epidemiology 5: Types of Epidemiological Studies Ecologic Case Series Cross-Sectional Cohort Case-Control Hybrid

observational studies descriptive studies vs ANALYTIC

Analytic Studies Used to quantify association between an exposure & a disease or health outcome Used to test hypotheses about causal relationships Provide a control group (baseline) Test hypotheses about determinants Determine causation

How Do Study Designs Differ?

Availability of subjects Data collection methods Directionality of exposure Timing of data collection Measure of association Number of observations made Study designs differ in a variety of ways that affect their conduct. Some things that differ include the number of observations one needs to make, the "directionality" of the exposure-disease relationship

cohort study design

Cohort A group of individuals followed or traced over a period of time Cohort studies are similar to randomized trials Evaluation of an exposed group versus a nonexposed group BUT, because exposure is NOT randomized, there is the chance that an observed association is actually due to some factor related to exposure A "cohort" is a group of individuals who share some common characteristic: military veterans, college graduates, coal miners, nurses, or just volunteers who wanted to be in the study (e.g., American Cancer Society's Cancer Prevention Studies). Risk factor data are recorded on all study subjects before any of them have died or developed the disease being studied.

what are the types of cohort studies (in depth)

Combination of Prospective & Retrospective Exposure determined from pre-existing records (i.e. retrospective or historical cohort study) Follow-up & measurement of outcome continues into the future (prospective)

what are the types of cohort studies (in depth)

Comparison based on exposed versus nonexposed group of individuals Inclusion of only incident or new cases of disease allows for: Determining a temporal association between an exposure & development of disease Direct calculation of risk of disease

What Determines the Choice of a Study Design?

Consider whether the disease or exposure is rare Rare exposure best evaluated using a cohort study Rare disease best evaluated using a case-control study Amount of knowledge about a disease Less known, may be best to use a case-control study Consider practicality In choosing a study design, the epidemiologist carefully considers several issues. First is the frequency of the disease or exposure of interest. Generally, a cohort study is best for rare exposures & a case-control study is best for rare diseases. Case-control studies are generally preferred when little is known about the etiology & risk factors for a disease because they are less expensive to conduct than cohort studies.

What is the Measure of Association used in Ecologic Studies?

Correlation Coefficient = r Examples: Positive correlation - Diastolic blood pressure generally rises with age Negative correlation - Heart rate is usually lower in individuals who exercise frequently

What is the Measure of Association used in Ecologic Studies?

Correlation Coefficient = r Measures the extent of linear relation between exposure & disease across the study units & indicates the direction of the group-level association Ranges between +1 & −1, where: 1 is total positive correlation, other positive values indicate that a high value in one variable is associated with a high value of a 2nd variable 0 is no correlation −1 is total negative correlation, other negative values indicate that a low value in one variable is associated with a high value of a 2nd variable Strength of association r indicates that for every unit change in the level of exposure, the disease frequency increases or decreases proportionately Ecologic studies are often an efficient design for generating new hypotheses about exposures. Interestingly, subsequent case-control, cohort & clinical trial studies of fat intake and breast cancer have produced very mixed results. Case-control studies suggest that there is an association. Cohort studies suggest that there is no association. A large clinical trial to date (the Women's Health Initiative) also produced inconclusive results.

observational studies DESCRIPTIVE studies vs analytical

Descriptive Studies Organize and summarize data according to time, place, & person to... Describe natural history of disease Determine extent of public health problem Identify populations at greatest risk Determine allocation of health care resources Suggest hypotheses about causation

Disease Cluster

Disease cluster may be identified through a case-series It is a perceived excess in disease or death in time and/or place & often is linked to some environmental etiology Supposition is that there is shared exposure Numerator with no denominator Investigation of a disease may include several stages Initial contact Assessment to determine whether an excess number of cases has occurred or is it due to chance & whether the excess is tied etiologically to an exposure Feasibility study to evaluate geographic & temporal analysis based on cases from a different locale or time period Etiologic investigation (case-control or cohort study)

ecological fallacy

Ecological Fallacy' Making a causal inference about an individual-level phenomenon from grouped data Individual variability for a factor of interest (exposure) is unknown & cannot be accounted for

observational studies

Evaluates associations regarding Etiology Prognosis Findings used to infer from a sample to the level of population Variables associated with exposure & disease not under investigator control due to ethical concerns or logistical constraints Differs from a randomized trial where study participants are randomly assigned into an experimental group or a control group

