Use of AI tools in healthcare has been accelerating, with growing trust in results produced and delivered by such tools.

RAMP, one such tool already in action at UVA Health, focuses on delivering actionable, verifiable and explainable machine learning, integrating it as a decision support tool into clinical workflow to improve insights into patient health trends and facilitate faster delivery of necessary care, improving patient outcomes.

AI-driven predictive analytics models use complex real-time and historical patient data to provide healthcare professionals with actionable insights, and to alert care teams if the patient is in need of immediate attention.

Valentina Baljak is a senior data scientist at UVA Health. She holds a doctorate in information science and technology, applied machine learning. UVA Health created and uses RAMP today.

Baljak and two of her colleagues will discuss AI, RAMP and much more at HIMSS25 in March in Las Vegas in their session entitled "Real-Time Analytics Monitoring Platform: Usable AI in Action." We spoke with Baljak to get an understanding of what she and her colleagues will be talking about in the session and what HIMSS25 attendees can hope to take away from their talk.

Q. What is the primary theme you will address in your session and why is it relevant to healthcare and health IT today?

A. With the recent emergence of generative AI models, this topic is getting more traction in the healthcare field. In this work, we are focusing on clinical decision-making support tools in real time. Artificial intelligence isn't a new term.

At UVA Health we have been developing real-time predictive systems for several years now, and one of the biggest lessons we have learned is that the shape AI should take is the one that addresses your needs the best. Clinicians won't get behind tools they can't explain. Building trust in our models and tools meant close collaboration at every step, right from Day One.

We want to provide a blueprint on how to build a system that works in your environment, and to raise awareness to the importance of transparency, accountability and explainability of your models. This is especially important in the medical setting, with real-time predictions that can have a significant impact on patient outcomes.

Q. You will be focusing sharply on AI. How is it being used in healthcare in the context of your session's focus?

A. The key aspect of RAMP is real-time data collection from the EHR and other data sources. The ability to write results back to patient records in an EHR and alert care teams in real time makes RAMP a crucial tool in the clinical setting.

Technologies used here are fairly established and all open source. Python provides a solid basis for our ML development, back-end connectivity and data processing. Connections to various data sources are built with FiHR, REST API and custom HL7. The website is built with Angular.

As our latest major expansion, we are building a new predictive model on top of our largest real-time data stream, built with Kafka to collect all vitals and EKG waveforms from bedside monitors.

Q. Attendees will come to your session looking to bring knowledge home. What is one takeaway they can expect?

A. AI is a fundamental part of modern healthcare, taking different shapes depending on the need. Selecting the right AI approach is crucial, given the high stakes.

If you have in-house expertise and resources, then developing a custom AI system is a powerful alternative to vendor-provided black-box systems.

Valentina Baljak's session, "Real-Time Analytics Monitoring Platform: Usable AI in Action," is scheduled for Tuesday, March 4, at 12:45 p.m. at HIMSS25 in Las Vegas.