AI Insights: New Methods for Old Challenges
By Peter Monteleone, MD, FACC, FSCAI, and Akash Patel, MD, MBA
Radiology Today  
Vol. 24 No. 4 P. 26

AI assists with pulmonary embolism management. 

The clinical presentation of pulmonary embolism (PE) has always created a diagnostic challenge. Systematic reviews have reported dramatic rates of misdiagnosis of PE patients, as high as 50% for inpatient cases. As a result, patients often were—and still too commonly are—sent home without receiving proper treatment that could prevent a life-threatening clinical deterioration.

But the challenges in PE therapy go far beyond simple diagnosis. Limited prospective, randomized data have resulted in an absence of clear guidelines dictating PE care. Without clarity, awareness variation persists across systems and geographies. It sometimes seems that half the job of PE response team (PERT) champions is enhancing awareness of the benefits of advanced therapies for PE. And it becomes almost impossible for a small team of PERT champions to simultaneously educate their community about PE, identify and risk stratify PE, and treat PE patients.

PE risk stratification is not always obvious. For example, a patient on a betablocker with a large PE may present with a normal range of blood pressure and a heart rate in the 90s. An unchecked troponin was elevated. An elevated right ventricle:left ventricle ratio was unmeasured. As a result, an intermediate high risk PE was undertreated. This scenario is frighteningly routine and may reveal itself only 48–72 hours later, when the clinical condition deteriorates after an optimal care window is missed. Or it may only be appreciated 90 days later, when an echocardiogram is finally performed, and a dilated, dysfunctional right ventricle is identified for the first time in a persistently symptomatic patient.

When we started our PE program in central Texas, a single PERT champion with limited resources could not identify and risk stratify every elevated risk PE in our multihospital network. For programs developing in 2023, that is no longer the case.

AI Changes the PERT Paradigm
In 2023, a single champion of PE therapies, with the support of their health care system, can install an application on their mobile device that is linked to software embedded within their facilities’ CT imaging hardware that will inform them instantaneously of every patient whom a CT scanner visualizes a PE. That champion can risk stratify these patients almost immediately and guide care based on the resources they have available. The same application can be used to track volumes of care practice to predict future program support requirements (staffing, equipment, procedural space availability, etc). On October 3, 2022, and every day before that, our PERT program spent 50% of its time and energy on weekly outreach to our emergency departments and inpatient teams in both our central market and our satellite centers, educating teams about how to risk stratify PE, and how to contact our PERT team. On October 4, 2022, our PERT team members had RapidAI on our devices, and we risk stratified every PE within minutes of its visualization. Our outreach program changed overnight from the broad constant casting of the widest of nets to targeted, patient-centric education of first-line providers.

“Hello, this is Dr Monteleone from the Ascension Texas PERT team. Your patient just had a CT scan completed which identified a PE. We reviewed her case, and her clinical syndrome is consistent with an intermediate high-risk PE. Can we discuss her plans of care because there may be value in consideration of advanced therapies?”

AI Will Transform Clinical Trials
We are running ahead of guidelines in advanced PE therapy. There is an absolute abundance of data guiding the therapies we use every day, but there is an equivalent paucity of prospective randomized evidence. Appropriately clinical guidelines for advanced therapy are awaiting these data ... but they are coming. The HI-PEITHO trial sponsored by Boston Scientific is randomizing patients to ultrasound-assisted catheter directed thrombolysis vs conservative medical therapy and has already achieved robust enrollment in the United States and Europe. PE-TRACT, sponsored by the National Institutes of Health National Heart, Lung, and Blood Institute, will randomize patients to endovascular vs medical therapy and is set to begin enrollment in 2023.

But enrolling in trials is challenging. And enrolling in trials in a field like PE, in which continuous outreach is needed just to achieve appropriate clinical care, is even more challenging. Busy emergency department clinicians barely have time for education about care of the PE patients in their wards. They do not have time to continuously review inclusion/ exclusion criteria for a clinical trial to guide the care of future generations.

AI tools fix this issue. Here in central Texas, we are a clinical site for the HI-PEITHO trial, and our research coordinator dedicated to the trial has RapidAI on her mobile device. Every time a CT scanner in our system visualizes a PE, the app identifies that PE, and then that patient is screened by our coordinator for trial inclusion. Screened numbers have gone up. Positive prediction for enrollment has gone up. Screening time has gone down.

“Hello, this is Dr Monteleone from the Ascension Texas PERT team. Your patient just had a CT scan completed which identified a PE. She appears to meet the criteria for enrollment in the HI-PEITHO trial. Can we discuss options for her therapy and also if you think it would be appropriate for our research team to discuss the trial with her?”

We feel an obligation to continuously improve the care of our PE patients. We have an absolute obligation to advance the science guiding these therapies. AI tools will be a revolutionary shift in enrollment in the current and next generation of clinical trials in PE. They will also serve as a model for similar tools to be used in the future in every field of modern medicine.

Addressing Enrollment Disparities
Limited progress has been made in the reporting and representation of underrepresented minorities in clinical trials science. It is of the utmost importance to acknowledge that failure to correct this issue today, or it will fail generations of patients tomorrow. When identification of possible trial participants lies in the hands of busy frontline providers, complexities of care, including socioeconomic challenges and comorbidities, may push clinical trial enrollment further to the bottom of a clinician’s priorities.

But when the first screen for trial participants is automated by AI and not dependent on a clinician, these complexities, as well as our inherent biases, are removed. Every PE seen by a CT scanner is screened for enrollment by a dedicated coordinator. Every patient is evaluated for trial enrollment. Automated screening also allows for screening beyond the “ivory tower” facilities that often host clinical trials. Easing broad screening with resultant broad identification is the next step in “bringing the trial to the patient,” which will be a huge step forward in broadening the applicability of our current studies to our future patients.

Challenges and the Future
This is the path forward, but it requires a journey. There is a cost associated with software and support. There is an implementation process for embedding software within a system’s imaging hardware. There is an adoption curve (albeit rapid) for clinicians. And perhaps most importantly, there is an operations requirement to help prevent frontline providers from being extremely confused by a clinician with an app on their phone calling them about a patient for whom the PERT team was not consulted to discuss a CT scan that was not yet read by radiology.

Nonetheless, these tools will help clinicians identify all patients without exception and at a point earlier in their course. More will be accomplished with greater efficiency and with fewer available resources. The patients in front of us will benefit, and through clinical trial implementation, the patients of the future will have their care optimized.

The days of AI in medicine being a solution looking for a problem have officially ended. The early adopters among us have reached a point where we have trouble remembering the “before times,” and we look forward to the rest of you joining us on this next step in the path forward for modern health care.

— Peter Monteleone, MD, FACC, FSCAI, is the director of the Ascension Seton Heart Institute Clinical Research Group and the medical director for the SHI Vascular Imaging Laboratory. He is also an assistant program director for the Cardiovascular Disease Fellowship at The University of Texas at Austin Dell Medical School. 

— Akash Patel, MD, MBA, is an internal medicine resident physician (PGY-2) at The University of Texas at Austin Dell Medical School.