Saltar al contenido
Inicio » Blog » Data Science for Emergency Medicine

Data Science for Emergency Medicine

Data Science for Emergency medicine

Philips IntelliVue MX450 portable bedside patient monitor at work. Image by PickleTech.

In Emergency Medicine, decisions need to be made fast, and with a limited view on the patient’s health status. While the time urgency is imperative, in Emergency Medicine not all the decision outcomes are visible immediately.  Actually, some may take quite some time to show up, turning feedback and learning processes into a challenge. Limited resources impose in addition a very severe constraint, which makes Emergency Medicine too many times a complex resource optimization problem in nature.

Framed like that, it should seem natural to expect Data Science integrated in the operation of an emergency department. But as it turns out, Data Science for Emergency Medicine is still underdeveloped. We discuss here how Data Science has the potential to improve Emergency Medicine, focusing on the opportunities in such a complex context.


Imagine you have suffered a cardiac event, with your heart yet not naturally reverting to a sinus rhythm. The moment you get into the Emergency Department (ED) without a previous -nor clear- diagnosis, the complexity of the situation becomes quite daunting. Emergency health practitioners start a race to acquire enough information to understand your situation, and bring you back into a stable and safe condition.

What is the right procedure for your emergency?

Which tests should be done to decide on the procedure?

Incomplete data for diagnosis

At the beginning, their focus is on acquiring as much information as possible on your condition and context, in order to decide on the best medical procedure. Emergency doctors are not your family -recurrent- doctor, they instead have to rely on your electronic health records (EHR) to understand if something in your history may hint for an explanation and a procedure forward. This puts pressure on having good EHR systems, where relevant information is optimally available.

We could stop here and talk about how advances on AI and Natural Language Processing (NLP) are helping to better manage EHRs. But sometimes it turns out not even the results of the most recent electrocardiogram (ECG) you just had on the ambulance on the way to the ED are accessible. Interconnectivity improvements between medical systems are still a key necessity in Health.

Coming back to the imaginary case, after entering the ED you are now connected to the emergency life monitor and the other ED relevant systems. Emergency doctors need to start making sense of all the physiological data that is being recorded on your condition: heart rate, oxygen saturation, breathing rate, blood pressure, continued ECG monitoring, temperature, discrete measurements like glucose levels etc. Technology has made amazing advances in ED, with multiple reliable systems being intensively used already.

But data can actually be overwhelming if you are not used to it. You may be probably asking yourself how they distinguish so easily the warning monitor sounds that are relevant, from the ones they keep safely ignoring. On top of all that data, emergency practitioners need to filter and find out what extra information is there on your description of the cardiac event relevant to what may have triggered your condition.

What is missing in the existing data?

What other data could fill the gaps to proceed forward?

One could think it is all about getting as much information as possible. But testing requires time, and it consumes the limited hospital resources. Both are too precious and need to be spent wisely.

Incomplete immediate outputs

In this race against time, after a while your basic blood test is completed, together with some extra tests, like e.g. some chest images. The initial drug-induced treatments you got when entering ED have not been successful, another piece of useful data. With all this information the doctors already have at hand, what comes next?

After some additional drug-induced treatment prolonged over a few hours, emergency doctors decide to try something else. They proceed and perform an electrical cardioversion. That brings your heart back to normal. This is actually really good news, an invaluable immediate reward nevermind! But this is yet only a part of the whole picture.

Medical knowledge is built on Statistics. Even if that may not be obvious, Statistics plays quite a key role in the decision of which is the best procedure to move forward with your treatment. However, Emergency Medicine faces a challenge here. In the ED, the focus is on saving and stabilizing a patient’s life. Hopefully, after that is a success, you will follow your diagnosis path outside of the ED and get a complete diagnosis and long term treatment. Then you will learn what was the underlying reason for that cardiac incident that required emergency medicine. Additionally, you will also be past the side effects and consequences you may experience because of the procedures in the ED.

In other words, part of the data that is missing during the emergency, is eventually collected with time. And once all that information finally makes it to your EHR, we can close the feedback loop and answer important questions for Emergency Medicine.

Was that the right treatment procedure for your emergency?

Was there a better path for your health?

Which tests could have been done to speed up the procedure?

Which steps (if any) were unnecessary?

Could you have been released earlier from ED saving time and resources?

Let’s build complete feedback processes!

It turns out, Emergency Medicine could benefit from Data Science and a statistical learning approach. All the processes above can be framed in terms of a data collection step, followed by feature engineering, feature selection, and a decision system leading to both immediate and long term outcomes. You may exchange outcomes by rewards if you prefer to think of it as a Reinforcement Learning system. What is important is that provided enough complete cases are recorded, a Machine Learning (ML) model could be trained to find the best statistical learning approach to your ED path.

