The potential of surgical data science

BY LENA MAIER-HEIN, MINU TIZABI AND CHRISTOPH MICHALSKI 

Modern surgery saves lives every day, but operations also come with risks. Worldwide, an estimated 4.2 million people die each year within thirty days of a surgical procedure. Despite a high level of care, surgery in Germany is not yet as safe as it could be. Particularly in complex operations, the less experienced the surgical staff, the higher the complication rate. Experienced surgeons make better decisions, have better technical skills and work more efficiently. Thus, how safe an operation is depends not only on the type of surgery, but also on where it is performed.

This is where the fledgling research field of surgical data science comes in. By systematically collecting and analyzing data, primarily using artificial intelligence methods and novel sensor technology, it is possible not only to significantly improve the quality and safety of surgery, but also to significantly reduce costs for the healthcare system. As procedures become more complex, the learning curves of less experienced surgeons could be shortened, thus avoiding complications due to inexperience. The path leads away from eminence-based surgery, in which the skills of individual experts benefit a few, to data-driven surgery, which can ensure consistent quality and the objectively best possible treatment for all patients.

Finding solutions is only possible together

AI-based image and data analysis opens up completely new possibilities for surgeons to make operations safer. This is demonstrated by numerous research approaches, such as intraoperative navigation systems, novel imaging of blood flow and other functional tissue parameters, and intelligent prediction of therapy outcomes. Despite initial successes, the big breakthrough is still missing. Of the first 520 AI products certified by the FDA, only five are from surgery.

The reasons: Data is limited, and it is of poor quality and highly variable. In addition, they must be aggregated from heterogeneous sources and AI algorithms must be integrated into clinical workflows. Many of the barriers to safe, AI-assisted surgery stem from the lack of digital infrastructure. But regulatory limitations also stand in the way of successful development. All these hurdles can only be overcome if all stakeholders and their interests are involved in finding solutions. In 2022, a roadmap for the clinical implementation of AI in surgery was published in cooperation with more than 50 institutes worldwide. It emphasizes the need to create a structured data foundation on a large scale.

Gaps in collection and processing

In a decentralized country like Germany, it is logistically difficult and expensive to collect clinical data. The German hospital IT system is strongly billing-centric. This means that - in addition to legally required quality data - only a few data necessary for cost accounting are recorded. Over the entire hospital stay, the systems miss out on what is essential for patients: a wealth of useful data points that could be used to analyze, individualize and improve treatment in real time. But when such data is recorded, for example in the context of minimally invasive surgical videos, it is hardly ever saved and even less often annotated, i.e. provided with the necessary annotations for further machine processing. In addition, there is a multitude of blind data spots: Haptic or acoustic information, for example, can play a major role in intraoperative decision-making, but is currently not captured by any sensor in the operating room.

If this wealth of structured clinical data were available, many things would be possible. Based on individually tailored patient wishes, an AI could select the most promising therapy in each case and help to implement it safely using the latest technologies in imaging and robotics. Intersectoral communication between healthcare providers could also be optimized. Initial goals, such as the automated creation of doctor's letters using AI, would be achievable in the near future with comparatively little effort and could already give medical staff more time for actual patient care. Similarly, AI could automatically schedule appropriate appointments with patients based on a synopsis of all diagnostic findings. Such advances could then catalyze further innovation.

Focusing on opportunities

That AI can achieve all this and more is hardly in doubt. Large Language Models such as the ChatGPT program now even pass the infamous Turing Test. This means that a human can no longer distinguish whether he or she is talking to a machine or a real person. The unthinkable has also been achieved in terms of creativity and originality: only recently, an AI discovered a new method for a basic mathematical arithmetic operation, matrix multiplication, 
which outperforms the best known method to date in terms of efficiency. It is reasonable to assume that AI will eventually clear just about every hurdle.

This rapid progress may seem frightening at times, but it offers a unique opportunity. It would not yet be too late for Germany to secure a leading position in the international competition for the medical technology of the future. Particularly in the combination of AI with innovative hardware solutions, which is indispensable for clinical translation, there is great potential. This is where policymakers have a duty to act quickly. In order to be able to play a leading role internationally, there needs to be a paradigm shift away from billing-centric and toward patient-centric IT infrastructures. Specifically, this requires implementing the digitization strategy of the German Federal Ministry of Health, including the assignment of a uniform identification number for patients. In addition, it must be mandatory to standardize data formats and to be able to automatically export data in systems from different manufacturers.

This will benefit not only the healthcare system, which is reeling from dwindling human resources and exploding costs, but above all the patients: they can go under the knife in the knowledge that they are receiving the best possible and safest surgical care - thanks to the data.

Reference and citation:

This article originally appeared in the Frankfurter Allgemeine Zeitung (FAZ).

Professor Dr.-Ing. Lena Maier-Hein is head of the Department of Intelligent Medical Systems at the German Cancer Research Center (DKFZ) in Heidelberg.

Minu Tizabi, MD, works as a research associate in the same department.

Professor Christoph Michalski, MD, is Medical Director of the Clinic for General, Visceral and Transplant Surgery at Heidelberg University Hospital.