Predicting non-response to urotherapy in pediatric bowel and bladder dysfunction: A machine learning approach
Jackson M Dunning1, Adree Khondker2, Christopher S Cooper1
1Department of Urology, University of Iowa Hospitals and Clinics, Iowa City, IA, USA.
Insights
Machine learning models can predict which pediatric patients with bowel and bladder dysfunction (BBD) are unlikely to respond to urotherapy alone. Identifying these non-responders early allows for timely adjustments to treatment plans, improving patient outcomes.
Area of Science:
- Pediatric Urology
- Medical Informatics
- Machine Learning in Healthcare
Background:
- Urotherapy is a primary treatment for pediatric bowel and bladder dysfunction (BBD).
- A significant portion of pediatric patients exhibit limited or no response to urotherapy alone.
- Early identification of non-responders is crucial for optimizing management and improving outcomes.
Purpose of the Study:
- To develop and evaluate machine learning models for predicting non-response to urotherapy in pediatric BBD patients.
- To identify key predictors of urotherapy treatment failure in children.
- To enable earlier clinical intervention for non-responsive cases.
Main Methods:
- Retrospective analysis of 123 pediatric patients (aged 5-10 years) with BBD.
- Utilized a validated 18-question BBD symptomology questionnaire for data collection.
- Developed and compared multivariable logistic regression and random Forest machine learning models to predict urotherapy non-response, evaluating performance with AUROC.
Main Results:
- Older age and the presence of dysuria were associated with a higher likelihood of response to urotherapy.
- Daytime incontinence at the initial visit was linked to a lower likelihood of response.
- The random Forest model achieved a superior AUROC of 0.71 compared to logistic regression (0.67).
Conclusions:
- Machine learning models effectively predict non-response to urotherapy in pediatric BBD.
- Age, dysuria, and daytime incontinence are significant predictors of urotherapy outcomes.
- Early identification of non-responders facilitates timely implementation of alternative or additional therapeutic strategies, enhancing patient care.
Introduction:
Urotherapy remains the first-line conservative treatment of pediatric bowel and bladder dysfunction (BBD), however, some patients show limited or no response. Early identification of patients likely to fail urotherapy alone could influence early management and outcomes. This study aimed to develop predictive models to identify pediatric patients unlikely to respond to urotherapy alone (Summary Figure).
Methods:
A retrospective cohort of 123 pediatric patients aged 5-10 years diagnosed with BBD who completed a validated 18-question BBD symptomology questionnaire at their initial and follow-up visit was analyzed. Patients underwent urotherapy as the primary intervention and symptom improvement was assessed at 6 months or less through a standardized scoring system. Machine learning (ML) models, including multivariable logistic regression and random Forest classifiers, were developed to identify predictors of non-response to urotherapy. Model performance was evaluated using area under the receiver operating characteristic curve (AUROC).
Results:
123 patients met inclusion criteria with 92 (75 %) females, and the median age was 6 years (IQR 5, 8). The median time from the initial to the next follow-up visit was 3 months (IQR 1, 4). Overall, 26 (21 %) patients had complete improvement, 28 (23 %) had moderate improvement, 23 (19 %) patients had minimal improvement (19 %), and 46 (38 %) had no improvement. Older age (OR 1.45, 95 % CI 1.09, 1.98; p = 0.01) and presence of dysuria (OR 1.54, 95 % CI 1.06, 2.37; p = 0.03) at initial visit were associated with an increased likelihood of response to urotherapy, whereas the presence of daytime incontinence (OR 0.67, 95 % CI 0.46, 0.97; p = 0.04) was associated with a lower likelihood of response. The logistic regression model achieved an AUROC of 0.67, while the random Forest model slightly outperformed it with an AUROC of 0.71.
Conclusion:
ML models using demographic and standardized questionnaire data predicted non-response to urotherapy in pediatric BBD patients. Age, dysuria, and daytime incontinence were identified as significant predictors. Early identification of potential non-responders could permit clinicians to implement additional therapeutic strategies sooner, improving overall patient care and outcomes.
Related Concept Videos
01:19Urodynamic Studies: Uroflowmetry
01:26The Micturition Reflex
The process begins with bladder filling, where the bladder wall stretches as urine accumulates. This stretching activates the urine storage reflex, mediated by the sacral spinal segments and the pontine storage center. Efferent sympathetic impulses stimulate the detrusor muscle to relax and the internal urethral sphincter to contract, facilitating...