1. Study Designs in Epidemiologic Research
South Asian Cardiovascular
Research Methodology Workshop
Basic Epidemiology
Study Designs in Epidemiologic
Research
Thomas Songer, PhD

2. Fundamental Assumption in Epidemiology
Disease doesn’t occur in a vacuum
Disease is not randomly distributed throughout a population
Epidemiology uses systematic approach to study the differences in disease distribution in subgroups
Allows for study of causal and preventive factors

3. Components of Epidemiology
Measure disease frequency
Quantify disease
Assess distribution of disease
Who is getting disease?
Where is disease occurring?
When is disease occurring?
Formulation of hypotheses concerning causal and preventive factors
Identify determinants of disease
Hypotheses are tested using epidemiologic studies

4. Types of primary studies
Descriptive studies
describe occurrence of outcome
Analytic studies
describe association between exposure and outcome

5. Basic Question in Analytic Epidemiology
Are exposure and disease linked? E D
Exposure
Disease

6. Basic Questions in Analytic Epidemiology
Look to link exposure and disease
What is the exposure?
Who are the exposed?
What are the potential health effects?
What approach will you take to study the relationship between exposure and effect?
Wijngaarden

7. Basic Research Study Designs and their Application to Epidemiology
• Society will be healthier
• Society can save money on health care budgets
• It will improve life expectancy
• It will improve the economy
Responses to this question vary. As outlined below, some believe that altruistically, society should pursue disease prevention to improve health and save money. These are noble goals, but I will demonstrate later that prevention does not always save money. The desire to prevent NCDs probably is a repercussion of our efforts to prevent infectious diseases. We have had great success in preventing communicable diseases. This has led to an initiative to prevent NCDs.

8. Big Picture To prevent and control disease
In a coordinated plan, look to
identify hypotheses on what is related to disease and may be causing it
formally test these hypotheses
Study designs direct how the investigation is conducted

9. What designs exist to identify and investigate factors in disease?

10. Study Designs Descriptive Analytic Case report Cohort study RCT
Case series
Case-Control
study
Study Designs
Descriptive
Epidemiology
Case-Crossover
study
Cross-sectional
As Koepsell has illustrated in the book, “Injury Control”, injury studies may be descriptive in nature (describing the frequency or characteristics of injury events) or analytic (testing relationships between common traits and injury). Differing forms of descriptive studies exist. These designs are outlined in the next slide.
Analytic studies include experimental designs (the randomized controlled trial) and observational designs (case-control studies, cohort studies, etc.). The case-crossover study design has received a lot of attention in the injury field in the last five years.
study
Before-After
study
Ecologic study

11. Timeframe of Studies
Prospective Study - looks forward, looks to the future, examines future events, follows a condition, concern or disease into the future
time
Study begins here

12. Timeframe of Studies
Retrospective Study - “to look back”, looks back in time to study events that have already occurred
time
Study begins here

13. Study Design Sequence Hypothesis formation Case reports Case series
Descriptive
epidemiology
Analytic
epidemiology
Animal
study
Lab
study
Clinical
trials
Hypothesis testing
Cohort
Case-
control
Cross-
sectional

14. Increasing Knowledge of
Develop
hypothesis
Descriptive Studies
Investigate it’s
relationship to
outcomes
Case-control Studies
Increasing Knowledge of
Disease/Exposure
Define it’s meaning
with exposures
To further illustrate, if one seeks to identify the etiologic factors (e.g. causal factors) behind an outcome (e.g. an injury), then each step in the epidemiologic framework provides new and important information.
Descriptive studies are useful for identifying hypotheses to test in analytic studies. Case-control studies are then usually applied to evaluate if the hypothesized factor is related to the outcome of interest. Subsequently, cohort or longitudinal studies are applied to further define the importance of exposure to the causal agent for the development of the outcome.
Cohort Studies
Test link
experimentally
Clinical trials

15. Descriptive Studies • Society will be healthier
• Society can save money on health care budgets
• It will improve life expectancy
• It will improve the economy
Responses to this question vary. As outlined below, some believe that altruistically, society should pursue disease prevention to improve health and save money. These are noble goals, but I will demonstrate later that prevention does not always save money. The desire to prevent NCDs probably is a repercussion of our efforts to prevent infectious diseases. We have had great success in preventing communicable diseases. This has led to an initiative to prevent NCDs.

