Predicting Motor Learning of an Upper Limb Task Based on Behavioral and Disease-specific Characteristics in Patients With Parkinson's Disease

In order to achieve personalised rehabilitation in PD, it is imperative to understand which factors predict whether a patient will benefit from targeted training. To date, only few studies have looked into the predictors of gait and balance training and identified cognitive function at baseline and initial motor performance as the most important determinants. However, these studies focused on global cognitive function, while cognitive subdomains, such as executive function and memory, which are most affected in PD, likely impact more seriously on learning. In addition, patients with cognitive impairments were excluded in these studies. Hence, little is known about how non-motor symptoms, which may also include apathy and depression, interact with the capacity to retain learning in PD.

So far, the determinants of motor learning retention have only been examined as secondary analyses of effect studies in PD. Hence, this study aims to identify predictive factors, including motor and non-motor symptoms, for sustained motor learning in a wide and large PD cohort. The primary dependent outcome is to identify which patients are able to retain the learning gains after 4 weeks without training.

Given the impaired touchscreen skills and the espoused difficulties with retention, training of touchscreen sliding motions will be delivered on a tablet enabling practice in the home setting. Both single (ST) and dual task (DT) conditions will be offered in a random order and feedback will be provided to enhance motivation and retention. This project aims to identify the predictive factors for effect maintenance after receiving this two-week training program. The investigators will recruit a broad cohort of PD patients. Sample size was estimated, using a squared multiple correlation coefficient of 0.17 based on a prediction model from earlier work, which included DT-motor performance and cognitive function as determinants. Using an α = 0.05 and β = 0.20, sample size was calculated for a linear multiple regression: fixed model, R² deviation from 0. Although the number of predictors is dependent on possible multicollinearity, sample size was computed on the assumption of 10 predictors. Sample size was estimated to be 89 and after accounting for a dropout of 20%, 107 patients will have to be included. Importantly, this sample size is also supported by the rule of thumb to include 10 participants per predictor, suggested by Harrell et al. (1996). As the investigators expect to include 10 predictors in our multiple linear regression model (see below), a total number of 100 PD patients will be recruited. The investigators will enroll a wide range of cognitive profiles. However, all patients need to be able to follow instructions and engage in the motor learning. During the instructions of how to perform the touchscreen task, the investigators will assess eligibility pragmatically. The investigators will exclude PD-dementia using level I MDS-diagnostic criteria. All patients will train the SSP-task for a duration of two weeks. On day 1 (T0), all participants will undergo an extensive screening session, including motor and non-motor tests performed at the subject's home or in a quiet room in our laboratory at KU Leuven, according to patients' preference.

The global cognitive screening will consist of the Montreal Cognitive Assessment. Moreover, 2 specific tests will assess each cognitive subdomain. Attention and working memory are captured by the digit and visual span forward and backward test. The trail making and alternating names tests will be used for executive function. Visuospatial function will be examined using the short form of the Benton's judgement of line orientation and the Rey Osterrieth Complex figure. The 30-min recall of the latter test will also be used to assess memory, together with the Rey Auditory Verbal Learning test. The Boston naming test and the Animal fluency test of the Controlled Oral Word Association test will be used to assess language. Other non-motor features, such as anxiety, depression, and sleep quality, will be tested using validated questionnaires for PD. Touchscreen skills will be assessed using the SSP-test in ST and DT condition, the mobile phone task (MPT, typing a predefined telephone number on a smartphone) and specific questionnaires. Following the baseline session, all patients will receive 10 training sessions of the SSP-task randomly in ST and DT condition (2 weeks, 5 days/week, 10 min/session) over a period of two weeks. The training is home-based and unsupervised. Both immediately after training (T1) and after a four-week retention period (T2), touchscreen skills will be assessed at home. To account for performance bias due to other training, other rehabilitation content will be recorded.

For model selection, the investigators will follow the recommendation of the PROBAST tool and reduce the number of cognitive tests accordingly. Other independent variables may include age, gender, disease duration, the New Freezing of Gait Questionnaire, and LEDD. The investigators will also calculate a measure of early acquisition. After exploratory univariate analysis, the investigators expect to include 10 independent variables, taking multicollinearity into account.

Additionally, the investigators will study the determinants of compliance and its association with retention, as compliance with unsupervised home-based training is likely to be associated with specific non-motor characteristics. Apathy and the anticipation of reward have been shown to determine the level of effort generated for exercise in PD unlike in HC. Also, one previous study investigated the relationship between the success of a balance training program and compliance in PD, establishing compliance to be one of the two most important determinants. Other work revealed that disease duration was found to be related to the degree of motor learning of the SSP-task and to exercise adherence altogether. Hence, compliance is likely associated with training outcomes in PD (long- and short-term) and may be modulated by (multiple) clinical characteristics. Compliance will be automatically logged by the digitized system and stored in a secured data cloud. Training compliance will be expressed as a percentage of the desired number of sessions and personalized time, with a maximum of 100%. The investigators will analyze the relation between compliance and clinical characteristics first with a multiple linear regression model. Next, a multiple mediation model will be constructed to examine the expected mediating influence of the clinical profiles on this association. Different models will be applied for retention of learning and immediate training effects, as well as for the different outcome measures.