Lund University Publications

Improving the Efficacy of Enuresis Alarm Treatment through Early Prediction of Treatment Outcome : A Machine Learning Approach

• Mark

Jönsson, Karl-Axel ; Andersson, Edvin ^LU ; Nevéus, Tryggve ; Gärdenfors, Torbjörn and Balkenius, Christian ^LU

Abstract

Introduction: Bedwetting, also known as enuresis, is the second most common chronic health problem among children and it affects their everyday life negatively. A first-line treatment option is the enuresis alarm. This method entails the child being awoken by a detector and alarm unit upon urination at night, thereby changing their arousal mechanisms and potentially curing them after 6–8 weeks of consistent therapy. The enuresis alarm treatment has a reported success rate above 50% but requires significant effort from the families involved. Additionally, there is a challenge in identifying early indicators of successful treatment.

Methods: The alarm treatment has been further developed by the company Pjama AB, which, in addition... (More)

Introduction: Bedwetting, also known as enuresis, is the second most common chronic health problem among children and it affects their everyday life negatively. A first-line treatment option is the enuresis alarm. This method entails the child being awoken by a detector and alarm unit upon urination at night, thereby changing their arousal mechanisms and potentially curing them after 6–8 weeks of consistent therapy. The enuresis alarm treatment has a reported success rate above 50% but requires significant effort from the families involved. Additionally, there is a challenge in identifying early indicators of successful treatment.

Methods: The alarm treatment has been further developed by the company Pjama AB, which, in addition to the alarm, offers a mobile application where users provides data about the patient and information regarding each night throughout the treatment. The wet and dry nights are recorded, in addition to the actual timing of the bedwetting incidents. We used the machine learning model random forest to see if predictions of treatment outcome could be made in early stages of treatment and shorten the evaluation time based on data from 611 patients. This was carried out by using and analyzing data from patients who had used the Pjama application. The patients were split into training and testing groups to evaluate to what extent the algorithm could make predictions every day about whether a patient’s treatment would be successful, partially successful, or unsuccessful.

Results: The results show that a large number of patient outcomes can already be predicted accurately in the early stages of treatment.

Discussion: Accurate predictions enable the correct measures to be taken earlier in the treatment, including increasing motivation, adding pharmacotherapy, or terminating treatment. This has the potential to shorten the treatment in general, and to detect patients who will not respond to the treatment early on, which in turn can improve the lives of children suffering from enuresis. The results show great potential in making the treatment of enuresis more efficient. (Less)