16. Case Reports
Detailed presentation of a single case or handful of cases
Generally report a new or unique finding
e.g. previous undescribed disease
e.g. unexpected link between diseases
e.g. unexpected new therapeutic effect
e.g. adverse events

17. Case Series Experience of a group of patients with a similar diagnosis
Assesses prevalent disease
Cases may be identified from a single or multiple sources
Generally report on new/unique condition
May be only realistic design for rare disorders

18. Case Series Advantages Disadvantages Useful for hypothesis generation
Informative for very rare disease with few established risk factors
Characterizes averages for disorder
Disadvantages
Cannot study cause and effect relationships
Cannot assess disease frequency

19. Houseboat Carbon Monoxide Poisonings on Lake Powell
Study design
Definition of injury
Data Sources
Population
Bias
Findings
Case series
CO poisoning
NPS EMS transport records
Lake Powell events
missing cases
outdoor exposures
The first paper to examine is the MMWR paper on carbon monoxide poisonings noted at Lake Powell. This paper attempts to identify all CO poisoning events identified from the EMS records of the National Park Service from It is thus a case series study design. The study was initiated because of case reports of CO poisoning death related to houseboats on the lake. The definition of injury used was based upon information in the transport records. Cases were defined as events with either (a) symptoms of CO poisoning and laboratory elevated carboxyhemoglobin values or (b) exposure to exhaust and symptoms. By design, only injuries receiving medical attention are included in this study.
The study is unique in that it found several cases of outdoor exposure to CO poisoning. CO poisoning is most often found in indoor settings. The report suggests that the placement of generators on houseboats creates a poisoning hazard. It implies causality related to these generators, but the design cannot prove it in the strict methodologic sense. Potential bias in the report may be the underascertainment of CO poisoning cases. It is not clear how accurate the NPS transport records may be.
Source: Houseboat associated carbon monoxide poisonings on Lake Powell. MMWR 49(49): , December 15, 2000.

20. cases with denominator
One case of unusual
injury finding
Case Report
Multiple cases of
injury finding
Case Series
Descriptive
Epidemiology Study
Descriptive study designs include case reports, case series, incidence studies, and ecologic studies. The case report is the most elementary study design in the literature. It generally describes an injury or injuries to one or two individuals that have been identified in a medical setting. There is also usually a unique feature to the noted injury (by cause, by nature of injury, etc.). The case series design is an extension of the case report. In a case series, a number of events are described. These events usually have been observed over a set period of time (such as one year) and are identified from one reporting source (e.g. a hospital).
The descriptive epidemiology study is noted by the collection of injuries over a defined population base and by the use of denominator data to determine rates. The most frequent information generated from these designs are incidence rates for injuries. The ecologic study is a hypothesis generating study. Usually using group-level data, it examines if two factors are correlated with each other.
Population-based
cases with denominator

21. Analytical Studies
Responses to this question vary. As outlined below, some believe that altruistically, society should pursue disease prevention to improve health and save money. These are noble goals, but I will demonstrate later that prevention does not always save money. The desire to prevent NCDs probably is a repercussion of our efforts to prevent infectious diseases. We have had great success in preventing communicable diseases. This has led to an initiative to prevent NCDs.

22. Study Designs - Analytic Epidemiology
Experimental Studies
Randomized controlled clinical trials
Community trials
Observational Studies
Group data
Ecologic
Individual data
Cross-sectional
Cohort
Case-control
Case-crossover

23. Experimental Studies
treatment and exposures occur in a “controlled” environment
planned research designs
clinical trials are the most well known experimental design. Clinical trials use randomly assigned data.
Community trials use nonrandom data

24. Observational Studies
non-experimental
observational because there is no individual intervention
treatment and exposures occur in a “non-controlled” environment
individuals can be observed prospectively, retrospectively, or currently

25. Cross-sectional studies
An “observational” design that surveys exposures and disease status at a single point in time (a cross-section of the population)
time
Study only exists at this point in time

26. Cross-sectional Design
factor present
No Disease
factor absent
Study
population
factor present
Disease
factor absent
time
Study only exists at this point in time

27. Cross-sectional Studies
Often used to study conditions that are relatively frequent with long duration of expression (nonfatal, chronic conditions)
It measures prevalence, not incidence of disease
Example: community surveys
Not suitable for studying rare or highly fatal diseases or a disease with short duration of expression

28. Cross-sectional studies
Disadvantages
Weakest observational design, (it measures prevalence, not incidence of disease). Prevalent cases are survivors
The temporal sequence of exposure and effect may be difficult or impossible to determine
Usually don’t know when disease occurred
Rare events a problem. Quickly emerging diseases a problem

29. Epidemiologic Study Designs
Case-Control Studies
an “observational” design comparing exposures in disease cases vs. healthy controls from same population
exposure data collected retrospectively
most feasible design where disease outcomes are rare

