Clinical data science
Researchers in the Mayo Clinic Kern Center for the Science of Health Care Delivery's Clinical Data Science Program have expertise in a wide range of mathematical, statistical and machine learning methodologies. They lead the development and integration of solutions based in artificial intelligence (AI) and machine learning. Such solutions are a hallmark of the center's unique, practice-transforming work. The Clinical Data Science Program develops and guides the infrastructure needed for fast, efficient integration of data and information technology-based tools.
Focus areas
The center's Clinical Data Science Program is focused on:
- Developing, implementing, and evaluating AI and machine-learning models for clinical applications.
- Implementing AI algorithms into practice workflows.
- Bayesian modeling for complex problems.
- Designing evaluations for testing the effect of AI-based tools in practice.
Projects
These are some examples of research projects carried out by the Clinical Data Science Program.
COVID-19 and influenza-like illness modeling
From the earliest days of the coronavirus disease 2019 (COVID-19) pandemic, the center's clinical data scientists focused on identifying the best data sources while advising Mayo Clinic's internal data collection processes. They built new models and updated them daily. These models considered the rapidly evolving knowledge surrounding the SARS-CoV-2 virus' behavior as well as clinical observations describing COVID-19 infections.
In support of local, regional and state pandemic responses, the team developed a Bayesian Susceptible-Exposed-Infected-Recovered (SEIR) model. This model can predict COVID-19 cases and hospitalizations across the country. The program team worked with and made the model available to internal and external collaborators.
Related work led to a hospital census prediction model. The team later expanded this model to encompass numerous patient services across Mayo Clinic's hospitals and clinics in Arizona, Florida, Minnesota and Wisconsin. The expanded model can adapt to constantly changing infectious disease dynamics and uncertainties in human behavior. Furthermore, it can predict the effects of emergent therapies or vaccines based on possible levels of adoption or observed uptake.
In 2023, the focus of this work shifted to a more general influenza-like illness model. This allows researchers to account for other respiratory illnesses, such as influenza and respiratory syncytial virus (RSV). Program researchers translated many of the lessons learned and solutions developed from 2020 to 2022 into a robust framework for a wider reaching predictive model. This new model can predict infection activity level, patient census and staff absences due to any influenza-like infections.
Related publications:
Remote patient monitoring
Mayo Clinic studies new and unique ways to reliably deliver high-quality care and medical knowledge to people wherever they are. The Clinical Data Science Program leads in working AI into solutions that help clinicians personalize that care.
One example of this work is a pragmatic clinical trial evaluating clinical decision aids that use AI. The trial tests the decision aids' usability, clinical usefulness and effectiveness to help enroll patients in real-world remote monitoring programs after they leave the hospital.
The study team's findings will guide final improvements to the function and delivery of these models. This work will provide the critical groundwork necessary to introduce the aids across Mayo Clinic. The team's findings also will benefit people everywhere through peer-reviewed publication and knowledge dissemination.
Complex patient identification algorithm
Program researchers, with collaborators at and outside Mayo Clinic, are developing an algorithm based on health insurance administrative claim data. This algorithm will identify patients with complex disease who may need to see a specialized healthcare team at a tertiary medical center. The team plans to use the tool primarily to support the care of patients with complex and serious conditions.
Sometimes these patients "churn," or move around the healthcare system, in care patterns that are less than ideal. Ideally, the algorithm will improve continuity of care, reduce the time to develop correct diagnoses and treatment plans, and improve both experiences and outcomes for patients.
Control Tower: Innovation framework for patient care support
The Control Tower project provides a support tool for healthcare professionals in the inpatient setting. It is a framework that is built on Mayo Clinic's unified data platform and combines engineering, design, knowledge management and analytics abilities. This framework provides the elements needed to build both a physical interface and the AI to power proposed solutions.
The first proof-of-concept case for the innovation framework was developed and tested by researchers from the Mayo Clinic Kern Center for the Science of Health Care Delivery and Mayo's Center for Palliative Medicine. Researchers used machine learning techniques to develop a new and unique risk score, a real-time dashboard and alerts.
This tool predicts the potential need for a patient to receive palliative care support, allowing palliative care specialists to proactively offer recommendations. The tool has decreased the time for specialists to provide patients with palliative care consultations by more than 40%. Using the tool has reduced 60-day readmissions by more than 25%.
Related publications:
Contact