A NEURAL NETWORK APPROACH TO IDENTIFYING ADOLESCENT ADJUSTMENT.

Nair, Jyotsna ; Nair, Satish S. ; Kashani, Javad H. 等

ABSTRACT

This study examined the relationship between the quality of adjustment in adolescents and a set of psychiatric diagnoses, personality traits, parental bonding, and social support variables. One hundred fifty adolescents were administered the Millon Adolescent Personality Inventory, the Parental Bonding Questionnaire, the Social Support Questionnaire, and the Diagnostic Interview for Children and Adolescents. A neural network approach was then utilized, and it was found that several of the variables (e.g., Major Depressive Disorder, Conduct Disorder, and Societal Conformity) had a significant role in classifying adolescents into three groups: maladjusted, nominally adjusted, and well-adjusted.

Few studies have identified risk and protective factors in both dysfunctional and well-adjusted adolescents. Kashani et al. (1987) found that well-adjusted youths had goad self-concepts, caring parents, and satisfactory social support systems. Using the same sample, the present study sought to determine whether a neural network approach would offer additional information. A neural network is a nonlinear regression model that can predict outputs (or effect variables) using several inputs (or cause variables) and quantify complex relationships between such cause-and-effect variables (McCord-Nelson & Illingworth, 1991). Neural networks have been applied in several areas, including mental health (Cohen & Servan-Schreiber, 1992; Kashani et al., 1996; Nair et al., 1996). Here, a neural network model was created to ascertain how various measures would relate to adolescents characterized as well-adjusted, nominally adjusted (some dysfunctions), and maladjusted (serious dysfunctions).

METHOD

Data Collection

One hundred fifty youths between 14 and 16 years of age were drawn from a sample of 1,700 midwestern public school students. There were equal numbers of boys and girls. Ninety-five percent were Caucasian, and the rest were Asian or African American. Other characteristics are described in the study by Kashani et al. (1987).

Diagnoses were made based on data collected from adolescents and their parents using the Diagnostic Interview for Children and Adolescents (DICA; Herjanic et al., 1975; Herjanic & Reich, 1982). The DICA diagnoses used in this study were Oppositional Defiant Disorder, Conduct Disorder, Anxiety, and Major Depressive Disorder (Depression). The Millon Adolescent Personality Inventory (MAPI; Millon et al., 1972) was used to obtain information about adolescents' personalities. The Millon scales used were Cooperative, Forceful, Sensitive, Personal Esteem, Social Tolerance, Family Rapport, Impulse Control, and Societal Conformity. Additional inputs to the neural network were the Parental Care and the Parental Overprotection scales of the Parental Bonding Instrument (PBI; Parker et al., 1979), and the Satisfaction Rating and Total Number of People scales of the Social Support Questionnaire (SSQ; Sarason et al., 1983). Gender was also included in the neural network model, bringing the number of inputs to 17 (see Figur e 1).

The 150 adolescents were divided into three groups, based on clinical interviews conducted by expert psychiatrists (Kashani et al., 1987), and these were used as the model outputs. Group 1 consisted of troubled adolescents with psychiatric disorders. These maladjusted adolescents had one or more DSM-III diagnoses, experienced impaired functioning, and were in need of treatment. Group 2 consisted of nominally adjusted adolescents. They were not free of symptoms, but were not in need of treatment. Group 3 consisted of well-adjusted adolescents. They were free of psychiatric syndromes or symptoms. Seven adolescents were dropped from the analyses due to missing data.

Neural Network Modeling

A multilayered back-propagation neural network was used (17 inputs from each of the 143 adolescents, comprising the input patterns, and the three binary outputs). The network was exposed to the data, and the parameters (weights and biases) were adjusted to minimize error, using a back-propagation training algorithm. The procedure for analyzing the data involved three stages: (1) The inputs (personality, DICA diagnoses, parental bonding, social support, and gender variables) were mapped to the three outputs (maladjusted adolescents, nominally adjusted adolescents, and well-adjusted adolescents) using a neural network. (2) Each input was varied, one at a time, across its minimum to maximum range to determine how the increase in that input affected the output (i.e., whether group membership would change). This perturbation process (also called contribution analysis) was used to identify the variables most related to troubled adolescents and those most related to well-adjusted adolescents (i.e., variables associ ated with risk and protection). (3) Statistical analyses were conducted to confirm the relative effect of the input variables on the outputs.