Retrospective Cohort Study

Exposure & outcome of interest have already occurred when the study is initiated Only studies prior previously recorded outcomes & not future ones Follow-up period begins at some date in the past Data has already been collected

prospective cohort study

Exposure may or may not have occurred when the study is initiated, but the outcome has definitely not occurred Follow up period extends into the future Outcome of interest occurs after the start of the study in a prospective cohort study Exposure is measured & follow-up continues into the future Exposure is ascertained now about the present and the past As the study continues, a profile is developed of the individuals' exposure during a lifetime Populations studied are usually general populations or a subgroup of the population Measure of Association is the relative risk (RR)

What Factors May Influence a Study Design?

Exposure measure Outcome measure Are the methods for exposure & outcome measure valid & reliable? Confounders Are there factors that may influence the association of exposure & disease? Sample size Is the necessary sample size feasible? Sample selection bias Measurement error

study designed explained

First, one must recognize a hierarchy of study designs. The simplest is the "case series", in which an investigator simply ascertains a series of cases of disease. This is most frequently done in the context of a disease outbreak. The objective is to trace the communication of the disease from one individual to another. There are no comparison groups or "controls", and hence the ability to infer causes of disease is weak. The "ecological" study design is conceptually similar with the exception that the unit of study is the group or population, not the individual case. In an ecological design, populations are generally cross-classified by prevalence of diseases & exposures. The ability to infer causes is weak. A "cross-sectional" study is like a snapshot in time of disease in a population. Individuals within a cross-sectional study can be categorized as cases & non-cases of a disease or health condition and associations with risk factors can be estimated, but because there is no element of time, the ability to infer causality is again weak. At the next level, is the case-control study design that is the work-horse for most epidemiological studies. A series of cases of disease is ascertained & compared to suitably selected controls without disease for recalled exposures. In a case-control study the exposure has always occurred in the past (retrospective). A "cohort" study begins with a sample of healthy individuals in whom a suspected causal exposure is carefully measured & followed over time (prospective) for the development of disease. The "randomized trial" or "clinical trial" is the epidemiologic equivalent to an experiment.

prospective cohort study - framingham heart study

Framingham heart study enrolled a 2nd generation group of 5,124 of the original participants' adult children & their spouIn 1948, the Framingham Heart Study under the National Heart Institute (now the National Heart, Lung, & Blood Institute-NHLBI) was initiated. Little was known about the general causes of heart disease & stroke, but the death rates for CVD had been increasing steadily since the beginning of the century & had become an American epidemic. The objective was to identify common factors that contribute to CVD by following its development over a long period of time in a large cohort of subjects who had not developed overt symptoms or suffered a heart attack or stroke. The researchers recruited 5,209 men and women between the ages of 30 and 62 from the town of Framingham, Massachusetts, and began the first round of extensive physical examinations and lifestyle interviews that they would later analyze for common patterns related to CVD development. Since 1948, subjects have been evaluated every 2 years for a detailed medical history, physical exam & laboratory tests. In 1971, the study ses to participate in similar exams. Over the years, monitoring of the study participants led to the identification of the major CVD risk factors (high blood pressure, high blood cholesterol, smoking, obesity, diabetes, physical inactivity) & a great deal of valuable information on the effects of related factors such as blood triglyceride & HDL cholesterol levels, age, gender, & psychosocial issues.

prospective cohort study design

Incidence of Disease Among Exposed: (a) / (a+b) Incidence of Disease Among Nonexposed: (c) / (c+d) The focus of the cohort study design is on selecting a "representative" sample from a target population. Because the goal is to select a "sample at risk", prevalent cases of the disease of interest must be excluded. Sometimes this is not so easy, especially for diseases that are latent and/or difficult to diagnose; a variable number of "false negatives" might be included. The "eligible cohort" (sample of non-diseased subjects) is carefully measured for exposures & risk factors, & followed over time for the development of the disease of interest. The key difference from the case-control design is that the exposure is measured before identification of the disease; hence this design is called "prospective". Because the putative cause precedes the result, the prospective cohort study is considered to be the strongest observational design for inferring causality.

