Project Spotlight: Aneurysm Prognosis Classifier

Abdominal aortic aneurysm (AAA) remains the 15th leading cause of death in the United States. Currently, clinical experts largely rely on information about the size of the aneurysm and its rate of growth to decide whether surgical intervention is warranted. The result is that some cases relegated to monitoring will, in fact, go on to rupture and result in a high risk of mortality and increased healthcare costs. Aneurysm Prognosis Classifier (APC) will use machine learning to predict the complication risk of AAA, thus providing clinicians with an objective, predictive tool to guide surgical intervention decisions before symptoms appear. David Vorp, PhD, is a Professor of Bioengineering in the Swanson School of Engineering at the University of Pittsburgh and leads a team of researchers developing APC.

Please share a little about your background and your research experiences.

I am the John A. Swanson Professor of Bioengineering at the University of Pittsburgh’s Swanson School of Engineering and I have been engaged in abdominal aortic aneurysm (AAA) research for more than 25 years. My work has been supported over the years by over $9 million of federal and institutional funding to perform cutting edge research in biomechanics and tissue engineering. Our team includes Dr. Nathan Liang, a UPMC vascular surgeon and Assistant Professor of Surgery at the University of Pittsburgh, and Dr. Timothy Chung, a postdoctoral researcher in my lab who has expertise in computational modeling of AAA. We are also working closely with Health Record Research Request (R3), a service of the Department of Biomedical Informatics (DBMI) that helps investigators at the University of Pittsburgh access UPMC data for research purposes. Together, our team possesses research experience in automation, biomechanics, and machine learning methods in a clinical setting – all necessary to make progress on this highly interdisciplinary project.

What led you to the PHDA?

I had the idea to use the biomechanical status of AAA as a clinical tool to predict rupture two decades ago. However, the amount of data required to undergo a statistically significant study became at the time a severe bottleneck to achieving this goal. The APC team recognized with a sufficiently large image and clinical dataset, and today’s exponentially improved computational power, the opportunity to implement patient-specific healthcare was within reach. We learned about the PHDA through a graduate student in our lab that was working with sciVelo, the Pitt Innovation Institute’s early-stage clinical translation support organization. We immediately realized the tremendous opportunity and the vision of the PHDA to convert data into clinical implementation and found that AAA would be a suitable use case to give clinicians a new decision-support tool.

 Walk us through your project.

Abdominal aortic aneurysm is a costly condition that carries the risk of morbidity and mortality, with an increasing prevalence among the elderly population. Rupture of AAA is a catastrophic event and remains a leading cause of mortality in the U.S. Surgical and even minimally invasive procedures to repair AAA are not without risk, so physicians must weigh potential benefits to the patient in order to decide on the optimal course of treatment.

Patients that present with AAA are subject to watchful waiting through surveillance. This means, patients have to come to the clinic – sometimes as often as every six months – to check whether their aneurysm has grown until either the growth rate or maximum diameter criteria limit is reached that would necessitate intervention. Once an aneurysm exceeds the typical “maximum diameter criteria” of 5.5 cm, patients will be recommended for open surgery or endovascular repair of their AAA. However, with approximately 25% of medium-sized aneurysms rupturing, the maximum diameter criteria – a 50-year-old standard – is simply unreliable and outdated.

Our project will take a large variety of data with known outcomes and use it to build models that can be used to predict risk in a more accurate and objective manner. This could enable a more personalized and effective approach to intervention decision-making. Depending on the risk score, we will be able to propose which patients should undergo intervention or continue with “watchful waiting but with extended surveillance intervals (saving costs, time and potential radiation exposure to the patient).

What are your project’s next steps?

We are currently training and validating our models on de-identified data sets and, in parallel, working with our clinical collaborators to think through how this solution would function within the clinical workflow and healthcare system. Looking ahead, we plan to continue working with the CCA/PHDA team to explore extensions of our current study and commercial avenues for our work. We also expect that our methodologies developed for AAA will translate to other types of aneurysm including cerebral and ascending thoracic aneurysms.

In what ways has UPMC played a role lending clinical and healthcare data expertise?

My team and I have greatly benefitted from our collaboration with clinicians from the world-renowned UPMC Division of Vascular Surgery, which includes several grants from the NIH and other sources to investigate abdominal aortic aneurysm. Additionally, we are working with the Health Record Research Request (R3) program at Pitt, a service of the Department of Biomedical Informatics, and UPMC Enterprises to identify the types of data that our modeling efforts require and, where appropriate, access anonymized data sets for our research.