4.2% SSI ortho / THR
9.8% Readmit bariatric / RYGB
1.1% VTE cosmetic / BBL
2.6% Reop spine / ACDF

See the complication.
Before the incision.

What it does

Four things every partner clinic and payor needs, packaged as one score.

Risk Engine turns the patient record, the procedure plan, and the network’s outcome history into a single calibrated output — and then keeps scoring, silently, through recovery.

Pre-op probability
A single calibrated number for 30-day complication risk, with a 90% confidence interval. No star rating. No guessing.
Driver decomposition
The top three to five factors pushing the score up or down — surfaced in plain clinical language, not SHAP values.
Category breakdown
Infection, VTE, readmission, reoperation, mortality. Individually estimated, rolled up when you need one number.
Longitudinal tracking
The score keeps updating from day −30 through day +90 as new vitals, labs, imaging, and check-ins land.
Risk Lineage

Every input that moved the score, with citations.

A per-patient timeline of every event that nudged the risk score up or down — with the source document one tap away. Surgeons get an explainable history. Underwriters get a defensible audit. Patients see the work the platform is doing on their behalf.

AR
Ana Ríos · 54F · Pre-op THR (R)
Risk Engine THR-MDE v2.4.1 · clinic Ortopedia Avanzada
8.4%At intake
3.1%Today · pre-op day −7
Apr 02
Day −66
A1c lab result · 8.4%
Imported via SMART-on-FHIR from PCP. First glycemic signal.
Source: PCP Lab · Quest Diagnostics
+1.2 ptsat intake
Apr 04
Day −64
Comorbidity confirmed · T2DM + HTN
Pulled from FHIR Condition resources; both flagged in Risk Engine driver model.
Source: Passport · Conditions
+0.6 ptsbaseline
Apr 10
Day −58
CGM enrolled · pre-op glycemic plan
Counterfactual workup activated for Lever 1. Plan: A1c < 7.0 by Day −7. Risk recomputed.
Source: Pathway · pre-op plan
−1.0 ptplan effect
Apr 18
Day −50
CGM data trending favorably · TIR 78%
Continuous glucose monitor 14-day rolling time-in-range up from 51%. Risk model accepts in-window glycemic signal.
Source: Apple Health · CGM stream
−0.8 ptsignal update
Apr 22
Day −46
Ferritin 32 ng/mL · pre-op iron deficiency
Triggered Lever 3 (iron repletion). Without treatment, transfusion risk rises.
Source: PCP Lab · Quest Diagnostics
+0.4 ptnew finding
Apr 28
Day −40
A1c repeat · 7.4% (down from 8.4%)
First confirmed glycemic improvement. Risk re-derived with revised input.
Source: PCP Lab · Quest Diagnostics
−2.6 ptssignal update
May 04
Day −34
Smoking cessation milestone · 14 days clean
Cotinine read at Day 14 confirms abstinence; pulmonary risk component lowered.
Source: Pathway · cotinine read
−1.4 ptsplan effect
May 10
Day −28
Iron repletion confirmed · ferritin 102 ng/mL
Lever 3 met threshold. Transfusion-risk component reverts to baseline.
Source: PCP Lab · Quest Diagnostics
−0.5 ptplan effect
May 14
Day −24
Prehab adherence · daily steps ≥ 4,000
Apple Health step count 14-day trailing avg. Pre-rehab target met.
Source: Apple Health · Activity
−0.4 ptplan effect
The clinic surface

The risk card your surgeon actually reads.

Change any input — procedure, age, BMI, comorbidities, functional status. The card recalculates in real time with probability, categories, Shapley-style drivers, and the recommended workup. This is the same surface that renders inside the AI Patient Brief before the case is committed.

Interactive prototype · v4.2.1

Launch the full clinical prototype.

Open the Risk Engine in a dedicated SaaS workspace — with pre-op scoring, a longitudinal monitor from day −30 through day +90, driver analysis, a workup engine, and a full audit trail. Twelve procedure families, eight comorbidities, recalibrated in real time.

