Chapter 7: Analytic Epidemiology: Types of Study Designs

Descriptive studies

describing the health of the population, used to identify a health problem that may exist

• Classify and describe based on person, place, and time

Analytic studies

Used to identify the causes/determinants of the health problem

Observational studies

not in control of who is exposed to variables

• Can not randomize variable

• Can not assign group

Experimental studies

manipulation, researcher controls who gets exposed to something (aka manipulating variable)

Can randomly assign participants (who gets treatment vs. placebo) (aka radom group assingment)

Clinical trials (aka randomized control trials)

manipulation of study factor by investigator and randomization in assignment of subjects to treatment vs. comparison group

• Ex. testing a new drug

• Used to test efficacy of preventative/therapeutic measures

• Focus on individuals

• Prospective = future focused

Outcome of interest/clinical end point

examined to evaluate efficacy

• Ex. Who got the disease, who died/lived

Single blind

particicpant doesn’t know if they received the treatment or the placebo

Double blind

neither the participant nor the researcher know if the participant received the treatment of the placebo

Phases of clinical trial

1 —> less than 100 subjects

2—> 100-200 subjects

3 —> 100s - 1000s of subjects

4 —> released to general public

Prophylactic trials

prevent disease

• evaluating effectiveness of a substance to prevent disease

• Ex. vaccines, PREP for HIV

• Using healthy, at risk cases (examining incidence over study)

Therapeutic trials

improve health/treat disease

• Can a drug cure or treat an existing disease?

• Using prevalent cases → everyone already has the disease

Community trials

have manipulation involved, but examining larger populations

• Quasi experimental → manipulation, no individual randomization

• Examining behavioral changes in population

• Focus on community

• Prospective = future focused

• Example. Sex ed program efficacy

Efficacy (of a vaccine)

how well did they work in the clinical trial

Effectiveness (of a vaccine)

how well does it actually work in real life (outside of controlled environment)

• Examine once released to general public

Observational studies

 measurement of patterns of exposure and disease in populations to draw inferences about etiology (Comparing disease frequencies between group with characteristic and group without characteristic)

• Can not randomize variable

• Can not randomize group assignment 

Temporality in observational studies

timing of information regards to cause and effect

? Did the information about cause and effect refer to the same point in time ?

? Or, was the information about the cause garnered before or after the information about the effect ? 

Cohort studies

prospective (future) and longitudinal (over a long period of time)

• Start with group of healthy subjects but all at risk (i.e. examining incidence rates, not prevalent cases)

• Check in (at least 2 times) to determine exposure statuses between exposure cohort vs. non exposure cohort

• Measuring incidence!!

Sampling and cohort formation: population-based samples

• cohort includes an entire population (or representative sample of the population)

• Exposures are unknown initially

• At each check in

  • Examine incidence of exposed ground

  • Examine incidence of non exposed group

Prospective cohort studies

follow cohort into future

• Characterized by exposure variable present

• Follow up for occurrence of disease at some point in future

Retrospective cohort studies

using historical data to determine past exposure levels

Historical perspective cohort studies

historical data + examine into future

Sampling and cohort formation: exposure-based samples

group that has an exposure being compared to a group that doesn’t have the exposure

• Know from very beginning if people have been exposed or not

• Still starting with healthy at risk people and examining incident cases

Ex. occupational exposures

Effect measures

a quantify the strength of the relationship between an exposure and a health outcome

Absolute effect measures

subtraction (subtracting disease frequencies from one another)

Risk difference

gives information about the effect of an exposure

Relative measures

division (dividing disease frequencies from one another)

relative risk + incidence rate ratio + odds ratio

Relative risk

Measuring probability of an event occurring with an exposure and the probability of an event occurring without an exposure

• can use when using cohort study or randomized control trial (MEASURING INCIDENCE)

• Tells you if risk of disease is different among the exposed as compared to the non exposed

