top of page

Thanks for alighting here.

We're honoured you'd like to know more

Here's the skinny on our Predictive AI and a little about our UK patent

A typical  TrackMyTac display with patient data in the left panel, the  next predicted Tac value in the centre and the measured and predicted Tac values on the right hand chart. Read the text below for more about that red line!

Tacrolimus level prediction using Artificial Intelligence

​

  1. Ideally, we'd use Generative AI to make the Tac level predictions that we need for our Intelligent Assistants.

  2. But there's simply not enough data. The largest reliable supplier of kidney and liver transplant data has complete data for around 180,000 patients meaning that around 4 million data points (mostly US based) are available.

  3. It’s at once apparent that we can’t use Generative AI to predict Health-AI’s patient Tac concentrations. We’d need 5,000 times the data that we have, just to start at ground zero. 

  4. Instead we use Predictive AI. This uses historic data to predict the next data value in a sequence. 

  5. AI is really a set of different predictive tools and from the  AI tool box we’ve chosen Predictive AI and adapted it to predict a patient’s next Tac concentration. In fact, we have a UK patent on our adaption.

  6. Simply put, our Predictive AI says: if you have an idea of what the curve or line that you're trying to predict should look like, use this shape to influence your next individual prediction.

  7. Look at the red line in the TrackMyTac example. The blue dots around it are actual Tac readings and the green dots predictions. As a new Tac level becomes available (blue dots) it's compared with the red curve and a new prediction is generated (green dots). The process repeats and gradually the predicted and actual values become closer. and closer. This is usually referred to as "Bayesian Prediction" and it's widely used in Predictive AI.

  8. How did we make that red line? From the 18,000 data sets that we have! They're siloed by patient weight and age and the one most closely matching a patient's age and weight is used to make it.

  9. And that's all you need to know about our AI. It's smart, accurate, incorporates our predictive know-how, is patented and produces predictions quickly. Brill!

Thanks for reading! 

Click on Home to return to our landing page.

bottom of page