CC BY
20
Section 3. Life science
https ://doi.org/10.29013/ELBLS-21-1.2-43-51
Zhou Rui,
Zhangjiagang Foreign Language School, China E-mail: [email protected]
DETECTING COMMON FACTORS INFLUENCING ADHD IN CHILDREN: DEVELOPMENT AND VALIDATION OF A PREDICTIVE MODEL
Abstract. Attention-deficit/hyperactivity disorder (ADHD) is a common chronic condition that is characterized as hyperactivity, difficulty sustaining attention, and impulsive behavior. Without radical caret, symptoms of ADHD patients can only be eased with treatments like behavior therapy, including training for parents, and medications [1].
The objective of this research is to build a predictive model to determine the possibility for a child to develop ADHD caused a variety of factors including gender, race, age, and parents' education. Two models, a logistic regression and an artificial neural network, were tested using the data from National Health Interview Survey (NHIS) 2017. And we compared the two models using various evaluation metrics like the receiver operating characteristic (ROC) curve and the area under the curve (AUC) score. We found that the AUC score for the artificial neural network is 0.69, while that for the logistic regression model is 0.66. Therefore, we concluded that, overall, the artificial neural network has better performance.
According to the logistic regression, older children, non-Hispanic children, and males are more likely to be a victim of ADHD. Asian kids and kids with more educated parents were less likely to have early interventions services. The neural network indicates that the most important predictors of developing ADHD are age, followed by Asian children, mother's education level, White children, sex, Black children, Hispanic origin, and father's education level. We believed those results are useful for doctors and parents to evaluate the possibility of having children with ADHD, and to provide suggestions schoolteachers and peers to change the disruptive behaviors of ADHD children.
Keywords: ADHD, children, a predictive model, treatment, suggestions
1. Introduction who suffer from ADHD. According to the latest
Attention-deficit/hyperactivity disorder study, more than 16 million (9.4 percent) children
(ADHD) is one of the most common chronic con- in the U.S. have an ADD diagnosis [2]. The disease is ditions that tortures millions of children. Over the first diagnosed in childhood, but ADHD can contin-years, it has serious impacts on the lives of children ue as children mature, resulting in their hyperactiv-
ity as well as their disability in maintaining attention and controlling impulsion. People with ADHD tend to be daydreaming, insomniac, and overly talkative. They usually have great difficulty in resisting temptation and getting along well with others. These conditions in children dramatically affect their school performance, including inattentive in class, excessive talking at an appropriate time, and frequently interrupt others' conversations.
According to a Swedish cohort study of 544 children, a considerable association was observed between symptoms of inattentiveness (as measured by the Conners 10-item scale) in children aged 7 and 10 years and academic underachievement at age 16 years [3]. Thus, it is likely that people with ADHD have comparably lower educations levels which leads to their hardship in finding well-paid jobs. In this case, the affecting factors of ADHD in children are in imperious need for people to know to conduct and ensure better prevention of the detrimental consequences triggered by ADHD.
The hypothesis of this study is that "the possibility for a child to develop ADHD is related to one or more common factors such as gender, race, age, and parents' educational level." The objective of this study is to develop a predictive model to detect the likelihood for a child to suffer from ADHD in a family. With the help of the model, parents can evaluate the possibility of having a child with ADHD or that of their children to develop ADHD over time, thus taking appropriate measures such as changing daily habits or parenting methods to prevent the disease.
2. Method
2.1 Data
The National Health Interview Survey (NHIS) is the principal source of information on the health of the civilian noninstitutionalized population of the United States and is one of the major data collection programs of the National Center for Health Statistics (NCHS) which is part of the Centers for Disease Control and Prevention (CDC). The Na-
tional Health Interview Survey (NHIS) Data 2017 was used in this study.
2.2 Model
2.2.1 Logistic regression model development
Logistic regression models were used to calculate the predicted risk. Logistic regression is a part of a category of statistical models called generalized linear models, and it allows one to predict a discrete outcome from a set of variables that may be continuous, discrete, dichotomous, or a combination of these. Typically, the dependent variable is dichoto-mous, and the independent variables are either categorical or continuous. The percentage of students who developed Attention Deficit/Hyperactivity Disorder was the dependent variable in this study.
The logistic regression model can be expressed with the formula:
ln
7
1" 7.
= w0 + wx +... + wmxm
0 11 mm
In the logistic regression, each feature x. has its specific weight w, where w0 is the intercept while w1 through w are the coefficients of the independent
O m tr
variables.
Our task is to find a set of parameters w0,..., wm such that the loss function between the output y and the actual values u
l (y,u Hy - u|l2
is minimized.