Validation

The ability of the neural network model (a 17 X 20 X 5 X 3 structure was used) to perform the classification was examined by setting aside 20% of the patterns (or observations) as validation (or testing) data. In this cross-validation approach, the training involved repeatedly exposing the network to the remaining 80% of the patterns (training data) for several epochs, where an epoch is one complete cycle through the network for all cases. (Data were normalized before training.) Simultaneously, the prediction errors in the testing data were monitored. Typically, the training errors (in this instance, for the 80% set) drop consistently while the testing errors (for the 20% set) drop and then increase with continued training. The optimal number of training epochs is achieved when the training and testing errors are both acceptable. After the optimal number of training epochs is determined, the data are pooled (i.e., the training and testing data are combined) and the network is trained for this number of epoch s using the combined set. A network trained in this manner is considered generalizable, in the sense that it can be used to make predictions.

Contribution Analysis

Contribution analysis identifies the group to which an individual will belong for each change in input. Contribution analysis considers one variable at a time, keeping the remainder constant. This is a powerful way to change one factor, or a group of factors, and see the overall impact of that change on the output.

Each of the inputs was perturbed from its minimum value (zero) to its maximum value (one) for every adolescent, and the outcome of the neural network was computed. The change in the number of predicted individuals in a group when an input was varied from zero to one in all the patterns represented the contribution of that input to the output. The change in adjustment could then be predicted when, for example, the diagnosis of Depression went from minimum to maximum. This methodology thus quantified the dependence of a particular classification, for example, well-adjusted, on a particular input. Multivariate analyses were also performed to validate whether the neural network model can be used as an "expert" to determine the group to which an individual would belong.

RESULTS

Contribution Analysis

After 30,000 epochs of training, the network correctly classified 81% of the patterns (validation observations) into the three groups. This was found to be the optimal number of training epochs; it had the smallest training and testing errors. The misclassification of patterns could be due to inadequate training (the variety of observations to which the network was exposed) or to diagnostic discrepancies (the latter being a strong possibility, because 100% accuracy in clinical diagnosis is not possible).

The results of the perturbation process are shown in Table 1. The set of dominant characteristics was different for each of the groups (i.e., variables that optimally identified one group of adolescents were not identical to those that identified another group). For example, Oppositional Defiant Disorder was a predictor for nominally adjusted adolescents but not maladjusted adolescents, while the reverse was true regarding Anxiety.

Table 1 shows the average change in membership in each group when the particular input went from minimum to maximum. A positive number indicates a propensity toward inclusion in that group, while a negative number indicates a tendency toward inclusion in one or both of the other groups. For example, the average change in membership when Depression was varied (minimum to maximum) was 0.68, -0.59, and -0.07 in Groups 1 (malajusted), 2 (nominally adjusted), and 3 (well-adjusted), respectively. In other words, the individual would tend to move to Group 1 and leave the other groups, especially Group 2.

Among the DICA diagnoses (Oppositional Defiant Disorder, Conduct Disorder, Anxiety, and Major Depressive Disorder), Depression and Conduct Disorder were most likely to place the adolescent in the maladjusted group, and to a lesser extent Anxiety. Oppositional Defiant Disorder was most likely to place the adolescent in the nominally adjusted group.

Among the personality inputs, Societal Conformity was most likely to place the adolescent in the maladjusted group, followed by Cooperative, Impulse Control, and Sensitive. However, Social Tolerance was correlated with improved adjustment (i.e., the adolescent leaving the maladjusted group). Sensitive and Family Rapport (i.e., such traits as touchiness and poor family relations) were most likely to place the adolescent in the nominally adjusted group. Cooperative and Forceful were most likely to place the adolescent in the well-adjusted group, and to a lesser extent Impulse Control and Social Tolerance.

In addition, Total Number of People (social support) and Parental Care were likely to place the adolescent in the well-adjusted group, and Satisfaction Rating (social support) was likely to place the adolescent in the nominally adjusted group. Oppositional Defiant Disorder, Anxiety, Sensitive, Family Rapport, and Satisfaction Rating reduced membership in the well-adjusted group. Gender, Personal Esteem, and Parental Overprotection had virtually no impact on whether the adolescent was troubled or not.

