14 5 / 2013

## Final blog post

**THE ASSOCIATION BETWEEN MENTAL HEALTH AND ALCOHOL USE DISORDERS: THE ROLE OF GENDER DIFFERENCES**

by Ellen Brock, Bangalore, India

**Introduction**

The relationship between mental health and alcohol use disorders (AUD) is extremely complex. Not only are there different types of alcohol use disorders such as alcohol use and alcohol dependence, there are many different types of mental health as well. Mental health can range from being perfectly “normal” to suffering from depression, anxiety disorders, multiple disorders, etc.

This research question is furthermore very complex since the association does not tell us much on the exact causation between these two variables and can go in both directions (see e.g. Shivani et al., 2002).

While some work has been done on whether mental health status effects alcohol use disorders and vice versa (see e.g. Passchal et al. (2005) among others, little is known about the effects of different types of mental health on alcohol use disorders. Moreover, I also expect gender differences to play a major role as women e.g. might be subject to different types of mental health problems than man and that at e.g. certain points in their life such as child birth, menopause, etc. Kessler et al. (1997), among others, find that depression can initiate alcoholism and that this link is mainly strong in women.

It is extremely challenging for those who diagnose and treat alcohol use disorders, to know whether a further psychological problem is present or not and which psychological problem is most likely to occur in the case of AUD. Understanding this relationship will help clinicians to make an exact diagnosis and prescribe the right treatment.

**Research questions**

*Is there an association between having a mental health disorder (MHD, YES/NO question) versus having an alcohol use disorder (AUD, YES/NO question)?**Is mental health status (MH_status as defined by 8 different categories including healthy individuals and individuals with multiple disorders) associated with having an Alcohol Use Disorder?**Is the association different among men and women?*

**Methods**

**Sample**

- All the individuals (n = 43 093) in the National Epidemiologic Survey of Alcohol and Related Conditions (NESARC)
- NESARC is a nationally representative sample of non-institutionalised adults in the US.

**Measures**

- Mental health (MHD) and alcohol use disorders (AUD) are assessed using the NIAAA Alcohol Use Disorder and Associated Disabilities Interview Schedule DSM - IV (AUDIS-IV).
- Alcohol use disorder (AUD) means a person suffering from alcohol use dependence (also called alcoholism) and/or alcohol abuse (individual’s activities and responsibilities are affected due to his drinking such as like driving under influence and/or have legal and social problems due to drinking (family arguments and/or being arrested due to drinking). This variable was recoded in SAS as a 0/1 type of variable and is considered as response variable in my analysis.
- Mental health disorder (MHD) means a person is suffering from one or more of the following conditions: depression, panic disorder exclusive of agoraphobia, agoraphobia, social phobia, generalised anxiety disorder, pathological gambling.This variable was also recoded in SAS as a 0/1 variable and is considered as a explanatory variable in my analysis and will be used for answering research question 1 as explained above.
- The mental health status (MH_status) is used for answering research question 2 of my analysis and serves as another explanatory variable. I have recoded this variable in SAS into 8 different categories: no mental health disorder during lifespan (0), depression (1), panic disorder exclusive of agoraphobia (2), agoraphobia (3), social phobia (4), generalised anxiety disorder (5), pathological gambling (6) and multiple mental health disorders (7).

**Results**

*Univariate results*

The following tables give the frequency distribution for the person suffering from Alcohol Use Disorders (AUD) and Mental Health Disorders (MHD). Also, the frequency distribution for the Mental Health Status (MH_status) and the gender (sex) is given below:

These frequency tables show that:

- 27.5 % of the adults are suffering from a Alcohol Use Disorder (AUD)
- 23 % of the adults are suffering from a Mental Health Disorder (MHD)
- When looking at the MHD in greater detail, most are suffering from depression (11.79% of the entire sample, MH_status = 1) followed by multiple disorders (6.88%, MH_status = 7) and then a social phobia (1.82%, MH_status = 4).
- The sample has 57% males versus females.