  • 12 procedure families across 6 specialties
  • Pre-op score, longitudinal monitor, driver decomposition
  • Automated workup, override banner, full audit trail
  • Light and dark themes · keyboard-first workspace
Illustrative demonstration · model output not used for clinical decision-making
The longitudinal surface

One number that keeps updating — from day −30 to day +90.

The same score rides alongside the patient through pre-trip prep, surgery, and every check-in. When it moves, we tell the clinic and the patient what moved it — and what the next check is.

Illustrative patient · Sleeve gastrectomy · Medellín hub
Day −3 · Pre-op
9.1%
30-day risk
90% CI: 6.4% – 12.8%

What moved the score this week

  • HbA1c down to 6.2 on pre-op labs Targeted pre-op glycemic plan executed over 4 weeks.
    −1.6%
  • BMI 41.2 → 38.9 Pre-op nutrition plan · −2.3 kg/m² over intake-to-pre-op window.
    −1.1%
  • Smoking cessation confirmed (> 14 days) Self-report plus nurse-coordinator check.
    −0.8%
  • ·
    Cardiac clearance documented ECG & stress echo normal; anesthesia sign-off received.
    0.0%
  • Travel stress window opens Long-haul flight, time-zone shift, first 48h post-arrival.
    +0.2%
How it works

Three layers, one loop. No black box.

Risk Engine is not a standalone app. It’s a scoring layer threaded through every other JetPatient AI surface that already touches the case.

01
Ingest
Passport normalizes the patient record against FHIR R4 before scoring. Labs, imaging, comorbidities, and functional status all arrive as structured inputs — never free text.
  • Passport · FHIR-native intake
  • Ambient Scribe · structured history at point of care
  • JetVision · pre-op visual signals
02
Score
One gradient-boosted classifier per procedure family, calibrated on the cross-border network’s outcome graph. Each output includes a point estimate, confidence interval, and an audit trail.
  • Per-family models · ortho, bariatric, cardiac, spine, cosmetic, dental
  • Calibration against recent cohorts
  • Auditable, not a black box
03
Act
The score doesn’t hang in space. It renders inside the AI Brief for the clinic, inside the patient’s Navigator view in plain language, and into Guardian when a threshold is crossed.
  • Clinic · risk card inside AI Brief
  • Patient · simplified explanation in Navigator
  • Guardian · escalation on threshold breach
Model & data

Per-procedure models, not a single monolith.

Complication profiles for a hip replacement, a gastric bypass, and a BBL aren’t interchangeable. Risk Engine ships a family of calibrated models — each trained against the outcomes of the surgeries it scores.

Training substrate

Structured records from Passport (FHIR-normalized US histories), partner clinic EHRs, and post-op check-ins collected through Navigator. Every outcome is anchored to a concrete event: readmission, reoperation, SSI diagnosis, imaging-confirmed VTE, or mortality.

Guardrails

Risk outputs are framed as decision support, never a decision. A clinician signs every case. Patients see a simplified explanation — not raw probabilities — in Navigator. Model cards, data-sheet-for-dataset, and version-pinned audit logs are part of the deployment.

What’s explicitly excluded

No claims-to-claims inference. No inputs derived from insurance status, country of origin, or clinic pricing. Demographic features used (age, sex, BMI) are clinically justified and documented in the model card.

Target performance · v4.2 release gate

Held-out validation on 2024–2025 network cohort
On target
DiscriminationAUC, aggregated across families
0.83≥ 0.80 target
CalibrationBrier score, lower is better
0.071≤ 0.08 target
Cohort-level accuracyPredicted vs. observed rate, network-wide
±11%±15% target
Workup-change rate% of flagged cases where plan was modified
27%≥ 25% target
Clinician override latencyMedian time from surfacing to sign-off
2.8mmedian
OrthoTHR · TKR · ACL
BariatricSleeve · RYGB
CardiacCABG · TAVR
SpineACDF · lami
CosmeticBBL · TT · rhino
DentalFMR · implants
Who it’s for

Two audiences. One consistent score.

