Objective: To use Multilayer Perceptron (MLP), an artificial neural network, to identify candidates for device-assisted therapy (DAT) among patients with Parkinson’s disease (PD)

Background: DAT (deep brain stimulation [DBS] and continuous apomorphine and levodopa infusions) are indicated in selected patients with intermediate/advanced PD when conventional treatment fails to control disabling motor fluctuations, dyskinesias, and/or tremor. Except in cases of tremor, when DBS is indicated, candidates for DAT share certain selection criteria, and the main predictor of a good outcome is an optimal “on” in response to levodopa. Negative prognostic factors include advanced age, moderate cognitive impairment, “on” axial symptoms, and major comorbidity. Identification of candidates requires specialist expertise but may be facilitated using artificial intelligence tools

Method: Our cross-sectional pilot study included consecutive PD patients attended in a Movement Disorders Unit during 3 months. Gold standard for DAT candidate identification was the expert neurologist’s decision. Predictive variables gathered were: age, clinical subtype (tremor-dominant, rigid-akinetic or mixed), Charlson comorbidity index, and responses to 15 dichotomic items of the CDEPA Questionnaire. MLP and a backpropagation learning algorithm were used to identify candidates with these variables as network inputs. Stratified cross-validation and balance techniques were applied to confirm discriminative capacity and improve minority class detection of the MLP. The relative influence of variables was estimated with ReliefF

Results: We studied 251 consecutive PD patients (70.5±9.2 yrs, 41% female) with median H&Y scale of 3 and mean duration of 9.1±6.5 yrs.  PD was advanced in 130 (52%), including 39 (30%) with previous DAT (28 DBS and 11 intestinal levodopa). We identified 24 candidates for DAT, 3 with intermediate PD (refractory tremor) and 21 advanced PD. Accuracy of MLP identification was 90% (Sensitivity=62.5%; Specificity=92.5%). Areas under ROC and precision-recall curves were 84% and 94%, respectively. All input variables were required for optimal identification; the most influential were need for assistance in daily living activities and motor fluctuations with off time> 25%

Conclusion: Within the limitations of the applied methodology, this Neural Network showed good discriminative capacity to detect PD candidates for DAT

References: [1] Latourelle JC, Beste MT, Hadzi TC et al. Large-scale identification of clinical and genetic predictors of Parkinson´s disease motor progression in newly diagnosed patients: a longitudinal cohort study and validation. Lancet Neurol 2017; 16 (11): 908-916. [2] Luquin MR, Kulisevsky J, Martínez-Martín P, Mir P, Tolosa ES. Consensus on the definition of advanced Parkinson´s disease: A neurologist – based Delphy study (CEPA Study). Parkinsons Dis 2017; 4047392. Article IS. [3] Martínez-Martín P, Kulisevsky J, Mir P, Tolosa E, García-Delgado P, Luquín MR. Validation of a simple screening tool for early diagnosis of advanced Parkinson´s disease in daily practice: the CDEPA questionnaire. NPJ Parkinsons Dis. 2018 Jul 2;4:20. doi: 10.1038/s41531-018-0056-2. eCollection 2018. [4] Escamilla Sevilla F, Olivares Romero J, eds. Recomendaciones de Práctica Clínica en la Enfermedad de Parkinson. Grupo Andaluz de Trastornos del Movimiento. Sociedad Andaluza de Neurología. Barcelona: Glosa, 2017 (ISBN: 978-84-7429-668-6). [5] Escamilla Sevilla F, Campos Arillo VM. Perfiles de los candidatos y adecuación de las terapias de segunda línea. En: Escamilla Sevilla F, Olivares Romero J, eds. Recomendaciones de Práctica Clínica en la Enfermedad de Parkinson. Grupo Andaluz de Trastornos del Movimiento. Sociedad Andaluza de Neurología. Barcelona: Glosa, 2017; pp 131-143. (ISBN: 978-84-7429-668-6). [6] Zhan A, Mohan S, Tarolli C et al. Using Smartphones and Machine Learning to Quantify Parkinson Disease Severity. The Mobile Parkinson Disease Score. JAMA Neurol 2018; 75 (7): 876-880.

« Back to 2019 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/identification-of-candidates-for-device-assisted-therapy-in-parkinsons-disease-using-artificial-neural-networks/