Aim
The primary aim of the trial is to investigate the efficacy of a new intensive cognitive rehabilitation protocol in a sample of Italian patients aged 7-25 years with congenital cerebellar diseases. The hypothesis is that the VR-Spirit rehabilitation protocol should:
- Enhance social prediction ability resulting in better understanding of other people’s intentions and behaviours
- Facilitate general-domain implicit learning ability
- Indirectly improve cognitive performance in specific domains (attention and executive functions, memory, visuospatial abilities, sensorimotor integration)
- Produce an amelioration of patients’ quality of life.
Contextually to the primary aim, the study would verify the feasibility and acceptability of this rehabilitative intervention for the target-population.
Participants
Participants are children, adolescents and young adults aged 7-25 years with congenital cerebellar malformations and with a FSIQ greater than 45. Cerebellar malformations refer to anatomical abnormalities affecting the vermis and/or the hemispheres not due to acquired lesions and not associated with progressive pathologies. Though, it is noteworthy that these patients could exhibit malformations in other cortical structures. As an example, patients with Joubert syndrome often present with malformations of pontine and medullary areas (46). Participants are recruited at the Child Neuropsychiatry and Neurorehabilitation Unit of the Scientific Institute, IRCCS E. Medea. With the aim to achieve the target sample size, associations of patients can be contacted. The following exclusion criteria are adopted:
- Severe sensorial, motor and/or behavioural problems that could interfere with the use of GRAIL technology;
- Being simultaneously involved in a different cognitive rehabilitation treatment, to avoid excessive demands to children and possible interference on training adherence rates;
- Having been involved in a different cognitive rehabilitation treatment in the last six months before training, to avoid confounding follow-up effects.
Parents of all known potentially eligible patients are contacted telephonically by the attending physician and are informed about aims and methods of the protocol. If they agree with the study, an administrative staff person contacts the parents to organize the recovery. Since participation to the trial has to be arranged according to other clinical needs (e.g., routine medical checks) of the patients and to parents’ availability, a variable time frame from 1 month to 1 year could occur before patients get access to the trial. Intervention assignment is carried out by a research assistant only when patients are admitted to the recovery. It is noteworthy that the administrative staff that attends to the recruitment and organizes hospitalization is blinded to group allocation. Before starting the baseline evaluation, a research assistant provides a description of material and procedures to parents and patients and asks them to sign an informed consent.
Design
The study applies a single-centre, randomized active controlled trial design. Patients are allocated to one of two groups undergoing two different rehabilitation programs through a stratified permuted block randomization procedure (47). Age and cognitive level (more recent available full-scale intelligent quotient, FSIQ) are chosen as stratification factors. In particular, we consider two levels for age, corresponding to 7.0-12.9 and ≥ 13.0 years, and three stages for cognitive level consistent with, respectively, absence of intellectual disability (FSIQ > 80), from borderline intellectual functioning to mild intellectual disability (80 ≥ FSIQ ≥ 61) and from mild to moderate intellectual disability (60 ≥ FSIQ ≥ 45) (48). Doing so, 6 blocks are generated and within each block an estimated number of 8 patients should be enrolled to achieve and overcome the established sample size. First, participants are allocated to a block depending on the two stratification variables. Second, patients are assigned to one of the two interventions according to specific permuted sequence. Details of the permuted blocks are reported in Table 1.
Table 1. Stratified permuted blocks randomization of the study.
|
|
Age
|
|
7-12.9
|
≥13.0
|
FSIQ
|
46-60
|
CCSSCCSS
|
SCSCSCSC
|
61-80
|
SCCSSCCS
|
SSCCSSCC
|
>80
|
CSCSCSCS
|
CSSCCSSC
|
Legend: FSIQ = Full-Scale Intelligent Quotient; S = VR Spirit training, C = Control training.
