A leading pediatric healthcare system is one of the nation’s largest, driven by a mission to make kids better today and healthier tomorrow. For the Manager of Advanced Analytics & Outcomes, the future of that mission is reliant on AI.
“We know artificial intelligence will play a big part in that tomorrow,” says the Manager. “Some of the tools we’ve built today are laying the groundwork to ensure safe and effective use of AI.”
The data science team at the system’s facility focuses on the highest-stakes applications. Their insights directly inform essential clinical decisions, such as using predictive models to forecast the onset of sepsis in emergency department patients.
The core challenge for their team is not just ensuring governance for these critical models, but empowering others to make data-driven decisions faster and at scale. Speeding up decision-making for front-line clinicians directly impacts the life and health of their patients.
Faster Decisions For a Healthier Tomorrow
Their Advanced Analytics team started out rooted in biostatistics, using a mix of proprietary tools such as SAS, Stata, and SPSS. As their scope expanded into operational data science, new use cases demanded the power and flexibility of R and Python. To provide fast, trustworthy, and well-governed insights, the team needed to move from ad hoc analyses and fragmented tool usage to a unified, enterprise-grade environment. For this critical transition, they selected Posit Team, which they have been using since 2021.
Speeding up Evaluation and Deployment of Artificial Intelligence
While their team builds custom models with R and Python in-house, they also vet models provided by vendors. Their Advanced Analytics Manager adds, “Before we can implement any of these machine learning models that may impact patients, we put them through a rigorous phase of evaluation.”
Models inform organizational and clinical decisions in various areas, including:
- Forecasting daily census for different units to ensure adequate staffing.
- Predicting the onset of sepsis in emergency department patients so timely antibiotics can be administered if needed.
- Identifying patients likely to need urgent escalation of care so they can receive earlier interventions.
Before users can see or act on predictions, dashboards developed in Shiny and Streamlit are critical. These dashboards help ensure the model’s performance is adequate, aid in selecting the appropriate threshold for eventual implementation in the electronic health record or other enterprise systems.
“In our work in artificial intelligence, our model evaluation dashboards have really quickened the time to insight and our ability to communicate.” shares the Manager.
Posit Team empowers their Advanced Analytics team to conduct real-time working sessions with key stakeholders, including clinicians and informaticists, enabling immediate decision-making about whether a project should proceed to deployment. By publishing interactive visualizations to Posit Connect, the team can rapidly demonstrate how various model versions perform, test different thresholds, and quickly iterate based on feedback – substantially speeding up the time from development to model deployment.
Using scheduled reports on Posit Connect, their team continuously monitors the model’s performance. Some of the solutions developed with Posit have been instrumental in allowing their team to stratify model performance across race, language, and ethnicity to ensure equitable results.