For more information how these variables are constructed, please see the SAS code given at the end of this section and/or go to the earlier blog posts (see Assignment 3 and the blog on “My variables of interest for the Alcohol Use Disorders and Mental Health Disorders) in greater detail.

**Bivariate results**

Remember that I have two research questions (see above). First, I look at the association between MHD and AUD where both variables are 0/1 type categorical variables (and hence is a much simplier analysis).

I have used a Chi-square test to look at the association between the explanatory (MHD) and the response (AUD) variable. The results are given here (see also blog post on Assignment 7):

Second, I look at the association between mental health status (MH_status where actually MHD is split up in more categories and hence my explanatory is also categorical but contains more categories in comparison to my first research question.

Both tables were obtain in SAS using the “proc freq” procedure. After conducting the chi-square test for my second research question, I went one step further and as explained in the lectures, I also ran post hoc tests. More specifically, I ran comparisons for each pair of MH_status categories. Remember that I have 8 categories in our MH_Status variable so this means we need to run 28 comparisons!

The results of these post hoc tests are given by the following table and graph (for which the results will be further explained below):

This Figure is similar as in the lecture and is made based on a) the column percentages from the big cross tabulation table (see red rectangle) and b) the results from the table above from which several groups are obtained. If categories of MH_status have the same letter, then they belong to the same group.

Before turning to the discussion of moderation in the next section, I also present the results of a Chi-square test to see the association between gender (the moderation variable) and AUD:

The results based on the above tables and figure for the bivariate analysis are:

- There is an association that is statistically significant between Mental Health Disorders (MHD) and Alcohol Use Disorders (AUD) with X2 = 1194.74, df = 1 and p < 0.0001.
- There is also an association between MH_status and AUD with X2 = 1370, df = 7 and p < 0.0001.
- The adjusted Bonferroni p-value is defined as 0.05 divided by the number of comparisons (28 in my case). When I calculate this value I get an adjusted Bonferroni p-value of 0.001786.
- Group 0 (healthy individuals) is significantly different in terms of AUD rates in comparison to all the other groups except group 3 (agoraphobia).
- Group 7 (multiple health disorders) is significantly different in terms of AUD rates in comparison to all the other groups except group 3 (agoraphobia).
- Males are significantly more likely to meet criteria for AUD (40.39%, see row percentages) than females (17.76%) with X2 = 2713.6, df = 1 and p< 0.0001.
- Group 6 (pathological gambling) is significantly different in terms of AUD rates in comparison to all the other groups except group 3.

**Moderation**

Before explaining the topic of moderation in more detail, I will first show the SAS results for the two different research questions: a) association between AUD and MHD and b) association between MH_status and AUD. In this section I have investigated whether a third variable, gender, moderates the association between these two associations.

For the first research question, the following results are obtained females and males (please look at the row percentages).

For the second research question, the following results are obtained for females and males respectively:

These are the results:

- Gender does not moderate the association beween mental health disorders and alcohol use disorders (AUD).
- Gender does not moderate the relationship between mental health status (MH_status) and AUD except for category 5 (generalized anxiety disorder) where I see a dip for the females in comparison to the males (see figure below and the last two tables above).
- Men are subject to higher AUD rates than females for each category of mental health status (see figure below).

**Discussion**

*What might the results mean?*

- First research question: There is a strong association between suffering from a mental health disorder and suffering from a alcohol use disorder both for men and women.
- Second research question: Different health status seem to be associated with different rates of alcohol use disorders.
- Men seem to be more subject to AUD than women when suffering from a mental health disorder.
- Moreover, for each level of mental health status men are subject to higher rates of alcohol use disorders.

**Strengths**

Results are based on a large sample of the US adult population.

**Limitations**

The following figures illustrates some of the limitations of this study:

- This study does not consider the direction of causation between alcohol use disorders and mental health disorders. As pointed out before in this blog post and in the blog post below on “The association between mental health and alcohol use disorders”, the causation can go both ways.
- Third variables such as facing a personal crisis or family issues (see last blog post) could ”confound” the analysis between AUD and MHD. In other words, we might find an association between MHD and AUD but this association could be due to the fact that this third variable (which one that could be here will be explained next) is causing both the response and explanatory variable to change. Since both variables change due to this third variable, one would possibly (wrongly) conclude that there is an association between MHD and AUD.
- AUD consists of several components. A person can be suffering from AUD when he/she is suffering from: a) alcohol dependence (alcoholism) or b) alcohol abuse or c) both alcohol dependence and alcohol abuse. This study does not distinguish between these different type of AUD.

