Liver transplant candidates may get a fairer shake from an artificial intelligence-based organ allocation system than that offered by the current Model for End-Stage Liver Disease (MELD) scoring system, investigators contend.
A retrospective analysis of data from candidates on the waitlist for a deceased-donor liver indicated use of a machine-learning model trained to predict the probability of a candidate’s death or removal from the list within 3 months would have resulted in an average annual reduction in deaths of 418 patients compared with the MELD score.
In simulations, the artificial intelligence model, dubbed Optimized Prediction of Mortality (
OPOM), was associated with improved survival across all candidate demographics, geographic regions, and diagnoses. OPOM was substantially more accurate than the MELD score at predicting risk across all disease severity groups, according to Dimitris Bertsimas, PhD, from the Massachusetts Institute of Technology in Cambridge, and colleagues.
“The application of an OPOM-based allocation system would more accurately adhere to the ‘sickest-first’ principle. Indeed, the decrease in waitlist mortality/removal achieved through utilization of OPOM would not only represent the potential for more equitable allocation, but also would represent an important facet towards alleviating the discrepancy between supply and demand,” they write.
The MELD score has been used since 2002 to rank liver transplant candidates by disease severity, but the system’s method of “exception points,” which are intended to account for patients at imminent risk for death or disease progression, has resulted in what the authors called “inequitable and undesirable outcomes.”
Specifically, the exception point policy gives too much weight to candidates with
hepatocellular carcinoma at the expense of candidates without exception points, the researchers maintain.
The investigators sought to determine whether a machine-learning approach using a technique known as Optimal Classification Tree modeling could be better than the MELD score at answering the following question: “What is the probability that a patient will either die or become unsuitable for
liver transplantation within 3 months, given his or her individual characteristics?”
They applied the OPOM to data from the Organ Procurement and Transplantation Network Standard Transplant Analysis and Research dataset, including information on patients on the waitlist from January 1, 2002 through September 5, 2016.
The researchers first trained the system to predict the probability of a patient dying or becoming unsuitable for transplant within 3 months as a dependent variable, given observations of certain patient characteristics as independent variables. The independent variables included demographic and clinical characteristics.
After applying the trained model to the data, the researchers determined that liver allocation according to OPOM scores would have resulted in 417.96 (17.6%) fewer deaths annually among patients on the waitlist compared with the Match MELD (ie, MELD with exceptions) score. Additional analysis showed that OPOM would reduce deaths compared with MELD across all United Network for Organ Sharing regions.
“Notably, a higher number of female candidates received transplants when OPOM allocation was utilized,” the researchers write.
Compared with the Match MELD score, the OPOM score would have decreased deaths among patients on the waitlist, patients removed from the list, and post-transplant by 23.3%, 21.5%, and 1.8%, respectively.
Although OPOM allocated more livers to patients without hepatocellular carcinoma than MELD, OPOM decreased both waitlist deaths and list removals for patients with and without hepatocellular carcinoma.
The OPOM model “considerably outperformed” MELD for all patient exception statuses at predicting the 3-month probability of death or becoming unsuitable for transplant, as evidenced by a higher area-under-the curve of receiver operating characteristics.
“OPOM more accurately and objectively prioritizes candidates for liver transplantation based on disease severity, allowing for more equitable allocation of livers with a resultant significant number of additional lives saved every year. These data demonstrate the potential of machine learning technology to help guide clinical practice, and potentially guide national policy,” the researchers write.
The study had no specified funding. The researchers have reported no relevant financial relationships.
Am J Transpl. Published online November 9, 2018.
Abstract
