10/3/2022

OBSERVATIONAL STUDIES

Week 5: Overview and Case-Control Studies

PPB 485: Drug Literature Evaluation

October 4, 2022
Mary Amato, PharmD, MPH, BCPS, FCCP

Objectives
• Describe how observational studies differ from clinical trials and
descriptive literature
• Explain how observational studies are used in research
• Discuss the general limitations of observational studies including
types of bias
• Explain what confounding is and how it is controlled for in studies in
both the methods and statistical analysis

Strength of Literature/Level of Evidence

Meta-
analysis

Systematic Review

Randomized Controlled
Trials (RCTs)
Observational
Studies Cohort Studies

Case-Control Studies

Case Reports/Series

Expert Opinion

1
 10/3/2022

Descriptive vs Analytical Reports/Studies (review)

• Descriptive
-Generate preliminary hypothesis of exposure/disease relationship
-Examples:
Surveys, Case Reports/Series, Descriptive Cross-sectional (prevalence ) studies

• Analytical studies
- Test hypotheses, look for relationships
- Examples:
• Observational Studies- Cohort, Case-control
• Experimental- Randomized Controlled Trial (RCT)

Observational Studies

• Analytical studies in which the investigators observe

and collect data but DO NOT INTERVENE in any way

• Includes both cohort and case-control studies

• Like RCTs, they attempt to answer a clinical question (such as identifying factors
associated with increased or decreased risk of a certain clinical outcome)
• They are not descriptive

• Unlike RCTs, observational studies do not involve an intervention

(randomization to a treatment group) by the investigators, only observing what
occurs or has occurred naturally in the real world and recording outcomes.

How Observational Studies Are Used to Evaluate Drug Therapy

• To investigate effectiveness of treatment
• Benefits of a treatment in everyday practice (real world)
• (RCTs are more controlled and establish efficacy (does drug work) under
“ideal” conditions (not real world))
• To investigate harm/risk of treatment
• (RCTs are not designed to assess harm)
• Ethical issues
• Limited by time and sample size
• To investigate prognosis and long-term or infrequent outcomes
• (RCTs are limited by time and sample size)

2
 10/3/2022

Some Limitations of Observational Studies

• Potential for bias in:
- selection of study participants
- recording exposure or outcomes
• Potential for confounding variables to affect outcomes
• Key point:
- Due to the above limitations, a cause- effect relationship cannot be established
in observational studies; only an association

Selection bias in Observational studies

• Selection bias may result when a study fails to select a representative sample from
the population of interest, which could lead to a result that is not true.
• To evaluate the risk of outcome in a treatment group vs control group, those in both
groups should have equal risk of getting the treatment and equal risk of the outcome at
baseline
• In the real world, patients don’t all have equal chance of getting a drug. There are
factors that affect decisions on whether to give a treatment to individual patients (age,
their other diseases and meds etc).
• A group of study patients may also not have an equal chance of getting the outcome as
those you wish to apply the results. (may be healthier, sicker, younger, etc)
• (RCTs control for this type bias by randomization of patients in a study to treatment
vs control, which can’t be done in observational studies)

Selection bias - Examples

Example 1: Target population: All patients in the general population with type 2 diabetes.
Investigators conducted an observational study to determine the risk of developing renal
insufficiency in those who were treated with Drug X (treatment group) compared to those who
did not receive drug X (control group)

Treatment group: All those with diabetes who were admitted to a medicine inpatient unit at a
hospital over a 1-year period and were treated with Drug X
Control group: Nurses with diabetes who worked full time at the same hospital during the same
time period who did not receive Drug X during the same time period

How might this affect the outcome you are measuring? Would you expect those who received
the treatment with Drug X to be healthier or less healthy at baseline than the control group?
Does either group represent the general population of those with diabetes?

3
 10/3/2022

Selection bias - Examples

• Example 2: Study looking at the risk of side effects with a new vs an older drug

• New Drug- providers used in more healthy patients due to their lack of experience
with the new drug
• Old drug- providers more comfortable using in sicker patients because they were more
familiar with the drug

How would this bias the outcome? More or less side effects with the new drug?

