Types of Studies in Epidemiology: Observational Studies

beta carotene (pro-vitamin A) among smokers

• thought to be protective bc of antioxidants but only for non-smokers

What is purpose of health research studies? (3)

1. exploratory research

2. descriptive research

3. causal inference

What is exploratory research?

to explore or pilot test new ideas or new methods; to generate hypotheses

What is descriptive research?

to understand who is affected by the health condition or known risk factors, what conditions are most common, where they might be located, and when they are affected

What is causal inference?

to test hypotheses about what factors may cause the disease outcome, what interventions may prevent it, and what treatments may improve prognosis

What kind of studies are epidemiology studies?

population studies

• can’t tightly control conditions for the most part

What are three different study designs for epidemiology?

1. observational studies

2. experimental studies

3. research synthesis studies

What are observational studies? What are examples?

analyzing the world, not really intervening or setting a specific experiment to test

What are five different types of observational studies?

1. case reports/case series

2. ecologic studies

3. cross sectional studies

4. case-control studies

5. cohort studies

What are experimental studies?

studies that test an intervention/new therapy/community program

What is a major example of an experimental study?

randomized control trials (RCT)

• optimal design for experimental studies

• different variations of RCT

What are the six different sub-types of RCTs?

1. individually randomized, parallel group

2. cross over trials

3. factorial trials

4. cluster randomized

5. stepped wedge trials

6. adaptive trials

What are research synthesis studies?

studies where you look at all of the evidence that has been published across journals and locations

• you come up with an estimate of what an overall pooled effect might be in a generalized way based on all the info

• used for DGA and DRIs

What are two examples of research synthesis studies?

1. systemic reviews

2. meta-analysis

What are case reports/case series studies?

• case report: description of one person with illness

• case series: description of many people with illness

  • purely descriptive studies with no comparison group

When are case reports/case series studies done?

often performed with a new illness during the initial stages of an investigation

What are the three possible goals of case reports/case series studies?

1. to inform clinicians or public health authorities of a new disease

2. to generate hypotheses about the cause of a new disease (can’t really test the hypotheses though)

3. clinical training purposes to inform clinicians about rare disease or uncommon presentations of (common)

What is the effectiveness of case reports/series?

considered the weakest form of evidence for disease etiology, preventive interventions, or efficacy of treatment

What are ecologic studies?

studies where we look at associations in a group level (countries, communities, schools)

• only possible when there is info on the exposure and disease burden for the group of interest

• ex: examining relationship b/w indices of dietary intake or nutritional status and population prevalence of disease across groups

What are the two pros of ecologic study designs?

1. may be used when individual level data are unavailable or impractical to gather

2. useful for generating hypotheses

What are the four cons of ecologic study designs?

1. do not reflect individual level associations

2. ecologic fallacy

3. usually lack data necessary to adequately control for confounding

4. insufficient to establish causality

What is the ecologic fallacy?

When inferences are made about individuals based only on data about the group to which they belong to

• data reported are usually the AVERAGE for the group and do not necessarily reflect the ACTUAL relationship at an individual level

What is confounding?

a factor that is associated with the exposure, potentially causally related to the outcome and not on the direct causal pathway between the exposure and the outcome

What happens when you don’t account for confounding variables within the design or analysis of a study?

may cause there to be an incorrect estimate of the association between the exposure and the outcome

In terms of ecologic studies in nutrition, where are two places where dietary data can come from?

1. Food Agriculture Organization (FAO) Food Balance Sheets

2. Nationally representative dietary surveys

What are Food Agriculture Organization (FAO) Food Balance Sheets?

datasets that present a country's national agrifood supply and utilization pattern

• useful data for ecologic studies

What are three places where we can get disease outcome data for ecologic studies?

1. state/national cancer registries, death registries, and other vital statistics

2. mortality or morbidity data from global databases (WHO Data)

3. population surveys

What are some example confounding factors using the ecologic example of showing the correlation between UPF availability and obesity risk?