**Recommended for future research**

- Based on another Coursera course, Data Analysis with Professor Leek, I understand there are statistical methods to investigate the problem of “third variables”. The NESARC dataset has some of these third variables such as variables related to financial and family problems.
- Look at the association between MHD and AUD more in detail by considering more detailed definitions of the AUD variable and distinguish between individuals suffering from alcohol dependence or alcohol abuse or both conditions. The NESARC dataset does provide this distinction.

**References**

Kessler, R.C., Crum, R.M, Warner, L.A., Nelson, C.B., Schulenberg, J. and Anthony, J.C. (1997), Lifetime co-occurrence of DSM-III-R alcohol abuse and dependence with other psychiatric disorders in the National Comorbidity Survey, *Archives of General Psychology, *54(4), p. 313-21.

Passchal, M.J., Freisthler, B. and Lipton, R.I. (2005), Moderate Alcohol Use and Depression in Young Adults: Findings From a National Longitudinal Study, *American Journal of Public Health, *95(3), p. 453–457.

Shivani, R., Goldsmith, J., Anthenelli, R.M. (2002), Alcoholism and Psychiatric Disorders. Diagnostic Challenges, *Alcohol Research and Health*, 26 (2). p. 90-98.

** Appendix: SAS code**

14 5 / 2013

## Some final thoughts on causation and the third variable

As professor Dierker points out in her class, we need to be very careful when we speak about association. Association in no circumstance means causation and below (in an earlier post) I had illustrated this for the case of the MHD and AUD.

After watching the last video on the so called “third variable”, I thought for my analysis what a so called third variable could be. More specifically, the following pictures tries to give some of the ideas I am writing here and points out to both the direction of causation and the influence of a third variable. In other words, we might find an association between MHD and AUD but this association could be due to the fact that this third variable (which one that could be here will be explained next) is causing both the response and explanatory variable to change. Since both variables change due to this third variable, one would possibly (wrongly) conclude that there is an association between MHD and AUD. This example is similar to the ice cream sales and the number of people drowning as was covered in the class.

When looking at the NESARC data into greater detail, I thought that the following variables could be third variables:

**S1Q2310:**Experienced a major financial crisis, bankruptcy or been unable to pay bills on time in the last 12 months.**S1Q2311:**You or family members had trouble with police, got arrested or send to jail in the last 12 months.

Exploring this influence of the third variable on the association between Mental Health Disorders and Alcohol Use disorders could be one nice area for future work.

13 5 / 2013

## Assignment 9: Moderation

In this exercise, we were asked to see if there is a difference in association according to different subgroups. For my analysis, I have defined the subgroup according the sex. More specifically, I see whether the association differs between man and women. The variable that refers to the sex of a person is called SEX”. Before running the moderation exercise, I slightly recoded this variable and defined a 0/1 variable called “Male”:

After running the SAS code for the frequency distribution (table not reported here), I found that there are 43% females and 57% males in the dataset.

Remember, I worked on two research questions. First, I see whether MHD (Mental Health Disorders) are associated with AUD (Alcohol Use Disorders). These two variables are 0/1 variables. The following code in SAS is run for this:

The results are given here for females:

And here for males:

Important to see is where the red rectangle is in the last two tables. The discussion is similar to what was discussed under the discussion of the Chi-Square statistic (assignment 7) before. For both subgroups (men and women), we can observe (by looking at the row percentages) that those who suffer from a MHD are more likely to suffer from a AUD too. Also, the Chi-square statistic is equal to 1192.95 (with p<0.0001) for females and 664.97 for males (with p<0.001).

