Personalized rehabilitation for people with Parkinson's disease via novel AI stratification strategies: the PREPARE project

Background and aims: Parkinson's disease (PD) represents the archetype of complexity. It affects multiple body functions, producing functional profiles different in severity and course, and strongly influenced by many individual variables, including response to rehabilitation. Although evidence of the effectiveness of motor training is available, the outcome observed after any rehabilitation approach varies widely among patients. A refined predictive capacity is fundamental to guarantee the right treatment for the right patient. PREPARE (1) is a European HaDEA-Horizon project with 20 partners from nine countries, involving the application of machine learning (ML) techniques on large-scale patient datasets to develop, validate and implement robust, clinically relevant and data-driven computational prediction and stratification tools. As partners of PREPARE, we contribute to the goal of developing prediction models for people with PD. Methods We started a retrospective study to collect data recorded on the database implemented at the Movement Disorders Center of a University Hospital. The Center is a referral facility for the medical diagnosis, treatment and rehabilitation of people with PD. After the diagnosis, patients undergo regular follow-ups every 4-6 months. Clinical information includes demographic data, BMI, professional activity, amount of daily physical activity, disease phenotype and duration, any drug or surgery treatment, degree of motor and non-motor impairment and functioning profile. Results: This retrospective study will provide data from 970 subjects (56% male), presenting, on their first referral to the Centre, an age of 69.8+10 years (range: 31-95), disease duration of 8.1+7 years (median 6.0), Hoehn & Yahr stage <2 in 58% of cases. The progression of the individual clinical profile has been monitored for 5 years, producing approximately 10,000 records. Analysing all the available explanatory variables will allow patients to be appropriately stratified by prognostic factors, to study disability evolution in a large and representative sample of people with PD. Conclusion: The PREPARE project is still ongoing. The predictive model based on the data retrospectively collected will be validated in a prospective study, and the generalizability of the results will be discussed within the PRM community.