• GDP

• income

• physical activity levels

What are cross sectional studies?

a snapshot of info of a group of people at a given moment of time (can be done multiple times, however each additional time, we get a new group of people in the pic)

• collect data on a variety of exposures and disease states

• exposure is showed simultaneously with disease

does not tell us RISK levels bc again its prevalence, NOT incidence

What is another word for cross sectional studies?

prevalence survey

What are three reasons why cross sectional studies are also referred to as prevalence surveys?

1. can determine prevalence of disease/risk factors in the population

2. cannot determine incidence (new cases)

3. cannot establish temporality (time series of events like if the exposure occurred before the disease or after)

What are five purposes of cross sectional studies?

1. to describe burden of disease in population

2. describe population characteristics in relation to disease

3. describe knowledge, attitudes, and practices related to disease

4. if repeated, can monitor trends

5. generate hypotheses about associations

What are five pros associated with cross sectional designs?

1. inexpensive and quick

2. do not require long follow up

3. can be used to describe population prevalence

4. if repeated, can be useful to monitor trends

5. can generate hypotheses about associations regarding risk factors and disease outcome

What are four cons associated with cross sectional design?

1. risk of cofounding

2. establishing temporality (don’t know which came first, disease or risk and this is bc individuals may change their habits in response to a disease diagnosis)

3. length based sampling bias

4. insufficient to establish causality

What is a common example of cross sectional surveys used in nutrition?

NHANES

• national health and nutrition examination survey

What kind of sampling method does NHANES use to make sampling < 10k people nationally representative?

multi-stage cluster sampling method

Describe the multi-stage cluster sampling method

• stratify all counties in the country by geography and randomly select 15 counties per year with selection probability proportional to size

• within counties, randomly select geographic segments of the county

• list all households and randomly select households

• randomly select individuals within households

What is length-based sampling bias?

individuals who have a disease for longer are more likely to be sampled than those who have a short duration illness (bc we are sampling prevalent cases, whats present, not incidence)

What is the issue with length-based sampling bias?

can be an issue when investigating diseases with variable survival times or variable recovery times

• they won’t be as likely to be captured in the prevalence survey, leaving them out

• leads to biased interpretations of exposure outcome relationships as it may be more related to lengthy disease rather than disease risk

What are case control studies?

individuals with disease (cases) are compared to those without disease (controls) and history of exposure is retrospectively assessed (look backwards in time for past exposure or risk factors)

When is it desirable to use a case-control study? (6)

1. when trying to understand the etiology of disease

2. little is known about the disease

3. disease is rare

4. long latency period/long induction period (years or decades to reach disease endpoint)

5. all exposure data are difficult/expensive to obtain

6. underlying population dynamic (ex: highly mobile pop) is difficult to track over long periods of time

When we do a case control study, how do we determine what is considered a “case”?

• case definition is specific, clear, and contains a consistent diagnostic criteria

  • need to specify

    • eligible population

    • geographic area

    • study period

  • cases can be identified through clinics, hospitals, or disease registries

  • must be selected independent of exposure history

Why is it preferred to also include incident cases (enroll cases as they are diagnosed) rather than just prevalent cases in case control studies?

would help to reduce length-based sampling bias

When we do a case control study, how do we determine what is considered a “control”?

controls should represent the source population that produced the cases and represent the exposure patterns in that population

• enrolled independent of their exposure history

• they can have other illness; they don’t have to be completely healthy

• may be selected from population list from the same clinic/hospital from which cases were recruited

• inappropriate control group selection can create selection bias

What are two major cons associated with case control studies?

1. random recall errors

2. recall bias

What are random recall errors?

when people have difficulty remembering events, particularly when the length of recall is long or the event was relatively routine

• can cloud the researcher’s ability to detect a possible relationship between the exposure and the disease (bias towards null)

What is systematic recall bias?

individuals with disease tend to spend more time thinking about what could have possibly caused their disease (rumination bias), therefore they have their own hypothesis about what could have caused their disease and may be more likely to report that they were exposed OR less likely exposed if the possible exposure is socially undesirable

In case control studies, what should we use instead of memory recall?

try to use objective measures of exposures that don’t require memory recall

What are four pros associated with case control studies?