Cohort Study Design & Incidence Rates of Disease

Incidence of Disease Among Exposed: (a) / (a+b) Incidence of Disease Among Nonexposed: (c) / (c+d) Relative Risk: (𝑨/(𝑨+𝑩)//(𝑪/(𝑪+𝑫)

Prospective Cohort Study Design Strengths versus WEAKNESSES

Large study populations necessary if rare outcome Expensive (costly & time consuming) Bias due to loss to follow-up (lack of participation, withdrawal, migration, death) May require long study duration Subject to 'drift' or changes in measurements of exposure & outcomes over time May be statistically & practically inefficient for rare diseases with long latency

Retrospective Cohort Study Design STRENGTHS versus Weaknesses

Less expensive than prospective cohort studies Time frame of follow-up can still be long but there is a reduction of study duration Good for diseases with long latency or time between exposure & disease onset Ability to study many diseases in relation to one exposure The strengths of the retrospective cohort study include: Lower expense Short study duration since subjects are followed in time through historical records. If the historical records are sufficient, the time-frame of exposure to disease can be very long, allowing one to consider the possible association of a childhood exposure with an adult-onset disease. This would take many years to conduct within a prospective study.

retrospective cohort study design

Not all cohort studies are prospective. There are also "retrospective" cohort studies that share some features with the case-control design. The difference is that one begins with a clear & precise defined target cohort (for example, all workers at a ceramic fibers plant who worked there between 1987-1995). One then identifies cases of disease in the cohort, searches HISTORICAL records for possible exposures, & compares cases with noncases for level of exposure. This type of design is popular in environmental/occupational epidemiology because historical records for exposure are often present.

What are the Types of Cohort Studies?

Prospective study Concurrent Follow-up period begins in present extending into future Longitudinal Retrospective study * Nonconcurrent prospective Follow-up period begins at a past point & ends in past Longitudinal

observational studies design

Select an appropriate study design to determine if an association is present Advantage Natural setting Addresses ethical concerns Disadvantage No randomization Investigator can only account for variables measured

How are Groups Selected for a Cohort Study?

Select study population on basis of exposure / nonexposure (often conducted for occupationally exposed) Select a defined population prior to evidence for exposure based on a factor not related to exposure such as geographic location Exposed versus Nonexposed based on baseline data, which may be collected as biologic specimens and/or detailed questionnaire history (i.e. family history, physical activity, diet, weight history, reproductive history, occupational exposure history, etc.)

randomized trials study designs

Study participants are randomly assigned into an experimental group or a control group. Focus is on the outcome variable Clinical trials focused on treatment (no treatment vs. treatment; new vs. current treatment) Community-based trials (no intervention vs. disease prevention program) Advantages Randomization Potential confounders are evenly distributed among exposure Potential for bias is low Disadvantages Ethical concerns Cost Length of study Not good for rare diseases

Prospective Cohort Study Design STRENGTHS versus Weaknesses

Temporal relationship between exposure & outcome can be established with relative accuracy Good for rare exposures (such as occupational, chemical, natural disasters) Multiple diseases/outcomes can be assessed Good for medium to long time-frame Less subject to certain selection biases

Retrospective Cohort Study Design strengths versus WEAKNESSES

Temporal relationship between exposure & outcome more difficult to establish than in prospective cohort studies Accurate historical records necessary to define exposure May be more susceptible to bias in measurement of both exposure & outcome Observer bias may be greater Confounding by other unmeasured factors may limit findings The weaknesses include: The quality & accuracy of the historical records for past exposures & risk factors, which may make it difficult to exactly determine the true temporal relationship between exposure & outcome. It may be impossible to determine what happened to people who were exposed but never became part of the cohort. Observer bias may be greater given that a researcher may need to make a decision on exclusion or inclusion of particular types of records & data. Therefore, it is important to have clear criteria for definition of exposure, disease, as well as inclusion/exclusion study criteria.

