Built for calibration
The goal: probabilities that mean what they say, tested before anyone relies on them.
IN DEVELOPMENT · PRE-VALIDATION
The crash has a signature.Aurora is being built to read it early.
Forecast, not hindsight. Aurora is being built to read the ICU’s earliest signals and hand back the one thing no unit has enough of: time.
THE DECISION STAYS WITH THE CLINICIAN
DEVELOPMENT TARGETS — NOT VALIDATED RESULTS
up to 48 hrs
AKI — DESIGNED FORECAST WINDOW
6–12 hrs
SEPSIS — DESIGNED FORECAST WINDOW
Explainable
Every forecast shows its drivers. An advisor, never a black box.
Validated before use
Not for clinical use until it has been validated.
RISK FORECAST
ILLUSTRATIVE — NOT CLINICAL DATA
Illustrative: risk as observed up to now, then the forecast ahead of it. The forecast crosses the intervention threshold later; the time between now and that crossing is the hours of warning.
The published evidence Aurora builds on.
INDEPENDENT RESEARCH — NOT AURORA’S RESULTSAurora is in development and pre-validation. These figures describe published research on comparable approaches; they are not Aurora’s performance, and Aurora is not a certified medical device.
- 0.825 AUROC1
- pooled across 73 studies of machine-learning sepsis prediction, ahead of traditional scoring systems
- 0.87 AUROC2
- a gradient-boosting model predicting stage 2 acute kidney injury within 48 hours
- 90.2%3
- of acute kidney injuries that later needed dialysis, predicted up to 48 hours ahead, with 2 false alerts for every true one
- 1. Yadgarov et al., Frontiers in Medicine, 2024. Systematic review and network meta-analysis. doi.org/10.3389/fmed.2024.1491358
- 2. Koyner et al., Critical Care Medicine, 2018. doi.org/10.1097/CCM.0000000000003123
- 3. Tomašev et al., Nature, 2019. doi.org/10.1038/s41586-019-1390-1
Follow Aurora from first principles to the bedside.
OPENS WHATSAPP · dr.jys@xumanai.com