Artificial neural network predicts sex differences of patients with advanced Parkinson’s disease under Levodopa-Carbidopa Intestinal gel

Abstract

Objective: Although Levodopa-carbidopa intestinal gel (LCIG) treatment has shown to be efficacious in motor and some non-motor symptoms (NMS), not all the patients with advanced Parkinsons disease (PD) are ideal candidates. To improve their selection analysis knowledge of prognostic factors is of great importance. We aimed to develop a novel machine learning model to predict the clinical outcomes of patients with advanced PD at 2 years under the LCIG therapy. Methods This was a longitudinal 24-month, observational study of 59 patients with advanced PD of a Greek multicenter registry under LCIG treatment from September 2019 to September 2021. Motor status was assessed with the Unified Parkinsons Disease Rating Scale (UPDRS) part III (off) and IV. NMS were assessed by the NMS Questionnaire (NMSQ) and the Geriatric Depression Scale (GDS), the quality of life by PDQ-39 and severity by Hoehn &Yahr (HY). Multivariate linear regression, ARIMA, SARIMA, and Long Short-Term Memory- recurrent neural network (LSTM-RNN) models were used. Results Dyskinesia duration and quality of life were significantly improved with LCIG (19% and 10% greater improvement for men than women, respectively). Multivariate linear regression models showed that UPDRS-III was decreased by 1.5 and 4.39 units per one unit of increase of the PDQ-39, UPDRS-IV indexes, respectively. Among all the time series models, the LSTM-RNN model predicts these clinical characteristics with highest accuracy (mean square error =0.0069) Conclusions Τhe LSTM-RNN model predicts with highest accuracy sex dependent clinical outcomes of patients with advanced PD after two years of LCIG therapy.