103 Analysis of clinical procedure activity by diagnosis. (23rd February 2023)

• Record Type: Journal Article
• Authors: Veiga, David Peinador; Pissaridou, Eleni; Bowyer, Stuart A; Booth, John; Briggs, Lydia; Bryant, William A; Key, Daniel; Shah, Mohsin; Spiridou, Anastassia; Sebire, Neil J
• Is Part Of: Archives of disease in childhood. Volume 108(2023)Supplement 1
• Page Start: A38
• Page End: A38
• Publication Date: 2023-02-23
• Subjects: Children -- Diseases -- Periodicals; Infants -- Diseases -- Periodicals; 618.920005
• DOI: 10.1136/archdischild-2023-gosh.103
• Languages: English
• ISSNs: 0003-9888

Abstract

Abstract : Great Ormond Street Hospital NHS Trust (GOSH) has gathered a range of electronic health record data on diagnoses, interventions, and outcomes for its patients. These resources represent a database of experience that can be used to develop innovative data-based tools that could complement more traditional information sources for clinicians to provide care. We explored how a data processing and analysis pipeline could be created that provides information to clinicians about how the current diagnoses of a patient affect probable future procedures. In particular, which procedures are most likely to be required, and when. Based on dummy data (i.e., not from real patients) of diagnoses and procedures generated by the GOSH Digital Research Environment (DRE), a machine learning model was developed. The model was based on an array of logistic regression models, each trained to predict the probability of one procedure. The weights learned by the models were used to identify which diagnoses made a specific procedure more likely. The pipeline also used survival analysis methods to predict and present the time scale in which a procedure might be required. This information is displayed on plots of how the risk of a procedure changes over time after a diagnosis. The modelling and analyses were carried out in Python on the Aridhia secure DRE. The model was trained on 4000 dummy patients and validated on 1000 dummy patients, achieving an average AUC score of 0.72 in the validation set. This study shows how routinely acquired electronic health record data can produce personalised predictions of which procedures a patient is likely to require, and when, to support clinical decision making, resource planning and patient education. These tools were developed using dummy data that respect real patient data properties, which showcases the potential of dummy data techniques to innovate while keeping high privacy standards.