1. faster and cheaper than cohort studies

2. good for studying rare diseases

3. sample size may be smaller than for a cohort study

4. possible to study multiple risk factors for a given disease outcome, not limited to just one exposure

What are four cons associated with case control studies?

1. high risk of confounding

2. subject to recall bias, errors in recall, or incomplete data due to retrospective analysis

3. not good choice when exposure is rare

4. selection of appropriate controls may be difficult

What are cohort studies?

investigator selects a sample of the population, some exposed and some unexposed, follows them through time, and compares the incidence of disease in the exposed vs. the unexposed

• population selection could be unrelated to their exposure OR it could be based on whether or not they were exposed to a risk factor

• samples have to be large bc of how rare disease is

What are the two purposes of cohort studies?

1. determine etiology of disease

2. study the natural history of disease (prognosis)

What are two types of cohort studies?

1. prospective (common)

2. retrospective

What is the difference between prospective cohort and retrospective cohorts?

prospective cohort (more common)

• identify sample of people form study population and then measure their exposure status at the time of enrollment (don’t know before hand)

• follow them over time to monitor development of disease

• compare incidence of disease among exposed vs unexposed

retrospective cohort

• define a historical cohort whose members were at risk at the historical starting point (still don’t have the disease)

• reconstruct subsequent disease occurrence and exposure history from records

• compare disease incidence among exposed vs unexposed

What is the difference between a retrospective cohort and a traditional case-control study?

traditional case control study has participants enroll based on disease status, NOT exposure status (retrospective), alongside controls form the same population

• will likely determine exposure history afterwards through participant recall, records, or stored specimens

• compare likelihood of exposure among case vs. controls

A

C

• measured at one point in time

• measured the prevalence

B

• shows average of consumption

• only country level information, not individual

C

What is something to take into account for cohort studies if the disease is rare?

if the disease is rare, then cohort studies need to be extremely large and expensive

• if incidence is 37.6/100k then we would assume that we need to enroll at least 100k people

• lab biomarker testing is also expensive, so imagine testing for 100k… thats expensive

What is a solution to cut down on cohort study costs (if the disease is rare)?

nest a case-control study within a cohort study

Describe how nesting a case-control study within a cohort study works.

major funding agencies (ex: NIH) fund very large cohort studies (ex: 100k or more), collect a whole variety of bio-specimen form the participants but don’t analyze the samples, they just enroll them and follow them overtime to see who developed the disease

• if disease is developed then researchers identified two matched controls (based on gender and age) from the same cohort alongside the person who developed the disease

• grabbed the previously taken bio-specimen from each respective participant to measure the biomarkers of the specific exposure and compare/analyze them

• saves time and money in terms of which one’s to study

• more modern way of doing a case-control study, just bc its a strong design

What are the benefits of nestled case control studies within a cohort study?

• efficiency (save money and time analyzing)

• analysis of biomarker status instead of participants recall of diet

  • objective, lower recall bias, lower memory errors

What are five pros of cohort studies?

1. can assess incidence of disease

2. prospectively collected info is more accurate than recalling it

3. allows us to assess unexpected effects of the initial exposures

4. useful when exposure is rare

5. nested case control cohort studies can improve efficiency

What are five cons associated with cohort studies?

1. long duration of follow up may be required

2. difficult to capture rare disease outcomes

3. large sample sizes may be required

4. tracing individuals over a long period of time may become difficult (high losses to follow up)

5. expensive

Who is James Lind?

Scottish surgeon who wrote the first known examples of a randomized trial as he did research on the study of scurvy

• scurvy at the time was a big issue for sailors, and in the British Navy, scurvy killed a lot

• British Navy did not believe him at first, but later on repeated the trials and realized

• made lemon and lime juice a standard part of the diet for navy officers at sea

In general, what are randomized trials/experimental designs?

