One of the sessions at the AIMed Cardiology virtual conference took place yesterday (4 November) was on real-time applications of artificial intelligence (AI). Dr. Kathy Jenkins, Senior Associate at the Department of Cardiology at Boston Children’s Hospital and Professor of Pediatrics at Harvard Medical School presented fellow panelists a challenge and together, they brainstormed how machine learning and data analytics could possibly mitigate it.

The real-world cardiology problem

Dr. Jenkin said many congenital heart disease (CHD) patients are falling out of care. These individuals, some of whom as young as 8-year-old, had experienced gaps or have fall out entirely, even though they knew staying in care would bring enormous health benefits such as reduced risks in hospitalization and increased mortality. The trend becomes conspicuous as individuals get older.

Early on, a Canadian study had revealed many CHD patients remained in primary care but only half of them would be in cardiac care once they reached 18-year-old and above. The lack of access, socio-economic and financial barriers, people’s perception that they are well or being told at some point they do not need care anymore were some of the factors keeping patients away. The situation had led the Congenital Heart Public Health Consortium to designate the return to CHD care as a major public health priority in the US.

Dr. Jenkin and the center where she works are trying to address it using AI and data science. She defined two problems here: One is finding CHD patients who are not engaged in care and tries to re-engage them. The other is looking out for CHD patients who are most at risks of falling out of care and take preventive actions so it does not happen. Personally, she feels the second problem is easier to handle but most people will take the first one as more important.

Besides, Dr. Jenkin added The American Heart Association and the American College of Cardiology had segmented the CHD population into mild, moderate and severe. In this case, the focus lies on the moderate to severe CHD groups. In response to Dr. Jenkin, Dr. Rashmee Shah, Assistant Professor in Cardiovascular Medicine at the University of Utah School of Medicine said she would approach it as the problem, challenge and the solution.

The potential of having patients to carry their healthcare data

The problem would be patients with moderate to severe CHD are falling out of care and therefore, they would have worse health outcomes as compared to those who did not fall out of care. The challenge is the healthcare system lose track of these patients because it is disconnected. Coupled with a haphazard insurance policy, there is no good way to find these patients and keep them in care. The solution will be what the group is heading towards now.

Dr. Shah believes the current healthcare delivery processes are fragmented and it will be great if patients, rather than the healthcare system are holding onto their electronic health records (EHRs), like a bank card, so they can take it with them wherever they go. So that anytime when they engage somewhere with the healthcare delivery system, an alert is raised and care providers can get them back into CHD care immediately.

The potential of AI-driven ECGs

Dr. Partho Sanguptha, Director of Cardiovascular Imaging and Professor of Medicine at the West Virginia University beefed up the suggestion by assuming if these is some form of software or application embedded in the EHRs or distributed to healthcare providers, that are able to identify risk features reflected in patients’ electrocardiograms (ECGs) and when combined with other clinical data, able to underline or predict potential CHD triggers. This will, at the bare minimal, facilitate the monitoring of CHD patients whenever they visit any providers and have their ECGs taken.

Dr. Anthony Chang, AIMed Founder and Chief Intelligence and Innovation Officer at Children’s Hospital of Orange County agreed but he warned the possibility of CHD patients having completely normal ECGs and the fact that patients are falling out of care not because they were undiagnosed. Dr. Jenkin said they have been using some retrospective data in the CHD database to conjure some predictive modeling to see which attributes that are driving patients out of care. However, they have not come to the part where they are able to locate these patients when they come into the emergency departments or primary care services and they envision of doing that.

The potential of Natural Language Processing (NLP)

Dr. Christ Lovejoy, Recent Medical Doctor at the University College London Hospital and Research Data Scientist at dunnhumby asked if there is some kind of ground truth to the identification of CHD patients such as records of patients that were known to have the condition or some kind of documentations of appointments. Dr. Jenkin replied it would be difficult to establish that because of the need to go through each and every single record to even algorithmically define them.

In fact, Dr. Jenkin admitted the algorithm built by her team failed miserably in detecting the CHD patients who are truly out of care. So, they are wondering if NLP can be brought in to understand text data and determine if the patients are truly out of care or if they are out of care because the disease status has changed and other factors. At the same time, Dr. Jenkin also assumed insurance database may be an excellent choice to improve their understanding of what drives patients out of care.

Dr. Chang agreed the use of NLP tools to start looking for terminologies that can capture the patients. He noted, data, particularly the type of data and where to get the data, will always be challenging those who are doing an AI or data-driven healthcare projects. Yet, if one is able to overcome that, they will find the rest gratifying.

More information about AIMed Cardiology can be found here. You may also re-visit the virtual conference on demand here.


Author Bio

Hazel Tang A science writer with data background and an interest in the current affair, culture, and arts; a no-med from an (almost) all-med family. Follow on Twitter.