Software created from these “building blocks” can incorporate artificial intelligence and support medical staff

Software created from these “building blocks” can incorporate artificial intelligence and support medical staff

This article has been reviewed in accordance with Science


A diagram showing how to configure and use EASUL-based tools in two different ways. (a) For research, quality and service improvement using static datasets, scripting and analytics in Python. (b) Create a prototype of the CDS tool by integrating the outputs/results into the clinical information system. *ADT = hospital admissions, discharges, and transfers. In all cases, the plan is initially defined using Python classes. These plans act as containers for available reusable components including the DataSource, Algorithm, State, and Visual classes. Key logic is included within the critical steps, which support different modalities algorithms – ranging from simple clinical risk scores and logical comparisons (if/then) to advanced machine learning and artificial intelligence models. The data received by the steps at certain points determines the specific patient journey undertaken. Once the plan is defined, it is executed using the engine – which includes the client, broker and clock. The client handles local storage of states and results, the broker provides/receives data to guide the plan and the clock handles the temporal aspects of flows. For example, in (a) the client was a SQLite database storing information for later analysis, the broker was a static SQLite database providing the input data and the clock was set to increment forward every hour within each CAP admission to simulate progress. Clipart from draw.io. credit: Frontiers in digital health (2023). doi: 10.3389/fdgth.2023.1237146

× Close


A diagram showing how to configure and use EASUL-based tools in two different ways. (a) For research, quality and service improvement using static datasets, scripting and analytics in Python. (b) Create a prototype of the CDS tool by integrating the outputs/results into the clinical information system. *ADT = hospital admissions, discharges, and transfers. In all cases, the plan is initially defined using Python classes. These plans act as containers for available reusable components including the DataSource, Algorithm, State, and Visual classes. Key logic is included within the critical steps, which support different modalities algorithms – ranging from simple clinical risk scores and logical comparisons (if/then) to advanced machine learning and artificial intelligence models. The data received by the steps at certain points determines the specific patient journey undertaken. Once the plan is defined, it is executed using the engine – which includes the client, broker and clock. The client handles local storage of states and results, the broker provides/receives data to guide the plan and the clock handles the temporal aspects of flows. For example, in (a) the client was a SQLite database storing information for later analysis, the broker was a static SQLite database providing the input data and the clock was set to increment forward every hour within each CAP admission to simulate progress. Clipart from draw.io. credit: Frontiers in digital health (2023). doi: 10.3389/fdgth.2023.1237146

New ‘building blocks’ approaches to creating digital tools involving data and artificial intelligence could play a key role in improving the operation of hospital wards and disease management, according to the findings of new research.

The study, led by Dr Robert Frey from the National Institute for Health and Care Research (NIHR) at the Leicester Biomedical Research Center (BRC), suggests that using a package of digital ‘building blocks’ to create clinical decision support software would make it possible to create digital tools. Helps medical staff prioritize patient care and workloads more effectively.

The study published in Frontiers in digital health It demonstrates how a system of computer software building blocks, developed by the research team, can enable faster and more effective disease management protocols to deal with local admissions for community-acquired pneumonia (CAP). It also explores how this approach can be applied across healthcare in general.

Their platform, Embeddable Artificial Intelligence and State-Based Understandable Logic (EASUL), can use historical data and electronic medical records and includes algorithms to develop digital platforms that accommodate different stages of patients’ clinical care and allow medical staff to examine this – including the likelihood. Patient outcomes.

In the study, researchers modeled scenarios using existing patient data and consultations with teams of specialist pneumonia intervention (SPIN) nurses to test how software created with EASUL could be used to help doctors manage patients admitted to CAP. The program obtained data on 52,471 adults admitted between April and June 2022, of whom 630 were diagnosed with CAP. The advice and information generated by the program was compared with clinical risk assessments provided by the SPIN team.

When scored, the EASUL risk assessment matched the SPIN difference 49.4% of the time. EASUL has never classified any patient as low risk who has been classified as high risk by the clinical team. EASUL also identified 57 cases which, when reviewed by the researchers, should have been classified as high risk but were only scored as low or moderate by clinical staff.

The paper’s authors stressed that the differences were likely due to individual clinical judgment where a comprehensive risk assessment was not considered clinically appropriate. Due to the lack of information available in existing patient data, it was not possible to include this item in the evaluation.

The researchers behind EASUL also believe that another potential advantage is its flexible design. It allows “fast” data, collected during treatment and research, to be easily included in the system. It has also been designed in a format that can be integrated with existing digital clinical decision support systems.

As a result, the EASUL can be modified to fit the needs of a variety of clinical settings. It is also designed to automatically adjust its calculations in the event of data loss, meaning it can provide robust and relevant information to clinical staff in a variety of different situations.

Dr Robert Frey, lecturer in health data science and lead researcher on the study, said: “This is a very exciting development. Our clinical proof-of-concept system has allowed us to show how our basic approach can handle algorithms of different complexity” across patient care. Using EASUL, we were able to embed both simple risk scores and a pre-existing AI model into a real-time data-driven workflow and then deliver it to clinicians – helping them make decisions about patients.”

Dr Pranabhashis Haldar, senior clinical lecturer in respiratory at the National Institute for Health Research in Leicester, and a contributor to the study, added: “The flexible nature of our approach means it can be scaled to support different types of data, and adaptive workflows including advanced AI models.” “And perhaps mobile applications. “In addition, they can also be used to support patient-directed healthcare procedures, such as remote monitoring.”

“We believe that EASUL and similar approaches are important steps to make better use of health data from multiple sources and will help strengthen trust and accountability in complex AI that enables clinical decision support,” Dr. Frye concluded. “However, we recognize that more research “They are needed before this can be deployed in active clinical settings.”

more information:
Robert C. Frey et al., A data-driven framework for clinical decision support applied to the management of pneumonia, Frontiers in digital health (2023). doi: 10.3389/fdgth.2023.1237146

Provided by University Hospitals of Leicester NHS Trust

You may also like...

Leave a Reply

Your email address will not be published. Required fields are marked *