Artificial intelligence tools can identify potential complications after bone marrow or stem cell transplants before symptoms appear, potentially allowing for more accurate patient monitoring and pre-emptive treatment.
The BIOPREVENT algorithm combines machine learning with immune biomarkers and clinical data to predict death after chronic graft-versus-host disease (GVHD) or hematopoietic cell transplantation (HCT).
Tools described in clinical research journal, Currently designed for risk assessment and clinical research.
But eventually, doctors may be able to receive personalized risk estimates for post-transplant outcomes in real time, based on clinical and biomarker data.
“By the time chronic GVHD is diagnosed, the disease has often progressed over several months and is silently damaging the body,” explains Dr. Sophie Patzesny, a researcher at the Medical University of South Carolina.
“We wanted to know if we could detect red flags early enough for clinicians to intervene before patients become unwell and before the damage is irreversible.”
GVHD occurs when donor cells injected to treat a blood disease react against the recipient, commonly affecting the skin, intestines, and liver. This is one of the leading causes of debilitating disease and mortality after HCT transplantation.
In an attempt to predict its occurrence, Paczesny and team developed BIOPREVENT using data from 1,310 stem cell and bone marrow transplant recipients across four well-characterized multicenter studies.
The data incorporates seven previously validated plasma biomarkers related to inflammation, immune activation and regulation, tissue damage and remodeling, measured from blood samples taken between 90 and 100 days post-transplant.
This was combined with nine key clinical factors including patient age, transplant type, underlying disease, and previous complications identified from the transplant registry.
Patients were divided into training and validation datasets, and several machine learning and deep learning models were evaluated for their ability to predict outcomes at various time points over a year and a half post-transplant.
The researchers found that deep learning performs, at best, similarly to other machine learning approaches considered. They suggested that this may be due to the deep learning model’s complex neural network structure not being able to effectively understand the relationship between biomarkers and chronic GVHD risk without a much larger sample size of tens of thousands.
Bayesian Additive Regression Trees (BART) consistently gave high results and was ultimately chosen as the final model.
BIOPREVENT showed the best performance among several machine learning models and demonstrated its real-world applicability as a biomarker-based prediction tool in squaring hypothesis testing under two patient scenarios.
The researchers believe their study is the largest chronic GVHD biomarker study to date, and can be further tested by making the web-based tool freely available.
“It was important to us that this not be just a theoretical model or tool limited to a single institution,” Pakchesny said. “Making BIOPREVENT freely available will allow researchers and clinicians to test it, learn from it, and ultimately improve care for transplant patients.”
