Benefits of a dance group intervention on institutionalized elder people: a Bayesian network approach

Let us consider a finite set X = {X₁, X₂ … X_n of discrete random variables and P a joint probability distribution over the variables in X. A Bayesian network (BN) is a directed acyclic graph that encodes a joint probability distribution over the set of random variables X. Formally, a BN for X is a pair B = (G, Θ) where G is the graph whose vertexes are the random variables X₁, X₂ … X_n, and the Θ represents the set of parameters that qualify the network. The graph encodes independence assumptions: each variable X_i is independent of its non-descendants given its parents in G, the parameters. Here Θ consists of a set of parameters θ_xi|Πxi = P_B(x_i|Π_xi) for each possible value x_i of X_i and Π_xi of Π_Xi, where Π_Xi denotes the set of parents of X_i in G. A Bayesian network B defines a unique joint probability distribution over X given by:

PB(X1,X2,…<!-- ? -->,Xn)=∏<!-- ? -->i=1nPB(xi|Π<!-- ? -->xi)=∏<!-- ? -->i=1nθ<!-- ? -->xi|Π<!-- ? -->xi.$$\begin{array}{} \displaystyle P_B(X_1, X_2, \dots, X_n) = \prod_{i = 1}^{n}P_B(x_i|\Pi_{x_i}) = \prod_{i = 1}^{n}\theta_{x_i|\Pi_{x_i}}. \end{array}$$(1)

These networks permit to summarize a number of conditional dependence relations graphically. The graph structure can also be used to represent cause-effect relationships through the edges and their directions between pairs of nodes. The structure can be determined by different algorithms such as expectation-maximization algorithm, evolutionary algorithms, Gibbs sampling-based algorithms, or expert knowledge, see [14, 15]. In the case under consideration, due to the nature of the experimental study, we employed expert criteria to propose the structure displayed in Figure 1. As it will be seen later, the dance intervention will increase the probabilities of presenting higher psychological well-being and a better balance and gait, as well as to decrease the probabilities of presenting depression and emotional distress in institutionalized elder people.

Fig. 1

A total of 26 elder adults institutionalized in a nursing home “City of Valencia” in Valencia (Spain) participated in the study. Their mean age was 81, 5 (SD = 6.5), and 69.2% of them were women. The average of their years of schooling was 9.27 (SD = 5.09). In order to be included in the study, participants must be permanent residents of the nursing home and present a functional status that enabled them to participate in the dance intervention with no serious risk for their health. Participants with any severe functional limitation either physical or psychological were excluded.

The benefits of the intervention will be measured according to the following four dimensions:

Psychological Well-Being was measured using the Spanish adaptation of the 54-item version Ryff’s Psychological Well-Being Scale [16, 17]. It is a 6-point Likert-scale ranging from 1 (“Totally disagree”) to 6 (“Totally agree”) with items formulated both in positive (e.g. “In general, I feel I am in charge of the situation in which I live”) and negative (e.g. “I gave up trying to make big improvements or changes in my life a long time ago”), although for the analyses negative items were inverted. The Cronbach’s α obtained for the reliability of the scale was .89, showing good internal consistency.

We developed a randomized controlled trial with two times of assessment and data gathering. The present investigation was favorably informed by the Human Research Ethics Committee of the University of Valencia, since meets the fundamental principles established in the Declaration of Helsinki and the Council of Europe Convention on Human Rights, and the established requirements in Spanish legislation in the field of biomedical research, the protection of personal rights and bioethics are respected.

The flowchart of the design and procedure followed is displayed in Figure 1. Potential participants were selected by the psychologist and the physician of the nursing home. The inclusion and exclusion criteria followed for the selection of the participants have been described above in the paragraph concerning the participants.

They were then informed about the program by the psychologist in charge of the assessments and the facilitation of the dance intervention and asked for voluntary participation. Participants that voluntarily accepted to participate signed the informed consent and were acquainted with the possibility to leave the program at any stage if they did no longer want to continue with no compromise.

