New algorithm predicts impact of heart transplants on survival in heart failure patients

por John R. Fischer, Senior Reporter | May 23, 2018
Artificial Intelligence Cardiology
A new algorithm can determine survival
rates and lengths for heart failure patients
when they receive a heart transplant
and when they do not
UCLA researchers have developed a new algorithm for predicting survival rates among heart failure patients awaiting heart transplants.

Designed to determine individual survival rates with or without a transplant and for how long, the algorithm, coined Tree of Predictors (ToPs) by its creators, could provide more personalized assessments of patients, enabling better and more cost-effective use of lifesaving resources.

“These are multivariate statistical models or machine learning algorithms commonly used in medicine and data science. Limitations include that either risk models would apply a single solution to a population as a whole or the same algorithm to the entire dataset,” Dr. Martin Cadeiras, a cardiologist at the David Geffen School of Medicine at UCLA and one of the authors of a study behind the algorithm, told HCB News. “ToPs identifies the best solution for each node in the tree, resulting in improved performance for an individual patient.”

Using machine learning, the algorithm acquires new information over time on the basis of 53 data points including age, gender, body mass index, blood type, and blood chemistry, predicting a patient’s chances of surviving with or without a transplant.

The data points enable ToPs to assess complex differences among people awaiting heart transplants and compatibility between donors and potential transplant recipients with 33 points centered on the recipient or potential recipient's information, 14 pertaining to donors, and six applying to donor and recipient compatibility.

Researchers evaluated ToPs’ performance using 30 years of data on patients registered with the United Network for Organ Sharing, a nonprofit that matches U.S. donors and recipients.

The algorithm outperformed machine learning approaches developed by other research groups and was 14 percent more accurate than current predictions models in determining lengths of survival, correctly choosing 2,442 more patients out of 17,441 recipients who lived for at least three years following surgery.

In addition to predicting survival rates and lengths, the algorithm also analyzes many possible risk scenarios for potential transplant candidates and can incorporate more data as treatments progress.

Cadeiras says use of the algorithm could enable physicians to tailor organ transplant procedures, among other things, by taking into account the variability of each individual patient. He warns though that ToPs should not be used as the sole decision maker in determining the best course of action for a patient, and that other elements in the treatment process must be taken into account.

“All methods, including ours, are limited by the limitations of the dataset used and will most likely perform better as information improves,” he said. “However, when medical decisions are made, scoring systems are only one portion of the information clinicians use to estimate the risk of a patient, [as well as] multiple other factors, such as the experience of the treating team, which plays a major role. Methods like ToPs are meant to support decisions, not to be used in isolation or by others than experienced healthcare professionals who can adequately understand the strengths and limitations of the information.”

UCLA researchers also developed another machine learning technique that utilizes fMR to determine if patients with OCD would benefit from cognitive behavioral therapy.

The study was published in PLOS One.

You Must Be Logged In To Post A Comment