analyst advised confirming NSAID adherence first. Confidence: Moderate.
ASAS/EULAR 2022ABILITY-1ATLAS
Illustrative worked example. Not a real patient. Criteria and response rates are drawn from the encoded ASAS/EULAR guideline and trials.

How they reach a recommendation

The analysts do not just vote. Each proposes a next step, then responds to the others, the way colleagues would around a table, and a moderator reconciles the discussion into one recommendation. When three or four agree, it is presented as a clear recommendation; when they split, the report shows each option, names the analyst who argued for it, and lowers its confidence. When the case is genuinely unclear, the moderator prefers the safer, more reversible course, and keeps a working therapy in place unless the record gives a reason to change it.

Specific to the patient

Before the analysts start, they receive the encoded guidelines, drug information and trial evidence for that disease, plus a structured summary of this patient: their trends, what is overdue, prior decisions, and constraints such as pregnancy plans or funding. Its key claims point back to their source, whether a guideline step, a named trial, or a value in the record.

Cited and traceable

The reasoning is laid out step by step, with its evidence.

Each analyst breaks its conclusion into a few steps. Every step is one claim about the patient, and every claim carries its evidence: a guideline, a trial, or a value in the record. Open a citation to see the source it rests on.

Guideline analyst · reasoning
  1. 1

    Two NSAIDs, naproxen and ibuprofen, have been taken at maximum tolerated dose for over four weeks without adequate response.

  2. 2

    ASDAS-CRP is 3.4: high disease activity, above the treatment target.

  3. 3

    The ASAS/EULAR criteria to start a first-line biologic are met: at least two NSAIDs failed, and ASDAS ≥ 2.1.

  4. 4

    In bio-naive non-radiographic axial SpA, a TNF inhibitor reached ASAS40 in 36% versus 15% on placebo.

  5. 5

    No contraindication to a TNF inhibitor is recorded; TB screen is negative and baseline bloods are current.

    Gap: chest X-ray and a pre-start pregnancy test are still outstanding.
Select a citation to open its source.
Guidelineverified
ASAS-EULAR recommendations for axial SpA, 2022 update
Escalation threshold: ASDAS ≥ 2.1 or BASDAI ≥ 4, after at least two NSAIDs over four weeks at maximum tolerated dose.
Opens the passage in the source viewer · PMID 36270658
Randomised trialverified
ABILITY-1: adalimumab in non-radiographic axial SpA
At week 12, ASAS40 was reached by 36% (36/91) on adalimumab versus 15% (14/94) on placebo; p<0.001.
Opens the passage in the source viewer · PMID 22772328
Patient recordthis patient
The value in the patient's own record
ASDAS-CRP · 12 Feb3.4
Jumps to this value in the patient's Data tab.

Verified against the source

Every citation is checked against the corpus. Ones that resolve are marked verified; a citation that cannot be resolved, or whose quote does not match the passage, is flagged, so a weak reference stands out rather than passing silently.

verifiedweak supportnot found

Reconciled with the record

Every score the narrative mentions is compared with the recorded value. A drift is corrected to the record before it reaches you, and a contradiction, such as continuing a drug the record shows had failed, is flagged for review.

Back to the document

The original letters and PDFs sit beside exactly what was extracted from them. Document access is time-limited, and every view is logged.

Built-in safety

Safety checks that run on every patient.

Before a recommendation reaches you, the record is checked against the things that cause harm, using the same guidelines you follow.

Pre-biologic screening
BSR 2019 · a biologic is held until the required screens are current
On hold: 2 outstanding
TB screen (IGRA / QuantiFERON)negative · 18 Jan
Hepatitis B & C serologycomplete
FBC · U&E · LFTcurrent
Chest X-raynot recorded!
Pregnancy test (β-hCG)required before start!

Interactions and contraindications

Every active drug is checked against an encoded interaction set, with brand names resolved to the molecule. It catches combinations such as allopurinol with azathioprine, methotrexate with trimethoprim, and a live vaccine on a biologic.

Pregnancy and teratogens

Teratogenic drugs are checked against pregnancy and conception plans, with the washout each one needs, methotrexate 90 days, leflunomide up to two years, and a note when a β-hCG is out of date. Pregnancy-compatible options such as certolizumab are surfaced.

Flares and overdue monitoring

A clinically important rise in disease activity is flagged as a possible flare, and each drug's monitoring cadence is tracked, so an overdue FBC or LFT surfaces on the dashboard.

Advisories never block you

Safety concerns are shown, not enforced. Dismissing an urgent one takes a documented reason, which is kept in an audit trail the whole practice can review.

What it knows

An encoded, versioned knowledge base, per disease.

The analysts reason from the actual guidelines and evidence for the disease in front of them, not from general knowledge, so the same patient always produces the same sources.

Guidelines as pathways

The algorithms, as real steps

Each disease carries its treatment algorithm as the actual escalation steps and thresholds, not prose. For axSpA that means the ASAS/EULAR ladder, down to two NSAIDs failed and ASDAS ≥ 2.1 before a biologic.

ASAS/EULAR · axSpAACR/EULAR · RAGRAPPA · PsAEULAR · SLEACR · GoutGCA / PMR
Drug knowledge

Down to the brand

Mechanism, monitoring, contraindications, and the difference between an originator and its biosimilars. Humira, Amjevita and Hulio are not treated as interchangeable, because immunogenicity and cost genuinely differ.

MechanismMonitoring scheduleContraindicationsOriginator vs biosimilar
Validated scores

Computed from the record

Disease-activity and organ-function scores are calculated from the underlying values, not read off. ASDAS, BASDAI, BASFI and DAS28, with eGFR (CKD-EPI 2021) and Child-Pugh.

ASDAS-CRP / ESRBASDAIBASFIDAS28eGFRChild-Pugh
Trial evidence

Grounded in numbers

Response rates from the trials that apply to each drug and disease, with their placebo comparators, so an expectation rests on a number rather than an impression.

ATLASABILITY-1COAST-VMEASURESELECT-AXIS

The patient in context

Guidelines are only useful against a full picture, so before every analysis the system also gathers the parts of the record that are easy to lose between visits.

Lab trends and what is overdue against its expected interval.
Symptoms that are active and persistent versus resolved.
Reproductive status, where it bears on prescribing.
Drug levels and anti-drug antibodies, kept separate from routine labs.
Coverage and funding, so a recommendation is one the patient can access.
Prior decisions, what you chose last time and why.
Advisory clinical decision support, not a medical device. It never blocks a clinical action, and the judgment stays with you.