This article is sponsored by SAIVA AI. This article is based on a Skilled Nursing News discussion with Joanie Antico, Director of Clinical Services at Wachusett, Dr. Itai Schalit, CMO at SAIVA AI, and Tim Tarpey, Executive VP of Sales at SAIVA AI. The discussion took place on August 19, 2025 during the SNN RETHINK Conference. The article below has been edited for length and clarity.

Dr. Itai Schalit: CMO, SAIVA AI. Medical Doctor, PhD, Biomedical/Medical Engineering, BSc, Mathematics and Computer Science

Joanie Antico: Director of Clinical Services for Wachusett Healthcare. 

Tim Tarpey: VP of Sales, SAIVA AI

Tarpey: I often hear two things: “We need AI” or “We already have AI.” And my response is: if you have a calculator, you already have AI. The question to ask is, how can we improve? 

AI gives you opportunities to improve efficiency, clinical outcomes, admissions, and more. Technology can help you do what you already do, but better. Today we’ll use SAIVA as a case study, but this type of technology applies broadly.

Let’s start with definitions. What is Machine Learning, which SAIVA uses, and AI more broadly.

Schalit: AI is an umbrella term that refers to machines doing things that normally require human intelligence. As Tim said, it spans everything from calculators to self-driving cars.

Machine learning is a subfield of AI where an algorithm learns from data without being explicitly told what to do. For example, say you want to predict whether a patient will be readmitted to the hospital. You divide patients into two groups: those who were readmitted and those who weren’t. Then you feed their electronic medical records into the model and let it learn the differences.

That’s called supervised machine learning because the model is trained on labeled data. Once trained, you can feed in a new patient’s record, and the model can predict the likelihood of readmission.

Tarpey: What possibilities does this open up—especially with large language models?

Schalit: The biggest leap in machine learning recently has been in text analysis. We all know ChatGPT and other language models. Language is complex and hard to analyze with traditional methods. Natural language processing used to fall short, but large language models can penetrate much deeper.

In medicine, this is huge. Much of the EMR is text-based. Our research, textbooks, and medical literature are all text. If we can use machine learning to analyze this, the possibilities are endless. That’s the real breakthrough that has the potential to transform healthcare.

Tarpey: Great. So let me ask Joanie—how does Machine Learning help your team?

Antico: It gives us a very clear picture of what’s happening with our patients. It reviews the 24-hour report and identifies the top 15 patients at highest risk of hospitalization, along with the diagnoses driving that risk.

It organizes information in a way similar to an SBAR, but automatically and visually. Nurses save hours of time. They can communicate with physicians more clearly—without panic, without scrambling for vital signs or labs.

The system shows new changes, keywords from progress notes, recent orders, and even graphs of vital signs—all in one snapshot. It helps us teach nurses at the bedside: If we had done this yesterday, this patient might not be in the hospital today.

At one of our highest-hospitalization facilities, we cut transfers in half by using this tool in daily clinical huddles. I used to spend hours manually reviewing 24-hour reports and making a critical log. Now the system does it instantly, even across all my buildings.

Tarpey: You said it provides a concise picture—diagnoses and risk factors. Itai, will you explain how that works?

Schalit: Machine learning finds correlations in the data, not causation. That can be tricky. For example, suppose the model says a patient is at risk because they’re treated by Dr. Smith. At first you might think Dr. Smith is a bad doctor. But what if Dr. Smith is actually the top expert who treats only the toughest cases?

That’s why context matters. The data fed into the model shapes its output. In medicine, we have a lot of domain knowledge to guide us. If we build the model correctly, it will identify clinically relevant risk factors—not just statistically valid ones. That’s how we make AI both accurate and useful.

Tarpey: At first, we were told it wasn’t possible to explain why a patient showed up on the report. But now we have that explainability built in. Joanie, what difference does that make for you?

Antico: It’s huge. For example, if a COPD patient is flagged, the system shows a bar graph of their elevated respiratory rate, correlating with their diagnosis. That makes it clear why they’re at risk and helps guide next steps—like ordering a chest X-ray or starting nebulizers.

It helps experienced and newer nurses alike focus on what matters most at that moment.

Tarpey: What’s the right frequency for reports? Why 24 hours and not more often?

Antico: The picture changes hour by hour, but generating a morning report helps identify who’s most relevant at the start of the day. Supervisors then use it throughout the day, and we update in real time as things change. It’s not just a daily report—it becomes part of daily practice.

Tarpey: Exactly. It was built for morning huddles, but it can be run at shift change, multiple times a day—whatever’s needed.

Joanie, how has this tool improved communication with providers?

Antico: Nurses often know something’s wrong but feel unprepared when calling a doctor—missing labs, flipping between EMR tabs, sounding disorganized. Providers then lose confidence and send patients out “just to be safe.”

This tool organizes everything—vitals, labs, progress notes, recent orders—in one place. Even a newer nurse can present clearly: Here’s what’s changed over the last 24–48 hours, here’s what I recommend. That builds provider trust and keeps more patients in place.

Schalit: Exactly. When communication is unclear, the safest option is often to transfer the patient. AI helps systematize communication, reduce uncertainty, and give providers confidence to treat in place.

Antico: We’ve made it part of our process. Medical directors review the report daily now, nurse practitioners use it, and buy-in has grown as they’ve seen hospitalizations reduced.

Tarpey: Let’s talk about efficiencies. A customer once said, “There’s less clicking and more caring now.” Joanie, how has this helped your teams?

Antico: Nurses spend less time digging through notes and more time at the bedside. Supervisors work shorter days. I used to make my own critical logs; now the system does it. We even prevented an elopement because the system flagged behavioral notes that staff had missed. It saves time, reduces stress, and improves outcomes.

Tarpey: Speaking of outcomes—what have you seen?

Antico: Across our four buildings, we’ve seen reduced hospitalizations and better communication. But we also stress to staff: the system is only as good as the documentation. If nurses don’t record accurate notes and vitals, the AI can’t catch issues. We use those moments as teaching opportunities to improve assessments.

Schalit: Ultimately, outcomes are the crown jewel. But achieving them requires the whole chain: good data, clear analysis, strong communication, effective action. AI can help at each step—but only if integrated seamlessly into clinical workflows.

Healthcare is changing. We can either resist it, or we can lead the change. Done right, AI will improve outcomes for patients.

Tarpey: Last question: does focusing on objective data reduce defensiveness among nurses?

Antico: Reducing defensiveness is always tough—but yes, objective data helps. It shifts conversations away from blame and toward improvement.

SAIVA AI-powered platform combines deep clinical expertise with advanced machine learning to proactively detect early signs of clinical decline. To learn more, visit: https://saiva.ai/.