My second research question focuses on whether there is a link between mental health status, MH_Status, and AUD. Subsequently, I ran the following SAS code to see if the relationship between MH_status and AUD is changing between females and males:

The results of the SAS code can be found here for females:

And here for males:

The results of the last two tables show that gender does not moderate the relationship between MH_status and AUD (please note the red rectangles). Looking at the F-statistic, one can find an association between MH_status and AUD for both females and males. More specifically, for females the F-statistic is 1370 (p<0.0001) and for males it is 712 (p<0.0001).

Similar to what Professor Dierker has done, I have plotted to graphs using the column percentages for both males and females. From these graphs I can also conclude that gender does not moderate the relationship between MH_status and AUD except for category 5 (generalized anxiety disorder) where I see a dip for the females in comparison to the males:

13 5 / 2013

## Assignment 8: Computing the Pearson correlation coefficient

In this assignnment, we were asked to compute the correlation coefficient between two continuous variables. Since both my main variables for the X- and Y-variable for this analysis are categorical, I had to think how to approach this problem. During the ANOVA analysis (see assignment 6), I had computed a continuous variable, the number of drinks a person has during the past 12 months (NUM_DRINKS_YEAR). Remember that my study is on the association between Mental Health Disorders (MHD) and AUD (Alcohol Use Disorders). As a continuous variable, I have taken the number of mental health disorders a person has (NuMentalHealth).

The following is the SAS code to compute the Pearson correlation coefficient:

The result of running this code is:

You can see that the correlation between the number of mental health issues (NuMentalHealth) and the number of drinks per year is 0.01671(p = 0.0064). This means that only 0.000279 % (0.01671 squared) can be explained by the number of drinks per year.

25 4 / 2013

## Assignment 7: The Chi-square test

In this blog post, I will go back to my original data unlike in the previous post where I had to construct a new quantitative response variable. Remember my research question: Is there an association between having a mental health disorder (MHD)(explanatory variable) and having an alcohol use disorder AUD (response variable). Both this response and explanatory variable were defined as 0/1 variables and the way they were computed I have explained in previous blog posts.

As explained in the lectures, when one has categorical variables for both the response and explanatory variables one has to perform a Chi-square test to test for association between the explanatory and response variable. The null hypothesis is that there is no difference in AUD between those that have an MHD versus those who have not. The way I have done this in SAS (as Professor Dierker also shows in the lecture) is as follows:

The result is the following:

First, let us interpret the contingency table. I always need to be very careful not to take the wrong statistic! Once more I looked at the example of the drunken driving and sex that was covered in the class and sorted the data in a similar fashion and our contingency table is set up similarly with the independent variable on the left and the response variable on top. First, our attention need to be on the column where AUD = 1 and if we want to see if there is a difference according MHD we need to consider the row percentages in the above table. We see that these row percentages is higher for those who have an MHD versus those who have not. In order to statistically see if there is an association, I looked at the Chi-square statistic. The chi-square test of independence shows those who suffer from MHDs are more likely to suffer from AUD too (42.02%) compared to those who did not suffer from MHDs (23.40%), X2 =1194.74, 1 df, p<0001.

Subsequently, I experienced with the second explanatory variable I have created. Rather than the MHD status which is a 0/1 variable, I ran a chi-square test for the MH_Status variable (which contains the “healthy” individuals and those with “multiple disorders”. When running a similar code like in the previous case, I obtained the following results:

As also explained in the lectures, the numbers that we need to consider in the contingency table are the numbers which are in a rectangle red box and refer to the “Column percentages”. These column percentages show that there are differences among the different categories of MH_status. If we look at the Chi-square statistic, we can observe that we can reject the null hypothesis that all AUD rates are equal across the different MH_status categories since the value for the Chi-square test equals X2 = 1361.26 with 7 df and with p < 0001.

Similarly in the case of the ANOVA where the explanatory variable has multiple categories, here we also cannot tell why the null hypothesis got rejected and hence explain how the AUD rates are different across categories. Unlike in the ANOVA testing, there is not one simple command we can do. More specifically, we will need to run comparisons for each pair. Remember that we have 8 categories in our MH_Status variable so this means we need to run 28 comparisons!