ecological study design (more depth)

This is a schematic illustrating the basic design of an ecologic study. Various independent populations are sampled to determine the prevalence of a disease & some exposure. These samples (groups) are then compared, often graphically, for the apparent rank relationship between disease prevalence & exposure. We cannot conclude that a causal association exists based on this type of data

Case Series Study Design

This is a schematic of a case-series study design. This type of study begins with an "index case" of disease (the red dot in the figure). The epidemiologist collects data from this case with the goal of determining who else may have been in contact & might therefore also develop the disease. This is depicted in the connected blue, green, & yellow dots that represented new cases who developed disease who had contact with the index case at some point in time. The epidemiologist traces a "network" of interconnecting cases that may extend in both place & time. In the figure, the green cases were found in New York, the brown in Los Angeles, the blue in San Francisco, & the yellow in Chicago.

Retrospective Cohort Study - Intrauterine Starvation & Subsequent Low Birthweight & Adult Hypertension

This is an example of a proposed retrospective cohort study to examine the association of intrauterine exposure to famine starvation, subsequent low birth weight, & hypertension in adulthood. The proposed design takes advantage of a "natural experiment" in which a well-defined population received a well-defined exposure. In this example, the exposure is the "Dutch Famine of 1944". During World War II, the German army attempted to suppress the people of Amsterdam by restricting all food supplies entering the city, thereby inducing starvation among the residents during the year of 1944. Nearly 30,000 people starved to death during the winter. Among the survivors were pregnant women who gave birth during or immediately following the famine. The proposed design compares the people born at this time (who had low birth weight due to intrauterine starvation) with sibling "controls" born immediately before or after the famine for the prevalence of hypertension in 2006. This design was proposed as a means of testing the "Barker Hypothesis" that intrauterine exposures influence adult disease outcomes (Stein et al., Int J Epid 2004 33(4):831-836). The current version of the hypothesis for effects of intrauterine starvation could not be ethically addressed in a prospective cohort study

Ecologic Study Design

Use population or group level data versus individual data Describes population Describes an environmental exposure Describes a global measure Usually defines groups by time and or place Often compares rates over time in one geographic location They are distinct from all other designs in that they are conducted entirely at the group or population level so the unit of measurement is the group & not the individual subject. An ecologic study might compare one city with another for the prevalence of disease by some measurement of air pollution level.

Why & When are Ecologic Studies Useful?

Uses existing data Requires NO direct contact with individual subjects Conducted quickly at low cost Suggest where further research may be relevant for etiologic or causal associations Ecologic studies are often an efficient design for generating new hypotheses about exposures. Interestingly, subsequent case-control, cohort & clinical trial studies of fat intake and breast cancer have produced very mixed results. Case-control studies suggest that there is an association. Cohort studies suggest that there is no association. A large clinical trial to date (the Women's Health Initiative) also produced inconclusive results.

ecological study design (more info)

We cannot conclude that a causal association exists based on this type of data Only average values of fat consumption at country level known Need data on fat consumption for individuals not countries Only breast cancer incidence at country level Need data on breast cancer diagnosis for each individual in a study An association observed between variables on the aggregate level does not necessarily represent an association that exists on the individual level One must be very careful with over-interpreting results from ecologic studies. An association observed between variables on the group level does not necessarily mean that a causal relationship actually exists on the individual level.

hierarchy of study designs

lowest to case series - no controls ecological- group level cross sectional - "snap shot" case - control- "retrospective" cohort- "prospective" clinical trial - randomized Highest

What are the Potential Biases in Cohort Studies?

selection biases - nonparticipation - nonresponse -loss to followup information biases - data abstracted from past records observer bias exposed subjects may be more closely followed than non exposed