Clinical Decision Support System for Multisensory Stimulation Therapy in Dementia: A Preliminary Study

Worldwide around 50 million people are affected by Dementia, causing a growing public health problem with significant impact not only on individuals but also on caregivers, families, and communities. The first-line therapy is pharmacological, based on the use of a few drugs with effects on brain neurotransmitters. Nevertheless, these therapies are contraindicated in some subjects, offer few results and associated side-effects are not negligible. Among various available non-pharmacological treatments, the Snoezelen one (i.e., multisensory stimulation) is particularly interesting, since its main goal is the reduction of pressure and tension experienced by the patient in the housing groups. However, there is no clear evidence right now that such non-pharmacological interventions are effective in subjects with dementia. The aim of this study is to design and prototype a Clinical Decision Support System (CDSS) that collects patient's neurovegetative parameters during stimulation sessions, and searches for patterns that are predictive of behavioral state change (e.g., from agitated to relaxed, or from apathetic to activated), allowing therapists to decide with greater reliability the best stimulation combination for a patient. The proposed algorithmic framework was evaluated using publicly available data, also in perturbed form to investigate more challenging patterns. The compared predictive approaches (i.e., multivariate time series classification) achieved accuracy rates greater than 85% with original data, and greater than 83% when complex combinations of both shape and temporal perturbations were present. Instead, in the case of only one kind of perturbation, either shape or temporal, the achieved accuracy was greater than 90%.