All participants were assessed at the entry-level and then assigned randomly (with envelopes containing the group code) to the experimental group or the control one. Both experimental and control group had a final size of 13 participants each. The participants in the experimental group attended dance lessons while the participants in the control group were assigned to the waiting list and took the usual leisure activities of the center: cognitive training, arts and crafts, table games, cinema, etc. All participants were assessed again after the intervention. Participants were assessed individually in an isolated room during the time periods scheduled as free time. For the sake of equity among all the participants, after the post-intervention assessment, the participants of the control group were given the opportunity to attend the dance intervention.

The dance intervention consisted on dance group lessons of ballet adapted to characteristic old people physical functional limitations. The dance lessons were complemented with brief group dynamics at the beginning and the end of the sessions to work on physical and psychological self-awareness and self-regulation both at the individual and group levels. The facilitator was a trained psychologist and retired professional ballet dancer. The duration of the intervention was of a total of 18 sessions of 45 minutes each with a frequency of 2 times per week during 9 weeks. The intervention was conducted in the time scheduled for leisure activities but in the area normally employed for physical and active group activities.

Since no assumption could be done concerning the distribution of the values on any of the variables of the study, we have used the Wilcoxon unpaired test for evaluating the results of the intervention. Mean, standard deviations, and the p-values of the study variables before the intervention (initial time) and after it (final time) are displayed in Table 1. We also include the detail of the effects on each subdimension:

Table 1

Mean, standard deviations, and p-values of the study variables before (time 1) and after the intervention (time 2) for the intervention (INTERV.) and control (CONTROL) groups

Looking at the p-value at the initial time we cannot conclude that there was a significative difference in any of the variables of study in both groups. In contrast, after the intervention there has been a significative difference between both groups in depression (GDS), in all items about psychological well-being except in autonomy, in the general health situation (GHQ-28), and in balance and gait (POMA), see Table 1.

Fig. 2

A BN does not need that the joint distribution follows a specific parametric distribution. It is also consistent in probabilistic associations between the variables. In fact, it is not necessary to maintain the joint distribution in memory for doing Bayesian inference. Besides, it also let us evaluate many structure function interactions without explicitly consider the multiple comparison problem.

For analyzing the BN, the first step was to discretize the variables in the line of [24]. For the variable “type of group”, the experimental group was codified with 1 and the control group with 0. As the variables were initially measured following a scale, we computed the histograms and frequencies for each variable in order to discretized them.

Then, we randomly divided the data into two sets: the training set (80%) and the test set (20%). We used the software Netica

https://www.norsys.com/

Mean, standard deviations, and Pearson correlation coefficients of the study variables at time 1 are displayed in Table 2, as well as those from time 2 in Table 3. As it can be seen in the tables, there was no correlation before the intervention between the type of group and any of the study variables, whereas after the intervention the type of group is significantly correlated with all the variables studied. As expected, the intervention increased the levels of psychological well-being (r = .823, p < .001) and balance and gait (r = .649, p < .001), as well as decreased the levels of depression (r = .728, p < .001), and emotional distress (r = .732, p < .001). As to the validation of the model, the goodness of fit statistics of the BN show a valid correspondence between the theoretically proposed model and the data for all the study variables.

Table 2

Mean, standard deviations, and Pearson correlations (bilateral) of the study variables before the intervention (time 1)

Table 3

Mean, standard deviations, and Pearson correlations (bilateral) of the study variables after the intervention (time 2)

Both goodness of fit statistics and the percentages of mutual information of the studied variables (i.e., the influence of the type of group on the rest of the studied variables) are displayed in Table 4. Regarding these percentages, and considering the direction of the correlations obtained and already discussed in the previous paragraph, we can infer that the intervention had a positive influence on the levels of psychological well-being and balance and gait of 46.3% and 16.4% respectively, whereas at the same time had a negative influence on the levels of depression and emotional distress of 46.3% and 19.6%, respectively.

Table 4

Bayesian network’s validation and percentage of mutual information between Group and the rest of the variables