As explained in the lectures, I first need to compare the adjusted Bonferroni p-value. This value is defined as 0.05 divided by the number of comparisons which is in my case 28. When I calculate this value I get an adjusted Bonferroni p-value of 0.001786.

Subsequently (as explained in the lectures too), I ran the Chi-square tests for the 28 combinations (code will be added at the end of this thread). The results are given in the following table:

Remember that I had the following categories for mental health status: 0 for healthy individuals, 1 for those suffering from depression, 2 for panic disorder, 3 for agoraphobia, 4 for social phobia, 5 for a generalized anxiety disorder, 6 for pathological gambling and 7 for multiple health disorders.

Similar to what was explained in the lectures, I also use a plot. The plot is derived from the table above where I plot a bar plot using the table of MH_status by AUD above where the bars are the column percentages (Col Pct) when AUD = 1. In order to have this plot, I ran the following command in SAS to extract the column percentages while setting the AUD = 1:

After I have run this SAS code, I used the bar chart wizard in SAS to create the bar plot. Based on the above table and the Chi-square tests of the previous table, I added letters which refer to groups. Two more MH_status categories share the same letter if they are in the same group based on the Chi-square test.

From this bar plot and the table above, we observe that:

- Group 0 (healthy individuals) is significantly different in terms of AUD rates in comparison to all the other groups except group 3 (agoraphobia).
- Group 7 (multiple health disorders) is significantly different in terms of AUD rates in comparison to all the other groups except group 3 (agoraphobia).
- Group 6 (pathological gambling) is significantly different in terms of AUD rates in comparison to all the other groups except group 3.

25 4 / 2013

## Assignment 6: The ANOVA analysis

In this blog post, I will look at the ANOVA analysis which studies the association between one quantitative response (Y-variable) and one categorical explanatory (X-variable). Since my variables of interest are both categorical, I had to construct/choose another response variable that is quantitative. More specifically, in the previous blog post I described how I had constructed a variable called the number of drinks per year. The explanatory variable is whether a person is subject to a mental health disorder or not (this is a 0/1 variable).

WhenI look at the association between the amount of drinks consumed over the last 12 months(quantitative response variable) and whether a person is suffering from a mental health disorderd (MHD,categorical explanatory variable), the Analysis of Variance (ANOVA) shows that those who suffer from a MHD drank on average more drinks per year (Mean= 249.95, s.d. ± 489.32 ) compared to those without a MHD(Mean= 268.62, s.d. ± 670.82), F(1,26 634)=5.99, p=0.0144.

Next, I perform another ANOVA analysis to see if the proxy I use for the Alcohol Use Disorders (AUD) is plausible. More specifically, I have used the number of drinks per year as an “imperfect” proxy for the AUD. In this ANOVA I see whether those who have an AUD are more likely to consume more drinks per year or not. When I run this analysis in SAS, I get the following results:

These results show that there is a very strong association between the status of AUD and the number of drinks per year a person consumes. More specifically, the F-statistic is F(1,26 634) = 2 182.49 with p < 0.001. Also, on average a person who has an AUD consumes on average more drinks per year (Mean = 454.27, sd ± 297.99) versus a person who does not suffer from AUD (Mean = 143.61, sd 297.99).

Then, I also ran an ANOVA analysis to see if the mental health status (MH_status) where I have categories such as “healthy individuals” and “individuals with multiple disorders” is associated with the number of drinks per year. I obtain the following results:

These results show that there is a strong association between mental health status and AUD. This can be seen from the F statistic with is statistically significant. More specifically, this F-statistic equals F(7,26 628) = 3.38 with p = 0.0013.

As told in the lectures, while we can say there is an association between the mental health status of an individual, this test does not tell us for which MH_STATUS categories the means of the number of drinks per year differ. In other words, while we can see that there are some differences in between the means with the category 1 (“healthy people”) with the lowest mean and category 5 (those with a generalised anxiety disorder) with the highest mean, the F-statistic cannot tell us further between which categories of MH_status the mean of the number of drinks per year is different.

