Vega Health, a startup helping health systems evaluate and deploy AI, is teaming up with Parkland Center for Clinical Innovation (PCCI) to license its AI models.
Five of PCCI’s AI models are now available on the Vega Health Marketplace, accessible to Vega customers. The models have been validated in a real-world hospital setting. Most of the models focus on clinical decision support, population health or the social determinants of health. Vega's goal is to elevate innovations that may otherwise go unnoticed.
“Part of what we want to do at Vega Health is bring to market a lot of these use cases that really wouldn’t be standalone companies, but would still have a big opportunity to improve patient care and population health,” Mark Sendak, M.D., co-founder and CEO of Vega Health, told Fierce Healthcare.
In 2012, PCCI was spun out of Texas-based Parkland Health, one of the country’s largest safety-net health systems. The ongoing collaboration between PCCI and Parkland focuses on identifying opportunities for AI and digital health, with a particular focus on the needs of vulnerable populations across North Texas.
The five PCCI models on Vega’s marketplace are:
- Inpatient Sepsis Prediction: Identifies patients at risk of sepsis in the next 12 hours in inpatient units, surfacing top clinical drivers of each prediction in the EHR.
- ED and Urgent Care Sepsis Present-on-Admission (POA): Identifies patients already septic at presentation in the ED or urgent care, triggering clinical alerts for intervention.
- Parkland Trauma Index of Mortality (PTIM): A predictive model that updates hourly to assess in-hospital mortality risk for polytrauma patients.
- Patients at Risk for Adverse Drug Events (PARADE): At admission, this model stratifies patients by risk of experiencing an adverse drug event during their stay, enabling pharmacist intervention.
- Workplace safety AI Model: This model screens admitted patients by identifying encounters most likely to proceed without a violent incident, drawing on EHR data, human resources records and social needs.
The models, which have been tested at Parkland, have achieved promising results thus far. For instance, the inpatient sepsis predictive model fires alerts to clinicians well before a patient needs antibiotics with the goal of early intervention and better outcomes. The model alerted clinicians, on average, 19 hours before typical antibiotic administration. That's compared to 1.5 hours before administration for current industry models, per PCCI. The alerts can be snoozed by clinicians as needed.
The trauma index correctly identified 89% of the high-risk trauma patients and 92% of the low-risk trauma patients. The adverse drug events model prevented over 2,000 events and delivered over $17 million in avoided costs at Parkland. And, the workplace safety model accurately predicted 77% of violent incidents within 30 minutes of admission.
Vega was spun out of Duke University, where Sendak was the former population health and data science lead at the Duke Institute for Health Innovation, with the idea of democratizing access to effective clinical AI models built alongside frontline clinicians. Besides curating the models on its marketplace, Vega helps customers with the work it takes to actually deploy them: evaluation and testing; workflow integration; finetuning each model to unique patient populations; and post-deployment monitoring. This is particularly important for under-resourced hospitals, Sendak explained.
“There are very few organizations with internal capabilities to build and implement their own tools built on their own patient data,” he said.
“We’re not set up to be a commercial entity,” Steve Miff, Ph.D., president and CEO of PCCI, told Fierce Healthcare. PCCI has a small marketing team and no sales team. “We have been looking for the right partners to be able to scale the impact of the work.”
Vega, launched in late 2025, is currently live with two community health systems that include critical access hospitals. It has revenue-share agreements with its AI supplier partners, which include Duke and PCCI for now. This offers a commercialization pathway for innovators.
Just because a model was developed within an academic medical center is not a guarantee that a model will be superior, Sendak acknowledged. “You don’t know which model is better performing until you test it,” he said. But the benefit of having an affiliated or in-house innovation arm is the relationship between the developers and clinicians, who share accountability.
In addition to Parkland, PCCI also works with the Dallas County Health Department, payers and other health systems. PCCI has 19 AI models fully deployed today. Since 2019, the models have identified nearly 3 million at-risk individuals for interventions.
Health systems interested in leveraging PCCI's models will work with Vega to evaluate them on their own local patient data before implementing them. The data is shared with the Vega customer and with the relevant AI partner. If the model works well, Vega supports clinical adoption and ongoing monitoring to track accuracy, adoption and real-world outcomes.
If it doesn’t work as expected, Vega would not recommend a hospital purchase that particular model, Sendak said. Vega’s goal is not to say which model is superior. It needs to be tailored to each institution and that’s why it’s also crucial for the models to have been trained on diverse populations.
“We are not trying to be a kingmaker,” Sendak said. “We want to help every health system find the model that works best for them.”
Both Sendak and Miff believe AI has a future in healthcare. “Healthcare is so complex, … there is no one physician or no one entity that has the top expertise in every single clinical domain,” Sendak said.
“AI is and will play a huge role in healthcare and we need AI to be able to augment what we do,” Miff echoed. But, he cautioned, administrative use cases are much more scalable and translatable between organizations. The complexity comes in when AI is being used for clinical decision support or population health management. That’s when models need to be co-developed with clinicians and tested in real-world settings.
“This is the hardest part, but also has the ability to make, clinically, the biggest impact,” Miff said.