The Cleveland Clinic Center for Clinical AI aims to change the course of how we predict diagnosis, prognosis, and treatment of patients
Launched in March 2019, the Cleveland Clinic Center for Clinical AI aims to develop innovative clinical applications of AI, leveraging machine learning technology to improve healthcare delivery in diagnostics, disease prediction, and treatment planning.
Under the leadership of its formal director, oncologist Dr. Aziz Nazha, the Center attracted international attention and started 20 research projects in its inaugural year.
The Center utilizes AI to analyze as many as 1.5 million patient records to better diagnose certain types of cancer, forecast the risks of patient readmission within 30 days of discharge, predict whether patients would respond to interventions including the oncology drug azacytidine and refine chemotherapy treatments. The Center encourages collaborations and communications between physicians, researchers, and data scientists and provides programmatic and technical support for other AI projects that are taking place within Cleveland Clinic. The aim is for it to be a melting pot where people come together to share ideas and develop models that will benefit patients and decrease healthcare costs.
The result was a personalized prognosis prediction model that’s specific for the survival of each patient. The research team took clinical and mutation data from a combined cohort of 1500 MDS patients from Cleveland Clinic and the Munich Leukemia Laboratory and used the information to train a machine learning model.
To open the AI Blackbox, the research team asked the algorithm to highlight clinically relevant variables affecting a patient’s outcome. This also enables the team to identify new variables that they can’t pick up in traditional scoring systems to better reflect the biology of MDS. The machine learning model was subsequently validated in a completely independent cohort of more than 800 patients from Moffitt Cancer Center and MDS Research Consortium.
The research team found that the accuracy of the machine learning outperformed the traditional scoring systems, including the International Prognostic Scoring System (IPSS) and Revised IPSS (IPSSR), in terms of predictability of the overall survival and cancer transformation. Nevertheless, they believe the personalized prognosis prediction model remains a relatively novel concept as people often talk about personalized medicine and treatment and they hope this notion will become more generalized in the coming years.
Unlike other similar efforts, the Center also bridges the gap between creating AI models and the actual implementation and reproducibility of these models. For example, a web application was built to facilitate fellow physicians in using the personalized prognosis prediction model, to input the clinical and mutation data of patients to obtain specific survival rates as well as their survival probabilities at different time points. Right now, the Center is working on incorporating this personalization into the workflow of more physicians.
“Some say if you sprinkle a little AI on your jacket, it becomes a cool jacket,” Dr. Nazha jokes “We are here not because AI is cool, but because we believe it can advance medical research. We want to build a new generation of patient-oriented physician-data scientists. Healthcare data is different from banking data. We can’t just say to physicians, we have this great algorithm, and we want you to use it. It’d be disastrous. That’s why we program everything in-house and build those projects from scratch. We want to inspire the world to work with us and witness the birth of AI-driven models that are building the future of healthcare.”