Before going to the next series of tests, let me show the commands I used in the data step (at the end of this blog I will post a link to the entire SAS code for this excercise:

Based on what is done in the lecture by Professor Dierker (please note that this is why I choose an explanatory variable with multiple categories to try this out), I will conduct Duncan post hoc (paired comparison) tests.

In order to run these tests in SAS, the previous code is adjusted as follows:

The results when running this code produce the following table:

The way this table needs to be interpreted is that for those categories where the letter is the same, the mean of the number of drinks per year is not different between these categories. When looking at this table, one can see that there is no difference between the means in the number of drinks per year for the categories of MH_STATUS 6, 3, 5 and 7 on the one hand and 0, 1, 3, 4, 5 and 7. So, where is the difference then? From this table, I can observe that category 6 (pathological gambling) is different from categories 0 (healthy individuals), 1 (depression), 2 (panic disorder) and 4 (social phobia).

24 4 / 2013

## Assignment 6: Preparing one quantitative variable for the ANOVA analysis

In this blog, I will perform an ANOVA analysis. As mentioned in the lectures, this is done when the dependent/response variable is quantitative, also called continuous and the explanatory variable is categorical. In my analysis, both the dependent and independent variables are categorical but since doing an ANOVA is part of the course requirement, I had to think a little bit what do. More specifically, I created a new explanatory variable where I look at the number of drinks consumed in year. In the class example, for the ANOVA the continuous variable is the number of cigarettes smoked per month. However, by the way the variable for the drinking frequency is constructed in the NESARC dataset it would make more sense to define the amount of drinker per year. More specifically, I have taken this directly from the NESARC code book:

Although one cannot be 100% sure that more drinks lead to alcohol abuse disorders, we can intuitively expect that there is a higher chance that if one drinks more alcohol one is more likely to develop an alcohol use disorder. Moreover, we can also test by using an ANOVA if this is indeed the case.

In what follows I will run 3 tests for ANOVA of which 2 are directly related to my research question and the third one is referring whether the amount of alcohol consumed would indeed be a good proxy for alcohol use disorders (AUD).

But before doing the ANOVA analysis, I had to do quite some data management in order to construct this variable of the number of drinks per year. I had to drop missing observations and do quite some recoding. More specifically, I first (similarly to the example covered in class but this was at the weekly level) converted the above variable, S2Q8A, to the number of days one drinks in a given year. Subsequently, I multiplied this variable with the a variable called S2AQ8B which stands for the number of drinks of any alcohol consumed on days when drank alcohol in the last 12 months. The above figure on the most left columns also gives the frequency distribution per category. Next, I had to take a call what to do with the variables that are coded ‘99’ and ‘BL’. It is clear that the data labelled as ‘99’ can be considered as missing as this variable means ‘unknown’ according to the codebook (see figure). For the value ‘BL’, this can be either NA (Not Available), fomer drinker or lifetime abstainer. As such, we could code this variable as a 0 but this variable also seems to contain ‘NA’ values so we really do not know well whether it is “NA” or an abstainer. Moreover, I am finding it hard to believe that 16147 individuals had no alcohol at all during the past 12 months!

Doing statistics is a lot about decision making. Hence, I decided to restrict the sample to those who had at least one day of drinking during the past few months and deleted all those observations with ‘99’ and ‘BL’ codes (or keeping the variables where S2Q8A is from the value 1 to 10). Doing this, left me with 26 741 observations. When looking at the second variable that I needed for constructing the number of drinks per year, the S2Q8A, which stands for the number of drinks of any alcohol usually consumed on days when drank alcohol in the last 12 months, I found the following summary statistics:

Well, that some people have 98 dranks per day does seem hard to believe for me. When I further looked at the frequency distribution of this variable, I found that 6 people had more than 30 drinks per day. I considered this all as outliers (although I still feel 30 drinks per day would be lots!). Cleaning up the data further and removing these outliers left me eventually with 26 636 observations.

