The aim of this session is to teach you about the possibilities of big data analysis of (retrospective) urodynamic traces, and to give you practical insights in how to apply this methodology in your own practice.

Learning Objectives

• Learning the possibilities of big data analysis for urodynamics
• How to prepare and process your data for analysis
• How to implement big data analysis in your clinical and research practice

Suggested Prior Reading/Learning

• Gammie A, Arlandis S, Couri BM, et al. Can we use machine learning to improve the interpretation and application of urodynamic data?: ICI-RS 2023. Neurourol Urodyn. 2023;1-7. doi:10.1002/nau.25319
• van Dort W, Rosier PFWM, Geurts BJ, van Steenbergen TRF, Kort de LMO. Quantifying bladder outflow obstruction in men: a comparison of four approximation methods exploiting large data samples. Neurourol Urodyn. 2023;42:1628-1638. doi:10.1002/nau.25270
• van Dort W, Rosier P. Classification of the Pressure-Flow Curve after Maximal Flowrate by using an Unsupervised Machine Learning Model. Continence 7S1 (2023) 100870. doi:10.1016/j.cont.2023.100870

Description
Urological departments performing ICS-standard urodynamic tests produce data on a daily basis that can be used for retrospective analysis. The only thing preventing departments from using this data for research is a practical understanding of the required methodology.

Why is big data analysis necessary
Traditional analysis of urodynamic traces often requires a certain degree of manual parameterization, which remains a valuable method of data collection, especially if entered into a registry for every new and historic patient. Creating a prospective registry is manageable, but parameterizing all historical data demands significant time and energy. Automated analysis can be an alternative method that makes big data analysis possible, and allows for the generation of robust evidence for long-standing research questions.
The ICI-RS hosted a think tank specifically on analyses using machine learning in urodynamics, and said that machine learning can be expected to have a range of useful applications in urodynamics due to the existence of extensive historical test databases, pattern recognition-based diagnostics, and assessments with inherent noisy data signals.[1] As urodynamics exhibits variability based on the skills of operators and interpreters, utilizing data processing methods that surpass human operators in objectivity and consistency may lead to more accurate diagnoses, ultimately resulting in potential benefits for patients.

How to start: data acquisition and quality control
If urodynamic traces are loaded into computational software, automated analysis can be performed without manual inspection of the traces. The process of exporting traces from a urodynamic system and loading them into computational software differs between the various urodynamics devices. Some systems require you to export each individual patient by hand. These repetitive tasks that require mouse and keyboard can be automated using downloadable software. Other systems allow you to export the entire dataset at once. There will also be variations in how the trace data is packaged, requiring you to tailor your algorithm to the specific manufacturer.
Before engaging in urodynamic trace analysis using classical processing methods, a data quality control check is imperative. Urodynamic data is known to exhibit a variability in measurement quality, even with experienced users, due to factors such as catheter artifacts or hindering rectal contractions on the abdominal pressure trace. While a measurement of lesser quality may suffice for manual diagnosis and analysis, automated parameterization increases the risk of obtaining inaccurate values, necessitating the exclusion of such traces. The workshop will dive into the automatic detection of catheter artifacts used in [2], which is also outlined in the supplementary material of [2].
After performing the necessary quality control steps, it is still possible that some low-quality data may persist in your dataset, but this should only be a small amount. Moreover, the advantage of utilizing a substantial amount of data (big data) is that the impact of a few low-quality measurements on the outcome is minimal.

How to process: Classical processing and machine Learning
Subsequently, the data can be analyzed to address your research question. The research question of [2] compared four existing evaluation methods for bladder outflow obstruction, which could be analyzed using classical processing methods. The research question of [3], however, aimed to describe unknown flow patterns, for which unsupervised machine learning was the ideal tool. Machine learning could also be used on noisy data, which includes data that was not quality controlled. The workshop will guide the participants through analyses using classical processing as well as using machine learning.

How to implement: challenges and outlook
Despite the methodology being available, applying these methods in one's own practice still requires some technical expertise. This aspect, along with ethical considerations, is addressed during this talk. The presentation concludes with a glimpse into the future, exploring the broader possibilities of big data analysis. As a whole, big data analysis can be employed to provide relatively robust evidence for long-standing research questions.
Ample time is allotted for discussion and practical questions during the workshop.

Key learning points

• Urodynamic traces can be exported to computational software for automated analysis
• When traces are not inspected manually, a data quality control check is imperative before using classical processing methods
• Continue using classical processing methods where sufficiently useful, but applications such as pattern recognition-based diagnostics are very suitable for machine learning
• Big data analysis can be employed to provide relatively robust evidence for long-standing research questions.

Take home message
With some technical know-how, automated analysis can be applied by all urological departments that have large retrospective urodynamic datasets.

Additional references
[1] Gammie A, Arlandis S, Couri BM, et al. Can we use machine learning to improve the interpretation and application of urodynamic data?: ICI‐RS 2023. Neurourol Urodyn. 2023;1‐7. doi:10.1002/nau.25319
[2] van Dort W, Rosier PFWM, Geurts BJ, van Steenbergen TRF, Kort de LMO. Quantifying bladder outflow obstruction in men: a comparison of four approximation methods exploiting large data samples. Neurourol Urodyn. 2023;42:1628‐1638. doi:10.1002/nau.25270
[3] van Dort W, Rosier P. Classification of the Pressure-Flow Curve after Maximal Flowrate by using an Unsupervised Machine Learning Model. Continence 7S1 (2023) 100870. doi:10.1016/j.cont.2023.100870