Thus, Conduct Disorder and Major Depressive Disorder, along with several other variables, were significantly related to group classification. It has been noted by other researchers (e.g., Robbins, 1966) that adolescents with Conduct Disorder are likely to display antisocial behaviors or other psychiatric problems.

As an example of the contribution analysis, Figure 2 illustrates the variation in the extent of membership in each group as one input, Sensitive, was varied from its minimum value to its maximum (note the nonlinearity). There was a propensity toward inclusion in the nominally adjusted group and a decline in membership in the well-adjusted group. The results suggest that clinical efforts to make maladjusted adolescents less pathologically sensitive would not be very beneficial. The other inputs were also varied in this fashion, and the changes in the outputs followed expected trends.

Statistical Analysis

A multivariate analysis of variance (MANOVA) was conducted for each of the three outputs to test whether the means of the 17 slopes from the contribution analyses were different from zero, and Hotelling [T.sup.2] was used to test whether the means were equal. For each of the 17 inputs, 99% simultaneous confidence intervals, adjusting for the number of variables, were computed.

For the first output (maladjusted), F(17, 125) = 28.09, p [less than] .0001, and [T.sup.2] = 570.14, p [less than] .0001. For the second output (nominally adjusted), F(17, 125) = 39.62, p [less than] .0001, and [T.sup.2] = 691.09, p [less than] .0001. For the third output (well-adjusted), F(17, 125) = 12.17, p [less than] .0001, and [T.sup.2] = 246.93, p [less than] .0001. The variables with significant 99% simultaneous confidence intervals are noted in Table 1.

DISCUSSION

A neural network was developed that functioned as an "expert" to classify adolescents as maladjusted, nominally adjusted, or well-adjusted. Contribution analysis identified the inputs that impacted the classifications to the greatest extent. The results point to Depression and Conduct Disorder as being related to maladjustment, along with Societal Conformity (the lack thereof).

Conduct Disorder was a diagnosis associated with an increase in the extent of membership in the maladjusted group. Interestingly, an increase in the diagnosis of Oppositional Defiant Disorder led to a decline in the extent of membership in the well-adjusted group and a corresponding increase in the extent of membership in the nominally adjusted group. Soltys et al. (1992) found that the two disorders lie on a continuum of severity, with Conduct Disorder being the more severe of the two. The findings of the present study are thus in accordance with theirs.

Anxiety in moderation can improve an individual's performance, but high levels are disabling. Here, increases in Anxiety were associated with increased maladjustment.

Among the personality variables, increases in cooperativeness increased the extent of membership in the well-adjusted group and, to a smaller degree, in the maladjusted group as well. Cooperativeness may be an indicator of the importance of peer relationships in this age group and how friends can be helpful (improve adjustment) or harmful (e.g., lead to gang involvement, stealing in groups). Being assertive (forcefulness) improved adjustment, and this supports the earlier finding that forcefulness is associated with lower hopelessness (Kashani et al., 1996). Results were in the expected directions for the other personality variables. Hypersensitive individuals, socially intolerant individuals, and those with limited family rapport demonstrated adjustment problems. Lack of social conformity also increased maladjustment.

As expected, good parental relationships (Parental Care) improved adjustment. It had previously been reported that overprotection by parents increased hopelessness (Kashani et al., 1996). Greater social support (Total Number of People) was also associated with better adjustment.

Clinical Significance

The results indicate the need for early interventions with adolescents diagnosed as being depressed, anxious, or conduct disordered. Socially acceptable cooperativeness and assertiveness (forcefulness) are skills that adolescents should be taught. Being "thick skinned," tolerant of others, and more conforming to social norms should also be encouraged. In addition, family issues should be addressed in order to improve adolescent adjustment.

This research was supported in part by the University of Missouri Research Board, Grant No. C-3-41465.

Jyotsna Nair, Satish S. Nair, Javad H. Kashani, John C. Reid, and Venkatesh G. Rao, University of Missouri--Columbia.

Reprint requests to Jyotsna Nair, Department of Psychiatry and Neurology, University of Missouri--Columbia, 1 Hospital Drive, Columbia, Missouri 65212. Electronic mail may be sent to nairj@health.missouri.edu.

