A Review of Artificial Intelligence and Cardiac Arrest

AI for In-Hospital Cardiac Arrest

A common concern about artificial intelligence (AI) in healthcare is where it will be profitable in addition to improving and/or saving lives. One promising area is the earlier prediction of in-hospital cardiac arrest. [1] The AI software monitoring of vital signs +/- labs can detect changes earlier and can help predict cardiac arrest more accurately and earlier than existing risk models. Finding out a patient is headed towards cardiac arrest earlier can lead to measures to stop and reverse this progression.

This can lead to fewer complications for the patient, less need for critical care resources in the hospital, and decreased length of stay in the hospital for the patient. A patient in cardiac arrest can divert staff for an hour or more, which delays other patients from being seen and treated as well. Decreased use of critical care resources and decreased length of stay in the hospital can save money for hospitals and insurance companies, as well as patients, and in addition, will improve the patient’s quality of life.

The Data for AI Predicting In-Hospital Cardiac Arrest

There is a growing body of data, of varying quality, available for early prediction of cardiac arrest with AI algorithms. Studies show machine learning (ML) and more specifically deep learning (DL) can help predict patients at risk for cardiac arrest earlier than traditional methods. [2] The more recent use of neural networks is leading to the development of better algorithms to predict cardiac arrest earlier versus previous traditional ML algorithms. [3]

ML algorithms could be used to better predict cardiac arrest in patients when they present in the emergency room, patients could then be better triaged based on these predictions which can save time and resources for the hospital, as well as be beneficial for the patient. [4, 5] ML algorithms have also been shown to predict cardiac arrest in patients admitted with acute coronary syndrome (ACS) better than existing risk scores and models. [6] One of the benefits of DL is its ability to detect very minor changes on imaging studies. Generally, most patients have an electrocardiogram (ECG) performed prior to admission to the hospital, one study utilized a DL algorithm that could effectively predict cardiac arrest using the patient’s ECG. [7]

The existing studies used a variety of different variables to predict cardiac arrest earlier. Many of the studies using ML algorithms to predict cardiac arrest earlier used the patient’s vital signs as the primary data. [5] One study showed that an algorithm consisting of vital signs and a brief interview could predict cardiac arrest or acute respiratory failure 1 to 6 hours earlier than without the algorithm. [8] Another study used a recurrent neural network that needed only the 4 main vital signs to better predict cardiac arrest earlier. [9] This likely would be the easiest type of algorithm for hospitals to integrate into their existing systems as it uses data that can be easily collected from patients’ electronic health records (EHR).

Startups in this Space

There are a few startups working in this space. One is Transformative whose AI technology predicts sudden cardiac arrest by analyzing data from patient monitoring devices commonly used in hospitals. Another startup VUNO has an AI-based cardiac arrest prediction software that analyzes vital signs stored in the EHR which then predicts the likelihood of a cardiac arrest occurring within the next 24 hours. Likely startups in this area will continue to publish more around the accuracy of early prediction cardiac arrest AI software and also in regards to the financial benefits of this type of software for hospitals.

Areas of Concern

In healthcare AI there are ongoing issues of training data bias because of skewed data from available patient databases and this is something that will need to be accounted for in algorithms for cardiac arrest as there are differences related to age, gender, and race to account for. Another area of concern is integrating new AI software into existing systems and the associated cost, setup time, and the availability of trained personnel to do this. For hospitals, the amount of money saved from shorter length of stays, fewer patient complications, and decreased need for critical care resources would need to outweigh the cost of integrating new AI software. More emphasis on the cost savings would be helpful as it seems more likely that hospitals will pay the costs to integrate additional software if they’re shown it will be financially beneficial for them.

Future Opportunities

Future opportunities include expanding the capabilities and accuracy of ML algorithms so that the same software can predict cardiac arrest and acute respiratory issues as well. [10,11] In the future this will help better allocate resources, save money on expensive critical care costs, as well as the obvious benefit to the patient. As algorithm accuracy improves, we may even be able to remove overnight vital sign checks for some hospitalized patients, this can help improve patients’ sleep and likely the overall speed of their recovery. [12] Eventually, ML algorithms could include the capability to be all-encompassing and help hospitals improve resource planning and managing capacity for all acute patients, medical and surgical, and related emergency procedures. One of the main benefits of AI in healthcare is in areas where it can help save time and resources with logistics and planning to save money and give staff more time to take care of patients.

For consulting services contact me @mydstrategy@gmail.com

References

1. Development of a Real-Time Risk Prediction Model for In-Hospital Cardiac Arrest in Critically Ill Patients Using Deep Learning: Retrospective Study
2. Machine Learning–Based Early Warning Systems for Clinical Deterioration: Systematic Scoping Review
3. Deep Learning in the Medical Domain: Predicting Cardiac Arrest Using Deep Learning
4. Prediction of Cardiac Arrest in the Emergency Department Based on Machine Learning and Sequential Characteristics: Model Development and Retrospective Clinical Validation Study
5. Value of laboratory results in addition to vital signs in a machine learning algorithm to predict in-hospital cardiac arrest: A single-center retrospective cohort study
6. Machine learning for early prediction of in‐hospital cardiac arrest in patients with acute coronary syndromes
7. Artificial intelligence algorithm for predicting cardiac arrest using electrocardiography
8. Predicting Cardiac Arrest and Respiratory Failure Using Feasible Artificial Intelligence with Simple Trajectories of Patient Data
9. An Algorithm Based on Deep Learning for Predicting In‐Hospital Cardiac Arrest
10. Machine Learning Methods to Predict Acute Respiratory Failure and Acute Respiratory Distress Syndrome
11. Prediction model for patients with acute respiratory distress syndrome: use of a genetic algorithm to develop a neural network model
12. Let Sleeping Patients Lie, avoiding unnecessary overnight vitals monitoring using a clinically based deep-learning model