Can artificial intelligence be part of the triage proccess

In recent years, it has been observed that the number of emergency room visits shows a worrying trend due to its accelerated growth. In the US alone, for 2018 (1), an average of 130 million visits to the ED system were reported, causing unprecedented levels of crowding and delays in care.

With such volume it is to be expected that the system will have flaws in the clinical routing of patients, a study published in 2018 (2) by the Johns Hopkins University Department of Emergency Medicine, estimates that on average 1 in 5 patients are misassessed by healthcare staff when making prioritization assignment. This can have serious consequences for patients. On the one hand, over-triage results in patients being sent to unnecessary, costly and time-consuming intensive care treatments, exposing them to dangerous antibiotic-resistant bacteria that can be rampant in hospitals. On the other hand, sub-triage means that a patient in need of care does not receive it in a timely manner, preventing proper monitoring of the condition and eliminating the possibility of stopping the patient's deterioration.

If we add to this the number of patients who make unnecessary visits that end up saturating the emergency rooms on peak days, we find that triage is an activity that represents great challenges for health institutions to optimize their resources and for patients to receive adequate service and treatment.

Population triage done by artificial intelligence?

The emergence of technological tools that can automate the prioritization of patients according to their need for medical care has the great potential to decrease the burden on emergency systems, free up time for nurses and physicians to treat patients, and mitigate the inconvenience of over- and under-triage.

In recent years, the application of Artificial Intelligence (AI) to the healthcare sector has been transforming the way medicine is practiced. Particularly, in the case of triage, the creation of Clinical Decision Support Systems (CDSS) has shown that by applying Machine Learning (ML) and Deep Learning (DL), two of the major components of AI, the predictive capacity of healthcare personnel in the prioritization of patient care can be equaled or even improved (3).

However, it is important to mention that we speak of Decision Support Systems because a complete automation of the triage process is impractical. Just taking as an example the International Statistical Classification of Diseases and Related Health Problems (ICD) we see that in just the last few years the number of conditions has increased from 12,420 to 68,000 making the variability in the conditions that are presented daily to be prioritized quite high.

Do you need to know how to code to make a clinical triage tool with Artificial Intelligence?

The technology behind these CDSS is quite complex in general terms. To build a CDSS, not only do you need a broad knowledge of the diseases and the triage levels associated with those conditions, but you also need to correctly manage the processes necessary to develop AI algorithms. This represents a barrier in the development of these tools, especially considering that physicians, and healthcare personnel in general, do not have this knowledge in AI.

Consequently, Arkangel AI has made available to the entire healthcare community Hippocrates AutoML, a tool that allows the creation of AI algorithms, following all quality standards, without the need to write a single line of code.

What inputs does Hippocrates need?

Like any other AI-based platform, algorithms trained with Hippocrates require data. In the case of a CDSS for triage, these data must follow a similar relationship to those obtained in the manual screening process, i.e., it is necessary at least to collect information on the patient's symptoms, this also represents a challenge for health institutions as they usually have not raised an organized data architecture (if this is your case we recommend you to talk to one of Arkangel Ai's agents to know what alternatives you have to start organizing your data even if they are not structured). However, the interesting thing about AI is that, unlike healthcare workers, it can analyze large amounts of information in short periods of time. Therefore, for the construction of the CDSS it is also relevant to include information that could be relevant for triage: patient's medical history, baseline conditions, previous assessments, among others.

With this information and the "gold standard", i.e. the expected prediction result of the algorithm, Hippocrates is able to build and test an algorithm to support the prioritization of patient care. This algorithm should group the different existing triage levels, so that the expert receives notifications that allow him to make decisions based on data and with a lower degree of subjectivity.

Where in the clinical workflow can Triage done by Ai be implemented?

Emergency departments are not the most ideal place to perform this triage process, even with the help of an AI tool. Therefore, a good idea for CDSS is to design them in the form of Chatbots or interactive platforms for remote consultation, allowing to reduce the high flow of patients in the hospital system. Likewise, when performed remotely, it is possible to impact the management of resources in a comprehensive manner, since not only the hospital care of patients is prioritized, but also the allocation of ambulances for their transfer. This is of vital importance, considering that in recent years ambulance reaction times have increased dramatically (4), affecting the quality of care in the emergency system.

Finally, by performing triage remotely, CDSSs help to adequately inform patients of the urgency of the medical care their condition requires. This not only alleviates the fear that patients feel when they feel sick, but also helps the healthcare system to intelligently assign variants in medical care, such as home visits, teleconsultation appointments, priority appointments, among others.

**The use of Artificial Intelligence for the prioritization of medical care represents a great advantage for patients, doctors and insurers alike. If you are interested in developing an Ai-based tool or implementing one of the tools developed by Arkangel Ai let us (meet you), there is a lot of value to be exploited.

(1) National Hospital Ambulatory Medical Care Survey: 2018 Emergency Department Summary Tables

(2) Accuracy of emergency department triage using the Emergency Severity Index and independent predictors of under-triage and over-triage in Brazil: a retrospective cohort analysis.

(3) Artificial intelligence algorithm to predict the need for critical care in prehospital emergency medical services

(4) Intelligent Call Triage System with Algorithm Combining Decision-Tree and SVM

If you want to know more about Arkangel Ai contact us here and one of our team members will contact you for a one-on-one session.

‍