Data version: Controlled Tier All of Us Curated Data Repository (CDR) v7 (2022Q4R11)

Analysis tool: SAS Studio

Authors:

Emily Goldmann, Ph.D., MPH 
New York University School of Global Public Health

Stephanie Cook, Dr.Ph., MPH 
New York University School of Global Public Health

Please note: This guide aims to demonstrate the utilization of PROC FREQ and PROC LOGISTIC procedures using SAS Studio. However, it is important to note that this guide is not comprehensive and does not encompass all facets of the scientific process which researchers are required to undertake nor does it assume that this is the only way to correctly perform these statistical procedures. Specifically, it does not delve into data cleaning and verification, assumption validation, model diagnostics, potential follow-up analyses, or any other possible approaches for performing these frequency and binary logistic regression procedures.

This analytics guide is available as a downloadable PDF.

Table of Contents

• Introduction
• Description of the dataset
• Statistical analysis procedures
• Additional resources

Introduction

This guide includes examples of statistical analysis processes which were used to assess concordance between self-reported lifetime depression diagnosis and depressive disorder diagnoses documented in available electronic health records (EHR) using survey and EHR data from the All of Us dataset.

More specifically, these examples demonstrate the prevalence and demographic correlates of not self-reporting a lifetime depression diagnosis among adult (18 years or older) respondents with evidence of depressive disorder in EHR.

The study described in this guide is descriptive in nature and provides overall prevalence of not self-reporting depression among those with EHR depressive disorder and demographic patterns within this prevalence. To describe these categorical variables, frequencies and percentages are reported.

To evaluate bivariable associations between self-reporting depression and demographic factors, a Pearson’s chi-square test is generally performed. Finally, one can employ a binary logistic regression model including all demographic variables as independent variables to identify those factors that are independently associated with the outcome, i.e., not self-reporting a lifetime depression diagnosis.

This type of regression is commonly used for binary (yes/no) dependent variables and yields odds ratios (OR) as a relative measure of association between the dependent variable and each independent variable, controlling for all other independent variables in the model. 95% confidence intervals (CI) also accompany each OR as an indication of estimated precision.

To begin our discussion, it is important to outline some definitions of terms frequent used throughout this guide:

• Outcome or independent variable: A variable of interest for which we want to understand overall prevalence (or proportion) in a population and might vary across demographic subgroups (dependent variables).
• Dependent variable:  A variable of interest for which we want to understand its association with the outcome variable. In descriptive studies, these variables are often referred to as correlates.

Description of the dataset

For this example, we identified two cohorts for analysis:

1. Case group: Respondents who had an electronic health record (EHR) depression code but did not self-report depression

   • Adults (Participants aged 18 years or older)

   • Did not report that they had a lifetime depression diagnosis in the Personal and Family Health History survey

   • Had at least one EHR diagnostic code for depressive disorder (SNOMED code: 35489007, depressive disorder)

2. Control group: Respondents who had an EHR depression code and self-reported depression
   • Adults (Participants aged 18 years or older)
   • Reported that they had a lifetime depression diagnosis in the Personal and Family Health History survey
   • Had at least one EHR diagnostic code for depressive disorder (SNOMED code: 35489007, depressive disorder)

Demographic correlates / dependent variables included:

• Date_of_birth, to calculate age in years at the time of data analysis (1=18-44, 2=45-64, 3=65-84, 4=85 or older)
  • Recoded variable name: age_cat
• Race and Ethnicity, to generate a variable that combines race and ethnicity (1=Non-Hispanic white, 2=Non-Hispanic Black, 3=Hispanic any race, 4=Non-Hispanic Asian, Native Hawaiian or other Pacific Islander, 5=Non-Hispanic multiple or other race)
  • Recoded variable name: race_eth
• Gender (1=Female, 2=Male, 3=Other) 
  • Recoded variable name: gender_new
• Sexual Orientation (1=Straight, 2=Gay or Lesbian, 3=Bisexual, 4=Other)
  • Recoded variable name: sex_orient
• Highest Education Level (1=Less than high school, 2=High school graduate/GED, 3=Some college, 4=College graduate, 5=Advanced degree)
  • Recoded variable name: edu_cat

Other variables:

• Person_id as a unique participant identifier, used to merge and deduplicate datasets

Statistical analysis procedures

Step 1: Describe all variables using frequencies and percentages among adults with an electronic health record (EHR) depression diagnosis (n=30,260).

Example code:

proc freq data=mydata.depress_final2;
tables self_report age_cat gender_new race_eth sex_orient edu_cat;
run; * 30.94% have EHR Depression Dx but did not self-report depression Dx;

Example results:

Step 2: Examine associations between the outcome (not self-reporting depression diagnosis) and each demographic variable among adults with an EHR depression diagnosis (n=30,260) using cross tabulation and Pearson’s Chi-square test.

Example code:

proc freqdata=mydata.depress_final2;
tables (age_cat gender_new race_eth sex_orient edu_cat)*self_report /chisq nopercent nocol;
run;

Example results:

These results suggest that among adults 18 years or older with EHR depression diagnosis, non-Hispanic Black (38.17%), Hispanic any race (37.14%), and non-Hispanic Asian (34.46) respondents had a higher prevalence of not self-reporting depression diagnosis compared to non-Hispanic white respondents (29.06%, p<0.0001).

Another example evaluates the association between the outcome and the intersection of two demographic factors, e.g., association between the outcome and race/ethnicity by gender.

Example code:

proc freq data=mydata.depress_final2;
tables gender_new*race_eth*self_report / chisq nopercent nocol;
run;

Example results:

For women only (gender_new=1)

For men only (gender_new=2)

This yielded a similar pattern in the prevalence of the outcome by race/ethnicity among women and men.

Step 3: Identify demographic variables independently associated with not self-reporting depression among adults with an EHR depression diagnosis.

Example code:

proc logistic data=mydata.depress_final2;
class age_cat (ref=“1”) gender_new (ref=“1”) race_eth (ref=“1”) sex_orient (ref=“1”) edu_cat (ref=“5”);
model self_report = age_cat gender_new race_eth sex_orient edu_cat;
run;

Example results:

All of the odds ratios (OR) highlighted above show an estimate of the odds of the outcome (not self-reporting depression diagnosis) in that group that is significantly different from the odds of the outcome in the reference group (OR>1 if the odds are higher, OR<1 if the odds are lower). 95% confidence intervals (CI) that do not include the value of 1.000 indicate statistically significant associations at p<0.05. 

For example, we see that compared to non-Hispanic white respondents with an electronic health record (EHR) depression diagnosis (race_eth=1, the reference group), respondents in most other race/ethnicity groups had significantly higher odds of not self-reporting depression diagnosis (e.g., compared to non-Hispanic white respondents with EHR depression diagnosis, non-Hispanic Black (race_eth=2) respondents with EHR depression diagnosis had approximately 63% higher odds of not self-reporting depression diagnosis; OR=1.628, 95% CI: 1.493-1.774).

This suggests that non-Hispanic Black respondents with an EHR depression diagnosis are more likely than non-Hispanic white respondents with an EHR depression diagnosis to not self-report lifetime depression diagnosis, despite EHR documentation of depression diagnosis.

Additional resources

For additional information about using SAS Studio in the Researcher Workbench, explore the following articles: Exploring All of Us data using SAS Studio and How to run SAS in the Researcher Workbench.

You can also download this analytics guide as a PDF.

• SAS Analytics Guide_How to perform binary logistic regression.pdf600 KB