Impact of Multimodal Cognitive Training on Cognitive Traits of Children: A Multicentric Interventional Study


 In this article, we evaluate the hypothesis that a multimodal cognitive training (MCT) program, the Brighter Minds, can enhance certain inherent traits of a child and thus bring changes in the external behavior. For the study, 186 children (randomized to 93 each in intervention and control group) aged 10-15 years were enrolled from three different locations. Psychometric tests, parental/caregiver interviews and EEG (electroencephalography) tests were conducted before and after the program. Intervention group showed strong statistical significance for improvements in Mini Mental Status Examination (MMSE) (P<0.01) but no significance for Raven’s Standard Progrssive Matrices (SPM) or Susan Harter’s test. The parental/caregiver reported satistically significant improvements in focus (P<0.05), empathy (P<0.05), intuition (P<0.05), comprehension (P<0.05) and understanding of abstract concepts (P<0.05) for the intervention group. For the control, Power Spectral Density (PSD) of the baseline eyes-closed (EC) EEG recording, the spectrum below 20Hz exhibited the characteristic “1/f” spectral scaling of the power-law. This signature matches prior reported evidence in literature of those in wakeful state with EC. The intervention group EC PSD, however, exhibited a signature similar to those in a slow sleep state; reflective of the possible transfer effect of the training on other skills like relaxation. We used unsupervised learning methods with dice distance, on the psychometric and interview data, to show the effect of location and the exposure of a few control children to the program.


Introduction
It is a long-held notion that learnings and impressions that are experienced at an early age, as a child, can have impact on the neuroanatomy and several neural structures for a long-term (Teicher et al. 2003).
There are several evidences that link genetic, environment and life-style factors to the development or lack of development of the structure and functional aspects of the brain. The cognitive exibility of a child or the ability to adapt the thoughts and behavior in response to changes in goals or the environment in life (Chevalier et al., 2012) could be attributed to these aspects. With changes in the environmental task demands, the working memory, attention, and response selection also changes accordingly. The degree of the change determines the degree of exibility. Among children, exibility develops rapidly during the preschool years and continues to improve across adolescence and young adulthood (Yeniad et al., 2013, Titz andKarbach, 2014). Such cognitive skills, abilities (Hinshaw, 1994) and responses that are developed due to either or both of intrinsic and extrinsic dynamics in the early part of life are bound to prevail as traits throughout the lifespan. However, as the age advances, the cognitive ability to change in response to new extrinsic experiences reduces. The interplay of genes and in-utero experiences might be responsible for some children being born with hypersensitive behavior facilitation (Gray 1987) or an underactive behavioral inhibition system (Scarpal & Raine, 2000). It is thus necessary to develop a balanced approach to emotional, social, cognitive and language development for preparing children to The biopsychosocial (BPS) framework (Grinker RR Sr. 1964, Engel 1981, was the earliest to provide an integrative, comprehensive model to human development, health, and functioning. It looks at the interconnection between biology, psychology, and socio-environmental factors, to understand how these in uence health, disease and wellness. It has now established itself as the status quo model for contemporary medicine. Engel declared that this model was developed to overcome the limitations and to bring contrast to the biomedical model which was prevalent at that time. It is the claim that the BPS model itself takes its roots from the general systems theory (GST). Critics of the BPS (Ghaemi 2009, Benning 2015and Bolton 2019 argue that the BPS model is very vague, eclectic and unfaithful to GST, and is not scienti c in its approach. Another contention, related to the above subject, is that the GST itself might not be a right framework to address the complexity of the mind (McLaren 1998). These theories and models must be understood in the context of trying to understand or identify the root cause of observed effects such as behavior, disease or health. We acknowledge that it is di cult albeit sometimes impossible to determine the exact causal basis to some of these effects (Karunamuni 2020); because the cause-effect relationship and pathways might be multivariate with latent variables, and highly nonlinear pathways. There are alternative models proposed recently (Ghaemi SN 2009), which focus on the humanistic element to medical treatment, but we focus our attention on the psychological component of the BPS. The idea is to seek a psychological foundation for a behavior or behaviors.
Cognitive Training (CT) aims at exploring the central idea that the neuronal structure and function can be changed by altering either the environment or experience or both. Some of the earliest results from literature (Douglas, V.I. et al. 1976) (Shaw, C., Lanius, R., & Doel, K.V., 1994) provide light along these directions. As an alternative to pharmaceutical interventions, parents are turning to CT for their children and adolescents to enhance cognitive abilities. Similarly, caregivers of elderly citizens are turning to CT for reducing the onset of age related dementia. Studies have reported the effects of CT on changes in cognition and behavior, across multiple disciplines: developmental psychology, social psychology, clinical psychology, educational psychology, and cognitive psychology (Jedlicka, 2017 (2014)) have also had success in either mild injuries or early onsets. Early training or differential experience (Rosenzweig 1996) can in uence health, wellness, brain development, or successful aging.
For children, a large body of prior studies is dedicated to improving learning struggles and cognitive de cits. The mode of delivery of such programs is online. The impact assessment and reception of online digital CT programs, in the scienti c literature, has been mixed. The FDA's recent approval of a CT game to treat ADHD might be an acknowledgment of the potential of a certain of these. Any article on cognitive enhancement cannot be complete without discussing on the ethics of cognitive enhancement. Much of the concerns raised on ethics looks at pharmacological interventions and advocates for increased discussion on proper public policy or regulation (Bostrom, N., Sandberg, A 2009).

