MORE EFFICIENT AND SAFER HOSPITAL EMERGENCIES WITH ARTIFICIAL INTELLIGENCE

MORE EFFICIENT AND SAFER HOSPITAL EMERGENCIES WITH ARTIFICIAL INTELLIGENCE

The ER is sadly notorious among the public for two things: its eternal struggle to get the resources it needs to function properly, and for the endless queues. The latter is the most obvious, but not the only, result of this lack of resources. Amalfi Analytics understands this problem, and therefore works to enable professionals to manage emergency services efficiently, and to anticipate problems that may occur in the future.

This is no easy task as the Emergency Department is a complex system par excellence. It is made up of different components (patients and professionals) with behaviours that individually are to some extent simple or anticipatable, but which generate, almost by surprise, global phenomena of great complexity. For example, patients almost always follow a similar and fairly simple path: they come in, go to triage, wait, have a visit, may undergo a test, and are either hospitalised or go home. But, if we make an imaginative effort to ‘zoom out’ and look at the whole picture from above, we will see queues that appear and disappear, patients following anomalous flows, particular trends in hospitals that are not found in others, and so on. Moreover, a small deviation, such as the absence of a few professionals, can mean big global changes, such as overcrowding in some waiting rooms, sudden drops in the quality of care and sharp rises in error and accident rates. Not to mention extraordinary and disruptive situations such as the pandemic we are currently experiencing, which put maximum stress on resources and professionals.

Another added difficulty is the service’s interrelationship with the outside world. Whether it be (a) other healthcare environments, such as the hospital’s bed supply or its crown of socio-health centres, (b) the social, political and economic context, such as the budget that the centre receives from the different public and/or private bodies, and (c) the physical context of the hospital, such as the weather, festivities, public gatherings or other events near the hospital.

We make a proposal, based on the use of existing data and Artificial Intelligence and Machine Learning techniques, that helps professionals who manage the emergency department to better visualise the complexity of their patients, and to prepare for situations of congestion or lack of resources.

Amalfi’s approach to designing a predictive management tool is simple. Work with the data available but above all with the professionals. The first step is always to identify the problems that the emergency team encounters on a daily basis. These are formulated together and a solution is imagined based on the hybridisation of expert knowledge of the clinicians with the technological possibilities.

Secondly, there is a thorough analysis of the available past data, both from a computational point of view and in terms of their actual meaning. After uncovering the problems qualitatively, they are explored quantitatively. The situation is analysed and described for the specific hospital being worked with. For example, to address problems of congestion and excessive waiting times, some first steps could be: (1) Identify the most common pathways among patients, according to processes to be attended, as each hospital has its own reality. For example, the type of patients arriving at an emergency department next to ski slopes will probably be different from those in an urban hospital less than five minutes from the beach in a large city. (2) Find out, with the help of data mining techniques, which are the key points of the flows, i.e. at which points and with which types of patients the waits are formed and from here can (3) Assess the impact that a change of protocol or care circuits could have, for which specific types of patients. Similar procedures can be designed for each of the problems identified.

Subsequently, data, professional knowledge and findings are used to create predictive models to support the ED team’s decision making. Advanced Machine Learning techniques should be used for these predictive models, as we know that traditional statistical techniques tend to give results that are too imprecise: they tend to be linear and we are dealing with highly non-linear phenomena. Models must also be able to “explain” to the practitioner the reasons for their predictions. They cannot function as a black box, which only generates mistrust and causes predictive tools to be quickly abandoned. This is necessary for at least two reasons 1) That professionals can make the best decisions in the real world, and not only in the mathematical world, since they have the ultimate responsibility and because they have additional information, experience and real conditioning factors that the algorithm does not take into account. It is not a matter of decisions being made by the algorithm, but of the professional, with the information from the algorithm and his or her ability to interpret it and put it into context, making the best decisions 2) The models must be understandable so that, a posteriori, professionals can analyse which variables have influenced correct and incorrect decisions in the past. This information can help the hospital team to design and implement new protocols for the future.

Amalfi Analytics is a startup that applies cutting-edge Artificial Intelligence techniques to healthcare management. In particular, in the case of emergencies, we have designed a platform that automates the above steps and is capable, among other features, of

1) Predict emergency inflow, stratified by triage levels and process typology, from the next few hours to a week or more.

2) visualising and predicting short-term occupancy in the different areas of the emergency department

3) predict inpatient bed requirements in the following hours to days.

4) perform retrospective analysis of patient flows within the department and calculate a battery of process and efficiency indicators.

This platform can also indicate the human resources that will be needed to meet a future peak workload, as well as make predictions of expected absenteeism among professionals, in order to be able to cover absences in time. The result is a more efficient emergency department, with fewer unforeseen events, safer for the patient and a less stressful working environment for the professional.

Visit our website for more details of our APIS solution for emergency services.