To bridge this gap, the consortium “Right Data, Right Now” organized their first “medical data & pizza” meetup. The consortium consists of the Amsterdam UMC, OLVG, Free University, PacMed, the Municipality of Amsterdam and the Amsterdam Economic Board (Board). Jointly, they have organized these meetings with the goal of connecting data scientists and doctors in order to improve medical data science and patient care in the Netherlands. These meetups are also meant as a platform where projects and ideas are shared, and they ultimately aim for new projects to be started. This initial meeting discussed the benefits of using machine learning for medical problems.
AI is transforming the clinical advice process
Luca Roggeveen, who takes part in the research project “Right Dose, Right Now”, presented an innovative solution to treating patients with blood poisoning (i.e. sepsis). Sepsis is a serious infection that can result in death if not adequately treated. To treat patients with sepsis successfully, doctors need to know the correct dosing over time. Roggeveen and his team have come up with a supportive tool that combines personalized medicine with big data. Autokinetics (AutoK) predicts, through calculations, how the body will react to antibiotic drug intake. AutoK has direct access to the large amounts of patient data that are routinely collected. It then uses that data in pharmacokinetic models, allowing for a quicker as well as more precise dosing advice. Rather than having to wait for the lab results after each antibiotic intake, AutoK offers a more efficient and accurate dosing advice with minimum wasted time. This advice is subsequently either accepted or rejected by the doctor in charge.
Applying machine learning in health
Machine learning is the process of using algorithms to discover patterns in data with which future predictions can be made. An essential aspect of machine learning is the process of, firstly, identifying a problem to, finally, implementing its solution – otherwise understood as data mining or end-to-end learning. As mentioned by data scientist, Mark Hoogendoorn, “the benefits of machine learning for health are that larger amounts of data can be processed in a shorter time period. As well as providing more accurate diagnosis through the personalization of patient related data.” On the whole, machine learning makes diagnosing more efficient and accurate seeing that it processes more information in less time, and by providing that data with the right context. For instance, it can look at a person’s heart rate and make predictions based on general temporal patterns on heart rate fluctuation, as well as considering other variables, such as age and weight.
Some challenges are that end-to-end processes are quite time-consuming, especially data preparation. The latter takes up to 60% of the entire process. Data preparation is essentially cleaning and organising the data: going from raw data to data that can be used as data sets in algorithms. Another challenge is generating accurate predictive models that are explainable as well. Some machine learning techniques can produce accurate outcomes, but we do not (yet) now why or how the algorithms arrive at their solutions. In the highly regulated, evidence-based medical sector this is problematic. However, most current research focuses on explainable AI.
The challenges of using AI in the health domain
Currently, not a lot of actions are taken based on input delivered by technological predictions. For actions to be taken, according to Roggeveen, big data projects need to be tailored to the medical practice: “you can make a prediction with a 100% accuracy, but if it’s not clinically actionable and doesn’t help me improve my patient care, it is useless.” Also, when doctors are presented with a solution without an explanation as to why this is be the outcome, according to the algorithm, it is hard to estimate the value of such an advice. While any algorithm will deliver an outcome, there are as yet only few examples that are both explainable and actionable. Creating useable solutions in the health domain, as argued by Hoogendoorn, is complicated since the amassed data within the health domain is mostly unstructured. Lastly, applying machine learning in healthcare is very complex since the data has to be provided of context in order for it to be interpreted correctly and result in useful and reliable predictions. Nevertheless, these are not impossible problems to overcome. The current innovations are proof that machine learning solutions are viable if they keep developing and improving.
Medical Data & Pizza
After the speakers’ talks, everyone got the opportunity to ask them questions as well as to connect with each other while enjoying pizza. Most of the attendees stayed behind and exchanged ideas on- or recapitulated the main points. Towards the end of the event there were hardly any pizza’s left! If this appeals to you, and you have an interest in new developments within the fields of AI and healthcare, make sure to attend one of the meetups organized by the “Right Data, Right Now” consortium. You can also join the “Everything medical data science” group to keep up to date.
Our dream is that by 2025 residents of the Amsterdam Metropolitan Area will have gotten an extra two healthy years. We can achieve this by committing ourselves to the prevention of disease. Our region is excellent in the collection and analysis of data as well as in some specific medical fields. By cleverly combining this knowledge, residents can take control of their own health and stay healthy for longer. Along with the consortium “Right Data, Right Now”, we connect data scientists and doctors to stimulate innovation in healthcare. The broader aim of the Board is to capitalize on the opportunities of AI in Healthcare in the Amsterdam Metropolitan Area. Interested? Please contact Jeroen Maas, Challenge lead Health at the Amsterdam Economic Board.
For more reports from previous Medical Data plus Pizza Meet-ups , click here.