Deepak Srikant

The Value of AI PERT Activation and Avoiding Alert Fatigue

The advent and continued development of clinical artificial intelligence (AI) has been groundbreaking. The technology, once seen as experimental, is showcasing its value in globally renowned institutions, creating downstream benefits unforeseen in its early development. Other than flagging suspected positive findings and prioritizing the radiologist worklist, the relevance of AI is now felt beyond the workstation, showing its impact on patient care paths in crucial service lines like the cardiovascular and neuro spaces. While there are countless benefits to AI adoption for physician workflows, there is one AI phenomenon that is not so warmly embraced by clinicians: alert fatigue. It is especially important to avoid this phenomenon in certain care scenarios, like PERT activations, for example.

What is Alert Fatigue?

Akin to Aesop’s “The Boy Who Cried Wolf”, alert fatigue is defined as the overabundance of notifications a specialist is receiving about a specific pathology, especially when a case does not warrant their intervention. This can lead to a less-than-serious interpretation for specialists when they’re inundated with AI notifications on their mobile device. 

Let’s look at it from the perspective of pulmonary embolism (PE) cases.

Avoiding Alert Fatigue & Unnecessary PERT Activations

The role of AI in pulmonary embolism cases is vital. Amongst the leading causes of death from cardiovascular disease, PE is often called the “silent killer” for its sizable mortality rate when not treated. This makes it all the more crucial for a facility’s  pulmonary embolism response team (PERT) to be made aware of potential PE cases as quickly as possible. This is where AI plays a critical role, since not all PE cases are created equal as far as the interventionalist is concerned. 

Consider the varying grades of severity for PE cases:

  • Massive – Obstruction of the pulmonary arterial tree exceeds 50% of a cross-sectional area, causing acute and severe cardiopulmonary failure from right ventricular overload. These cases require immediate specialist intervention and warrant immediate PERT consultation.
  • Sub-Massive – An acute PE with notable myocardial necrosis evidenced by elevated troponin. In most cases, a sub-massive PE is treated with anticoagulants. Thrombolysis is only considered when RV dysfunction is present. Therefore, PERT consultations are rare. 
  • Non-massive – A PE that is stable with normal RV function. Non-massive PEs do not call for a PERT consultation, but anticoagulant therapy.

Now consider the role of AI and the potential for alert fatigue. We can see that in many scenarios, there is no necessity to trigger a PERT activation for the presence of a PE, as this can cause fatigue and for the value of AI-driven notifications to diminish with each case that doesn’t actually justify a PERT consultation.

To offset PE alert fatigue in a facility, AI must:

  1. Have additional risk stratification measures in place. For example, the ability to view the RV/LV ratio.
  2. Utilize this information to assess the severity of the PE and, based on a facility’s thresholds, only notify the PERT when they meet specific criteria.
  3. Factor in EHR integrations. This would empower AI with real-time lab values. Not a single pull from an hour or a day ago, for example, but providing live lab values that enable care teams to triage and, when necessary, elevate relevant patients to the PERT.

The image below is an example of how, over an 18-month period, AI can surface the patients you want to see, drastically reducing the amount of unnecessary PERT activations team from over 2,200 to just 364.

Top down funnel graphic showing how clinical AI can reduce PERT activations by only alerting on patients in need of consultation.

Ending the Fatigue

Without AI accounting for additional risk stratification measures, clinicians are destined to run into the headache of alert fatigue, with each notification taken less seriously than the one before it. An AI workflow without this added layer can cause more than a minor disturbance: it can impact a patient’s treatment path. 

For a new technology to truly break through in this space, it has to be useful, usable and used. In a use case like PE, time to treatment decisions can have serious implications for patient outcomes. For AI to be useful, it must activate the PERT at the right time. To do this, parameters need to be in place to limit alert fatigue, giving specialists a sense of confidence that they can trust that any PE notification is important and warrants their immediate attention.

Click here to learn more about Aidoc’s Cardiovascular Solutions.

Explore the Latest AI Insights, Trends and Research

Deepak Srikant