AI Case Study

MIT researchers developed a model that could decrease mortality rate of people on liver transplant lists by 20% with machine learning

MIT researchers have leveraged artificial intelligence to develop a system to more efficiently allocate organs to patients on a liver transplant list. Their system which was designed to more accurately account for the severity of each patient's state and predict the chance that their condition will significantly worsen within 3 months. It managed to outperform the current system being used in accuracy and fairness and promises to reduce mortality rate by 20%.

Industry

Healthcare

Healthcare Providers And Services

Project Overview

"MIT Sloan School of Management Prof. Dimitris Bertsimas and Prof. Nikos Trichakis utilized machine learning to create a model that reduces mortality by 20%, averting nearly 400 deaths each year. Their model, Optimized Prediction of Mortality (OPOM), also provides a fairer and more equitable allocation to candidate groups, including women.

“There are many significant benefits to using this new model over the current system. Unlike the current system, which makes some arbitrary choices and results in bias against certain populations, OPOM’s methodology for prioritization is clear and understandable to surgeons — and it can save hundreds of additional lives every year,” says Bertsimas.


Trichakis noted, “OPOM fixes many of the current system’s problems because it was designed specifically for liver patients using real data. As a result, it can accurately prioritize patients across all populations without bias. This shows the potential of machine learning technology to help guide clinical practice and national policy on transplants.”

The researchers explain that the current model created in 2002 depends on the Model for End-Stage Liver Disease (MELD) score to rank disease severity and priority for receiving a liver transplant. As certain patient populations are at risk of death or of becoming too sick or unsuitable for transplantation based upon disease progressions that are not captured in their MELD score, the system arbitrarily grants them “exception” points. While the overall MELD score has led to a more objective ranking of candidates awaiting liver transplantation, the process of MELD exception point granting has resulted in inequitable and undesirable outcomes.

More specifically, the MELD exception points policy has disadvantaged women. “Data shows that women have historically had less access to liver transplantation and have had higher death rates on the wait list,” notes Trichakis. “This is due to the awarding of exception points to cancer patients, as more than 75% of those patients are men. Women also tend to have lower muscle mass and higher sodium levels, which lowers their MELD scores.”

Using a state-of-the-art machine learning method developed at the MIT Operations Research Center and real historical data from liver patients, the researchers sought a better way to prioritize the allocation of organs. With OPOM, they asked the question: What is the probability that a patient will either die or become unsuitable for liver transplantation within three months, given his or her individual characteristics?

They found that the OPOM allocation outperformed the MELD-based prediction method in terms of accuracy and fairness. In simulations, OPOM averted significantly more waitlist deaths and removed the bias against women. As a result, it allowed for more equitable and efficient allocation of liver transplants."

Reported Results

According to the authors: “If we use this model to change how we measure mortality and allocate livers, the death rate will decrease by 20%, which is very significant. We’re hopeful that our findings will affect the national policy.”

Technology

Function

Background

"Demand for liver transplants is much higher than organ supply, resulting in approximately 2,400 deaths every year. Also problematic is the current model used to identify and prioritize the “sickest” patients, which does not allow for equitable access to all waitlisted candidates, with a particular disadvantage to women."

Benefits

Data