(UroToday.com) At the American Urological Association Annual Meeting, Dr. Katherine Fischer presented her and her team’s development of a machine learning model for spontaneous ureteral stone passage. Fischer et al. are attempting to overcome the hurdle that prior studies have had difficulty efficiently defining specific factors that predict the likelihood of spontaneous passage.
The team did a retrospective cohort study to gather a patient dataset to train their random forest model. 256 patients were retrospectively chart reviewed for clinical CT scans and patient characteristics that do or do not indicate spontaneous ureteral stone passage. Due to the major differences between pediatric and adult patients (n=103 and 153, respectively), the team created two distinct models for each group.
Figure 1: Feature Importance for pediatric (A) and adult (B) models.
The pediatric model demonstrated 70% accuracy with important features to be stone size (area), age, prior stone episodes, nausea, and renal pelvis dilation. Similarly, the adult model demonstrated 63% accuracy with the important features to be stone size (area), age, location of stone, dilated renal pelvis, and fever. Dr. Fischer shares that stone area was the most important feature for both models as seen above.
Attendees were surprised that stone composition was not one of these factors for either model; however, Dr. Fischer explained that stone composition was in fact incorporated into the model, but no significance was seen for it to be considered an important feature. Upon speaking with Dr. Fischer, she is optimistic about the future of this model, especially in its potential utilization in clinical practice as a supplement to other investigative measures. She concluded her talk by stating that “more data is always better”, and shares that the team is aiming towards further improving their algorithm. “Our goal ultimately is to incorporate clinical data with automated deep learning with image”, she says, which will certainly bring the field closer to determining when surgical intervention is necessary (or not) for ureteral stones in patients.
Presented by: Katherine Fischer, MD, Division of Urology, Children's Hospital of PhiladelphiaWritten by: Amanda McCormac, Junior Research Specialist at Department of Urology, University of California Irvine, @Mccormacamanda on Twitter during the 2023 American Urological Association (AUA) Annual Meeting, Chicago, IL, April 27 – May 1, 2023
References:
- Fischer, K., Singh, A., Logan, J., Schurhamer, B., Cao, B., Daniel, R., McGregor, R., Nadeem, I., Uppaluri, C., Xiang, A., Choi, E., Li, Y., Fan, Y., Ziemba, J., & Tasian, G. (2023). MP16-01 MACHINE LEARNING MODEL TO PREDICT LIKELIHOOD OF SPONTANEOUS URETERAL STONE PASSAGE. In Journal of Urology (Vol. 209, Issue Supplement 4). Ovid Technologies (Wolters Kluwer Health). https://doi.org/10.1097/ju.0000000000003236.01