Group 1 (S) receives the social prediction VR training for two weeks (four daily sessions in a week). In each 1-hour session, 80 trials of the experimental program and one of four motor games, selected among the applications available in the GRAIL, are administered; for each weekly session, a different game is administered in random order. Group 2 (C) receives a control VR training of the same duration (two weeks, four 1-hour sessions per week) as the experimental training; the control training involves, for each session, a navigational game and the daily repetition of all the four games from the GRAIL suite that are also presented, one per day, in the experimental session; the social prediction experimental program is not presented in the control training.
Before the training (T0), a battery of neurocognitive tests from the Developmental NEuroPSYchological Assessment 2nd edition (NEPSY-II; (49,50)) spanning different domains, and specifically social perception abilities, are administered to all participants. Both groups also receive a 10-minute training of how to move within the GRAIL environment using a custom-made navigational application. Then, a pre-training evaluation through a VR game session, based on the same paradigm of the VR experimental training but in a different scenario, and a computer-based Action prediction task (51) are administered. Moreover, at T0, both patients and parents compile questionnaires on quality of life (TACQOL, TNO Quality of life / LUMC, 2001; (52)) and parents also complete the Child Behaviour Check List (CBCL;(53,54)).
In order to verify and compare the effects of the experimental and control training sessions, at the end of the two-week training (T1) all participants are re-evaluated with the same neurocognitive tests, the VR evaluation scenario and the Action prediction task.
With the aim to investigate the far transferability of the effects, a follow-up evaluation is provided after two months (T2) with the same protocol used at T0 except the GRAIL evaluation scenario. Details of the study design are set out in Figure 1 according to the “Standard Protocol Items: Recommendations for Interventional Trials” (SPIRIT) statement (55,56) (see also Additional file 1).
Intervention and study setting
The rehabilitation trainings are administered in the GRAIL Lab at the Scientific Institute (IRCCS) E. Medea (Bosisio Parini, Italy). The GRAIL system is an integrated platform equipped with a treadmill on a motion frame, a Vicon motion-capture system (Oxford Metrics, Oxford, UK) and a 180° cylindrical projection screen. The D-flow software controls the relationship between the patient, the scenery and the interactive feedbacks and stimulations. This software runs on Microsoft Windows and it was used to develop the interactive virtual reality applications with a block diagram approach. For the creation of the GRAIL scenes, objects and scenario were modelled separately by means of Google SketchUp while the avatars were created by using MakeHuman and then modified in Blender. The modelling process was first dedicated to the creation of three-dimensional geometries and then to the application of selected materials and textures. Files generated in SketchUp and Blender were exported in the COLLADA interchange format and then imported into Autodesk 3ds Max software. The latter allowed to convert models in Wavefront OBJ format and to assemble all the models created within the scenery. The whole scene was exported in Ogre format to be used within the D-flow software: the final scene contained the environment and the individual objects.
Two different scenes were developed specifically for this study: the “sweet stands” environment for the pre- and post-training evaluations and the “playground” scenario for the social prediction training. Both scenes are designed with a linear 9-meter long path that branches into three 3-meter long streets. At the end of each branch, one of three objects are located in a semicircle at the same distance from the starting point: the “playground” setting includes a swing, a circular carousel and a rocking carousel, while the “sweet stands” setting includes an ice cream, a donut and a lollipop stand. Furthermore, four different avatars, two males and two females, were designed: they are adolescents, clearly identifiable by body and clothing features (i.e., hair and t-shirt colours). An example of the two scenarios and of the avatars, respectively, for the evaluation and for the training sessions, is reported in Figure 2.
Before the beginning of the session, the patient wears two reflective markers on the posterior superior iliac spines, that allow to trace patient’s movements and control the virtual environment: to go faster, the patient has to move forward, to slow down he/she has to move backward, while to turn right or left he/she has to shift the pelvis right or left. Then, the patient come up the GRAIL system, the trainer calibrates his/her starting position and the session can start. Before starting the first evaluation session, a short and effective navigational training is administered to participants with the aim to learn to navigate within the GRAIL VR environments. A physical therapist specifically patented for using GRAIL technology administers all the applications.