RR > 1

GREATER than the risk of the disease among the nonexposed

Exposure variable is related to the health outcome

Exposure could be a risk factor for the disease

RR = 1

EQUAL to the risk among the nonexposed

The exposure is not associated with the health outcome

Exposure is not associated with the disease

RR < 1

LESS than the risk of the disease among the nonexposed

Exposure could be a protective factor for the disease

Incidence rate ratios

Compares the incident rates among the 2 groups (exposure and non exposure)

IRR > 1

GREATER than the incidence rate of the disease among the nonexposed

IRR = 1

EQUAL to the incidence rate among the nonexposed

IRR < 1

LESS than the incidence rate of the disease among the nonexposed

Risk difference aka attributable risk

Difference between the incidence of disease in the exposed group (Ie) and the incidence of disease in the nonexposed group (Ine) 

• How many cases of disease would be eliminated in the population if we were able to remove the exposure from the population

• How many cases is that exposure contributing to disease occurrence in the population

Etiologic fraction

Measures of potential impact → impact of exposure removal on exposed

proportion of the rate of disease in the exposed group that is due to the exposure

• ? If we removed the exposure, what would happen to the exposed group ?

• Usually expressed as percentage

• Tells you how much the particular exposure accounts for the disease etiology in the exposed group

Population risk difference

Measure of potential impact —> impact of exposure reval on population

difference between the incidence rate of disease in the nonexposed segment of the population (Ine) and the overall incidence rate in the total population (Ip)

1. Must first measure overall incidence rate in total population

2. Then subtract incidence rate of disease from non exposed group

Case control studies

• Trying to determine if there are different factors that lead to the development of disease in 1 group but not the other → exposure determined retrospectively

  • type of analytic study → observations

• Recruiting subjects based on presence or absence of a particular disease status

  • Case group = have disease

  • Control group = group without the disease

• Single point of observation (snapshot)

• Unit of observation and analysis = individual 

• Examining PREVALENT cases of disease

Selecting cases of case-control studies (2)

Conceptually: what constitutes a case in theory

• Diagnostic criteria, test, list of symptoms

• What does the case look like

Operationally: is there a measurement that can be taken/completed to determine if disease is present

Selection of controls for case-control studies

Population based controls: pulled from list

Applications of case-control studies (3)

• Investigating outbreak of infectious disease

• Chronic disease when etiology is unknown

• Have hypothesis of what etiology/cause is, able to test that theory

Advantages of case-control studies (4)

• Smaller sample sizes

• Quick and easy to complete

• Cost effective

• Useful to study rare diseases

Limitations of case-control studies (4)

• Human error when recalling past exposures

• Unclear temporal relationships between exposure and disease

• Not useful for rare exposures

• Use of indirect estimate of risk

Nested case-control studies

Type of case-control study in which cases/controls are drawn from the population in a cohort study

• cohort study, follow healthy people, determine who developed disease in exposure and non exposure group. At end, recruit from cohort study (healthy and sick)

Advantages of nested case-control studies (4)

• Helps reduce cost

• Confident candidates will continue to participate

• Able to collect data quickly

• Have degree of control over confounding variables, helps clarify temporal relationships

Odds ratio

• Measure of association between an exposure and an outcome

  • Tells you if odds of disease are different among the exposed as compared to the nonexposed

OR = AD/BC

OR > 1

GREATER than the odds of the disease among the nonexposed

Exposure is associated with higher odds of disease The exposure may be a risk factor for the disease

OR = 1

EQUAL to the odds among the nonexposed

Exposure does not affect odds of disease. The exposure is not associated with the disease.

OR < 1

LESS than the odds of the disease among the nonexposed

Exposure is associated with lower odds of disease. – The exposure may be protective against the disease.

