Digitalized Data Acquisition and AI for Safer Surgery
Complications can always occur during operations, sometimes resulting in death. The KIPeriOP project aims to minimize the risk of such complications. It seeks to develop digitalized decision-making support and self-learning algorithms that are designed to provide reliable risk assessment based on individual patient data: What is the probability that certain complications will occur, and how can they be avoided?
The KIPeriOP research project is set to receive funding of 1.5 million euros from the German Federal Ministry of Health. It is coordinated by Professor Anja Hennemuth from the Fraunhofer Institute for Digital Medicine MEVIS and Professor Patrick Meybohm from the University Hospital of Würzburg. Physicians from Asklepios Medical School GmbH, University Hospital Frankfurt, and Charité-Universitätsmedizin Berlin are also involved in the project. They are collaborating with specialists from the fields of AI, user guidance, ethics, and health economics, including Professor Saskia K. Nagel, chair of the Applied Ethics group in the Department of Society, Technology and Human Factors at the RWTH Faculty of Arts and Humanities.
Every year, more than 16 million surgical operations are performed in Germany. Complications occur time and again, not infrequently resulting death: In Western industrialized nations, 0.4 to 0.8 percent of those operated on die during or after surgery. One of the ways doctors seek to reduce the number of complications is by taking possible risk factors into account: What other conditions does a patient have, what medications are they currently taking? What complications could arise as a result, and how can the risk of complications be minimized?
There are guidelines that support medical staff in assessing the risks involved in surgery, which, among other things, list the type and number of useful preliminary examinations. However, in practice, these guidelines are not easy to apply: They are typically quite complex, and their application requires doctors to take a wealth of information into account that is not always easy to obtain.
This is where the KIPeriOP research project comes in: The aim of the interdisciplinary consortium project is to develop a clinical decision support system, or CDS for short. The software, developed by Börm-Bruckmeier Verlag, will first collect possible risk factors on a patient-specific and guideline-compliant basis, correlate them with each other, and provide a resulting risk assessment: How likely is it for a particular patient that serious complications will occur during or after surgery? "Based on this risk assessment, doctors can decide, for example, whether further examinations are necessary and what measures can be taken to optimally prepare the patient for surgery," explains Meybohm.
The system should be fed with as much information as possible about the patient, including lab values, medication schedules, vital signs, and information about lifestyle habits. In addition to taking the relevant guideline into account, KIPeriOP will deploy articificial intelligence methods to analyze all digitally collected data. Self-learning algorithms search for patterns and correlations to identify what constellations of risk factors are likely to result in what complications. The AI method would make it easier, for example, to detect a patient’s undiagnosed heart condition and thus their increased risk for surgery.
The project explores different AI methods to find an optimal model. For the AI to work reliably, the algorithms must first be trained, that is, fed with many data sets on preliminary examinations and outcomes of surgical interventions. These data are collected by the four clinical project partners. "We're not just collecting data that's already available, but we are able to tailor the data collection process specifically to our needs," explains Meybohm.
One of the challenges in developing the CDS system is to optimize it in terms of usability. "We have to design the AI-based solution in such a way that it supports the doctors in their activities and is not perceived as a burden," says Hennemuth. "Trust in the new technology can only be built if we clearly explain how and with what certainty the algorithms arrive at their results."
Accordingly, the AI system should not function as a black box – uncertainties and possible sources of error must be transparently communicated. The CDS system is developed in close coordination with the clinical partners, but also with RWTH’s ethics team.