Finally, I constructed the variable NUM_DRINKS_YEAR which is nothing but the product of the number of drinks per day in the past 12 monhts, S2AQ8B, and how often one person drank alchohol in the past 12 months expressed in the number of days. This variable is called ALCFREQYEAR and refers to the number of days a person has alcohol during the past 12 months and is constructed from S2AQ8A which is explained above.

The following SAS (only a part is shown and later in will give the entire link with the ANOVA code) shows how I recoded this variable S2Q8A into the number of days one person drank per year and what other data management steps I took to finally arrive at the NUM_DRINKS_YEAR variable.

As an example, when the drinking frequency has a code 6 (meaning S2AQ8A = 6) this according to the NESARC code book means that the individual drinks 2 to 3 times a month. Drinking 2 times a month means that he drinks 24 drinks per year (2 times 12) and 3 times a month means that he drinks 36 drinks per year. I took the average of these two numbers which is 30 and took that as the number of drinks per year that individual has.

Before turning to the ANOVA statistics, I think it is always useful to plot the data. That way we not only get an idea of what we the data show us visually but also let us have an expectation to some extent what the ANOVA statistics will tell us. I have gone back to the previous lecture where we look at the bivariate plots in case the response variable is a continuous/quantitative variable. More specifically, I obtained the following plot:

This plot shows that those who have a MHD have a slightly higher alcohol consumption expressed as the number of drinks per year than those who do not have a MHD. Next, I will look at what the Chi-squared statistic tell.

24 4 / 2013

## Assignment 5: Plotting the association between mental health disorder/state and alcohol use disorders

Here, I will present some bivariate plots. As a reminder, I am interested in the association between mental health disorders (MHD) and alcohol use disorders (AUD). Both the AUD and MHD are 0/1 variables. I have used the SAS bar chart wizard to draw these graphs.

This graph shows that those who suffer from a MHD (see the bar labelled 1) are on average showing higher AUD (one can see a higher bar). In the subsequent blogs, I will further explore this association using statistical measures such as the Chi-square statistic.

Next, I have also used at the bar plot where I have split up mental health in 8 categories. “Healthy” individuals and individuals having “multiple mental health disorders” are considered as separate categories.

Interestingly, this graph shows that pathological gambling is associated with a more alcohol use abuse. An explanation I can come up is that people who are addicted to betting are also more likely to drink during the betting which itself could develop into an alcohol use disorder (AUD). The next step will be to understand through a Chi-square test whether mental health status is associated with alcohol use disorders.

24 4 / 2013

## Assignment 4: Plotting of the variables of interest (univariate case)

In this assignment, I will plot the variables of interest. As a reminder, I am trying to answer the following three questions:

**Is there an association between mental health disorders (MHD) and Alcohol Use Disorders (AUD)?**Although there can be reverse causation, we assume that MHD is the independent variable (X-variable) and AUD the dependent variable (Y-variable).**Is mental health status which we labelled MH_status (including “healthy state” and “multiple disorders”) associated with AUD?**

First, I will give the “univariate” plots. In other words these are plots that do not look at any association between variables. I will do so for the following three variables: MHD, AUD and MH_status. All these plots were done with the “bar chart wizard” in SAS under the tab “tasks”.

Here is the bar plot for the Mental Health Disorders (MHD):

This bar chart shows that 23.17% of the people in the total dataset of 43 093 individuals are suffering from a mental health disorder.

The following bar chart gives a more detailed split-up of the mental health disorder status where “healthy” and “multiple disorders” are defined as one category (with a total of 8 categories):

From this graph you can see that after being “healthy” as the biggest category of mental health status, the second largest category is people suffering from a depression. More specifically, 11.79% of the total sample 43 093 people have suffered from one depression during their lifetime. Subsequently, the second largest is that of multiple health disorders (6.88%) followed by the category of generalised anxiety disorders (1.82%) and agoraphobia (1.58%). The other categories all are having a frequency of less than 1 %.

The last bar chart is referring to the Alcohol Use Disorders (AUD).

More specifically, you can see 27.48% of the 43 093 people in the NESARC sample are suffering from Alcohol Use Disorders (AUD).