Axios Pro Exclusive Content

Exclusive: Preemptive AI launches biomedical AI model

Illustration of a phone wearing a doctor's coat and a stethoscope.

Illustration: Aïda Amer/Axios

PreemptiveAI, which is developing AI models to interpret biomedical signals, emerged from stealth with $6.4 million in funding, CEO Jamien McCullum tells Axios exclusively.

Why it matters: The company aims to interpret biomedical data from smartphones and wearable devices to help predict events like stroke or heart attack.

How it works: Preemptive's models measure photoplethysmography (PPG) signals, used for measuring vital signs, to provide physiological insights.

  • In its partnership with Duke, for example, discharged patients get a text from Preemptive AI prompting them to place their finger over their phone camera lens for 30 seconds. That biometric data is collected, analyzed and connected to a patient's EHR, and used to identify likely readmissions.
  • "Open AI is built on the language foundation model. We're about building up a biomedical foundation model, which is mapping human physiology," says McCullum.

Zoom in: Inspired Capital and Meridian Street led the seed round, with participation from the Allen Institute, AI2 and Precursor Ventures.

  • Preemptive also inked a year-long strategic partnership with Duke University, primarily focused on predicting hospital readmissions.

What we're watching: Preemptive AI hopes Duke will become an official customer, McCullum says.

  • The company also has "six or seven partnerships coming together right now" with digital health entities, including health tracking apps like fertility tracker Ovia Health.

Between the lines: Preemptive co-founder and chief science officer Leon Gatys was a founding member of Apple's AI division.

  • Gatys and his team "spent years training and building models" using Apple Watch's optical sensor and the blood flow signal derived from it, McCullum says, which was foundational to Preemptive AI's model building.

Context: Many PPG sensors and pulse oximeters are known to be less accurate when it comes to taking measurements on darker skin.

  • McCullum calls a pulse oximeter a "dumb device," noting smartphones offer variable controls like exposure to garner a better signal.
  • Preemptive AI's database has more than 35,000 patients worth of data, which McCullum says is highly diverse, and the company assessed the model on patients from Cameroon and India to validate efficacy on darker skin tones.

The big picture: Inspired principal Charlotte Ross predicts wearables will become more ubiquitous, but notes Preemptive AI's smartphone-focused approach was particularly significant for Inspired.

  • PreemptiveAI's model shows "you don't have to have the fancy wearable around your wrist or the Oura ring," Ross says.
  • "This is kind of a democratized way to access the same type of valuable data without having to spend a lot of money."

What's next: Preemptive AI and its investors expect to raise a Series A this year, McCullum and Ross tell Axios.

Go deeper