Scientists at Heidelberg University Hospital develop “Cognitive Medical Assistant” / Algorithm aims to identify individual surgical risk of the patient in advance, facilitate therapy decisions and prevent complications
In order to be able to assess and take into account a patient’s individual risk of complications as accurately as possible even before surgery, scientists at Heidelberg University Hospital want to use “machine learning” methods. As part of the “Cognitive Medical Assistant (KoMed)” project, an interdisciplinary team from the Clinics of Anesthesiology and General, Visceral and Transplant Surgery will train an algorithm over the next two years to evaluate a variety of clinical data from patients using Big Data analyses. The goal is to detect patterns in the data and identify correlations that can be used to create individual risk profiles. In the future, KoMed, which was developed together with industrial partners, will provide a sound decision-making aid to avoid complications through adapted treatment and care.
Previous risk scores are based on age, gender, and preexisting conditions, for example. They do not adequately reflect the actual complication risk of the respective patient. The KoMed will analyze a variety of available patient data to identify which characteristics are associated with increased or low risk of complications, such as wound infections or heart attacks. “This not only gives patients and treatment teams more certainty when making treatment decisions,” explains project leader Dr. Jan Larmann, senior physician at the University Department of Anesthesiology. “Assessing risk as accurately as possible also allows for targeted use of resources and thus provides economic benefits.”
“The risk of complications can only be reduced to a certain extent through further development of surgical techniques and anesthetic procedures. We urgently need more information about which patient characteristics are associated with an increased or reduced risk of complications in order to be able to treat patients in an individualized manner in the future,” says Professor Pascal Probst, M.D., Senior Physician at the University Department of Surgery and Medical Director of the Study Center of the German Society for Surgery (SDGC). Routine data and treatment histories of an initial 600 surgical patients will be collected as part of an initial observational clinical study. This data is processed in a structured and analyzable form and provides the basis on which KoMed learns to identify potential risks. Although data on underlying and concomitant diseases, from imaging, on the type and course of surgery, medication and blood values, as well as a large number of other measured values from clinical routine are already being digitally recorded, only a fraction of these are used for risk prognosis – the systems used for processing do not allow any analysis.
In addition, so-called proteome analyses are carried out on the patients in the study: These provide an overview of all proteins currently active in the body and thus an insight into metabolic processes, their alteration or disruption. “From the combination of the proteome data and the routine clinical data, we hope to gain a better understanding of the circumstances that lead to complications and the disease mechanisms that trigger them. This will make it possible to take targeted countermeasures in the future,” says Larmann.
At the end of the training phase, the system should be able to predict complications with an accuracy never before achieved. “We assume that this knowledge alone will help prevent complications because high-risk patients can be monitored more intensively in a targeted manner and treated earlier,” Larmann says confidently. While intensive medical care is often indicated for high-risk patients, KoMed, on the other hand, is intended to spare low-risk patients an unnecessary stay in intensive care: For example, if today a patient is automatically assigned to a high-risk group due to his age or the type of surgery, in the future KoMed will recognize a stable state of health and include it in the risk analysis. Before clinical use, however, KoMed must be trained with additional patient data and validated in an independent group of patients.
Source: Anesthesiology University Hospital Heidelberg (D)
