Photo (L-R): Rahul G. Krishnan (Assistant Professor, Computer Science and Laboratory Medicine and Pathobiology, Faculty of Arts & Science, Faculty of Medicine, University of Toronto and Faculty Member, Vector Institute); Mamatha Bhat (Assistant Professor, Division of Gastroenterology, Temerty Faculty of Medicine, University of Toronto and Clinician-Scientist, Multi-Organ Transplant Program, University Health Network)
By: Cormac Rea
Few experiences inspire panic and fear as much as a time spent in a hospital waiting to be seen for a serious medical procedure.
Yet, despite ongoing advances in medical science and modelling, patients often remain dependent on limited assessments and data modeling to determine if they even qualify for certain medical interventions.
Liver transplantation is a critical intervention for patients with end-stage liver disease. But current systems for prioritizing patients on the transplant waitlist create inequities, particularly for women, older patients, and those with some advanced conditions like non-alcoholic steatohepatitis (NASH) or cholestatic liver disease.
Supported by a Data Sciences Institute catalyst seed grant and co-led by investigators’ Rahul G. Krishnan (Assistant Professor, Computer Science and Laboratory Medicine and Pathobiology, Faculty of Arts & Science, Faculty of Medicine, University of Toronto and Faculty Member, Vector Institute) and Mamatha Bhat (Assistant Professor, Division of Gastroenterology, Temerty Faculty of Medicine, University of Toronto and Clinician-Scientist, Multi-Organ Transplant Program, University Health Network), DynaMELD and DynaCOMP is a coordinated effort between clinicians and computer scientists to address specific issues with liver transplant wait-times and patient selection.
“By applying advanced deep learning techniques to large and often complex datasets, the DynaMELD and DynaCOMP models aim to better predict patient outcomes, reducing mortality on the liver transplant waitlist, and using data sciences to offer a more just allocation process for all patients,” said Gary Bader, DSI Associate Director, Research and Software.
The project blends data sciences with health research and modelling as a driver for positive social change, a mandate also at the core of DSI funding ethos through catalyst seed grants.
The team has published part of their work on DynaCOMP at the 2024 Machine Learning for Healthcare conference. Leveraging their DSI seed funding, the researchers were awarded Canadian tri-agency funding and are currently in the process of external validation, using new data sets from different hospital systems and provinces.
“As of February this year, we were awarded a five-year CIHR Grant to expand the scope of DynaMELD to collect data from across Canada,” said Krishnan. “It has really launched a pan-Canadian idea to collect data from Alberta, from Quebec, from BC and the Atlantic provinces, in order to see how different risk scores perform on their data as well.”
But how exactly will DynaMELD and DynaCOMP address issues of inequalities in the current system with respect to liver transplant wait-times and patient selections?
“Let’s say you have 50 individuals who are all waiting for a liver,” said Krishnan. “Doctors need some number to guide them as to who should be ranked first or second or third on the transplant wait list. It’s a number that clinicians sat around the table and came up with about two decades ago.”
“So you have this score that’s been developed, and over the course of time, the score has become less calibrated since the population it was originally designed for has changed. It does not assess risk of mortality as well on women as it does on men, or for patients whose clinical condition deteriorates rapidly. We started rethinking how to calculate this score and, using what we know about AI and machine learning, wondered – what would a new score look like?”
The existing metric, known as the Model for End Stage Liver Disease (MELD)-Na score, can sometimes fail to accurately capture the severity of illness in certain groups, leading to a higher risk of waitlist mortality. Using clinical data from the University Health Network, Krishnan and Bhat used machine learning tools to develop DynaMELD, a more precise and equitable scoring system. The focus of this study included the development of new data science methodology on how changes in patients’ physiological status could be incorporated into risk scores predictive of mortality on the liver transplant waitlist.
“DynaMELD captures not just a patient’s risk of mortality but also their risk of accelerating in terms of likelihood of mortality through changing dynamics over time,” said Krishnan.
“In addition, we wanted to provide clinicians with an early warning system if the subsequent soft tissue graft was not functioning as intended – to create a similar risk score – and that motivated the DynaCOMP part of the project.”
After an individual receives a liver transplant, a common problem that clinicians are often faced with is the likelihood of soft tissue graft failure; DynaCOMP addresses this question.
“We’re very grateful to have received funding from DSI to pursue this project,” Krishnan concluded.
“You need to show evidence that in some sense you put in an effort to de-risk the project before applying for funding and the initial results that we’ve got supported by DSI were very important towards that end.”