Weeks 13 and 14 Guided Exercise: Bias and Misclassification

Introduction

Weeks 13 and 14 Guided Exercise seeks to demonstrate the nature of misclassification and how it influences a measure of association.

Part 1: Misclassification of Exposure

The data in Table 1 contains results in the source population (gold standard) from a case-control study of stroke and type A behavior. Presence of Type A behavior represents the exposure.

Table 1. Unbiased Data from Case-Control Study True Distribution
Type 1	Disease Status
Case	Control
Yes	56	42
No	33	81

Question 1: Name and estimate the unbiased relative measure of association from the true distribution shown in Table 1. Show your calculations.

Question 2:Suppose that the exposure status was misclassified with the sensitivities and specificities shown on Table 2.

Table 2. Sensitivity and Specificity
	Cases	Controls
Type 1 present - Exposure sensitivity	1.0	0.6
Type 1 absent - Unexposed specificity	0.9	0.7

Apply the corresponding sensitivity and specificity to the cells in Table 1 to determine the observed distribution.Fill each of the cells shown on Table 3.This is what the investigator observes as a result of exposure misclassification. Round cell numbers to the nearest integer (XX.X).

Report your results using this format:

1. Case/Type A present: XX.X
2. Control/Type A present: XX.X
3. Case/Type A absent: XX.X
4. Control/Type A absent: XX.X

Table 3. Biased Data from Case-Control Study Observed Distribution
Type 1	Disease Status
Case	Control
Yes	A	B
No	C	D

Question 3: Estimate the biased relative measure of association from the observed distribution using the data from Table 3. Show your calculations.

Question 4: Compare the unbiased and biased odds ratios. Comment on the difference between the odds ratio from the true distribution versus the odds ratio from the observed distribution. Is the observed distribution under or overestimating the measure of association? Why is that? Justify your answer.

Question 5: What type of misclassification is observed in this data - is it differential or non-differential disease misclassification? Why is that? Justify your answer.