Partner clinics see the full decision surface. Employers see the actuarial rollup. Neither sees the other’s view — but the numbers reconcile.

For partner clinics
Catch the risk the imported chart didn’t flag.
Your surgeons already adjust pre-op plans informally. Risk Engine turns that adjustment into a documented, auditable signal — without adding a tool to the OR.
  • Risk card inside the AI Brief — no new login, no new dashboard.
  • Drivers in clinical language — “elevated HbA1c, 7.8” not “SHAP = +0.04”.
  • Recommended workup — pre-filled checklist that matches your protocols.
  • Post-op monitoring — check-ins auto-score and escalate to your team.
  • Your outcomes, your credit — clinic-level improvements flow into your Guardian score.
Target workup-change rate ≥ 25% · Median time-to-decision < 3 min
Partner apply →
For employers & payors
Model the complication tail, not just the sticker price.
Surgical claims live and die on the complication curve. Risk Engine exposes that curve at the population level — with defensible, calibrated probabilities.
  • Cohort risk rollup — expected complication rate for your covered lives, by procedure.
  • Claims-model inputs — plug calibrated probabilities into your actuarial model.
  • Steerage intelligence — see how risk shifts when you route to a higher-quality clinic.
  • Reconciliation — predicted vs. observed rates, refreshed quarterly.
  • No individual scores leave the clinical surface — rollups only.
Cohort accuracy target ±15% · Refresh cadence quarterly
Employer model →
Inside Guardian

The fifth pillar of a four-pillar platform.

Risk Engine is the scoring half of Guardian. Credentialing verifies the clinic. Outcomes track what actually happened. Risk Engine tells you what’s likely to happen next. Together they form JetPatient’s third-party quality layer — the trust substrate that cross-border cash-pay has always lacked.

Credentialing
Who’s allowed to operate — license, specialty, volume, accreditations · See credentialing →
Outcomes
What happened — tracked per clinic, per procedure, quarterly · See quality report →
Risk Engine
What’s likely to happen — pre-op and longitudinal, calibrated per family. You’re here.
Questions we get

Things clinicians, patients, and payors ask first.

Is this clinical decision support? Do I need to deprecate my own workflow?
No. Risk Engine surfaces decision input, not a decision. A clinician signs every case and can override the recommended workup without friction. We log overrides to improve the model, not to audit the clinician.
How is this different from ACS NSQIP or the ASA physical status classification?
Both are excellent for elective surgery in U.S. settings. Risk Engine is tuned for the cross-border cash-pay context — procedure families we operate in, travel as a time-varying input, and the specific recovery patterns the network sees. We don’t replace NSQIP; partner clinics that prefer it can continue using it, and Risk Engine will still surface the longitudinal component they can’t get from a one-shot tool.
What data does a partner clinic have to share?
Structured FHIR-normalized inputs: demographics, comorbidities, relevant labs, and the procedure plan. Imaging and operative notes are optional inputs; they improve the score but aren’t required. All data transfer is covered by the partner clinic data agreement and routed through Passport.
Do patients see the probability?
Not the raw number. Patients see a simplified explanation inside Navigator: tier (low/moderate/elevated/high), the top two drivers they can act on, and concrete prep steps. Showing raw probabilities without context tends to backfire clinically; we keep that discussion in the consult.
How often is the model retrained?
Quarterly on accrued network outcomes, with a version-pinned audit trail. Every prediction a clinician sees is tagged with the model version that produced it — so post-hoc audits can reconstruct exactly what the clinician was looking at.
What does this cost the clinic?
Risk Engine is bundled inside the partner clinic platform. There’s no per-score charge and no per-case margin on top of the procedure. Our incentive is outcomes across the network, not scoring volume.

Bring Risk Engine into your pre-op workflow.

Partner clinics get the risk card inside the AI Brief they already use, plus longitudinal monitoring for every case the JetPatient network routes to them. Employers and payors get a calibrated cohort-level view that plugs directly into the surgical-claims model.