Brighter Minds
The study presented in this article is related to examining a multimodal cognitive training ( work and community. We look at the cause-effect relationship in the framework of focused BPS by varying one or more parameters and observing the effect on traits and short-term behaviors. Traits in a child may include focus, memory, intuitive thinking, self-con dence, comprehension, and others. We assume that all others in uencing the effect such as social or biological parameters are constant during this study as all the participating children are from a similar background. In (

Methods
Study Design (Controlled before and after study) In Fig. 2, we show the set of inclusion criteria for selection of the intervention and control group based on a randomized design. Both the groups were tested before and after the BM training program. The terms intervention and case are used interchangeably during this article.

Sample Size
The sample size is selected based on some assumptions from prior literature (Madhavi Bongarala, 2019) that at least 33% will show positive improvement in the sample for the majority of the cognitive abilities.
In considering this, a sample size of 168 students (84 intervention and 84 controls) were chosen for the study. Sample size was increased by 10% to accommodate sample loss due to various reasons and the nal sample size included for the study were 186 (93 intervention and 93 controls) students.

Inclusion and Exclusion Criteria
Children, between the age of 10-15 years of age, who had signed up for an ongoing BM program were included for the study. Those children who might not be regular or had a risk of discontinuing the program mid-way were excluded.

Sampling Technique
Three locations (Gangtok, Mahabubnagar, Bangalore) were selected with purposive sampling from different sites of India, where children were enrolled into the training program. Within each of the study locations, multistage random sampling technique was used to select the children for the study

Study Tools
We have used several tools to assess the impact and these are listed below: C) Electroencephalogram (EEG) tests were conducted among 26 randomly selected children from the sample of control and intervention. To account for dropouts, about 10%, the nal numbers enrolled were 28.
Self-esteem might refer to an individual's subjective evaluation of his or her worth as a person. It might not refelect on that child's innate skills, talents or abilities or even how the child might be perceived by others.
In Fig. 2, the sampling frame and work ow is illustrated.

Statistical Analysis and Unsupervised Learning
Given the two groups, the control and intervention, the paired t-test on the pre and post 8-week data is used to test the null hypothesis that the two groups have identical average (expected) values (or the difference between the expectation is zero) for the Psychometric and interview data. The p values for the statistical analysis provides an evidence to reject or not reject the above null hypothesis, and the threshold is xed at 0.05.
The Factor Analysis of Mixed Data (FAMD), like the Principal Component Analysis (PCA), is usually used to analyze a dataset having mixed variables that are both qualitative and quantitative. The qualitative variables, obtained from both the Psychometric tests and the interview results, are converted to numerical values by encoding them. The corresponding loadings and scores, in the reduced dimension, is further analysed for insights.
Although the data is labelled as control and intervention, in order to study the effect of the exposure of the control to the training program or study the effect of location, we assumed that the data is unlabeled and analyzed the data using unsupervised learning methods.

Electroencephalogram Studies
Electroencephalogram (EEG) was selected in this study for the ease of administering in remote locations where laboratory access is limited and for the high temporal resolution. There is a wider acceptability of EEG to study improvements in cognitive functions ( placed at the following locations: AF7, AF8, AF9, TP9 and TP10 as per the 10-20 International Standards, and with the electrode Fpz marked as the reference electrode. The data sampling rate for the acquisition was xed at 256 Hz ( > > Nyquist frequency) for all the subjects. The participants were seated on a chair comfortably, in a noise-free room, with no external stimulus. The room where the recordings were taken were kept ambient noise free but were not sound-proof, and so any external disturbances or external auditory inputs were manually edited out or the readings entirely discarded. The baseline EEG data was recorded, with eyes closed, prior to the BM program and also post the training program. The studies were administered directly by the lead researchers who were blind to the control and intervention group list. Pre-intervention EEG readings were carried out in December 2018, and the Post-intervention EEG studies were carried out with both the control and intervention groups in March 2019. The pre-processing of the entire dataset was done using MATLAB® (MathWorks®, Natick, MA, United States) with EEGLab Toolbox (Delorme A 2004), and the pre-processing protocol was adopted from the HAPPE pipeline (Gabard-Durnam LJ 2018) with minor changes. The only difference with the HAPPE pipeline was that if there was any noise while collecting the data, rather than interpolating the signal or dropping out the channels, the entire dataset was discarded. We lter the data with a high pass lter from 1 Hz, which removes the nonstationary signal drift across the recordings. As mentioned in (Maess et al. 2016), under noisy conditions, as in our case of measurements obtained from experiments outside an electromagnetically shielded lab, the usage of a high-pass lter is the better option than detrending or baseline correction. Any noise or artefacts in the channels were checked for by using wavelet-enhanced ICA (wICA), as mentioned in (Gabard-Durnam LJ 2018).

