CVD counts辅导、Homework 5. Confounding讲解

Homework 5. Confounding

Use information in the table below to answer questions 1-10.  Combine NonCVD and CVD counts to get counts of all deaths.  

Cause ofNumberPerson-Years

GenderHave to NapDeathPersonsof follow-up

MaleYesTotal300013640

DeadNon-CVD61

DeadCVD122

NoTotal500020040

DeadNon-CVD70

DeadCVD67

FemalesYesTotal340014646

DeadNon-CVD59

DeadCVD170

NoTotal510022400

DeadNon-CVD40

DeadCVD31

1. From the table above, what is the Crude incidence density ratios (combining males and females) for napping and death (non-cvd and Cvd combined)?    

A.   1.83

B.  2.97

C.  3.011

D. 3.465

Now we will work on getting the adjusted incidence density ratios from Proc Freq for death.      

data one;

input gender $ nap $ count py ;

datalines ;

Males   AYes  183  XXXX

Males  No137 20040

FemalesAYes XX14646

Females  No71 XXXX

;

Fill in all  the XX’s (no credit yet), and copy and paste into SAS.

Continuing this exercise, study, paste into SAS and run the following code.

data two; set one ;

dead ='AYes' ;

output;

dead ='No'  ;  

count = PY-count ;

output;

proc print;

run;

Note: If you’re wondering why the letter “A” is placed before “Yes”, this is one way to get the Yes-Yes Cell in the top left hand corner of the 2X2 table when we run Proc Freq since, “A” comes before “N” in the alphabet.

2. How many observations are in Data 2?

A. 4

B. 6

C. 8

D. 10

3. What does the following code do?

dead ='No'  ;  

count = PY-count ;

output;

A. For each gender-exposure group, it calculates the number of survivors and outputs the observation. (It creates the  “dead = no” column that will be used in proc freq).

B. It renames the count variable PY_count so that we can identify it.

C. It counts up all the people who died across genders and exposures and outputs it to the log.  

D. It tells SAS the person years are not statistically significant and should be thrown out.  

Now run the following code.  

proc freq order =data ;

tables gender*nap*dead/cmh ;

Weight count;

run;

4.   Look at the output.  What is the Logit estimator (also called Woolf’s method) for the incidence density ratio for death (non cvd and cvd combined) adjusting for gender?  

A. 0.787

B. 1.443

C. 2.192

D.  2.85

5.  What do questions 1 and 4 have to do with confounding?

A.  One provides the crude IDR while the other provides the adjusted IDR. The comparison between the two will tell us whether gender is a confounder.

B.  They don’t have a lot to do with confounding, because we can only assess confounding by examining whether gender is associated with both death and napping.

C. The crude is helpful, but the logit estimator has nothing to do with adjustment, which is done by modelling.

D.   The answers to 1 and 2 have nothing to do confounding;   We would need the gender specific death rates.    

6. Using the 10% rule, is gender a confounder of napping and death rate?  Hint: If the ratios are less than 1, remember to take the reciprocals of them before comparison.

A. Yes

B.  No

C.  Can’t tell because this exercise was designed to produce exactly a 10% difference

D. Can’t tell without modelling.  You would need to adjust for the variable in one model and not adjust for it in the second.  

7.   Concerning gender being a confounder or not being a confounder, which of the following is true?

A. In this study (note this is fake data!) women have a much higher rate of death and a much higher propensity to nap than men.

B.  In this study (note this is fake data!) women have a much higher rate of death, but the difference between men and women in the propensity to nap is slight.

C.   In this study (note this is fake data!) women have a much higher rate of death, but a much lower propensity to nap than men.

D. In this study, although women have a much higher propensity to nap than men, they have a similar death rate.  

8.   The formula for the Mantel-Haenzsel estimator to get an adjusted IDR  is:      

        ∑ [ai(PY2i)/(PY1i)]                                                                                                                            

        ∑ [bi(PY1i)/(PY2i)]      

where the letter i denotes the Table (aka strata: males or females) and the first subscript on the person years denotes whether or not the person is exposed (1 =exposed,  2 = unexposed)

In this example, if Table 1 is for males and Table 2 is for females what is PY22 ?