30. Case-Control Studies Cases: Disease Controls: No disease
Case-Control studies represent one form of analytic study that provides information on the relationship between causal factors and injuries. In a case-control study, subjects who have been injured are identified and their past exposure to suspected causal factors is compared with that of controls (persons who have not been injured).
Many case-control studies ascertain exposure from personal recall, using either a self administered questionnaire or an interview. The validity of such information will depend in part on the subject matter. People may be able to remember recent events quite well. On the other hand, long term recall is generally less reliable.
Source: Chapter 8: Case-control and cross-sectional studies, Epidemiology for the Uninitiated

31. Case-Control Design Study begins here factor present Cases (disease)
factor absent
Study
population
factor present
Controls
(no disease)
factor absent
Case-Control Design
present
past
time
Study begins here

32. Case-Control Study Strengths Limitations
Less expensive and time consuming
Efficient for studying rare diseases
Limitations
Inappropriate when disease outcome for a specific exposure is not known at start of study
Exposure measurements taken after disease occurrence
Disease status can influence selection of subjects

33. Seismic, structural, and individual factors associated with earthquake related injury
Case-control study
fatal or hospital-admitted
coroners office/hospital records
moderate to severe
Los Angeles County
controls identified by phone
higher risk in elderly, women, and apartments
Study design
Definition of injury
Data Sources
Severity of Injury
Population
Bias
Findings
Many believe that the health status of older drivers places them at greater risk for motor vehicle accidents. This study examines the association between several medical conditions and motor vehicle accidents resulting in injury. It uses the case-control study design as the method of study as crashes are rare events and several disease conditions are relatively rare in their occurrence. The study was based upon members in the Group Health Cooperative HMO. Cases were individuals who received medical care for injuries sustained in a crash where they were the driver. Accidents were identified from police reports. Controls were a random selection of HMO members who had not been injured in a crash. Collision injuries were identified from a review of the HMO database. Information on existing health conditions were also gathered from the database. All subjects were mailed a questionnaire to ascertain driving patterns, driving exposure, and health habits.
Overall, the study found higher crash rates in subjects with diabetes, but not in subjects with several other medical conditions, including cardiovascular disease, neurological disease, arthritis, and depression. This study was based upon police accident reports, so events with injuries that were not reported are not included. Also, the sample sizes of subjects with several of the conditions were small (less than 10) indicating limited ability to detect small to moderate associations.
Source: Koepsell TD, Wolf ME, McCloskey L, Buchner DM, Louie D, Wagner EH, Thompson RS. Medical conditions and motor vehicle collision injuries in older adults.

34. Earthquake Injuries

35. Case-Crossover Each participant is a case acting as their own control
Accounts for effect of potential confounders (e.g. matches on age, sex, genetic susceptibility)
Exposure status immediately before event/outcome compared with exposure some time prior to event
Acute exposures and outcomes (e.g. anger & MI; driving while using cell phone & injury)
Recall of prior exposures

36. Hypothesis Testing: Case-Crossover Studies
Study of “triggers” within an individual
”Case" and "control" component, but information of both components will come from the same individual
”Case component" = hazard period which is the time period right before the disease or event onset
”Control component" = control period which is a specified time interval other than the hazard period

37. Cell phones and crashes
Case-crossover study
property damage crash
phone records, survey
moderate, no severe injury
Ontario
volunteers, control time frame
4 times higher risk for crash when using the phone
Study design
Definition of injury
Data Sources
Severity of Injury
Population
Bias
Findings
Are you distracted when using a cellular telephone in a car to the extent that you crash more often? That is the basic question that this study seeks to address. This work was one of the first examples of the case-crossover study design in the injury literature. It examined a population of drivers in Toronto who brought their vehicles into a collision damage assessment center following a crash. They obtained informed consent from the drivers and got copies of their phone records. The cases served as their own controls. Case events included phone use at the time of the crash. Control events included phone use at the same time on the day before the crash. Alternative evaluation days were also considered. The data examines moderate crashes producing property damage to the cars. No serious of fatal injury crashes are included.
The authors found a higher risk for crash (4 times higher) at the time of phone use. This was a fairly well done study for its time. The study subjects, though, were all volunteers and may not be representative. The driving patterns in the control period may also have been different than in the case period.
Source: Redelmeier DA, Tibshirani RJ. Association between cellular telephone calls and motor vehicle collisions. NEJM 336: , 1997.
N Engl J Med 1997 Feb 13;336(7):453-8

