HSCI 330 Lecture Notes - Lecture 1: Logistic Regression, Control Variable, Sampling Frame
16 May 2018
School
Department
Course
Professor
2-1 IMPORTANT METHODOLOGICAL ISSUES
Epidemiology: is primarily concerned with identifying the important factors or variables that influence a health
outcome of interest
Research Study: are persons who swim at Sydney beaches at increased risk for developing an acute
infectious illness?
Summary:
• The Sydney Beach Users Study is an example of the application of epidemiologic principles and methods to
investigate a localized public health issue.
• The key question in the Sydney Beach Users Study was:
o Does swimming at the beaches in Sydney, Australia (in 1989-90) pose an increased health risk for acute
infectious illnesses?
o The conclusion was yes, a 33% increased risk.
METHODOLOGICAL ISSUES
Issues to consider when planning an epidemiological research study:
Question
Define a question of interest and key variables
-include specifying key variables measured
Ex. What is the relationship of 1+ hypothesized determinants to a disease or health outcome of interest?
Variables
What to measure and how; (E) Exposure, (D) Disease, (C) Control variables
-exposure variable = determinant (E)
-disease = health outcome (D)
-variables known to predict health outcome = control variable (C)
Measuring variables (Data): surveys, interviews, samples, laboratory
Design
What study design and sampling frame? (select design and devise sampling plan)
-depends on feasibility and cost as well as variety of characteristics of pop. and study purpose
Frequency
Measures of disease frequency
-provides quantitative information about how often a health outcome occurs in subgroups of interest
Effect
Measures of effect
-allows for a comparison among subgroups
Bias
Flaws in study design, collection, or analysis (could lead to spurious conclusions about exposure-disease
relationship?)
-selection bias, information bias, confounding bias
Analysis
Perform appropriate analyses
-includes stratification and mathematical modeling as appropriate
-take into account other previously known risk factors for health outcome
-failing to do so can distort the results and lead to incorrect conclusions
Ex. Logistic regression, risk factors, confounding, effect modification
❖ What is the study question?
❖ How should the study variables be measured?
❖ How should the study be designed?
❖ What measures of disease frequency should be used?
❖ What kind of bias are likely?
❖ How do we analyze the study data?
THE STUDY QUESTION
The study question needs to be stated clearly and early, particularly to indicate the variables to be
observed/measured
❖ Describes the relationship between a health outcome variable (D) and exposure variable (E)
find more resources at oneclass.com
find more resources at oneclass.com
❖ Take into account the effects of other variables already known to predict the outcome (C)
❖ An important first step in an epidemiologic research study is to carefully state the key study question of
interest.
❖ The general question: To what extent is there an association between one or more exposure variables
(Es) and a health outcome (D), taking into account (i.e., controlling for) the possible influence of other
important covariates (Cs)?
❖ We can expect a complicated set of analyses to be required to deal with all possible relationships among
the variables involved.
VARIABLES
D
Health outcome variables (1)
E
Exposure variables (1)
C
Control variables (1+)
THE STUDY QUESTION
PREDICTORS
OUTCOME
E
→ D
C1
C2
C3
.
.
.
STUDY QUESTIONS (Q2.1)
In the Sydney Beach Users study, exposure was alternatively defined by distinguishing those who swam in polluted water from those
who swam in non-polluted water and from those who did not swim at all. Based on this scenario, fill in the missing information in
the following statement:
1. The exposure variable has ??? categories, one of which is ???
Choices: 2 3 4 5 did not swim polluted water swam water not polluted
2. When considering both swimming and pollution together, which of the following choices is appropriate for defining the exposure
variable in the Sydney Beach Users study: ???
Choices:
a) E=O if did not swim, E=1 if swam in polluted water
b) E=O if did not swim, E=1 if swam in non-polluted water
c) E=O if did not swim, E=1 if swam in polluted water, E=2 if swam in non-polluted water
d) E=O if did not swim, E=1 if swam
In the Sydney Beach Users study, the illness outcome was whether or not an acute infectious illness developed 1 week after
swimming at the beach. Also, in addition to age, another control variable was whether or not a study subject swam on days other
than the day he or she was interviewed. Fill in the missing information:
3. The health outcome has ??? categories.
4. There are at least ??? control variables.
5. Which of the following choices is not a control variable: ???
a) Age
b) Swimming status on other days
c) Swimming status on day of interview
Choices: 2 3 4 5 a b c
2-2 METHODOLOGICAL ISSUES (CONTINUED)
find more resources at oneclass.com
find more resources at oneclass.com
Document Summary
Epidemiology: is primarily concerned with identifying the important factors or variables that influence a health outcome of interest. Issues to consider when planning an epidemiological research study: Define a question of interest and key variables. What to measure and how; (e) exposure, (d) disease, (c) control variables. Variables known to predict health outcome = control variable (c) What study design and sampling frame? (select design and devise sampling plan) Depends on feasibility and cost as well as variety of characteristics of pop. and study purpose. Provides quantitative information about how often a health outcome occurs in subgroups of interest. Flaws in study design, collection, or analysis (could lead to spurious conclusions about exposure-disease relationship?) Take into account other previously known risk factors for health outcome. Failing to do so can distort the results and lead to incorrect conclusions. The study question needs to be stated clearly and early, particularly to indicate the variables to be observed/measured.
