Cohort Studies

Observational Studies: Cohort Studies

Key Features and Design

• Observational studies include cohort and case-control studies.
• They investigate etiological risk factors and associations between independent (exposure) and dependent (outcome) variables.
• Cohort studies assess exposure-outcome relationships over time, establishing temporality (exposure precedes outcome).
• Case-control studies, discussed later, start with the outcome and retrospectively assess exposure.

Drawbacks Compared to Randomized Controlled Trials

• Randomized controlled trials (experimental) involve researchers actively assigning exposures, controlling for confounding variables through randomization.
• In observational studies, researchers observe self-selected exposures without intervention.
• Example: Studying tobacco smoking and lung cancer; researchers cannot randomly assign smoking.

Cohort Study Design Sequence

• Start with a broad population.
• Sample taken and non-randomly sorted into exposed and unexposed groups based on self-selection.
• Follow individuals forward in time to observe outcome development in both groups.
• Assess if the outcome is more likely in the exposed group compared to the unexposed.

Key Findings and Calculations: Incidence

• Incidence is key because cohort studies follow people forward in time, identifying new (incident) cases.
• Temporality (exposure precedes outcome) is crucial for assessing potential causal relationships.
• Incidence helps determine the risk of developing the outcome given exposure; often expressed as a percentage.
• The incidence of the outcome in the exposed group is compared to the incidence in the unexposed group to see if the numbers differ statistically.

Sampling and Recruiting Methods

• Both methods calculate incidence by starting with exposure and tracking forward to the outcome.

Method 1

• Recruit individuals known to be exposed and individuals known to be unexposed.
• Example: Study smoking during pregnancy and baby birth weight.
• Recruit pregnant women who smoke and pregnant women who don't.
• Follow pregnancies and assess for the outcome of low birth weight babies.

Method 2

• Recruit a sample of pregnant women regardless of smoking status.
• Divide into exposed (smokers) and unexposed (non-smokers).
• Follow forward in time to see if incidence of low birth weight differs.

Visual Depictions

• Cohort study designs can be depicted in various ways, but exposure assessment must always precede outcome assessment.

Cohort Study Design Options

• Design options are differentiated on the timing of data collection relative to exposure and outcome.
• Cohort studies start with a population sorted into exposed and unexposed groups via self-selection.
• Exposure measurement occurs at one point, and then outcomes are observed over time.

Prospective (Concurrent or Longitudinal) Design

• Collects information from individuals in the present and follows them into the future.
• Most common cohort design.

Benefits

• Ensures outcome has not occurred before exposure data is collected.
• Allows assessment of temporality by assessing exposure and looking for any signs or symptoms of the outcome that might be present when exposure is assessed.
• Individuals with outcome symptoms at the start are excluded.

Drawbacks

• Problems with loss to follow-up (increases with longer time periods).
• High expense due to long follow-up duration, requiring personnel, tracking movements, and participant incentives.
• Time: Long periods introduce other factors unrelated to the exposure that might influence the outcome.

Retrospective (Non-concurrent Prospective or Historical) Design

• Begins with a cohort defined in the past.
• Recruit individuals from that cohort and collect data on exposure from historical records (e.g., medical records).
• Look forward from that exposure point to see if the outcome is recorded at a later time.

Benefits

• Does not require long study duration for the researcher.
• Avoids loss to follow-up.

Drawbacks

• Difficult to recruit the entire historical cohort (e.g., name changes, relocation).
• Data quality issues with past records.
• Reliance on individuals to recall past exposures accurately, with potential for recall bias.

Design Similarities

• Both prospective and retrospective designs start with a population, assess exposure at a point in time, and assess the outcome at a later point.

Example Study Designs

• A twenty-year prospective cohort study begins in 2017.
• Recruit a population and follow them, assessing exposure in 2027, ensuring no outcome has occurred by this time.
• Continue following until 2037 to see who develops the outcome.
• Compare incidence in exposed and unexposed groups.
• A retrospective twenty-year cohort study also begins in 2017 but looks backward.
• Identify a cohort from 1997 (e.g., children born at a hospital).
• Assess exposure (e.g., gifted program participation by 2007) using records.
• Assess the outcome by 2017 (e.g., university admission) using records.
• An option is to combine both: Identify an old study population, assess past data, and follow the population until present to see whether or not an outcome develops.

Calculations from Cohort Studies

Key Terms

• Exposure.
• Outcome.
• Incidence.