10

Bias in recording exposure or outcomes

• Unequal measurement of outcomes in two study groups could occur if:
- One group was seen more often than the other
- The provider treating the patient was more likely to
record outcomes in one group vs the other
- One group was better able to recall or report exposures or
outcomes than the other (those with birth defects may
better recall taking a drug that could cause birth defects)
• (RCTs control for this type of bias by blinding investigators to treatment group
assignments so that all variables are measured in the same way, which can’t be done in
observational studies)

11

Example: Bias in Measuring and Recording Outcomes

• Study comparing patients who received warfarin to those who had received a
direct oral anticoagulant (DOAC); Outcome = bleeding side effects
• Warfarin group- was seen every month to get monitoring blood tests; had
outpatient visit note in chart that recorded lab results and any patient’s response
to question about bleeding side effects every month.
• Other anticoagulant group- seen once per year, note recorded in chart if patient
called to report bleeding side effects during the year.

12

4
 10/3/2022

Confounding
• A causal relationship between two variables is often described as the way in
which the independent variable affects the dependent variable. The independent
variable can take different values independently, and the dependent variable
varies according to the value of the independent variable.
• Confounding is the distortion of the association between the independent and
dependent variables because a third variable is Independently associated with
both.
• If the third variable is not present equally among the two groups, it could be
that variable, not the one being studied that is affecting the outcome/dependent
variable
https://s4be.cochrane.org/blog/2018/10/01/a-beginners-guide-to-confounding/

13

Confounding
• Let’s say you want to find out how alcohol consumption affects mortality
• You want to compare the mortality rates between two groups –
Group 1 = heavy users of alcohol
Group 2 = those who drink little or no alcohol

alcohol consumption = exposure (independent variable)

mortality = outcome (dependent variable)

https://s4be.cochrane.org/blog/2018/10/01/a-beginners-guide-to-confounding/

14

Confounding
• If you find that Group 1 (heavy alcohol users) are more likely to die you may
conclude that alcohol use increases the risk of death.
• However, it is possible that alcohol use is not the only mortality-affecting
factor that differs between the two groups.
• People who consume less alcohol might be more likely to eat a healthier diet
or less likely to smoke, for example.
• Eating a healthy diet or smoking might affect mortality.

https://s4be.cochrane.org/blog/2018/10/01/a-beginners-guide-to-confounding/

15

5
 10/3/2022

These other influencing factors are called confounding variables. If you ignore them and assume that any differences in
mortality must be caused by a difference in alcohol consumption, you could end up with results that don’t reflect reality.
You might find associations where in reality there are none or fail to find associations where they do in fact exist.

https://s4be.cochrane.org/blog/2018/10/01/a-beginners-guide-to-confounding/

16

Confounding- A Major Limitation of Observational Studies

• Confounding variable: patient characteristic that may be associated with some outcome
(e.g., disease) independent of the exposure itself

• Examples: age, sex, weight, diet, medical history, social history (e.g., smoking, ETOH
use), other medications

• Since there is no randomization of subjects to treatment groups, characteristics of

exposed group may differ from unexposed group, leading to confounding and masking
true effect of exposure

• Cannot rule out possibility that it is some confounder, and not the exposure itself, that
resulted in an outcome. Confounding can distort the “truth!”

17

When reading/evaluating a study,

How do we look for potential confounders?
• Look at results (often available in Table form) in a study that list the
characteristics of two groups with several variables (example: age, gender,
smoking history comorbid diseases, concurrent medications)
• Look to see if the two groups differ in the frequency of these variables, if no,
then not a confounder, if yes, then could be a potential confounder if associated
with the variables being studied
• We will look at this in some study examples

18

6
 10/3/2022

Three Characteristics of a Confounder

Confounder

Exposure Outcome

To be considered a confounder, a variable:

1. Must be associated with both risk factor (exposure) of interest and outcome being studied
2. Must be distributed unequally among groups being studied (if it was equal among the two groups it would
not sway the results one way or the other)
3. Must NOT be a result of exposure (not part of the causal pathway, not an intermediate on the way to
outcome (ex diet and coronary ht disease; cholesterol level is not a confounder, diet can lower or raise
cholesterol)

19

Study: Statins and Coronary Heart Disease

Blood pressure

Statin (Lipitor) Coronary Heart Disease

1. High Blood pressure is a known risk factor for CHD

2. High BP more common in those taking statin than not taking statin
3. High BP is not a direct result of taking statins

An observational study could give the false impression that taking statins is associated with increased
CHD (or have no reduction in CHD) due to the confounder BP if more subjects had high BP the statin
group compared to the non statin taking group.