Participants Pro le
The distribution of intervention and control cases in the three different study locations and the sociodemographic pro le of the sample is as given in Table 1.

Parent/ Caregiver Interview Results
The statistical analysis of the interview results of the parents and caregivers are listed in Table 2 for both the groups and also the signi cance values. Among the intervention group, statistically signi cant improvements were seen in child's focus (P < 0.05), empathy (P < 0.05), intuition (P < 0.05), comprehension (P < 0.05) and child's abstract concept understanding (P < 0.01). Study of psychometric test results shows that statistically signi cant improvements were seen in tests of MMSE (P < 0.01), and Memory pro le (P < 0.01) for the intervention group. However, statistically signi cant improvements were also seen in the control group for Memory Pro le (P < 0.01) ( Table 3).

EEG Test Results
In Fig. 4, we show the power spectral density (PSD) plot of the signal collected from two electrodes located in the pre-frontal lobe (locations AF7 in purple and AF8 in green) for a child who underwent the BM program. This participant, a twelve-year-old female child, was randomly chosen from among all the children who had participated in the program. We do not normalize the power spectrum by spectral attening or baseline normalization as we would like to compare the rate of the '1/f' drop for control and intervention.

Unsupervised Learning Results
In this section, we learn and draw insights from the entire data by adopting some recent trends in unsupervised learning approaches. One such approach is the dimensionality reduction algorithm such as Principal Component Analysis (PCA). However, the data comprised of a mixture of both qualitative and quantitative variables.
In Fig. 5 (below), we show the scatter plot of the components after Factor analysis (Esco er Brigitte & Pagès Jérôme (2008)) of the data which is a mixture of quantitative and qualitative. FAMD was chosen to include also the categorical variables along with the continuous variables. The columns that speci ed the location of the training program was removed prior to factorization, and only used for visualization and color labelling.

Discussion
In recent years, literature shows a growing interest among researchers in studies related to early childhood development and the role of brain and cognition enhancement during the formative years of childhood. The objective of this article is aimed at exploring the hypothesis that multi-modal cognitive training can in uence the inherent traits and bring a change in a child's external behavior. It is one of the rst of its kind that used a controlled before and after study design to assess effectiveness of MCT among children under general settings, and evaluate their implications. The tests that are canvassed in this study were chosen to cover most of the spheres of cognitive abilities, because of its reliability and the validity tested in the previous research conducted (Madhavi and Jayanna, 2019).
The parent and caregiver interviews in our study reported improvements in cognitive traits such as the child's focus, empathy, memory, intuition, comprehension and understanding of abstract concepts, and improvements in behavior traits such as self-con dence, empathy, and participation. These ndings are comparable to the results of the study conducted recently in (Madhavi & Jayanna, 2019), who also reported within the educational settings.
The Psychometric Test results revealed statistically signi cant improvements in the MMSE (P < 0.01) for the intervention group. However, for both the control and the intervention group, statistically signi cant improvements were seen for Memory (P < 0.01). It had been reported (Nezla et al., 2017.) that self-worth and self-competence in different settings like scholastics, social acceptance, athletic competence, physical appearance, and behavioral conduct improved with multimodal brain training programs. From our results, it could not be concluded if the BM program brought any signi cant difference in Memory, Raven's SPM or Susan Harter's Test to the intervention as against control participants. A similar conclusion can be drawn from the Violin Plots in Fig. 3, where we see signi cant improvements in MMSE but not the other tests.
One possible explanation in the case of Susan Harter's test is that it was observed that children had in general very low self-esteem scores. New ndings on this subject believes that self-esteem is relatively stable trait and it is similar to some basic personality characteristics. Another reason for the difference could be attributed to the lower number of samples of Control than Intervention participants, and hence seeming to appear from a different statistical distribution. This might also be explained by a new theory (Jaeggi, S. M., 2020) that was put forth suggesting that the outcomes may be averaged across all individuals who received the intervention, and bene t experienced by a few might be averaged out.
Another possibility is that the program does not improve uid intelligence due to the shorter time span between training intervals (< 7days) (Wang, Z. et al. 2014). The SPM results might be pointing out to the fact that while some abilities are enhanced after the training program, a few others might have impaired or diminished (Colzato, L. S. et al. 2020). Figure 4a is the PSD calculated on the baseline EEG signals for the control, and Fig. 4b was the PSD calculated from the processed EEG signal for the intervention after the eight weeks training program. It was reported that in the waking state with eyes closed, the frequency band below 20Hz scales as 1/f, and with possible peak around 20Hz, while in the slow-wave sleep state, the '1/f' scaling is not visible (Bedard et al. 2006). We observed a similar trend as well, except this difference was observed between the control and the intervention group. In the control group, the '1/f' scaling is visible (see Fig. 4a red line for local slope) but in the case of the intervention, the drop is much steeper than for the control (see Fig. 4b orange line for local slope). Our test results of EEG on a selected sub-sample of children, showed that the spectral signatures of the EEG in the intervention group were quite similar to those with spectra from EEG recorded during a relaxed or slow sleep state. We observed that, in general, the spectrum for the intervention group exhibited a drop in the higher frequencies, especially above 12 Hz (see Fig. 4b) frequencies. This raises the interesting re ection if the intervention program might have led to transfer effect on other skills like relaxation and a permanent change in their inner traits in comparison to the control group. It is known that Power-Law scaling governs the brain's surface electric potentials (Miller et al. 2009, Bedard et al. 2006) and is also an intrinsic property of complex dynamic systems (Marković, D. and Gros, C. 2014, Meisel, C., Bailey, K., Achermann, P. et al., 2017).
During FAMD, for visualization, we can see in Fig. 5, that the ellipses separate the two data coming from the two locations Mahabubnagar and Gangtok very distinctly. The location clustering could be an indicator to a hidden latent variable-the effect of the training facilitator (here teachers) on the training program. When we removed the in uence of the locations in the analysis, for the data from the Gangtok site, there is a clear grouping happening between the Intervention and the Control groups (see Fig. 6) after the Factor Analysis. However, for the Mahabubnagar location (see Fig. 7), the distinction between the Control and intervention data is not distinct. This could be explained by the fact that the control group children from the Mahabubnagar program were living in hostels and the program leaked through the intervention group.