We are not talking about a superintelligent AI system that automates all the decision process and removes the emergency doctors from the picture, NOT AT ALL! Instead, we devise a system that will use all the data that we are already recording on cases and procedures, to bring the statistical reasoning behind Emergency Medicine one step up. Using ML, this system will eventually provide live recommendations on possible ED procedures. This includes recommendations on which tests and data to record next. Overall, the system will objectivize even more the risks and expected outcomes of every action practitioners take, given the known-visible patient data at every moment.

This type of AI solutions already exist when troubleshooting technology, Health should benefit from analogous solutions. There are of course quite some challenges ahead specific to the Health context, particularly around data access, patient privacy, system interconnectivity, plus the extra sensitivity in the field to the dangers of untrustworthy AI systems. However, we have regulations that should be followed to facilitate impactful developments that are beneficial to society. And we possess the technical maturity that enables the development of trustworthy AI.

ECG excerpt containing 15 seconds of a sinus arrhythmia as recorded with a standard wireless heart rate monitor (Polar H10). The bottom panel depicts the heart rate of the same cardiac incident prolonged during 20 minutes as read by the same device. This information can be easily recorded and accessed through a mobile app (e.g. Polar Sensor Logger). This provides basic heart monitoring even before receiving direct care by an emergency medicine specialist. Image by PickleTech.

Further Innovation

Resource Optimization

The worrying limitation of resources in an ED is quite the elephant in the room when talking about Health, and particularly Emergency Medicine. Depending on the country and the circumstances (e.g. covid crisis), the imaginary experience described above could have had a very different -undesired- ending. And probably it would have involved quite some more hours of waiting for care. Managing the resources of an emergency department involves dealing with human resources including a wide spectrum of health practitioners of multiple types; the availability of facilities and resources for tests and treatments; plus all possible context knowledge on external factors and events that may lead to unusual high pressure on an ED. This is a complex optimization problem that can be approached taking advantage of Data Science improvements.

AI diagnosis systems

Data Science developments are also focusing on other solutions relevant to Emergency Medicine. For instance, early diagnosis solutions based on AI can speed up the decision process of emergency practitioners for health events, see e.g. [6],[7], and [8] in relation to brain incidents. In emergency cases where the speed of the decision has a direct effect on the subsequent incident injuries, having AI systems to provide fast assessments can be life-changing.

Remote Monitoring

When interconnectivity between different hospital and ED systems is still a big issue, adding patient managed devices to the picture may seem way too ambitious. But health wearables are spreading. They are starting to enable the remote monitoring of physiological data. Nowadays, we have access to for instance portable ECG devices that can track our heart with relatively good precision. They can provide relevant information on that cardiac incident described above, without the need to be physically present inside the ED. Have a look at the figure above for a very illustrative example. AI solutions help process that data and look for early detection signals. Finding a way to integrate wearable data and AI in the operation of hospital departments may be a way forward to save resources and speed up diagnosis and treatment processes.

Powered by Data, Driven by Science

Emergency Medicine is a fascinating field requiring emergency specialists to have a broad knowledge and multiple procedural skills. Integrating Data Science solutions within the procedures of an Emergency Department has the potential to improve the outcome of such a vital care system.

At PickleTech, we work developing tailored solutions to improve competitive aspects related to Salud, Deporte, and DeepTech. We believe Data Science y advances in Machine Learning coupled with domain knowledge and experimentation have the potential to provide new tools to better understand, monitor, and systematically improve the performance of organizations.


Header Image: Philips IntelliVue MX450 portable bedside patient monitor at work. Image by PickleTech.


[1] Artificial intelligence and machine learning in emergency medicine: a narrative review, https://onlinelibrary.wiley.com/doi/full/10.1002/ams2.740

[2] Artificial intelligence in emergency medicine: A scoping review, https://onlinelibrary.wiley.com/doi/full/10.1002/emp2.12277

[3] Priorities to Overcome Barriers Impacting Data Science Application in Emergency Care Research, https://onlinelibrary.wiley.com/doi/full/10.1111/acem.13520

[4] Training for a Reward, https://pickletech.eu/blog-rl/

[5] Algorithm Audit for a Trustworthy AI, https://pickletech.eu/blog-audit/

[6] https://tbicheck.com/company-profile

[7] http://www.tibtimeisbrain.com/

[8] https://www.methinks.ai/

[9] https://www.polar.com/en/

[10] https://play.google.com/store/apps/details?id=com.j_ware.polarsensorlogger

ES