Clinical Decision Support
The integration of AI algorithms into clinical routine has so far been limited. CAIMed investigates methodological challenges in the evaluation of informatics diagnostics, the synthesis of machine-learned and evidence-based decision models, the handling of continuously learning algorithms and the integration of AI models into clinical processes, including hybrid models with ontological domain knowledge ( Human-centered AI) and explainability components.
The evaluation is based on studies with users, usability tests with prototypes, prospective diagnostic studies, simulated randomized controlled trials (RCTs), synthetic data and comparative clinical studies in the field of infectious diseases in pediatrics. Semantically integrated, heterogeneous medical data is used, in particular from the Medical Data Integration Center of the MHH. The creation of hybrid models and their interdisciplinary interpretation is carried out in collaboration with the Human-Centered AI group. During evaluation, the group cooperates closely with the Statistical Evidence in AI Systems.
Use Cases
The care of patients in the pediatric intensive care unit poses a particular challenge due to the heterogeneity, different age groups, complexity and unpredictability of the course of the disease. Severe and complicated clinical courses in critically ill pediatric patients often lead to partial or complete multiple organ failure (MODS). If this is not just temporary but terminal, it can lead to the death of the individual. Length of stay and ventilation, morbidity and mortality, as well as the occurrence of severe, sometimes life-limiting long-term complications are significantly influenced by the occurrence of MODS.
One of the greatest challenges is to identify early predictors of complicated disease progression and subsequent morbidity and mortality from the large amount of quantitative routine clinical data. It is also difficult to reliably predict the specific effects of the disease on the various organ systems affected. This uncertainty prevents efficient and rapid treatment of organ dysfunction and progression to organ failure, which can have undesirable consequences for the patients concerned.
Based on a multimodal and interoperable, highly complex pediatric intensive care dataset that has been established in recent years, this project aims to address precisely these questions. The aim is to identify possible causal relationships in the course of the disease that determine the development and progression of MODS.
Infection and the resulting sepsis are a significant problem in the management of critically ill intensive care patients and affect both adults and children of all ages. Sepsis as a nosocomial infection, i.e. as a complication of treatment, by definition affects patients who are already seriously ill for other reasons. The occurrence of nosocomial sepsis has a significant influence on morbidity and mortality in close interaction with the underlying illness of the intensive care patient. Early detection is particularly important to prevent progression to septic shock, making this application of great clinical relevance.
Some risk factors for the development of nosocomial infantile sepsis, such as the presence of a central venous catheter or invasive ventilation, are now well known. However, the exact interaction of other possible influencing factors and the mechanisms of the transition from an infection to a septic clinical picture are not yet sufficiently understood.
The implementation of AI-based prediction models could enable early therapy and mitigate the negative effects of the disease on the outcome of critically ill children. Based on extensive preliminary work as part of the ELISE project funded by the Federal Ministry of Health from 2020-2023, causal methods are now to be applied in this use case in order to improve the understanding and knowledge of the pathophysiology and causes of nosocomial sepsis. The aim is to identify possible early signs that could contribute to the development and improvement of clinical decision support systems for the prediction of sepsis.
Cardiac surgery in children with the aid of the heart-lung machine (HLM) leads to a profound modulation of the maturing child's immune system and often produces a complex mixed picture of inflammation and immune paralysis. As a result, some of these highly vulnerable patients develop postoperative infections and associated organ dysfunction, which have a significant impact on morbidity and mortality. The exact clinical, procedural and immunologic factors that contribute to this imbalance and its consequences are not yet fully understood.
To address these challenges, we plan to analyze standardized and interoperable processed and enriched routine clinical data. The use of the already existing, large machine-readable dataset with outcome labels should lead to new insights through the continuous enrichment of the database and the application of causal methods. These include:
- Identification of patient groups at high risk of postoperative infections.
- The identification of sensitive early indicators in complex intensive care settings after HLM use.
- The data-supported planning and evaluation of molecular characterization studies, for example using multi-omics analyses.
These approaches aim to define predisposing factors for postoperative infections more precisely and to better identify early signs of infection. This could enable targeted, individualized treatment at an early stage in the future.
Team
Prof. Dr. Philipp Beerbaum, MHH
Mentor
PD Dr. med. habil. Thomas Jack
Gruppenleiter
Louisa Bode
Group member
Marcel Mast
Group member