20

Study: Moving to Florida and Decreased Memory

Older Age

Those who moved to Decreased short term

Florida memory

1. Age is a known risk factor for decreased short term memory

2. Moving to Florida (or warm climate) is common after retirement
3. Older age is not caused by moving to Florida

An observational study could give the false impression that moving to Florida is associated with short
term memory loss if those who moved to Florida included more who were older than those who did not
move to Florida.

21

7
 10/3/2022

Study: Alendronate and Fractures in Elderly females

Drive a car that is

blue

Alendronate Fractures

1. Color of a car one drives is not associated with taking alendronate or getting fractures
2. Color of car one drives would not be predicted to be more common in those prescribed alendronate
than those not prescribed alendronate.
3. Alendronate does not affect the color of car one drives, and the car color does not affect fracture rate
(not along the causal pathway)

22

Study: Alendronate and Fractures in Elderly females

Change in BMD
(bone marrow
density)

Alendronate Fractures

1. Low BMD is a risk factor for Fractures

2. A greater change in BMD is more common in those prescribed alendronate than those not prescribed
alendronate
3. BMD is increased by alendronate, part of the causal pathway, NOT considered a confounder
here but a result of the medication treatment

Another question: In this study, could gender be a cofounder? Why or why not?

23

Controlling for Confounding in Study Design

X Randomization
• Not possible in Observational studies- no intervention by investigators so
can’t randomly assign patients to treatment group; (will see this in
Interventional studies such as RCTs covered in later lectures)
✓Restriction
• Exclude subjects with a certain confounding characteristic from study (e.g.,
smokers, ETOH users, people with certain medical conditions)
• May limit generalizability of study (external validity)
• Can use in both observational and experimental studies (RCTs)
✓Matching- Next slide

24

8
 10/3/2022

Controlling for Confounding in Study Design

• Matching
• Match subjects by certain characteristics (e.g., age, sex) between study groups
• Commonly used in case-control study design
• Match by propensity score (PS)
• PS = probability of being in the exposed (or case) group conditional on
observed baseline characteristics (covariates/confounders)
• Summarizes many confounders into a single score (from 0-1)
• PS estimated using regression techniques
• Used mainly in cohort study designs but also case-control studies

25

https://pubmed.ncbi.nlm.nih.gov/23613075/

26

Study Example- Finding information in studies:

Example/review of sections of an observational study paper
• Background
• Study methods
• Results
• Discussion
• Conclusion

London MJ, Her K, Schwartz GG, Henderson WG. Association of perioperative β- blockade with mortality and cardiovascular
morbidity following major noncardiac surgery. JAMA. 2013; 309(16:):1704-1713.

https://pubmed.ncbi.nlm.nih.gov/23613075/

27

9
 10/3/2022

Question
What is the exposure being studied?
What is the outcome being studied?
Are there any potential confounding variables/confounders that could affect the
results of this study? (see Table 1)

https://pubmed.ncbi.nlm.nih.gov/23613075/

28

Example of Matching by Propensity Score in Cohort Study

Study: Association of perioperative β-blockade with mortality and CV

morbidity following major noncardiac surgery. JAMA 2013;309:1704-13

In full cohort group there are significant differences in baseline characteristics

between exposed (to β blockers) and unexposed (no β blockers) groups.
(note P values!) Exposed group has more risk factors.

29

Matched Cohort Based on Propensity Score

Matching by propensity score makes groups more comparable. What do

you notice about P values now?
Notice that not everyone in original cohort could be matched, so sample
size of matched cohort is smaller.
Limitation of PS matching: does not address bias from unmeasured
confounders.

30

10
 10/3/2022

Postoperative Outcomes
Full Cohort Matched Cohort
(% of patients) (% of patients)

Outcome All Exposed Not P value All Exposed Not P Value

Variables N=136,745 N=55,138 Exposed N= 75,610 N= 37,805 Exposed
N=81,607 N= 37,805

30 Day 1.1 1.4 1.0 <.001 1.3 1.1 1.5 <..001

Mortality

Cardiac 0.6 0.8 0.4 <.001 0.7 0.6 0.7 .01

Arrest

Q-wave MI 0.3 0.5 0.3 <.001 0.4 0.3 0.5 <.001

Cardiac 0.9 1.2 0.7 <.001 1.0 0.8 1.2 <.001

Complication

CVA 0.3 0.4 0.2 <.001 0.4 0.4 0.3 .45

Notice that in full cohort, the % of patients experiencing an event in exposed group is greater than
in unexposed group. However after PS matching, perioperative beta-blockade is shown to be protective
for most events. This suggests that the unmatched results may be due to confounding (e.g., subjects
exposed to beta-blockers had more risk factors than unexposed group).

