Article Text

Download PDFPDF
373 Enhancing usability of a clinical decision support system in endometrial cancer, using ENDORISK
  1. Anna Kleinau1,2,
  2. Marike Lombaers3,
  3. Steffen Oeltze-Jafra1 and
  4. Johanna Ma Pijnenborg3
  1. 1Peter L. Reichertz Institute for Medical Informatics, Hannover Medical School, Hannover, Germany
  2. 2Otto-von-Guericke University, Magdeburg, Germany
  3. 3Dept. Obstetrics & Gynecology, Radboud University Medical Center, Nijmegen, The Netherlands

Abstract

Introduction/Background Advances in artificial intelligence (AI) can greatly enhance clinical decision making. However, it is crucial for Clinical Decision Support Systems (CDSS) integrating AI to ensure high usability, and transparency in explaining the model’s predictions. The ENDORISK-model is a Bayesian Network (BN) that integrates clinical and molecular biomarkers to predict the risk of lymph node metastasis in patients with endometrial cancer preoperatively [3]. It has been validated in independent cohorts with an AUC of 0.81 and aids in clinical decision making [1,3,4]. In preparation of implementing ENDORISK in clinical practice, the CDSS for ENDORISK was iteratively improved to ensure high user friendliness and explainability.

Methodology The initially developed and formally evaluated user interface of the CDSS for ENDORISK served as a starting point [2]. The user interface was iteratively improved for both the usability and explainability. User-tests with clinicians (n=6) with expertise in endometrial cancer treatment from the Netherlands, Germany and Italy were conducted using a semi-structured test guide. Tests were recorded and documented to allow for direct implementation of changes after each test.

Results Usability improvements include an enhanced tutorial for novel users, presets for common tasks, and complete translations in English, Dutch, and German. Explanations of model predictions have been improved by adding new detail views of individual BN nodes, replacing color-coded percentages with more accurate bar charts, and adding support for minimizing likelihoods in addition to maximization. The user interface is compatible with similar AI models for other diseases. It will undergo long-term evaluation in Dutch hospitals over the next few years, and can be accessed at https://www.endorisk.eu/en.

Conclusion We have iteratively improved the usability and explainability of a clinical decision support system based on the ENDORISK-model for endometrial carcinoma therapy.

Disclosures -

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.