38. Epidemiologic Study Designs
Cohort Studies
an “observational” design comparing individuals with a known risk factor or exposure with others without the risk factor or exposure
looking for a difference in the risk (incidence) of a disease over time
best observational design
data usually collected prospectively (some retrospective)

39. Cohort Design Study begins here disease Factor present no disease
population
free of
disease
disease
Factor
absent
no disease
Cohort Design
present
future
time
Study begins here

40. Timeframe of Studies
Prospective Study - looks forward, looks to the future, examines future events, follows a condition, concern or disease into the future
time
Study begins here

41. Prospective Cohort study
Exposed
Outcome
Measure exposure
and confounder
variables
Non-exposed
Outcome
Baseline
Case-control studies are perhaps the most frequent form of analytic study designs used in the injury field. These designs are very good for events that are rare in occurrence, and one could argue that severe injuries are relatively rare.
Still, there are some situations where cohort study designs would be appropriate in the injury field. Most notably for less severe forms of injury. The classic design in a cohort study is shown here. The study begins by assessing baseline levels of the exposure and other variables. Study subjects are then followed on a regular basis to identify the outcome. The frequency of outcomes are tested between persons who had exposure to the possible risk factor at baseline and persons with no exposure.
time
Study begins here

42. Timeframe of Studies
Retrospective Study - “to look back”, looks back in time to study events that have already occurred
time
Study begins here

43. Retrospective Cohort study
Exposed
Outcome
Measure exposure
and confounder
variables
Non-exposed
Outcome
Baseline
An alternative form of the cohort study is something termed the retrospective cohort study. Other researchers may also call this a historical prospective study. This design is nearly identical to the prospective cohort study. The sequence of baseline exposure determination and longitudinal follow-up for outcomes is similar. The difference lies in the time in which the study begins. In this retrospective design, the researcher constructs the cohort study by looking back in time and placing data in the appropriate order and sequence. These studies are possible to do with large medical databases, such as the membership files of the Health Maintainance Organizations, or the medical files in the Scandinavian countries.
time
Study begins here

44. Cohort Study Strengths Limitations
Exposure status determined before disease detection
Subjects selected before disease detection
Can study several outcomes for each exposure
Limitations
Expensive and time-consuming
Inefficient for rare diseases or diseases with long latency
Loss to follow-up
-- Exp. Measured before disease - so no temporal ambiguity
-- Exposure measured before disease - so disease cannot influence the amount of error with which exposure status is measured
-- Subject selection before disease, disease status does not influence of subjects

45. Experimental Studies investigator can “control” the exposure
akin to laboratory experiments except living populations are the subjects
generally involves random assignment to groups
clinical trials are the most well known experimental design
the ultimate step in testing causal hypotheses

46. Experimental Studies
In an experiment, we are interested in the consequences of some treatment on some outcome.
The subjects in the study who actually receive the treatment of interest are called the treatment group.
The subjects in the study who receive no treatment or a different treatment are called the comparison group.

47. Epidemiologic Study Designs
Randomized Controlled Trials (RCTs)
a design with subjects randomly assigned to “treatment” and “comparison” groups
provides most convincing evidence of relationship between exposure and effect
not possible to use RCTs to test effects of exposures that are expected to be harmful, for ethical reasons

48. Experimental Design time Study begins here (baseline point) outcome
RANDOMIZATION
Intervention
no outcome
Study
population
outcome
Control
no outcome
Experimental Design
baseline
future
time
Study begins here (baseline point)

49. Epidemiologic Study Designs
Randomized Controlled Trials (RCTs)
the “gold standard” of research designs
provides most convincing evidence of relationship between exposure and effect
trials of hormone replacement therapy in menopausal women found no protection for heart disease, contradicting findings of prior observational studies

50. Randomized Controlled Trials
Disadvantages
Very expensive
Not appropriate to answer certain types of questions
it may be unethical, for example, to assign persons to certain treatment or comparison groups

51. Thromboembolism and Air Travel
Study design
Outcome
Treatment
Population
Findings
RCT
DVT
Elastic hose
high risk for DVT
lower frequency of DVT in those wearing hose
The authors found no significant association between marijuana use and the risk for injury. This is a unique study approach, but several bias’ may influence the results. One, the study population are members of an HMO, and may have a higher SES than the general population. Two, the assessment of marijuana exposure was crude and was by self-report. Three, no measures of marijuana use are available at the time of the injury.
Source: Braun BL, Tekawa IS, Gerberich SG, Sidney S. Marijuana use and medically attended injury events. Annals of Emergency Medicine 32: , 1998.
Angiology 52(6): , 2001