31

Controlling for Confounding in Statistical Analysis

• Stratification
• Regression analysis

32

Controlling for Confounding in Statistical Analysis: Stratification

• Stratify subjects based on confounder

• For example: group together by smoking history, gender, age
• Creates homogenous groups (e.g., people with same smoking history, gender
or age group compared together)
• May result in strata with very small numbers
• Statistical test (Mantel-Haenszel) used to provide an “adjusted result”
(weighted average) according to strata which is then compared to crude
(unadjusted) result. If results differ, confounding is likely.

33

11
 10/3/2022

Example: Data Not Stratified

Scenario: Case-control study investigating association between alcohol abuse and
lung cancer in 300 subjects (150 cases and 150 controls)

Alcohol Abuse Cases (Lung Cancer) Controls

Yes 88 68

No 62 82

OR of lung cancer associated with alcohol abuse:

88/62 = 1.70 (95% CI: 1.08-2.70)

68/82

Data suggests that alcohol abuse is significantly associated with an increased risk of
lung cancer. What about smoking as a potential confounder? It is associated
with both alcohol use and lung cancer, and is not in causal pathway between exposure
and lung cancer (lung cancer doesn’t cause one to smoke).

34

Stratifying Data by Smoking Status

Non-smokers (n=81) Smokers (n= 219)
Lung cancer No lung cancer Total Lung cancer No lung cancer Total

Alcohol abuse

Yes 53 15 68 35 53 88

No 10 3 13 52 79 131
Total 63 18 81 87 132 219
Stratum specific OR =1.06 (95% CI: 0.26-4.35) OR= 1.01 (95% CI: 0.58-1.74)

After stratifying by smoking, there is no significant increase in lung cancer with

alcohol abuse in either smokers or non-smokers. The link between alcohol abuse
and lung cancer found in unstratified analysis was likely due to the confounding
effect of smoking.
Note: using M-H formula, the overall OR of lung cancer associated with alcohol
abuse after adjusting for smoking = 1.01 (95% CI: 0.60-1.69). No increased risk!

35

Controlling for Confounding in Statistical Analysis: Regression

Analysis
• Used when there are many potential confounders
• Adjusts simultaneously for several confounders
• Confounders treated as predictor variables
• Variable that is used to predict some outcome
• Naturally occurring variables and not manipulated by investigator (unlike an
independent variable)
• Results then “adjusted” based on confounders
• Controls for the effect of confounders
• Crude and “adjusted” results compared to see if confounding may exist
• A large difference suggests confounding exists

36

12
 10/3/2022

Using Regression Analysis to Assess For Confounding

Hypothetical results of a case-control study reporting the odds
ratio of hip fracture associated with Drug X. After matching
for age and gender, OR was adjusted for history of
osteoporosis, dietary intake of calcium and body mass index.

Crude OR (95% CI) Adjusted OR (95% CI)

3.0 (1.50-4.50) 2.0 (0.50-3.50)

Note: When multiple regression was used to adjust for

confounders, the risk is no longer significant. This indicates
that confounding was present!

37

Summary: Types of Bias in Studies of Drug Therapy and How to Address

Type of Bias Observational Studies Experimental studies (RCTs)
Bias in selection of study Randomization of subjects to
subjects treatment groups
Bias in recording Blinding of subjects and
outcomes investigators to which treatment
group subjects are in
All measurements the same in both
groups
Bias due to Confounding Study Design: Study Design:
variables - Restriction (exclude all pts with a certain - Restriction
potential confounding variable)
- Matching (by a certain characteristic or by
propensity score)
Analysis:
Stratification
Regression analysis

38

Huang et al; Pneumonia and Antipsychotics study- (link below)

Question:
Which of the following was done in this study to control for confounding?