Evaluation sessions
Pre- and post-training evaluation sessions in the “sweet” stand scenario are administered to both the experimental and the control groups. The paradigm exploits a probabilistic design that has already shown its reliability in assessing social prediction abilities, since children with Autism Spectrum Disorder (ASD), who show clinically relevant social deficits, were impaired in using contextual priors to predict the unfolding intention of observed actions ahead of realization (45). Within a session, events take place in a pseudorandom way in respect to the pre-established probabilities. Specifically, in each trial, one of four avatar moves from the starting point to one of the stands with pre-established probabilities as shown in Table 2:
Table 2. Example of Event probability in a evaluation session with the sweet stands scenario.
Session A
|
Stand
|
Avatar
|
Ice cream
|
Donut
|
Lollipop
|
Avatar A
|
80%
|
10%
|
10%
|
Avatar B
|
10%
|
80%
|
10%
|
Avatar C
|
10%
|
10%
|
80%
|
Avatar D
|
33%
|
33%
|
33%
|
Two different evaluation sessions were generated, changing the avatar-object associations and event sequence, and are presented, respectively, at T0 and T1 in random order. In this way, we avoid repetition of the same events in the two evaluation sessions in order to minimize learning effects. The order of the two sessions is counterbalanced between patients of the same group (e.g., for patient 1 session A at T0 and session B at T1, for patient 2 session B at T0 and session A at T1 etc.). Considering twenty trials per avatar, eighty trials are administered in each session.
The Grail therapist asks the patient to move toward the sweet stand chosen by the avatar and activate it before him/her. In each trial the patient has to reach one of the stands in maximum 15 seconds and his/her maximum speed is 2 m/s. The avatar, one per trial and visibly positioned next to the patient, moves towards a stand, reaching it in ten seconds. When the participant reaches a sweet stand, the stand is activated providing a visual reinforcement, otherwise the event is interrupted five seconds after the avatar has reached the object and patients are invited to try again. Furthermore, when the participant anticipates the avatar in reaching the correct stand he/she also receives an auditory reinforcement (clapping sound), which signals the scoring of a point in the game in addition to visual reinforcement (activation of the object). The object reached by the avatar is always visible to the participant, for both successful and unsuccessful trials, in order to provide information on the avatar’s preferences that can be used in the next trial. Specific features of the application force the patients to move according to the anticipation of the avatars’ preference rather than following his/her movements. Indeed, the path is a 9-meter straight-line trajectory and, then, it splits into three ways. Therefore, participants are not exposed to motion cues concerning avatars’ directions until the crossroad. Moreover, after this division the speed of the avatars equals the maximum available for the patients, so that they cannot be surpassed anymore. This way, patients are prompted to implicitly learn the probabilistic associations between the avatar and the most chosen sweet stand, thus allowing this paradigm to evaluate and improve the ability to form predictive models of other’s behaviour.
For each trial, the D-flow software automatically saves one raw with the following measures in a .txt file:
- duration of the trial,
- mean speed of the subject,
- speed of the subject at specific points of the path (e.g. at 0.5, 1, 9 meters from the starting point),
- specific avatar,
- object selected by the avatar,
- object selected by the subject
- victory/no victory
- incremental score.
At the end of the 80 trials, the D-flow automatically saves the total score of the session in a different .txt file.
VR-Spirit training
Training sessions adopt the same logic of the evaluation sessions but in the playground scenario. With the aim to balance the association between avatars and objects, four diverse sessions (A, B, C, D) were obtained, such that avatars’ probability of moving toward a specific object is equally distributed across the four session. The four sessions are randomly administered during the first week and repeated in the same order in the second week.
Every day, the experimental group is administered with one of the four diverse sessions of the Spirit Training so that avatars’ preferences change day by day both in the first and in the second week. Moreover, after the participants have completed the eighty social prediction trials, they play also one of four selected games from the GRAIL kit (see below for the description of the games).
