How a Medical Center Accelerates Clinical Research with AI

“Tasks that used to take hours or were impossible can now be calculated with the flick of a finger. But it’s not just about working faster, DataRobot enables us to do work we couldn’t before.”

Optimizing Patient Outcomes with the Experts Who Know Them Best

Every day, clinicians at University Medical Centre Mannheim set out to optimize patient outcomes. Part of the esteemed University of Heidelberg, the medical center sees more than 250,000 patients annually.

To deliver the best possible care, physicians and doctoral students strive to understand patient outcomes relative to medication and other factors. They know that the data holds answers, but they lack experience in modern data science techniques to unlock them. That’s what led them to augment their team with an AI platform.

“What excites me most about AI is the potential to find explanations for diseases or for therapy efficacies that we are too blind to see with our existing statistical methods,” explained Victor Olsavszky, PD Dr., Clinical Cooperation Unit – Healthy Skin and Joints, one of the primary teams using AI at the medical center. “We want to uncover new influences on disease progression, predict disease flares, and promote patient treatment adherence.”

Empowering Physicians and Researchers with Data Science

University Medical Centre expanded its research potential by partnering with DataRobot. For the teaching hospital, the DataRobot AI Platform offered the chance to quickly enhance their understanding of factors influencing disease and mortality across various clinical areas.

“The biggest selling point with DataRobot is it’s really easy for people without a data science background to pick up,” Olsavszky said. “You don’t learn data science in medical school, and DataRobot has been easy for us to understand.”

DataRobot’s AI capabilities empower users to analyze various datasets, build predictive models, monitor the health of those models, and govern them in a highly-regulated industry. Critically, DataRobot provides metrics on model accuracy, which the medical center needs to report clinical results or publish as part of their research. 

Understanding Treatment Outcomes

DataRobot’s intuitive user interface, plus expert data science guidance from the DataRobot team, allows clinical teams to tackle many use cases. 

The Clinical Cooperation Unit – Healthy Skin and Joints brings together the dermatology and rheumatology departments to look for links between the conditions they see, such as psoriatic arthritis. They consider patients’ symptoms, quality of life, and medication to understand what factors influence chronic skin and rheumatic conditions. Patients use a smartphone app to log their symptoms, creating a rich dataset to analyze.

The team taps into DataRobot, particularly Time Series, to analyze data from clinical studies with patients who are impacted by chronic eczema or psoriasis. Using DataRobot advanced Visual AI, they can factor in medical images.

With the platform, the team discovered findings that will influence how they treat patients going forward, including 29.6% of patients had an improved quality of life at six months, and change in therapy and disease activity at onset were the most influential features affecting quality of life improvement.

“We demonstrated the effectiveness of applying machine learning to existing data sets to discover new relationships we could not have uncovered using traditional regression analysis,” the published research paper states.

In another published research paper, the team highlighted how medication and lifestyle affect disease progression. For example, they learned that the use of certain medications or biologics showed a lower propensity for a therapy change at six months.²

“It’s hard to know which patients benefit the most from different therapies,” said Daniel Schaffert, a doctoral student and one of the primary DataRobot users. “That’s where DataRobot is helping. Tasks that used to take hours or were simply impossible can now be calculated with the flick of a finger.”

Pinpointing Cardiac Risks with AI

In another use case, the Department of Internal Medicine is looking at lipid metabolism relative to heart attack or heart disease risk. 

“We found laboratory markers that correlate with cardiovascular events. This could lead to medication coming soon to help lower the biomarker and decrease the risk of cardiovascular mortality,” Olsavszky said.

So far, the team has found DataRobot flexible for a wide range of use cases and is excited about the possibilities going forward.

“The fact that we can work with DataRobot on entirely different projects from different fields shows that this can be applied to even more areas and medical professionals should use it more,” added Igor Bibi, a doctoral student.

Ensuring Accuracy with Clear Documentation and Easy Validation

University Medical Centre clinicians find that DataRobot makes analytics accessible for the medical center. The platform helps identify the top-performing models and ensures they remain accurate.

“It’s not just about working faster,” Schaffert said. “DataRobot enables us to do work we couldn’t before.” 

“When we publish research papers, the most important things are reproducibility and documentation,” Olsavszky added. “The advantage with DataRobot is having triple validation of models and a plethora of metrics on the accuracy of them. They have been trained, retrained, validated, and revalidated, which helps us sleep better at night.”

The teams at University Medical Centre see their results as just the start of what’s possible for medicine.

“DataRobot has led us to predict insights that promise to improve patient outcomes. That experience is replicable across the entire healthcare sector. By adopting tools that are within our reach, we can collectively push the boundaries of what’s achievable in medical science.”

¹ Automated Machine Learning Analysis of Patients with Chronic Skin Disease Using a Medical Smartphone App: Retrospective Study https://www.jmir.org/2023/1/e50886/.

² Automated Machine Learning Predicts Necessary Upcoming Therapy Changes in Patients With Psoriasis Vulgaris et Arthritis And Uncovers New Influences On Disease Progression: Retrospective Study https://preprints.jmir.org/preprint/55855/accepted.