• similar to cohort, but exposure is randomized to one group and not the other

  • main purpose is to evaluate opportunities for prevention or treatment of disease

• strongest evidence of causality because confounding and bias are minimal

• gold standard (most rigorous)

What does double-blind mean in an experimental study?

the allocation is unknown to the participant AND to the investigator

What does randomized mean in an experimental study?

the allocation is randomly assigned (computer generated random codes, drawing cards from a hat, etc.)

What does controlled mean in an experimental study?

usually the control group is offered some comparable intervention to maintain blinding and control for the “placebo effect”

• minimizes confounding and bias

What does clinical trial mean in an experimental study?

term more commonly used if a medical intervention is used (can be interchangeable with community trial though)

What does community trial mean in an experimental study?

term usually used to refer to the fact that communities were randomized, not individuals BUT can also mean that the study was community based, not hospital/clinic based

What does treatment group/allocation group/intervention group mean in an experimental study?

the group that was randomized to certain intervention

What is residual confounding?

what is left over after we adjusted for confounding as best as possible

• unable to adjust for things that we did not measure in observational studies

• tend to have less residual confounding with randomized trial design bc they are randomly assigned intervention/exposure and the characteristics of the two groups are random (won’t have healthy user bias)

What are the pros of randomization?

randomization tends to balance measured and unmeasured prognostic characteristics (chance imbalance can remain though)

• reduces risk of confounding and risk of selection bias

What is it ethical to randomize interventions?

when it is truly unknown if there are benefits or risks of the intervention (clinical equipoise: genuine uncertainty about the comparative benefits and harms of the study interventions)

Why should investigators not be involved in whether a group receives the treatment or not?

if the investigators or study staff have a preconceived idea about the likely effect of the intervention, they may wish to allocate that intervention to certain groups over others

What is selection bias?

when researchers selectively choose that certain individuals are allocated one of the treatments

What is surveillance bias?

when researchers actively monitor one group for the outcome

What is bias in the analysis?

when researchers interpret the results in such a way as to favor one of the groups

What is the placebo effect?

when subjects develop a reaction based on what they think should be the effect of the intervention

When is it ethical to use a placebo? (3)

1. no known effective treatment is available

2. when withholding an established treatment would result in, at most, only temporary discomfort

3. compelling methodologic reasons for the use of a placebo and participants are not at risk of excessive harm

What are the two strengths of RCTs?

1. considered gold standard for research study design

2. minimal confounding and bias if properly implemented

What are the two limitations of RCTs?

1. many more ethical considerations than in observational studies

2. expensive

What are three different types of RCTs?

1. cluster randomized trials

2. cross-over designs

3. factorial designs

What is a cluster randomized trial?

• unit of randomization is a cluster rather than an individual

  • clusters can be communities, neighborhoods, schools, and classrooms

What are three major challenges with cluster randomized trials?

1. needs to be independence between the intervention cluster and the control cluster (community members can’t travel from one place to another)

2. if number of clusters is small, there may be an imbalance on population characteristics in groups (resulting in confounding)

3. intra-cluster correlation/design effect (need to have larger to account for it)

What is intra-cluster correlation/design effect?

idea that people within a cluster are usually more similar to each other than they are to people in other clusters

• sample in each group is less diverse that what you might expect

What are the two advantages to cluster-randomized trials?

1. useful when intervention is impractical to deliver on an individual basis

2. cost efficient if resources are all directed specific areas instead of spread out

What are three disadvantages of cluster-randomized trials?

1. intra-cluster correlation

2. larger total sample sizes are usually required

3. possibility of imbalance between groups

What is a crossover design?

a randomized trial in which each group receives each intervention, but the order with which the interventions are delivered is randomized

• each group “crosses over” to receive the other intervention during the course of the trial

How does the participant serve as their own control in crossover design?

the individual’s response to intervention A can be compared to that same individual’s response to intervention B

What are two advantages of crossover RCT designs?

1. subjects serve as their own control (reduces confounding)

2. can be used for dietary interventions if the outcome responds quickly to the intervention

What are the two limitations of crossover RCT designs?

1. carryover effects of the interventions (need to have a sufficient washout period)

2. most suitable when intervention period is short adn teh likely effects are rapid