Active control training
The control group is exposed for the same amount of time (1-hour session per day, four sessions per week for two weeks) to sessions requiring the participants play a navigational game, in which they have to conduct a ball out of three mazes, and all four selected GRAIL games. The four selected games are “skiing”, “balloons shooting”, “world soccer” and “traffic jam”. These games have been chosen because they do not present social agents and do not require any form of prediction ability. In the “skiing” game, participants have to do a slalom between snowmen, scoring a point when they pass each snowman on the right side. In the “balloon shooting” game, participants have to hit balloons appearing in a natural environment simply by pointing at them. In the “world soccer” game, children kick a virtual ball toward a goal: they score points when they hit targets put inside the goal. In the “traffic jam” game, participants are in the middle of a crossroad and they have to raise the left or right foot according to the cars’ movements.
Action prediction task
We adopt a validated computer-based Action Prediction task as an experimental outcome measure for social prediction and implicit-learning abilities. This experimental paradigm consists of a probabilistic learning task (familiarization phase) followed by an action prediction task (testing phase). During familiarization, participants are exposed to videos showing a child actor performing two different grasping-actions associated with specific contextual cues and they are asked to recognize actor’s intention. Notably, in this phase the association between contextual cues and actions was implicitly biased with pre-established probability of co-occurrence. During testing, the second half of the same videos is occluded and patients are asked to predict the final outcome of the action. Since movement kinematics are ambiguous, responses should be biased toward the contextual priors acquired during the familiarization phase. For each participant and separately for the two phases, a standardized beta coefficient is calculated across trials at the individual level using a regression analysis with probability and accuracy as the independent and dependent variables, respectively. Indeed, the beta coefficient could be considered as a direct index of the strength of the contextual models of others’ intentions, thus providing a measure of social prediction ability.
Neuropsychological assessment
A neuropsychological assessment is administered at each stage of the study using the Italian version of the NEPSY-II battery. The NEPSY-II is designed to evaluate six different cognitive domains in children and adolescents aged 3-16 years. In our study, we administer tests that assess visual attention and executive functions, visuo-spatial memory and functions, sensorimotor integration and social perception skills. The Visual attention test assesses speed and accuracy of patients in focusing and maintaining attention on visual targets among a series of distracting stimuli. To assess executive functions, we adopt the Inhibition test, in which participants are asked to denominate different figures respecting diverse rules, thus inhibiting automatic responses. In the Memory for drawing test, children are exposed for ten seconds to a table representing drawings in diverse spatial position and then they are asked to choose the correct stimuli in a series of cards and place them in a panel in the same position they have seen before. The recall is asked immediately later the exposition and after 20 minutes. The Picture and the Geometric puzzles use, respectively, concrete and abstract examples to evaluate visual-perceptual and visual-spatial representation abilities. For sensorimotor integration, we administer the Finger tapping test, which measures the ability to repeat fast finger movements and maintain a motor program. To assess social perception skills, we administer the Theory of mind and the Affect recognition tests. The first is composed of two parts resulting in one score. In the verbal part, verbal or pictorial descriptions of social situations are presented in order to evaluate the ability to understand mental constructs, such as beliefs and intentions, and how other people could have thoughts, emotions and perspectives, which might be different from ours. Conversely, the contextual part assesses the ability to infer others’ emotion and mental state by social context. The Affect recognition test provides a measure of the ability to recognize affective states from emotional facial expressions using pictures of children. Moreover, the baseline assessment includes a full cognitive evaluation with the Wechsler Intelligence Scale for children 4th edition (WISC-IV) to estimate IQ scores (57).
Questionnaires
At the baseline (T0) and follow-up evaluations (T2) the TACQOL is administered to parents and children. This questionnaire has been primarily designed for research and evaluates quality of life in diverse domain: body, movements, autonomy, cognitive abilities, sociality, positive and negative emotionality. At the same time points, parents are asked to compile the CBCL, the most adopted questionnaire about the behaviour of children and adolescents. This questionnaire provides scores for eight empirically-based syndrome scales, namely aggressive behaviour, anxious/depressive symptoms, attention problems, rule-breaking behaviour, somatic complaints, social problems, thought problems and withdrawn. These scales are further aggregated in three main dimensions: internalizing, externalizing and total problems.