1. Controlling for confounding in study design:

- Restriction
- Matching
2. Controlling for confounding in statistical analysis
- Stratification
- Regression Analysis

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8436093/pdf/ijmsv18p3565.pdf
https://pubmed.ncbi.nlm.nih.gov/34522183/

39

13
 10/3/2022

STROBE Statement
• Developed by an international group of experts (epidemiologists, researchers,
statisticians who conduct observational research
• Aimed at strengthening the reporting of observational studies in epidemiology
• Includes a checklist of items that should be included in reports of observational
studies
• Available at: https://www.equator-network.org/reporting-guidelines/strobe/

40

Questions to ask when Evaluating Observational Studies

• Were the subjects in the study treatment and control groups comparable at
baseline? (did selection bias exist?)
• Were confounding variables obviously present?
• Was the information on exposure and outcomes collected consistently in both
groups? (was there bias in recording exposure or outcomes?)
• How good is the quality/completeness of the collected data?
• Was study design modified (restriction or matching) or statistical analysis done
(stratification or regression analysis) to correct for potential confounders?

41

Summary
• Observational studies are useful in assessing real world
effectiveness and harm/risk of treatment
• Observational studies are often the best approach when long
periods of observation are needed, for rare exposures or
outcomes, or when clinical trials would be unethical
• Because they are not experimental (like RCTs), only associations
and not cause-effect relationships can be determined
• Understanding how to critically evaluate cohort and case-control
studies is important to utilizing them for making patient care
decisions- both types of studies will be outlined in detail in the
next few lectures

42

14
 10/3/2022

OBSERVATIONAL STUDIES (CONTINUED)

Case-Control Studies

PPB 485: Drug Literature Evaluation

October 4, 2022
Mary Amato, PharmD, MPH, BCPS, FCCP

43

Review questions- Observational studies:

• If observational studies have so much bias, why do them instead of RCTs?
- Real world data
- Cost (RCTs more expensive)
- Time (RCTs may take longer)
- Ethics (may not be ethical to randomize subjects to dangerous or ineffective
treatment)
• Where do I look to find out what type of study was done and how it controlled for
bias?
- Methods- tells step by step how the study was done (type of study, use of
restriction, matching, statistical methods to control for bias)
- Statistical tests used listed at end of methods

44

Objectives – Case Control Studies

• Describe how case control studies are designed and used in
research
• Demonstrate ability to critically evaluate case control studies

45

15
 10/3/2022

Case-Control Study
• Begins AFTER individuals have developed the outcome of interest
• Therefore, these studies are retrospective
• Two groups: cases (with outcome of interest) and controls (without
outcome) who reflect exposure pattern from same source population from
which cases arose
• Cases and controls are matched for key characteristics (e.g., sex, age) to
reduce bias/confounding
• Longitudinal: goes back in time (retrospectively) to determine differences
between the groups in some exposure (e.g., drug) prior to event occurring
• Like cohort studies, used to define risk factors associated with the
development of particular disease

46

CASE CONTROL DESIGN – starts with outcome

Exposed +
CASES- Those with the
outcome (ex. disease)
Not Exposed -

Exposed +
CONTROLS-
Those without the
Not Exposed - outcome (ex disease)

Past Present

47

Compared to COHORT DESIGN (next lecture)- exposure groups

Outcome +
EXPOSED

No Outcome -
COHORT

Outcome +
NOT EXPOSED

No Outcome -

48

16
 10/3/2022

Selecting Cases and Controls

• Some common sources of cases:
• Patient rosters at medical facilities
• Death certificates
• Disease registries
• Health plan databases
• Selection of controls
• Should be representative of population that produced the cases
• Should be sampled independent of exposure (not be any more or less likely to have
exposure of interest)

49

What if Case-Control Study Rather than Cohort Study?

Suppose investigators used a case-control study design to assess the
relationship between Vitamin C intake and gout based on information
from a health care database

1. Who would be the cases?

2. Who would be the controls?
3. How should they be matched?
4. What statistic would be used to report the association of Vitamin C
with the outcome variable gout?