OR = food approximation of risk when (3)

• Control are representative of a target population

• Cases are representative of all cases

• The frequency of disease in the population is small

Cross sectional study

• Survey of a population, used to estimate prevalence of a disease

• Exposure and disease status report obtained at single/same point in time (snapshot)

• Examining PREVALENCE

Applications of cross-sectional studies

• evaluate/compare trends in health/disease

• plan/evaluate health services/intervention

• Identify problems for analytic studies/hypothesis generation

Advantages of cross-sectional studies

• Generalizability 

• Can be large + short period of time

• Completed at low cost

Disadvantages of cross-sectional studies

• Not proving causation

• no info on incidence of disease

• Not great for rare diseases/low prevalent diseases

• Focusing on snapshot in time i.e. can’t determine temporality

Ecological studies

• can be descriptive or analytic

• Unit of analysis = group (NOT the individual)

  • Level of exposure for each individual is unknown

• Using secondary data sources → not collecting ourselves

Ecologic fallacy

• observations made at the group level may not represent the exposure-disease relationship at the individual level

• Incorrect inferences about the individual are made from group level data

• Conclusions may be the reverse of those from a study that collects data on individual subjects

Simpson's paradox

• an association in observed subgroups of a population may be reversed in the entire population.

  • association between 2 variables emerges but then disappears when population is divided into sub groups

Applications of ecologic studies (4)

• Test specific etiologic hypothesis

• Develop new etiologic hypothesis

• Suggest mechanisms of causation

• Testing efficacy of certain programs/outcomes

Advantages of ecologic studies (2)

• Quick, simple, inexpensive

• Approach for generating hypothesis

Disadvantages of ecologic studies (2)

• Ecological fallacy

• Imprecise measurement of exposure and disease

Internal validity

established first, make sure we got our data accurately and wholly, was the research done right

External validity

need to make sure subset examined is representative of the population as a whole/data is generalizable

Truth

proving causation (does the association represent a cause-and-effect relationship, throwback to necessary and sufficient variables)

Chance

random error, association occurring by chance

• Type 1 and type 2 errors

Bias

systematic error → happens because of error in study design/recruitment

Random error

reflect fluctuations around a true value of a parameter

•  issue with sampling or random variability within subject/observer

Factors contributing to random error

1. poor precision

2. sampling error

3. variability in measurement

How to reduce random error

increase sample size

systemic error

error results in incorrect estimate of the  measure of association 

• Can happen throughout study (design, data collection/analysis, interpretation, reporting, publication)

• Issue of accuracy (clustering)

Hawthorne/observer bias

 participants behavior changes with the knowledge of knowing they are being observed

survival/selection bias

• different sizes of populations being examined can lead to over estimation of effect of exposure on disease

• occurs when trying to identify who is going to participate in the study

Healthy worker effect

employed populations tend to have a lower mortality compared to general population (missing sick workers from sample)

Techniques to reduce selection bias (3)

• Explicit/objective case definition

• Capture all cases of disease in specific time/region

  • Want to have high participation rates

• Make sure population is representative

Recall bias

hard to recall things in the past

• Easier for cases to recall than controls

Interviewer bias

occurs when interviewers probe more thoroughly for an exposure

• Pry more in cases than in controls on suspected exposure

Prevarication (lying) bias

occurs when participants have ulterior motives for answering a question and thus may underestimate or exaggerate an exposure

• Common with behavioral exposures

Techniques to reduce information bias

• Address recall bias: memory aids

• Address interview bias: blind interviewers to subjects study status + use standardized data collection forms + standardize training sessions + ensure questions are clearly written

• Address lying bias: blind participants to study goals and classification status 

confounding variable

 An alternate explanation for observed association between an exposure and disease

Criteria for a confounding variable (3)

1. Be a risk factor for the disease.

2. Be associated with the exposure.

3. Not be an intermediate step in the causal path between exposure and disease.

Ways to control for confounding variables (3,2)

Study design

1. Randomization

2. Restriction

3. Matching

Analysis

1. Stratification

2. Multivariate techniques

Randomization

attempt to ensure equal distribution of the confounding variable in each exposure category

• Creating comparable populations

• Need large sample size

• Can control for unknown confounders

Restriction

restricting who is allowed in the study

• Complete control over confounders

• Cant control for unknown confounding

Matching

matches subjects in study groups according to confounding 

• Can be hard to make matches

• Requires less people

Stratification

analysis performed to evaluate the effect of an exposure within strata of the confounder

• Examine differences once divided into strata

• Easy and logical thing to do

• Makes group sizes smaller

Multivariate techniques

statistical methods

• Can control for multiple confounding variables at once