2.2.2 Artificial neural network
An artificial neural network is a computational model vaguely inspired by the biological neural networks that constitute animal brains. An artificial neuron that receives a signal then processes it and can signal neurons connected to it. The "signal" at a connection is a real number, and the output of each neuron is computed by some non-linear function of the sum of its inputs.
A typical artificial neural network consists of one input layer, several hidden layers, and one output layer. The input layer is the first layer, the output layer is the last layer, and any layers between them are hidden layers. The data are passed into the input
layer, processed by the hidden layers, and finally transformed into predicted labels in the output layer. In this study, the model has one hidden layer.
A package called "neuralnet" in R was used to conduct neural network analysis. The package neuralnet focuses on multi-layer perceptrons (MLP,
Table 1.- Variables
2.4 Data pre-processing
The data set is pre-processed in this step to improve both the training speed and accuracy. As most machine learning algorithms are not able to deal with missing values, all the data points with missing entries are excluded from training. Then we center and scale each feature variable independently. Feature standardization transforms differ-
Bishop, 1995), which are well applicable when modeling functional relationships. 2.3 Variables
The outcome variable is the percentage of students who developed Attention Deficit/Hyperactiv-ity Disorder (LAHCC13).
used in this study
ent features into comparable scales and ensures all features weigh equally in the training process. The standard scaler function standardizes features by scaling the feature means to zero and standard deviations to unit variance. Finally, we partitioned the dataset into two sets, the training dataset (50%) for model development and the test dataset (50%) for model test.
LAHCC13 Attention Deficit/Hyperactivity Disorder (ADD/ADHD) causes limitation
SEX 1: male 2: female
ORIGIN_I Hispanic Ethnicity: 1: yes; 2: no
RACRECI3 1: White 2: Black 3: Asian
AGE_P Age <18 years old 0-17
MOM_ED 01 Less than/equal to 8th grade 02 9-12th grade, no school diploma 03 School graduate/GED recipient 04 Some college, no degree 05 AA degree, technical or vocational 06 AA degree, academic program 07 Bachelor's degree 08 Master's, professional, or doctoral degree
DAD_ED 01 Less than/equal to 8th grade 02 9-12th grade, no school diploma 03 School graduate/GED recipient 04 Some college, no degree 05 AA degree, technical or vocational 06 AA degree, academic program 07 Bachelor's degree 08 Master's, professional, or doctoral degree
The nominal variable is one kind of categorical variables whose levels are simply labels and thus does not contain any meaning of order. For example, in the variable "RACRECI3", "White" is encoded as 1, and "Black" is encoded as 2, and "Asian" is encoded as 3. Even though we want these three levels to be equally weighted, it is usually problematic, as the logistic regression will mistakenly assume that Black is greater than White. A way to solve this is to use the one-hot encoding technique. The idea behind this approach is to create a separate variable for each race group. Here, a new binary feature was created whose value was used to indicate the particular race group of a student.
3. Results 3.1 Chorogram
Basically, a chorogram is a graphical representation of the cells of a matrix of correlations. The idea is to display the pattern of correlations in terms of their signs and magnitudes by using visual thinning and correlation-based variable ordering. Moreover, the cells of the matrix can be shaded or colored to show the correlation value. The positive correlations are shown in blue, while the negative correlations are shown in red; the darker the hue, the greater the magnitude of the correlation.
Figure 1. Matrix of correlations among variables
According to the chorogram above, ADHD had the strongest positive correlation with age, followed by non-Black people and non-Hispanic origin. Meanwhile, it has the strongest negative relationship with dad's education level, followed by sex, mom's educational level, Asian people and White people.
3.2. Logistic regression model
From the table above, sex, origin, and father's educational level are significant predictors of the dependent variable. Females are less likely to develop Attention Deficit/Hyperactivity Disorder (ADD/ADHD) than males. Non-Hispanic children were more likely to have Attention Deficit/Hyperactivity Disorder
(ADD/ADHD) than Hispanic kids. Kids are more kids and kids with more educated parents were less likely to be a victim of ADHD when they aged. Asian likely to have early interventions services.
Table 2. - Logistic Regression for Having Attention Deficit/Hyperactivity Disorder (ADD/ADHD)
Estimate Std. Error z value Pr(>|z|)
(Intercept) -2.402 0.553 -4.347 0.000 ***
SEX -0.606 0.135 -4.498 0.000 ***
ORIGIN I 0.121 0.153 0.788 0.431
AGE P 0.116 0.014 8.203 0.000 ***
MOM colleage -0.168 0.164 -1.024 0.306
DAD colleage -0.624 0.191 -3.266 0.001 **
White 0.624 0.452 1.381 0.167
Black 0.699 0.468 1.491 0.136
Asian 0.469 0.600 0.782 0.434
3.3. Artificial Neural Network
The structure of the neural network used in this study is shown in (Figure 2).
