Investigators claim that liver transplant candidates could benefit from an artificial intelligence-based organ allocation system that is more accurate than that proposed by the Model for Final Stage Liver Disease (MELD) system. according to the investigators.
A retrospective analysis of the data of the candidates on the waiting list of a deceased donor liver showed that using a trained machine learning model to predict the likelihood death or expulsion of the candidate within 3 months would have resulted in an annual average reduction in the number of candidates. deaths of 418 patients compared to the MELD score.
In simulations, the artificial intelligence model, called optimization of mortality prediction (OPOM), was associated with improved survival in all demographics, geographic areas, and diagnoses. According to Dimitris Bertsimas, Ph.D., of the Massachusetts Institute of Technology in Cambridge, and his colleagues, OPOM was significantly more accurate than the MELD score in predicting risk in all disease severity groups.
Researchers published their findings online November 9th at American Journal of Transplantation.
"The application of an OPOM-based allocation system would adhere more specifically to the principle of" sicker first ". Indeed, the reduction of mortality / deletion of the lists of The expectation obtained through the use of OPOM would not only represent the potential for a more equitable allocation, but would also represent an important aspect of narrowing the gap between supply and demand. request, "they write.
The MELD score has been used since 2002 to classify candidates for liver transplantation according to the severity of the disease, but the system's method of defining "exception points", intended to take patients into account presenting an imminent risk of death or progression of the disease, resulted in what the authors called "unfair and undesirable results."
Specifically, the exception points policy gives too much weight to candidates hepatocellular carcinoma at the expense of candidates without exception points, maintain researchers.
The investigators sought to determine whether an automatic learning approach using a technique known as Optimal Classification Tree Modeling could be better than the MELD score to answer the following question: "What is the probability that a patient will die or become unfit for liver transplantation within 3 months, given its individual characteristics? "
They applied the OPOM to data from the standard data set of analysis and transplant search of supply and transplant organs, including patient information. registered on the waiting list from 1 January 2002 to 5 September 2016.
Researchers first trained the system to predict the probability of a patient dying or becoming unsuited to transplantation within 3 months as a dependent variable, with observations of some patient characteristics constituting 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 the OPOM scores would have resulted in 417.96 (17.6%) fewer annual deaths among patients on the list of patients. MELD match score (ie MELD with exceptions). Further analysis showed that OPOM would reduce the number of deaths compared to the MELD in all regions of the United Network for Organ Sharing.
"Notably, a higher number of female candidates have received a transplant while using the OPOM allowance," the researchers write.
Compared to the MELD Match score, the OPOM score would have resulted in a decrease in the number of deaths among patients on the waiting list, the number of patients removed from the list and 23.3%, 21.5% and 1.8%, respectively.
Although the OPOM has allocated more livers to patients without hepatocellular carcinoma than the MELD, the OPOM has decreased the number of deaths on waiting list and elimination lists for patients with and without hepatocellular carcinoma.
The OPOM model "significantly outperformed" the MELD for all patient exception statuses in terms of the probability of death at 3 months or becoming unsuitable for transplantation, as evidenced by a larger area under the curve of operating characteristics of the receiver.
"More specifically and more objectively, OPOM gives priority to liver transplant candidates based on the severity of the disease, which allows a more equitable allocation of the liver with a significant number of extra lives saved each year. potential of machine learning technology to guide clinical practice, and potentially guide national policy, "write the researchers.
The study had no specified funding. The researchers have not reported any relevant financial relationship.
Follow J Transpl. Posted online November 9, 2018. Abstract