Medical decision-making especially in emergency situations is not easy. Tools like machine learning or the decision tree can help to evaluate the opportunities. We talked to Dr Mirjam Jenny, Head Research Scientist at the Harding Center for Risk Literacy, also about the digitalisation process in the medicine industry.
Mirjam Jenny is the Head Research Scientist of the Harding Center for Risk Literacy at the Max Planck Institute for Human Development (MPIB). Her research in decision making and risk literacy is located at the intersection of psychology, computer science, and medicine.
Currently, she is developing decision aids for the medical domain for example, fast and frugal decision trees for emergency medicine using machine learning methods. Being actively engaged in the German health care landscape she hopes to improve the medical decisions of patients, doctors, and institutions. After receiving her PhD in psychology at the University of Basel, Switzerland, she spent her postdoc at the Center for Adaptive Rationality at the MPIB where she won the Otto Hahn Medal awarded by the Max Planck Society. She then moved into the health care sector and spent one year at the National Association of Statutory Health Insurance Physicians as a data scientist.
Decision Trees
How can the process of decision-making for a doctor in an emergency situation be improved?
Emergency physicians face the challenge of having to make both accurate and quick decisions about a very diverse patient population. Some patients that present at the emergency department are in life threatening conditions while others only have minor ailments. Today, emergency physicians already make many very good decisions and judgments on a daily basis. At the university hospital in Basel we could show for example that physicians’ intuitive judgment of how ill a patient looks reliably predicts patient outcomes in many patients. Some patients, such as patients with nonspecific symptoms like weakness or fatigue, are very difficult to triage and diagnose, however.
What we need to develop for these patients are simple and quick heuristics that tell the physicians which pieces of information to focus on and how to combine the most relevant pieces of information into reliable decisions. In such situations, less is often more. For example, simple decision trees consisting of 3-5 questions can help physicians to quickly assess whether their patient with nonspecific symptoms suffers from something serious or not. While these fast-and-frugal decision trees are developed based on large medical data sets and complex machine learning algorithms, they are transparent and easy to use and do not require a computer at the point of decision. To our experience, physicians therefore readily accept them. Physicians’ intuitions about the patients can even be included into the decision tree’s structure.
Algorithm versus Physicians’ Experience
What are the consequences and possibilities of combining machine learning, psychology and medicine?
Combining machine learning, psychology, and medicine allows the development of decision tools for physicians that are as predictive as highly complex machine learning algorithms yet tailored to how human beings think. Fast-and-frugal decision trees are one such class of tools. They allow physicians to make highly accurate decisions quickly without having to turn away from the patients towards the computer. Due to their transparency they can further be analysed and their decisions can even be refuted if necessary. Refuting black box algorithmic decisions impossible and poses a legal problem in the age of informed consent.
“While computers can process large amounts of data in little time, they lack relevant context knowledge.”
Human oversight is therefore paramount. If our algorithm would deliver a decision tree that goes against physicians’ experience, we would take a step back to determine why this is the case. The algorithm may have been fed wrongly coded data, it could have missed data, or something else may have gone wrong. To develop truly powerful tool we therefore need interdisciplinary teams consisting of computer scientists, psychologists or decision scientists, and field expert – in the field of medicine physicians and other medically trained experts.
Industry 4.0 in the Medicine Sector
The digital revolution is changing a lot of different industries. What impact has the digital transformation in the health care system?
“The digitalisation process has just started.”
In many countries including Germany, the digitalisation of the health care system has barely started. The biggest novelties are still Dr. Google and health apps. If we manage to collect reliable medical data via electronic health care records in the future while maintaining patients’ privacy, research will be able to make new discoveries and hopefully translate them into improving medical practice. These data sets will hopefully allow us de develop reliable decision tools, diagnostic tools, and management tools.
Further, communicating with an expert physician will hopefully become possible for patients in rural areas via online consultation. And if we manage to learn from the emerging electronic data sets using reliable methodologies and algorithms, we can hopefully detect illnesses, particularly rare diseases, more reliably and detect currently unreliably diagnosed conditions such as celiac disease earlier. States must get organised and enable public organisations to reliably collect and analyse data, however. Otherwise we run the risk that reliable diagnosis of disease will become a private business.
Global Female Leaders 2017: A Thorough Look Back
To sum up this great event in a single blogpost would be nearly impossible. Fortunately, we don’t have to try this, because our dear partner Accenture provided us with an in-depth White Paper, that you can now download on our website!
Just fill out the form below and enjoy over 50 pages of insights and key takeaways from the 2017 summit. We thank Accenture for their contribution and hope you enjoy the read!
