Assignment 1: Logistic Regression

This assignment called for a logistic regression utilizing a dataset of our choosing. I chose the National Health Interview Survey(NHIS), which is a cross-sectional household interview survey conducted annually in the U.S. I was initially interested in looking at the effect of job insecurity on hypertension. The 2015 survey year is unique because it included a supplement on occupational health questions. Ultimately, I ran the regression looking at the effect of job insecurity on body mass index with age and sex as additional predictors. I’ve included relevant printouts in the captions in the event that the dataset is not easily accessible.

summary(tab2)

     ### Summary of categorical variables ### 

strata: Overall
                 var     n  miss p.miss  level  freq percent cum.percent
             JOB_INS 33672 14279   42.4    Yes  2147    11.1        11.1
                                            No 17246    88.9       100.0
                                                                        
 Hypertension_Status 33672    43    0.1    Yes 11745    34.9        34.9
                                            No 21884    65.1       100.0
                                                                        
                 Sex 33672     0    0.0   Male 15071    44.8        44.8
                                        Female 18601    55.2       100.0
                                                                        

The 2015 survey is unique because it included an occupational health supplement. Variables, such as job insecurity, were included and could be tested against health outcomes. Job insecurity is captured by work stress in occupational health, but it is often left out of studies on social determinants of health. For this survey year, there were 33,672 sample adults. Sex was evenly distributed in the sample (women: 55%). Age was normally distributed with the majority of respondents between 25 and 64 years of age (67%). Approximately 1/3 of participants reported hypertension. Job insecurity, our exposure of interest, was only reported by 11.1%. However, this may have been expected if we assume that job insecurity is linked to level of education. This is a highly educated sample - only 12% of the sample reported less than a HS diploma. Future studies may want to consider a more robust work stress variable that includes job insecurity, as well as harassment and low task control.

Figure 1: Job Insecurity and Hypertension among NHIS Participants by Sex Group (2015). The bar plot is organized by sex according to reported job insecurity and stacked by hypertension status. Men and women who report job insecurity seem to have higher proportions of hypertension than those who are not job insecure. This data is from 2015. It would be interesting to repeat this analysis with data from 2020, given the high levels of job insecurity and the high unemployment rate (assuming NHIS runs the occupational health supplement again). (Author’s Note: I was going to attempt to eliminate my “7” and “9” responses, but I am always curious about health outcomes among people who opt out of a question. Here, hypertension is higher for both men and women who did not respond to the question about job insecurity.)

#run a logistic regression
glm(formula= Hypertension_Status ~ JOB_INS + SEX, family = "binomial", data=alldata)

Call:  glm(formula = Hypertension_Status ~ JOB_INS + SEX, family = "binomial", 
    data = alldata)

Coefficients:
(Intercept)    JOB_INSNo          SEX  
     0.5718       0.3040       0.1997  

Degrees of Freedom: 19372 Total (i.e. Null);  19370 Residual
  (14299 observations deleted due to missingness)
Null Deviance:      21480 
Residual Deviance: 21410    AIC: 21410

fit1 <- glm (Hypertension_Status ~ JOB_INS + SEX, 
             data = alldata, 
             family = binomial("logit"))
bcoef <- round(coef(fit1), 2)
bcoef

(Intercept)   JOB_INSNo         SEX 
       0.57        0.30        0.20 

#test for interaction terms
fit1.int <-glm (Hypertension_Status ~ JOB_INS + SEX * AGE_P,
                data=alldata, family=binomial ("logit"))
anova(fit1, fit1.int, test="LRT")

Analysis of Deviance Table

Model 1: Hypertension_Status ~ JOB_INS + SEX
Model 2: Hypertension_Status ~ JOB_INS + SEX * AGE_P
  Resid. Df Resid. Dev Df Deviance  Pr(>Chi)    
1     19370      21406                          
2     19368      19064  2   2342.8 < 2.2e-16 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Logistic Regression: The GLM and fit1 code above produces the equation for a line where hypertension is an outcome variable with job insecurity and sex as predictor variables. The results are the same - it is just two ways to run it. The intercept is 0.5718. The beta coefficients are 0.3040 for job insecurity, and 0.1997 for men. These coefficients could be utilized to project someone’s hypertension given their values for the predictors. I was curious about the relationship between job insecurity and hypertension given the high levels of unemployment related to COVID-19. I would like to utilize this code with future datasets. The interaction model and ANOVA highlighted that age is an effect modifier - results were statistically significant at p<0.001 (prevalence chi square: <2.2e-16). Additional models might include SEP indices and cardiovascular health indicators, such as BMI, smoking status, and leisure time physical activity level. A future analysis might also present a more robust work stress variable, including harassment and task control. ```