REFERENCES

Cohen, J. D., & Servan-Schreiber, D. (1992). Introduction to neural network models in psychiatry. Psychiatry Annals, 22, 113-118.

Herjanic, B., Herjanic, M., Brown, F., & Wheatt, T. (1975). Are children reliable reporters? Journal of Abnormal Child Psychology, 3, 41-48.

Herjanic, B., & Reich, W. (1982). Development of a structured psychiatric interview for children: Agreement between child and parent on individual symptoms. Journal of Abnormal Child Psychology, 10, 307-324.

Kashani, J. H., Nair, S. S., Rao, V. G., Nair, J., & Reid, J. C. (1996). Relationship of personality, environmental, and DICA variables to adolescent hopelessness: A neural network sensitivity approach. Journal of the American Academy of Child and Adolescent Psychiatry, 35(5), 640-645.

Kashani, J. H., Rosenberg, T., Beck, N. C., Reid, J. C., & Battle, E. F. (1987). Characteristics of well-adjusted adolescents. Canadian Journal of Psychiatry, 32, 418-422.

McCord-Nelson, M., & Illingworth, W. T. (1991). A practical guide to neural nets. New York: Addison-Wesley.

Millon, T., Green, C. J., & Meagher, R. B. (1972). Millon Adolescent Personality Inventory Manual: Interpretive scoring systems. Minneapolis: National Computer Systems.

Nair, J., Nair, S. S., Kashani, J. H., Mistry, S. I., & Reid, J. C. (1999). Analysis of the symptoms of depression: A neural network approach. Psychiatry Research, 87, 193-201.

Nair, S. S., Reid, J. C., & Kashani, J. H. (1996). Neural network models in psychiatry. In M. J. Miller, H. W. Hammond, & M. J. Hile (Eds.), Mental health computing (pp. 365-385). London: Springer-Verlag.

Parker, G., Tupling, H., & Brown, L. B. (1979). A parental bonding instrument. British Journal of Medical Psychology, 52, 1-10.

Robbins, L. (1966). Deviant children grown up. Baltimore: Williams and Wilkins.

Sarason, I. G., Levine, H. M., Basham, R. B., & Sarason, B. R. (1983). Assessing social support: The Social Support Questionnaire. Journal of Personality and Social Psychology, 44, 127-139.

Soltys, S. M., Kashani, J. H., Dandoy, A. C., Vaidya, A. F., & Reid, J. C. (1992). Comorbidity for disruptive behavior disorders in psychiatrically hospitalized children. Child Psychiatry and Human Development, 23(2), 87-98.

Table 1. Results of the Contribution Analysis
 Average Change in Membership
Inputs Maladjusted
Gender 0.00
Oppositional Defiant 0.03
Conduct Disorder 0.34*
Anxiety 0.16*
Depression 0.68*
Cooperative 0.22*
Forceful 0.04
Sensitive 0.10*
Personal Esteem ss -0.08
Social Tolerance ss -0.30*
Family Rapport ss 0.03
Impulse Control ss 0.17*
Societal Conformity ss 0.48*
Parental Care 0.01
Parental Overprotection 0.04
Satisfaction Rating -0.10*
Total Number of People 0.02
Inputs Nominally Adjusted Well-Adjusted
Gender -0.02 0.01
Oppositional Defiant 0.16* -0.18*
Conduct Disorder -0.37* 0.02
Anxiety -0.02 -0.14*
Depression -0.59* -0.07
Cooperative -0.47* 0.35*
Forceful -0.30* 0.24*
Sensitive 0.37* -0.43*
Personal Esteem ss 0.01 0.06
Social Tolerance ss 0.12 0.15*
Family Rapport ss 0.26* -0.32*
Impulse Control ss -0.33* 0.17*
Societal Conformity ss -0.39* 0.04
Parental Care -0.10 0.12*
Parental Overprotection -0.03 -0.02
Satisfaction Rating 0.14* -0.l0*
Total Number of People -0.22* 0.20
(*)99% simultaneous confidence intervaldid not contain zero.
(ss)Reversed scales (i.e., higher valuesindicate greater pathology).

[Graph omitted]