Limitations of the study
The study has a few limitations. Identi cation and recruitment of children who are naturally seeking admission into the Brighter Minds program was a challenge in study sites outside of Mahabubnagar. Fewer children were enrolling into the program in the learning centers on these sites. Even though we had identi ed and recruited 186 participants (93 intervention and 93 controls) in the baseline study, only 131 participants (82 intervention and 49 control cases) were available for end line assessment. The remaining cases (about 30%) were lost from the sample due to exposure of controls to intervention programs and drop out because of illness. This 30% loss was much higher than our anticipated loss of 10%. Many children, from the schools in Mahabubnagar district, were residing in hostels, and there are chances of exposure of the control participants to the intervention (see Fig. 7 and Fig. 8). It is also likely that some of the facilitating teachers from these rural schools do understand the importance of proving an environment that can improve the children's con dence and self-esteem. The logistical complexity of conducting the controlled studies in community settings such as children's hostels will need to be taken into account in future studies. This also violated our initial assumption that the biological and social conditions during the study will not change or in uence the study. It was also not possible to evaluate clearly what cognitive skill was impaired after the training program.
For the EEG analysis, in order to simplify the recordings for children, we restricted the number of electrode sensors to 5 and were dry electrodes. Dry electrodes can bring in additional noise due to a temporary loss in contact with the skin. This was partly mitigated by using gel but we lost the signal from the dry electrodes and had to redo the experiments. As the number of electrodes are only a few, the recordings might not be precise for source localization using methods such as ICA. The recordings are done in a relatively noise-free room, but it is not shielded completely from outside noise. Advanced EEG analysis in a laboratory-controlled settings and functional MRI in future studies have potential to shed light on trait changes in the children that undergo cognitive and behavioural changes as a result of cognitive training.

Conclusions
Brighter Minds' multimodal cognitive training program has a positive effect on self-worth and selfcompetence as well as cognitive skills such as comprehension, observation, intuition, and focus. Ability of children for abstract thinking, verbal and spatial abilities, and mathematical reasoning when improved, will in turn bene t achievements and comprehensive abilities of children. The program duration and feasibility to implement in diverse settings including the schools, shows that the bene ts of this program potentially outweigh the efforts. Early childhood education requires a paradigm shift as emphasized by India National Education Policy (2020) to advance human potential and development, and the present study provides new evidence and direction in this regard. Figure 1 Multimodal Cognitive Training can in uence traits such as focus, observation and empathy, and have transfer effects on behavior.     Scatter Plot after Factor Analysis on the Mixed Data (Categorical and Continuous) and on the dataset collected from the location Mahabubnagar. The purple color represents the Intervention group while the orange color represents the Intervention Group.