Figure 2. Artificial Neural
In the figure above, line thickness represents weight magnitude and line color weights sign (black = positive, gray = negative). The network is essentially a black box, so we cannot say that much about the fitting, the weights, and the model. However, it suffices to say that the training algorithm has converged. Therefore, the model is ready to be used.
Network in training sample
3.3.1 Variable Importance in Artificial Neural Network
Variable importance represents the statistical significance of each variable in the data with respect to its effect on the generated model. It ranks each predictor based on the contribution predictors make to the model. This technique helps data scientists weed out certain predictors that are con-
tributing to nothing and that instead add time to neural network. The importance for each variable processing [4]. Garson's algorithm was used to cal- is shown in (Figure 3). culate relative importance of input variables in this
0 0,05 0,1 0,15 0,2
Figure 3. Variable Importance in Artificial Neural Network
The most important predictor was age, followed by Asian children, mother's education level, White children, sex, Black children, Hispanic origin, and father's education level. 3.4 ROC
A receiver operating characteristic curve (ROC curve), is a graph showing the performance of a classification model at all classification thresholds [5]. The x-axis of ROC plot represents false positive rate, while the y-axis represents true positive rate. When the decision threshold changes, a better classifier will have a lower false positive rate and a higher true positive rate. In other words, when the false positive rate of two models are the same, the better one will have a higher true positive rate, which makes the curve come closer to the upper left corner of the ROC space. We can use the ROC curve to compare the performance of logistic regression model and the neutral network. Sometimes, it might be hard to identify which model performs better by directly looking at ROC curves, as one curve may not completely encompass the other. Area Under Curve (AUC) overcomes this drawback by finding
the area under the ROC curve, making it easier to find the optimal model. Usually, a better model has a higher AUROC score.
3.4.1 ROC in the training sample
For the training sample, the AUROC was 0.70 for the logistic regression and 0.72 for the artificial neural network. Artificial neural network performed better than logistic regression did.
3.4.2 ROC in the testing sample
In the testing sample, the AUROC was 0.66 for the Logistic regression and 0.69 for the artificial neural network. Artificial neural network had better performance in the testing sample.
4. Discussion
4.1 Logistic regression analysis and interpretation of the results
Of the 1752 school students participating in the study, 21.2% had Attention Deficit/Hyperactivity Disorder (ADD/ADHD), and 79.8% did not have. Among those patients, about 15.8% were female and 24.2% male. At the significance level of 0.01, the predictive model ofAttention Deficit/Hyperactivity Disorder is:
False positive rate
Figure 4. ROC in the training sample for Logistic Regression (Red) vs Neural Network (Blue)
Figure 5. ROC in the testing sample for Logistic
Predicted logit of adolescent depression: -2.402-0.606*SEX+0.121*ORIGIN_ I+ 0.1 16*AGE_P-0.168*MOM_colle-age-0.624*DAD_colleage+0.624*White+0.699*Bl ack+0.469*Asian
The coefficients of the parameters were interpreted as follows. At the significance level of 0.01: • On average, controlling other variables, a female student is 45 percent less likely to report
Regression (Red) vs Neural Network (Blue)
to develop Attention Deficit/Hyperactivity Disorder (ADD/ADHD) than a male student.
• On average, controlling other variables, a non-Hispanic student is 13 percent more likely to report to develop Attention Deficit/ Hyperactivity Disorder (ADD/ADHD) than a Hispanic student.
• On average, controlling other variables, a one-year older student is 12 percent more
likely to develop Attention Deficit/Hyper-activity Disorder (ADD/ADHD) than a younger student.
• On average, controlling other variables, a student who has a higher educational mother is 15 percent less likely to develop Attention Deficit/Hyperactivity Disorder (ADD/ ADHD) than a student whose mother is poorly educated.
• On average, controlling other variables, a student who has a higher educational father is 46 percent less likely to develop Attention Defi-cit/Hyperactivity Disorder (ADD/ADHD) than a student whose father is poorly educated.
• On average, controlling other variables, a White student is 87 percent more likely to develop Attention Deficit/Hyperactivity Disorder (ADD/ADHD) than a student from other race groups.
• On average, controlling other variables, a Black student is 101 percent more likely to develop Attention Deficit/Hyperactivity Disorder (ADD/ADHD) than a student from other race groups.
• On average, controlling other variables, an Asian student is 60 percent more likely to develop Attention Deficit/Hyperactivity Disorder (ADD/ADHD) than a student from other race groups.
