OBJECTIVE: Due to high recurrence rates of urolithiasis, many attempts have been performed to identify tools for predicting the risk of stone formation. The application of Artificial Neural Networks (ANNs) seems to be a valid candidate for reaching this endpoint. The aim of this study was to find a set of parameters able to predict recurrence episodes immediately after clinical and metabolic evaluation performed at the first visit in a 5-year window. MATERIAL AND METHODS: Data were collected from 80 outpatients who presented idiopathic calcium stone disease both at baseline and after 5 years; patients underwent treatment including both general measures and medical therapy. After 5 years, patients were classified into two subsets, namely SSFs (without recurrence episodes), consisting of 45 subjects (56.25%) and RSFs, with at least one episode of recurrence after the baseline, consisting of 35 subjects (43.75%). Helped by conventional statistics (One-way ANOVA and three Discriminant Analyses: standard, backward stepwise and forward stepwise), an Artificial Neural Network (ANN) approach was used to predict recurrence episodes. RESULTS: An optimal set of 6 parameters was identified from amongst the different combinations in order to efficiently predict the outcome of stone recurrence in approximately 90% of cases. This set consist of serum Na and K as well as Na, P, Oxalate and AP (CaP) index from urine. The results obtained with ANN seem to suggest that some kind of relationship is present between the identified parameters and future stone recurrence. This relationship is probably very complex (in the mathematical sense) and non-linear In fact, a Logistic Regression was built as a comparative method and performed less good results at least in terms of accuracy and sensitivity. CONCLUSIONS: The application of ANN to the database led to a promising predicting algorithm and suggests that a strongly non-linear relationship seems to exist between the parameters and the recurrence episodes. In particular, the ANN approach identifies as optimal parameters serum concentration of Na and K as well as urinary excretion of Na, P, Oxalate and AP (CaP) index. This study suggest that ANNs could potentially be a useful approach because of their ability to work with complex dynamics such as recurrent stone formation seems to have.
OBJECTIVE: Due to high recurrence rates of urolithiasis, many attempts have been performed to identify tools for predicting the risk of stone formation. The application of Artificial Neural Networks (ANNs) seems to be a valid candidate for reaching this endpoint. The aim of this study was to find a set of parameters able to predict recurrence episodes immediately after clinical and metabolic evaluation performed at the first visit in a 5-year window. MATERIAL AND METHODS: Data were collected from 80 outpatients who presented idiopathic calcium stone disease both at baseline and after 5 years; patients underwent treatment including both general measures and medical therapy. After 5 years, patients were classified into two subsets, namely SSFs (without recurrence episodes), consisting of 45 subjects (56.25%) and RSFs, with at least one episode of recurrence after the baseline, consisting of 35 subjects (43.75%). Helped by conventional statistics (One-way ANOVA and three Discriminant Analyses: standard, backward stepwise and forward stepwise), an Artificial Neural Network (ANN) approach was used to predict recurrence episodes. RESULTS: An optimal set of 6 parameters was identified from amongst the different combinations in order to efficiently predict the outcome of stone recurrence in approximately 90% of cases. This set consist of serum Na and K as well as Na, P, Oxalate and AP (CaP) index from urine. The results obtained with ANN seem to suggest that some kind of relationship is present between the identified parameters and future stone recurrence. This relationship is probably very complex (in the mathematical sense) and non-linear In fact, a Logistic Regression was built as a comparative method and performed less good results at least in terms of accuracy and sensitivity. CONCLUSIONS: The application of ANN to the database led to a promising predicting algorithm and suggests that a strongly non-linear relationship seems to exist between the parameters and the recurrence episodes. In particular, the ANN approach identifies as optimal parameters serum concentration of Na and K as well as urinary excretion of Na, P, Oxalate and AP (CaP) index. This study suggest that ANNs could potentially be a useful approach because of their ability to work with complex dynamics such as recurrent stone formation seems to have.