**KEY Chapter 11 Exercises KEY**

11.1 TB-RATES [Click here for analyses in Excel spreadsheet; data as of Apr. 2002]

11.2 Injuries in farm worker

(A) What are the explanatory variable(s) in this analysis? race {Caucasian / Af. Am} and ownership (owner / worker). What is the response variable? Injury
(B) Explain why experimentation is impossible when addressing these issue.  The factors of race and ownership are not assignable.
(C) Explain why this is an observational cohort design. The is exposure is not assigned. All individuals in the cohort are followed to ascertain the outcome. 
(D) Are data prospective or retrospective? prospective
(E) Determine what measures of association can be estimated from these data. You could use a rate difference (RD) or rate ratio (RR).
(F) Calculate the rate ratios of injury associated with groups 1 and 2 (i.e., let the Caucasian Owners serve as the non-exposed group). See below. Is there a risk or benefit associated with race? Neither. Is there a risk or benefit associated with being a non-owner (worker)? Risk  

Group	Cases	Pers-years	Rate (per 1000)	RR
Caucasian Owners (0)	67	2047	32.7	reference
Af-Am. Owners (1)	27	821	32.9	1.01
Af-Am. Workers  (2)	37	359	103.1	3.15

11.3  Case-control study of influenza vaccination and primary cardiac arrest. 

(A) This is a case-control study because subjects were selected from the population based on their disease status.
(B)  OR = (79 × 373) / (236  × 176) = 0.71. This indicates a negative association between PCA and heart attacks, specifically, a 29% decrease in risk.
(C) The spouses of controls were interviewed to make exposure information comparable to that of cases. The assumption is that data will be equally inaccurate in cases and controls (see comments about non-differential misclassification in Chapter 12).

11.4 Case-control study of pancreatic cancer and meat consumption. OR = (53  × 85) / (53  × 43) = 1.98 = 2.0. Eating fried or grilled meat doubled the risk of pancreatic cancer. (Equivalently: a 100% increase in risk.)

11.5  Doll & Hill, 1950.  OR = (647 × 27) / (622 × 2) = 14.04. This indicates a strong positive association: the risk of lung cancer is 14 times higher in smokers.

11.6  IUDs and infertility. OR = (89 × 3193) / (640 × 194) = 2.29. IUD users 2.29 times the risk as nonusers (equivalently: 129% greater risk).

11.7 OC studies from the 70s.

(A) The first study is a case control-control study because subjects were selected based on their disease status. All cases in the source population were studied while only a sample of non-cases (controls) were studied. Exposure of various factors were compared in the cases and controls.

(B) The first hint that this is a cohort study occurs in sentence 1. The study is described as prospective. Case-control studies can never be fully prospective. Cohort studies my be prospective or retrospective. (At the time this abstract was written, the terms prospective and cohort were used more or less interchangeably--a fact we now recognize as an error.)  In addition,  the abstract references a rate difference. Cohort studies can determine rate differences, but case-control studies cannot.

(C) Controls may not be typical of non-cases in the source population. They may be "sicker" than is typical, and may suffer from multiple chronic conditions (Berkson's bias).

(D) A 40% increase corresponds to RR = 1.4. 

(E) This represents are rate difference.

(F) 
1) Smaller total sample size - case-control
2) Shorter time to complete - case-control (since it used retrospective data)
3) Case control studies cannot estimate incidence directly.
4) Less prone to recall bias - Royal College (cohort) study, since it was prospective
5) Less prone to selection bias - Royal College (cohort) study, since it is tricky to select controls in a case-control study
6) Less prone to loss-to-follow-up - Mann (case-control)
7) Case-control studies are generally done before cohort studies since they are more efficient and generally take less time to complete.