It is indicated in the model that being male, being non-Hispanic, being older, having a more educated mother, having a more educated father, being White or Black or Asian are significant predictors in predicting Attention Deficit/Hyperactivity Disorder (ADD/ADHD). Having a more educated mother or father will decrease the probability of suffering from Attention Deficit/Hyperactivity Disorder (ADD/ADHD); while being a male, being non-Hispanic, getting older may increase the probability of developing Attention Deficit/Hyperactivity Disorder (ADD/ADHD). Moreover, compared to be-
ing White or Black, Asians are less likely to develop Attention Deficit/Hyperactivity Disorder (ADD/ ADHD).
According to the study, parents can reflect on themselves, and then take appropriate measures to reduce the possibility of having children with Attention Deficit/Hyperactivity Disorder (ADD/ ADHD). For example, for parents' with a relative low education level, they can improve their prenatal and postnatal care by pursuing a higher education, which provides them with well-paid jobs and thus enough money to take better care of their children. However, since most factors involved in the study are innate and generally unchangeable, treatment and mental comfort are more essential in the face of ADHD patients.
Treatments can be divided into drug intervention and human intervention. For children younger than 6, the American Academy of Pediatrics (AAP) recommends that human intervention like parents' help should be the first line of treatment before medication is tried. However, for older children, both behavior therapy and medications can be applied to help treat ADHD [6]. To achieve success, it is the parents' responsibility to receive training in behavior management first to treat their children's disruptive behaviors. This requires both time and effort, but the benefits to their children will be considerable. Despite parents' commitment, schoolteachers and peers play important roles. School-based management training led by teachers can have an enormous impact on students' constructive manners. For example, a reward system can be applied to encourage positive behavior such as concentrating in class, actively but appropriately participating in a discussion, making progress in academic performance. For peers, they can help classmates with ADHD by caring for them more, playing some concentrating games with them instead of isolating them. In this way, they can help these ADHD patients integrate into the collective, thus providing mental comfort.
4.2 Limitations and future study
Although this study gives predictions on developing ADHD for kids, and a great deal of treatments and suggestions can be evoked from the results of the study, it has limitations. In this study, factors leading to ADHD are innate and determined after birth, which can only be treated by the environmental and medical intervention. If we want to lower the incidence ofADHD more effectively, factors that trigger ADHD even before birth play an important role. In other words, a life's formation and growth in a mother's embryo are susceptible to external influences. A pregnant woman's behavior may have a colossal impact on her child's possibility of developing ADHD. Therefore, to better prevent the development of ADHD in children, factors such as the mother's smoking or drinking history during pregnancy may be included as prime suspects for future study.
5. Conclusion
In this study, logistic regression was used to develop a predictive model to evaluate the prob-
ability ofAttention Deficit/Hyperactivity Disorder (ADD/ADHD). The National Health Interview Survey (NHIS) Data 2017 was used, and factors such as gender, race, age, and parents' educational level were included in the study. The results indicated that this model can be used for doctors and parents to evaluate the possibility of having children with ADHD. With this model, doctors can provide suggestions for parents, schoolteachers, as well as peers to change the disruptive behavior of ADHD children, help them concentrate in class, and offer them better mental care. An application of this model can also be distributed to parents who are worried about their children's potential mental problems and to teachers who deal with ADHD children every day. With this diagnosis model, we believe that the percentage of children with ADHD will decline and fewer children will suffer from ADHD. In the future, more factors such as the mother's smoking or drinking history during pregnancy may be in need of further study.
Refernces:
1. "Treatment of ADHD." Centers for Disease Control and Prevention. Centers for Disease Control and Prevention, September 21, 2020. URL: https://www.cdc.gov/ncbddd/adhd/treatment.html
2. ADHD Editorial Board Medically reviewed by Sharon Saline, ADHD Editorial Board, and Sharon Saline. "ADHD Statistics: New ADD Facts and Research." ADDitude, August 6, 2020. URL: https://www.ad-ditudemag.com/statistics-of-adhd
3. Holmberg, Kirsten, and Sven Bolte. "Do Symptoms ofADHD at Ages 7 and 10 Predict Academic Outcome at Age 16 in the General Population?" Journal of attention disorders. U. S. National Library of Medicine, November 2014. URL: https://www.ncbi.nlm.nih.gov/pubmed/22837550
4. Chauhan, Avkash. "What Is Variable Importance and How Is It Calculated? - DZone AI." dzone.com. DZone, June 15, 2017. URL: https://dzone.com/articles/variable-importance-and-how-it-is-calculated
5. "Classification: ROC Curve and AUC | Machine Learning Crash Course." Google. Google. Accessed February 3, 2021. URL: https://developers.google.com/machine-learning/crash-course/classification/ roc-and-auc
6. "Treatment of ADHD." Centers for Disease Control and Prevention. Centers for Disease Control and Prevention, September 21, 2020. URL: https://www.cdc.gov/ncbddd/adhd/treatment.html
