Medical Intelligence in healthcare: insights from Roel van Reij

What is medical intelligence?  

By medical intelligence, we mean supporting healthcare providers with smart, AI-driven solutions. The word ‘support’ is key here. Because unlike in other domains where AI sometimes makes decisions independently, medical intelligence always requires a human to monitor what’s going on. Decisions are made by a nurse or doctor. A good example is an algorithm that predicts the likelihood of complications after surgery based on risk profiles.

Additionally, medical intelligence must always be transparent. This is sometimes referred to as explainable AI. There has to be complete transparency as to how the algorithm arrives at a particular recommendation. However, this does not mean that deep neural networks (known as black boxes) cannot be applied in healthcare. They may prove useful in discovering completely new relationships in very large data sets. But then the next step is always to find an explanation for those relationships through research, so that a white box can be used when applying new insights in clinical practice.

What are the benefits?  

Medical intelligence is being used to address the three major problems in healthcare: staff shortages, growing demand for care, and rising costs. By providing AI systems to support decision[1]making, doctors and nurses can make better decisions faster. One example of this are the computer vision algorithms mentioned above, which are widely used in radiology and pathology departments.

AI can also help reduce the administrative burden; think of the use of speech-to-text technology that allows a nurse to record a reading by speaking rather than typing it in. Or conversational AI that summarizes action points for a doctor based on a conversation and creates a lay description for the patient, with links to web pages with more explanations about the disease and treatment. Or, taking it a step further, a chatbot that patients can use to ask questions they would otherwise have to ask the doctor or nurse. Of course, as mentioned earlier, a human is always involved.

What makes medical intelligence so complex?  

First of all, the many different systems and a lack of standards create a multitude of data islands: patient data is scattered across many different systems. It is therefore extremely difficult for a hospital to piece back together patient data in the data warehouse, let alone be able to centrally access and analyse information from different partners in the healthcare chain. Furthermore, domain knowledge is crucial. You need to know exactly what data are relevant to a diagnosis or decision. And you need to know how that data relate to each other. For example, is a particular blood value the cause of a disease or the result of a particular medication? And finally, the context also determines the relevance of the developed AI solution. For example, a predictive algorithm that works very well in a teaching hospital in the Randstad may not be of any use for the patient population of a regional hospital in Limburg or Drenthe.

Medical Intelligence in practise 

A good example of AI in practice is a dashboard that Conclusion Mediaan developed for a hospital in the south-east of the Netherlands. It predicts patient movements within the hospital based on statistical simulations. The need for such a solution arose during the Covid-19 pandemic, when it was critical for hospitals to be able to correctly predict bed capacity. Even now, at a time when hospitals are trying to work through a backlog of patients, bed capacity is a crucial factor in planning operations. To gain an insight into the inflow, throughput, and outflow of patients, Conclusion Mediaan developed a dashboard that uses large amounts of internal and external data (including CBS data on the patient population and weather data) to predict bed occupancy in different departments. The model not only predicts the number of patients admitted per department and specialization but also provides upper and lower limits with a certain degree of reliability. This allows the hospital to strategically plan for worst-case scenarios.