Outcome measures
Primary outcome measure:
- Social prediction ability: performance during the pre (T0) and post-training (T1) evaluation in the “sweet stands” scenario; standardized beta coefficients of the regression between accuracy and probability in the testing phase of the validated pc-based Action prediction task administered at every time point (T0, T1, T2).
Secondary outcome measures:
- Social cognition: Theory of mind Part A and B and Emotion recognition of NEPSY-II testing battery.
- Implicit learning: accuracy and reaction time in the familiarization phase of an Action prediction task.
- Executive functions (inhibition and flexibility): Inhibition test of NEPSY-II.
- Visual attention: Visual attention test of NEPSY-II.
- Visuospatial and visual-perceptual abilities: Geometric Puzzle and Picture Puzzle tests of NEPSY-II.
- Memory: Memory for drawings test of NEPSY-II.
- Sensorimotor functions: Fingers-tapping of NEPSY-II.
- Behavioural problems: CBCL 6-18, Parent version.
- Overall functioning and quality of life assessed using the TACQOL questionnaire, presented in two forms: the self-compiled and the parent-compiled one.
To assess training feasibility:
- Number of dropouts: number of children who renounce to complete the two-weeks training
- Number of sessions completed per child: total number of sessions done in front of the total number proposed of eight sessions.
To assess training acceptability:
- Acceptability questionnaire: an ad hoc questionnaire completed by participants and another one by their parents after training conclusion (T1) to assess subjective evaluation of training accessibility and efficacy. It is to note that the same questionnaires are fulfilled by patients of the active control group and their parents.
A psychologist, who is not blinded to the intervention assignment, administers all the neuropsychological tests and questionnaires and records the performance during the Grail sessions.
Statistical methods
Demographic, clinical and neuropsychological variables of the two groups of patients are inspected through descriptive statistics. T-test and χ2 are used to assess differences between the experimental and control training groups at baseline for continuous and categorical variables, respectively, thus allowing us to verify successful randomization. For each outcome measure, we calculate the change between T0 and T1 (training effect delta) and between T0 and T2 (follow-up effect). Delta is calculated as the arithmetic difference between the second/third time points and the first time point. The delta values of the two groups for the primary outcome measure are compared using independent sample t-tests (two-tailed). For the primary outcome measure, a Bonferroni correction procedure will be used to control for multiple comparisons. Multivariate analysis of variance will be used to explore differential effects of the trainings in the secondary outcome measures. As what concerns missing data, a modified intention to treat analysis approach will be adopted, including in the analyses all the participants that had completed the pre- and post-treatment evaluation sessions, even if they had not completed all the training sessions. No imputation of missing data, however, will be used considering the limited sample size and observation points (58,59). The stratification factors (i.e., age and FSIQ), will not be considered in the statistical analyses, assuming that the stratification procedure has ensured a balanced distribution across the two groups.
Estimation of sample size
Estimation of sample size was based on the distribution of the standardized beta coefficients of the regression between accuracy and probability in the PC-based, contextualized action prediction task in children with typical development (M = 0.154, SD = 0.215) and with Autism Spectrum Disorder (M = 0.008, SD = 0.112), with a between-group difference of moderate-to large size (Cohen’s d = 0.87; see Amoruso et al., 2019 (45)). Thus, we estimated that a moderate increase (0.13 mean change or 0.8 SD) of the standardized beta coefficient after as compared to before the experimental training had clinical significance. Accordingly, a final sample of 21 patients per group has been set for our study in order to detect a between-group difference (independent sample t-test, two tailed) between the effects of the experimental vs. control training (T1-T0) of moderate effect size (Cohen’s d = 0.8) with a power of 0.80 and alfa level set at 0.05. The software G Power 3 was used for this estimation.