A. Person-years for males who had a death

B.  Person-years for females who are still alive.  

C.  Person-years for males who did not have to nap

D. person-years for females who did not have to nap  

9.  Regarding answers to questions 1 and 4, which of the following is true?  Hint, for the crude IDR, the variance of the log IDR =  1/a +1/c , where a and c are the number of events in the exposed and unexposed groups respectively.  You can look at the SAS output for the adjusted IDR.  

A. Neither the crude IDR or the adjusted IDR are highly significant.  

B. The crude IDR is highly significant but the adjusted is not.

C. The adjusted IDR is highly significant but the crude is borderline.

D. Both the crude and adjusted are highly significant.  

Feedback:  How does SAS know it’s an IDR and not an RR?

 It doesn’t.  But the formula for the IDR and RR are the same.  

SAS will use the formula     +       for the variance of the log RR, but because b≈py1 and d≈py2, this is extremely close to  1/a +1/c.    

For the 2x2 table set up as follows:

SAS does not care

Does not matterabN1 =a+b

cdN2=c+d

SAS calculates quantities as follows:

Odds ratio      ad/bc

Column 1 relative risk     (a/N1)/(c/N2)

Column 2 relative risk     (b/N1)/(d/N2)

10.  Based on the information above why is it recommended that the tables be situated in SAS Proc Freq as exposure*outcome rather than as outcome*exposure?  

A.  In order to provide an odds ratio, SAS requires those with exposure and disease to be in the upper left hand cell.

B.  As long as the exposure is the row variable, either the correct Risk ratio or its reciprocal will be given

C.  It’s just the convention outside of SAS; it makes no difference to SAS

D. Two answers are correct.

Base questions 11-18 on the article

Neal D. Freedman, Ph.D., Yikyung Park, Sc.D., Christian C. Abnet, Ph.D., Albert R. Hollenbeck, Ph.D., and Rashmi Sinha, Ph.D. .Association of Coffee Drinking with Total and Cause-Specific Mortality. N Engl J Med. 2012 May 17; 366(20): 1891–1904. doi:10.1056/NEJMoa1112010

11. What is the primary purpose of this research?

A. To assess association of coffee consumption with dietary and lifestyle factors

B. To assess association between dietary and lifestyle factors and all-cause mortality

C.  To assess association between coffee consumption and all-cause and cause-specific mortality

D. To determine whether coffee consumption is causally linked to lower risk of total and cause-specific mortality.

12.  What is the study design?

A. Case-cohort

B. Nested case-control

C. Prospective cohort

D. Historical cohort

13. From Table 1, which characteristic was NOT significantly associated with coffee drinking (did not differ significantly across levels of coffee drinking)?

A. Vigorous physical activity in women

B. Family history of cancer in men

C. Self-reported health status in men

D. It’s impossible to tell based on Table 1

14. In Table 2, find and interpret results related to drinking 6 or more cups of coffee and heart disease in men. Write the results and interpretation below (2-4 sentences).  (3 points)

15. In Table 3, the multivariate-adjusted hazard ratio for all causes of death listed under 4-5 cups of coffee is listed as 0.84 (0.79-0.90). Write a clear and accurate interpretation of these numbers. (3 points)

16. Select the correct statement about adjustment for confounders in this study:

A. The authors adjusted for a large number of confounders, including all possible variables that may confound the results, therefore the results could not be explained by confounding

B. Although the authors adjusted for a large number of confounders, some of these variables may have been poorly measured, resulting in residual confounding

C.  Although the authors adjusted for a large number of potential confounders, there may be other variables they did not measure that may have confounded the results

D. Both B & C are correct

17. In the discussion section, the authors mention the possibility of reverse causality. Which statement below is an example of reverse causality in this case:

A. 60 year old man who rarely drank coffee in his younger years, hears of a study that showed coffee drinking to be protective against heart disease, and starts drinking coffee on a daily basis

B. 50 year old woman with heart disease, advised by her physician to reduce her caffeine intake, so she decreased her coffee consumption from 3 cups to 1 cup per day  

C. 35 year old athlete increases his coffee consumption in order to boost his energy level

D. All of the above

Question 18. In 2-3 sentences, summarize the main findings of this study. (2 points)