Swedish researchers have developed a computer model that is better at matching transplant candidates to living donors than traditional methods, and as result could improve long-term survival rates for transplant recipients.
Weight, gender, age, blood group (of both donor and recipient), and the time when no blood flows to the heart during a transplant are just some of the numerous variables that can affect a patient’s survival chances after transplantation.
However, analyzing the interaction of these six individual risk factors alone requires 30,000 distinct combinations. Using traditional open source software methods, simulating these combinations for hundreds of thousands of patients can take up to three to four weeks.
Finding a ‘survival model’ that optimizes a successful match between recipients and donor could help guide decision-making for physicians, which is particularly important since the number of patients needing heart transplants far exceeds the number of donors. With this new ‘survival model’, analysis is significantly faster, and can be completed in less than five days.
The Swedish research team, from Lund University and Skåne University, utilized MATLAB and Neural Network Toolbox to develop a computing system that can learn on its own - an artificial neural network (ANN) - to explore the complex non-linear relationships among multiple variables.
ANN models have pattern-matching and learning capabilities that can tackle problems that are difficult or impossible to solve by standard computational and statistical methods.
In this case, the ANN model had to analyze around 57 risk factors in thousands of patients to make an optimal donor/recipient match. The resulting simulation predicted survival rates for heart transplants based on the suitability of the donor match.
“In a simulated randomized trial, we found that the ANN models could increase the five year survival rate by 5 to 10% compared to the traditional selection criteria,” said Johan Nilsson, associate professor in the Division of Cardiothoracic Surgery at Lund University.
According to the International Society for Heart and Lung Transportation (ISHLT) registry nearly 90% of heart transplantees survive longer than two years after their operation. At five years, the figure is 75% decreasing considerably to a 50% survival rate after 10 years.
With increasing numbers of patients awaiting heart transplantation, there is a need to expand the donor pool by stretching the margins of donor acceptability. Nilsson’s model may also help broaden the eligibility criteria for donors and recipients.“Our preliminary results show that the ANN model would transplant approximately 20% more patients than would have been considered using traditional selection criteria,” said Nilsson, who had programmed the application himself.
“What we have found is that both the number of transplanted patients and the long-term survival rate could be increased using ANN models. This means for an individual patient, not only does their chance for transplantation increase, more than likely their survival time will increase as well.”
The application has not yet been implemented in the clinical setting, which is one of the main goals of the research, said Nilsson. The team’s short-term plan is to start a randomized trial in two years with the software being used in a clinical setting within the next five years. “Eventually, we hope that the application could be used on any desktop computer, and the matching process would take only a few minutes.”
“The ultimate goal is that eventually every donated heart could be used, and the risk of early and long-term graft failure could be decreased by 50%. Today (in Sweden) we only transplant about one of every three available organs,” he said.
Understanding how various risk factors affect survival rates involved thousands of computationally and data intensive operations. Due to the immense combinational load of the recipient-donor variables, the models require an enormous amount of computer horsepower. This optimisation problem was solved using Parallel Computing Toolbox and MATLAB Distributed Computing Server to accelerate the simulation of more than 200 ANN configurations.
The ANN models were then trained, which involves correlating donor and recipient data, so the risk factors are weighted accurately. To do this, data was obtained from two global databases: the International Society for Heart and Lung Transportation (ISHLT) registry and the Nordic Thoracic Transportation (NTTD).
After training, the accuracy of the algorithms was verified by running simulations in a sub-set of 10,000 patients whose data had been omitted from the initial training set. The ANN model results were then compared against actual survival rates. The research team then conducted thousands of simulations in parallel to rank the 57 risk factors for predicting long-term survival. Finally, the team identified the best and worst possible donors for any particular recipient based on survival rates after transplantation.
Predicting how a transplant candidate's immune system will respond to a specific transplanted organ is vital for forecasting rejection. Around 40%of heart transplantees can experience an acute rejection after one year. Differences in a set of genes called the Human Leukocyte Antigen (HLA) system can influence a person's ability to accept foreign tissue (i.e a transplanted organ). For the next phase of the project, the Lund researchers intend to use the ANN model to investigate the use of HLA genetic profiles to match donors with recipients.