Informatorische Komponenten eines Lernenden Gesundheitssystems für die Versorgung von Patienten mit chronischen Wunden: Elektronische Patientenakten, Interoperabilität und KI-gestützte Prognosemodelle

Please use this identifier to cite or link to this item:
Open Access logo originally created by the Public Library of Science (PLoS)
Title: Informatorische Komponenten eines Lernenden Gesundheitssystems für die Versorgung von Patienten mit chronischen Wunden: Elektronische Patientenakten, Interoperabilität und KI-gestützte Prognosemodelle
Authors: Hüsers, Jens
ORCID of the author:
Thesis advisor: Prof. Dr. Swen Malte John
Thesis referee: Prof. Dr. Ursula Hübner
Abstract: Introduction: In the recent years, a growing body of clinical and health information has been stored digitally, e.g., in electronic health records. This trend is expected to continue and will contribute to the crucial role of digital information as an essential source for clinical data-driven knowledge generation. Artificial intelligence systems that analyze clinical data are employed to accomplish learning and knowledge generation. This knowledge can then be used to support clinicians in clinical decision- making, thereby closing learning loops from the provision of health care, the production of more clinical data, to the extraction and modelling of knowledge. The concept of Learning Health Systems describes this idea of knowledge generation through its aim to learn from each patient’s data and continuously improve the caring of patients. Learning Health Systems were first proposed in 2007 as a socio- technological concept where social and technological components are synchronized to enable data- driven learning. In 2021, 14 years after the initial publication on Learning Health Systems, the German Council of Economic Experts on Health Care calls for its implementation in the German health system. The question remains if Germany is ready to do so. Thus, this thesis investigates components necessary to establish Learning Health Systems. This thesis’ investigation focuses on three crucial information-technological components: electronic health records as data containers, a standardized terminology for health data to be shared and analyzed, and artificial intelligence systems providing algorithms to analyze the data. Those components are investigated from the viewpoint of chronic wound care as an example. Wound care is knowledge intense, but evidence is lacking because national guidelines are sparse and outdated. Thus, wound care promises to benefit from data-driven continuous learning in Learning Health Systems. Research has been conducted for each of the three investigated information-technological components: The first two investigations focus on the availability of electronic health records and interoperable data structures that form the foundational building blocks of Learning Health Systems as they provide the necessary data for data-driven learning. The third investigation focuses on developing and validating an artificial intelligence-based prediction model that estimates wound complication risks, i.e., lower leg amputation, and can learn continuously from available health data. Based on this research results, this thesis evaluates and discusses whether the requirements – for the wound care use case – exist in Germany to build a Learning Health System on a local or national scale as suggested by the expert council. Methods: The first investigation modelled the diffusion process of electronic health records (EHRs) in German hospitals using Bass’ formula based on Rogers’ Diffusion of Innovation theory. This bass model describes its uptake in terms of innovation and imitation. To measure the diffusion of EHRs they were conceptualized as basic records that also comprise wound related information. The diffusion process was captured by five surveys conducted between 2007 to 2019. The second study investigated to which degree wound-related data items can be depicted in a standardized reference terminology, i.e., SNOMED CT, as a high degree of depiction would lead to what Learning Health Systems require for learning: interoperable health data. In the third initiative, a prediction model for wound care is developed, which focuses pars pro toto on patients with a diabetic foot ulcer. This model aims to use a-priori existing knowledge, which is subsequently updated by data to mimic continuous learning cycles of Learning Health Systems. A Bayesian modelling approach is evaluated to investigate the potential of this learning principle. Results: All three studies demonstrate that a Learning Health System for the care of patients with chronic wounds is in principle feasible. For the question of electronic health care diffusion, the results showed that after 2007, the electronic health records leaped to a high initial uptake to an adoption rate of 42.4 % in 2009. This leap was followed by stagnation leading to an adoption rate of 49.3 % in 2017. For 2020, the Bass model forecasted that 55.0 % of hospitals will have adopted an electronic health record. These data showed that about half of the German hospitals are in principle capable of analyzing basic wound related data digitally. However, Germany has a lower adoption than the US that reached a margin of 80% in 2015 The second research revealed that two-thirds of a consented wound care dataset could be mapped to the standardized reference terminology SNOMED CT with satisfying reliability. It thus proved that based on SNOMED CT, wound related information could – at least part– become interoperable, sharable, and analyzable. Third, the Bayesian risk model to predict amputations in diabetic patients achieved high validity using local data in combination with a-priori knowledge from a European study. Thus, Bayesian models were found to be promising to serve for data-driven, continuous learning in Learning Health Systems. Discussion: Combining the three findings, a clearer picture emerges of how ready Germany is to build a Learning Health System for wound care. The current availability of EHRs is likely too low to fuel a Learning Health System with data to learn from at a national level, However, those 50 % of the hospitals with an EHR may spearhead the process at a regional or local level. Furthermore, the German federal government has enacted laws that stipulate the use of those records from which Learning Health Systems can profit nationwide on the long run. EHRs are necessary but not sufficient conditions for Learning Health Systems, as they must contain information that is interoperable – here in this case interoperable wound information. Interoperability can be achieved using SNOMED CT. Thus, wound- related EHRs should use this reference terminology to be able to share information across institutions and settings. Third, the results encourage the use of Bayesian modelling approach to data-driven learning because it supports the continuous integration of data and the foundation of learning cycles that learn from every patient treated. It also allows the integration of knowledge from external studies and thus enables a local Learning Health System to rely on more than its own data. Although German hospitals as major hubs of EHRs must catch up as seen in the first study, Germany seems to be on the right way to approach the 100 % EHR adoption margin. The same applies to the use of SNOMED CT. Germany recently acquired a national SNOMED CT membership which puts German health care providers in a position to code wound related information in SNOMED CT and share them via EHRs. Despite these positive developments, the establishment of a national Learning Health system for wound care will take some time. However, regional, and local LHS seem to come soon within reach and can be leveraged using Bayesian models to avoid local or regional bias. In summary, this thesis shows what a path toward Learning Health Systems - not only in wound–care - might look like: interoperable domain-specific electronic records that provide data for Bayesian learning to improve health care delivery.
Subject Keywords: Lernendes Gesundheitssystem; Elektronische Patientenakte; eHealth; Interoperabilität; Künstliche Intelligenz
Issue Date: 4-May-2023
License name: Attribution-NonCommercial-ShareAlike 3.0 Germany
License url:
Type of publication: Dissertation oder Habilitation [doctoralThesis]
Appears in Collections:FB08 - E-Dissertationen

Files in This Item:
File Description SizeFormat 
thesis_huesers.pdfPräsentationsformat1,55 MBAdobe PDF

This item is licensed under a Creative Commons License Creative Commons