50

Calculating Risk from Case-Control Studies: Odds Ratios

• Determines odds that cases were exposed to the

risk vs odd that controls were exposed to the risk
= Odds Ratio (OR)

51

17
 10/3/2022

Risk Ratio or Relative risk (RR) is used in cohort studies:

Can’t use for Case Control studies
• RR = Rate of outcome in exposed Outcome Outcome
present not
Rate of outcome in unexposed + present -
RR = a/(a + b) / c/(c + d) Exposed a b
Not Exposed c d

In case control studies you don’t start with exposed, therefore don’t know
how many were exposed so can’t calculate RR
To estimate RR in case control studies- Use Odds Ratio (OR)
OR estimates RR if outcome not too common; (more on this in cohort lecture)

52

Odds Ratio in Case Control studies vs Cohort

Case Control Outcome Outcome Cohort study Develop Do not
study present+ absent - Outcome develop
(cases) (controls) + Outcome -
History of a b Exposed a b
exposure
No history of c d Not Exposed c d
exposure

CASE CONTROL STUDIES COHORT STUDIES

• OR = Odds that a case was exposed OR = Odds of developing outcome if exposed
Odds that a control was exposed Odds of developing outcome if not exposed
OR = a/c / b/d = ad/bc OR = a/b / c/d = ad/bc

53

Odds Ratio in Case Control studies vs Cohort

Case Control Outcome Outcome Cohort study Develop Do not
study present+ absent - Outcome develop
(cases) (controls) + Outcome -
History of a b Exposed a b
exposure
No history of c d Not Exposed c d
exposure

CASE CONTROL STUDIES COHORT STUDIES

• OR = Odds that a case was exposed OR = Odds of developing outcome if exposed
Odds that a control was exposed Odds of developing outcome if not exposed
OR = a/c / b/d = ad/bc OR = a/b / c/d = ad/bc

54

18
 10/3/2022

Remember: Interpreting Odds Ratios

OR<1: decreased odds of event in exposed group

OR>1: increased odds of event in exposed group

OR=1: no increased or decreased odds of event in exposed group

Need to also know confidence interval (CI) to evaluate whether

there is a statistically significant difference between the groups;
• If OR +/- 95% CI includes 1.0- no difference
• If OR +/- 95% CI all >1.0→ associated with increased risk
• If OR +/- 95% CI all < 1.0→ associated with decreased risk

95% Confidence Interval = 95% chance that the true value falls
within this range

55

Case Control Study Example

• Cases: Depression clinic patients who had Sexual Dysfunction
• Controls = Depression clinic patients who had Sexual Dysfunction
• Looked back to see if had taken antidepressant medication

56

Odds Ratio Calculation

Odds of sexual dysfunction in those on antidepressants:

OR = a/c / b/d = 8/3 / 32/33 = 2.75
Hypothetical OR +/- 95% CI is 2.75 (1.6-3.5)
Increased odds of sexual dysfunction in those treated with antidepressant

Crude OR (Unadjusted) can also be adjusted (Adjusted OR) for variables that weren’t matched
such as presence of other diseases, smoking etc using Regression analysis- see next example

57

19
 10/3/2022

Study Example:
Bruderer et al. Use of Diuretics and Risk of Incident Gout: A Population-
Based Case–Control Study; ARTHRITIS & RHEUMATOLOGY 2014;Vol.
66 (1):185–196
https://www.ncbi.nlm.nih.gov/pubmed/24449584 or
https://onlinelibrary.wiley.com/doi/epdf/10.1002/art.38203

Another example/review of sections of an observational study paper

Introduction/Study objectives, Methods, Results, Discussion

58

Bruderer et al study-
Which of the following was done in this study to control for confounding?

1. Controlling for confounding in study design:

- Restriction
- Matching

2. Controlling for confounding in statistical analysis

- Stratification
- Regression Analysis

59

Study Example and Participation Activity:

Bruderer et al. Use of Diuretics and Risk of Incident Gout: A Population-Based Case–Control Study; ARTHRITIS &
RHEUMATOLOGY 2014;Vol. 66 (1):185–196
https://www.ncbi.nlm.nih.gov/pubmed/24449584 or https://onlinelibrary.wiley.com/doi/epdf/10.1002/art.38203

Questions:
1. What matching was done with the controls to make them similar to cases?
2. What variables were adjusted for in the analysis?
3. Was thiazide diuretic current use overall associated with an increased risk of gout?
What was the OR and 95% CI? Corrected (adjusted) OR? (see Table 3)
4. Was Potassium sparing current use overall associated with an increased risk of
gout? What was the OR and 95% CI? Corrected (adjusted) OR? (see Table 3)