Calculations of Risk

• Absolute Risk.
• Relative Risk.
• Attributable Risk.
• Population Attributable Risk.

Incidence and Risk

• Cohort studies assess new cases to determine if incidence varies with exposure.
• Temporality strengthens evidence for causal relationships.

Statistical Calculations

• Look at the association between exposure and outcome.
• Absolute risk: Incidence in the exposed group.
• Relative risk: Ratio of incidence in exposed vs. unexposed.
• Odds ratio: Primarily for cross-sectional or case-control data but can be used.

Absolute Risk

• Incidence of disease in the exposed group.
• Limited information without comparison to the unexposed group.

Relative Risk

• Assess the increase of the outcome in the exposed group versus the unexposed group.
• Calculated by \frac{\text{Risk of incidence in exposed}}{\text{Risk of incidence in unexposed}}.
• Measure of how many times more likely the exposed group is to experience the outcome than the unexposed group.
• If relative risk (RR) = 1, the probability of the outcome is the same in both groups.
• If RR > 1, the incidence is higher in the exposed group.
• If RR < 1, the incidence is higher in the unexposed group.

Two by Two Table

• Outcome (disease) as columns, exposure as rows.
• Cells labeled a, b, c, d.

Incidence Calculation

• Exposed group: \frac{a}{a+b}.
• Unexposed group: \frac{c}{c+d}.
• Total population: \frac{a+c}{\text{Total population}}.

Relative Risk Calculation

• \frac{a/(a+b)}{c/(c+d)}.

Fictional Cohort Example

• Research question: Is smoking during pregnancy associated with having a low birth weight baby?

Ethical Considerations

• Randomized controlled trial unethical because it would involve randomly assigning women to smoke.

Recruitment Methods

• Recruit pregnant women smoking, assess infant birth weights, compare to non-smokers.
• The goal of both retrospective and prospective studies is to follow patients and assess outcomes.

Prospective Designs

1. Recruitment Method A: Large sample, sort into smoker/non-smoker, follow to assess birth weight.
2. Recruitment Method B: Recruit a fixed number of smokers and non-smokers and assess birth weights.

Retrospective Design

• Find a past retrospective cohort, obtain medical records to determine smoking status, and assess birth outcomes.

Numerical Example

• Recruit 5,100 pregnant women: 1,035 smokers and 4,065 non-smokers.
• Of smokers, 404 had low birth weight babies.
• Of non-smokers, 775 had low birth weight babies.
• Calculate the total number of healthy babies in each group.
• Incidence can be calculated for total population, exposed, and non-exposed groups.
• Relative, attributable and population attributable risk can all be calculated.

Initial Data Interpretation

• The most extensive data shows the largest number of healthy pregnancies come in the non smoker group so there appears to be a lower risk.
• Interventions require the risk calculation.
• Cannot conclude recommendations of smoking to reduce low birth weight.
• Requires calculation that smokers compared to non-smokers.

Incidence

• Number of individuals w/ outcome / number of individuals in that same group.
• Total Sample, Exposed Group and Unexposed Group.
• Total sample: Low birthweight - 1,179/5,100 = 0.23.
• Exposed: 404/1,035 = 0.39 this is also absolute risk.
• Unexposed 775/4,065= 0.19.

Relative Risk

• Incidence in exposed/incidence in unexposed.
• 0. 39/.19 = 2.05.

Attributable Risk

• (Incidence in exposed group – Incidence in unexposed group) / Incidence in exposed group.
• (0. 39-0. 19)/.39 = .51.
• Multiply by 100 = 51% of low birth weights seen among smokers are directly a result from smoking.

Population Attributable risk

• Incidence total POP-Incidence exposure divided by incidence total population.

• Multiply by 100 is 17 % of the total population attribute low births to maternal smoking.

  Methodological Considerations

External Validity

• Relates to generalizability.
• The cohort should be representative of that population.
• If it cannot be generalized, the study is limited.

Internal Validity

• The measurement of exposure/outcome should not harm internal validity.

• If you carefully create specific categorizations, internal validity is sound.

• May be measurement problems if using old data.

• Dropout rates lead to internal validity problems.

  Study Biases

• With cohort designs.

Selection Bias

• In a cohort study, we may have selection bias that occurs due to loss to follow-up.

Information Bias

• Occurs if the quality of info for the unexposed and exposed differs.

Assessor Bias

• Very likely, the researcher will know all that information which cannot be blinded.

Statistician Bias

• Statistical bias is another biased, and they will search for data internally that proves their points.