Machine-learning–based classification of nocturnal enuresis: The contribution of maternal anxiety in a case-control study
Eda Çelik Güzel, Murat Güzel, Cenk Murat Yazıcı, Birol Topçu, Tülin YıldızObjective
Nocturnal enuresis is common during childhood; however, the contribution of maternal psychological well-being remains underexplored. We applied interpretable machine learning to determine which routinely available factors best distinguish children with nocturnal enuresis from controls and to quantify the contribution of maternal mental health.
Methods
In this single-center case-control study, we compared 20 clinical and family-related variables of 112 children with primary nocturnal enuresis and 54 controls, including maternal anxiety and depression scores. Maternal anxiety and depression scores were missing in 50/112 cases (44.6%) and 0/54 controls. Four models were trained using 10×5 repeated stratified cross-validation with all preprocessing refitted inside each training fold, and SHapley Additive exPlanations analysis was performed to identify key predictors.
Results
The random forest model had the best discriminative ability (area under the receiver operating characteristic curve = 0.971, 95% confidence interval: 0.947–0.990). In the full-sample analysis, the largest adjusted associations were constipation (odds ratio = 4.68, 95% confidence interval: 3.60–8.50) and maternal anxiety (odds ratio = 2.87, 95% confidence interval: 1.59–5.09), and their interaction was the strongest (SHapley Additive exPlanations interaction = 0.676). Under median imputation, maternal anxiety and depression together accounted for approximately 29% of total SHapley Additive exPlanations feature importance—a measure of contribution to model output, not of effect size.
Conclusions
Maternal anxiety was associated with case status in this sample; however, this association should be regarded as provisional given the differential missingness of maternal scores and their measurement after the child's diagnosis was established. These hypothesis-generating classification findings establish neither causality nor clinical utility and require prospective external validation.