60

20
 10/3/2022

Nested Case-Control Studies

• A case-control study “nested” within a cohort group

• The cohort group is followed until a certain event occurs (e.g., MI,
stroke, death)
• “incident” or new cases are identified

• Cases are then matched to a sample of controls (event free) within

the same cohort group

61

Example of Nested Case-Control Study

• Objective: to assess the risk of recurrent pneumonia associated with
inhaled corticosteroids (ICS) in elderly patients
• Cohort group: adults > 65 with hx of pneumonia but currently w/o
pneumonia identified in a clinical database
• Cases: subjects who developed recurrent pneumonia during a 5 year
follow-up period
• Controls: subjects free of pneumonia matched for age, male sex and
COPD history
• Exposure: ICS use in both groups

Eurich et al. Inhaled corticosteroids and risk of recurrent pneumonia. Clin Infect Dis 2013;57:1138-44

62

Study Results
ICS Use Controls Cases Unadjusted Adjusted OR
(n=6244) (n=653) OR (95%) CI (95%) CI

Never 4208 395 Ref Ref

Past 1289 135 1.23 (.92-1.65) 1.15 (.85-1.55)

Current 747 123 1.92 (1.47-2.50) 1.90 (1.45-2.50)

OR was adjusted for comorbidities, total meds, use of PP inhibitors or H2 blockers,

vaccinations, smoking, nursing home residence, functional status and pneumonia
severity index

1. What can you conclude about past use of ICS & pneumonia based on
these results?
2. What can you conclude about current use of ICS & pneumonia based on
these results?
3. Was there evidence of confounding?

63

21
 10/3/2022

Another Example:
Daneman N et al. Metronidazole-associated neurologic events: A nested case control study
https://academic.oup.com/cid/article/72/12/2095/5821517

Which of the following was done in this study to control for confounding?

1. Controlling for confounding in study design:

- Restriction
- Matching

2. Controlling for confounding in statistical analysis

- Stratification
- Regression Analysis

64

Strengths of Case-Control Studies

• Relatively inexpensive/can be completed in a short time

• Data associated with many variables (e.g., monitoring

parameters) can be collected at once

• Can be used to study rare conditions or diseases (because

you start with people that already have the disease rather than
waiting for it to develop)

65

Weaknesses of Case-Control Studies

• Information on exposure collected by self-report, interview, or recorded
information is susceptible to recall, interviewer or information bias
• Cases may be more likely than controls to recall prior exposures
• Interviewer may have inconsistent approach between cases and controls
• Difficult to validate information (e.g., compliance)
• Selection of appropriate controls may be difficult
• Cannot determine incidence of disease
• Cannot prove cause-effect relationship, only association

66

22
 10/3/2022

Comparison of Observational Studies

Cohort Studies (next lecture) Case-Control Studies

Starts with exposure and goes forward to Starts with outcome and goes backward to
outcome exposure
Typically larger sample size Typically smaller sample size

Longer time to complete Can be completed in short period of time

Not suitable for rare diseases Suitable for rare diseases

Yields incidence rates, RR Yields only OR

Can provide information about multiple Provides information only about outcome
outcomes under study
Expensive Inexpensive

67

Comparison of Observational Studies

Cohort Studies (next lecture) Case-Control Studies

Starts with exposure and goes forward to Starts with outcome and goes backward to
outcome exposure
Typically larger sample size Typically smaller sample size

Longer time to complete Can be completed in short period of time

Not suitable for rare diseases Suitable for rare diseases

Yields incidence rates, RR Yields only OR

Can provide information about multiple Provides information only about outcome
outcomes under study
Expensive Inexpensive

68

Evaluating Case-Control studies

• Were the cases and controls chosen from the same study population? How were cases,
controls, and exposure defined?
• If there are differences between the cases and controls in characteristics that could
contribute to the outcome, independent of the exposure being evaluated, then
confounding exists
• Were cases and controls matched for some of these key characteristics (age, sex etc) to
prevent confounding?
• Were subjects with any characteristics excluded from the study?
• Was regression analysis used to adjust the OR to control for additional confounders?

69

23
 10/3/2022

Questions?

• Cohort studies- next lecture

• Participation Quiz on Blackboard available until the end of class

70

24