Epidemiology Lecture 3 - Observational Studies: Cohort Studies

Example: Does low Nut Consumption CAUSE HTN?

Associated? YES Causal? 1.) Biological Plausibility: Maybe 2.) Strength of Association: Close (RR = 1.8) 3.) Dose-Varying: Not in this analysis 4.) Temporal Association: Yes

Relative Risk Example Interpretations 2

Consuming nuts ≥1 time per week is associated with *50% lower incidence* of HTN compared to consuming nuts <1 time per week

Cohort Studies: Dose Varying Relationships

Examples of what you WANT to see to indicate Dose Varying Relationships in studies

Cohort Example: Incidence Proportion of HTN

Incidence Proportion = (new events)/(pop free of disease at beginning) Pro: gives a good number to tell patients (e.g. 10% risk of HTN if eat nuts >1/week) Con: Could be differences in follow up time between groups, so this is not the best number for comparison --> need to use person-time at risk to find Incidence Rate

How does ONLY Reporting Relative Risk Mask Reality?

Jupiter Study (Interventional Study) - Incidence Proportions --> Relative Risk = "Rosuvastatin reduces the risk of MI by 55%" - Should it say "associated with" instead of "reduces?" No, a large study should be well randomized and exposure is black and white yes/no so any change in risk is ok to assume is causal - *HOWEVER: The 55% does not capture that BOTH groups had a Very low of MI, so risk is actually very low in both cases*

Cohorts (Definition)

Measure exposure and divide the study pop into groups based on exposure - assignment to cohorts can be arbitrary --> can create as many cohorts as you want

Indicence

Measures the development of new disease over time

Cohort Study Populations

Must be individuals free of the disease at the start of the study so that NEW ONSET can be measured

Cohort Studies

Type of observational study design - Begins w/ individuals who are free of the disease outcome of interest (so that new onset can be measured) - Measure exposure and divide the study pop into groups based on exposure - Follow for the development of new (incident) disease over time - Helps establish temporal association between exposure and outcome: *clarifies that exposure precedes outcome in time* - Important puzzle piece for addressing causes of disease

Cohort Example: Incidence Rate

Use person-time at risk to find Incidence Rate = New cases /Person-Time at Risk

Incidence Rate of Disease in the Entire Study Population

(New cases in exposed group + new cases in unexposed group)/(Person-time at risk in exposed group + Person-time at risk in unexposed group)

Attributable Risk

*AR = Incidence Exposed - Incidence Unexposed* - tells you the extra amount of disease that happens that is attributable to the exposure in the exposed group - could be other reasons for the excess risk, which is why you need to be confident about causal relationship!

Cohort Studies: Example

*Background*: Peanuts and other nuts rich in magnesium and unsaturated fatty acids. These nutrients can reduce BP in experimental models. Nuts are part of healthy diets linked with lower BPs (e.g., Mediterranean diet) *Hypothesis*: Lower nut consumption (exposure) --> HTN (outcome) *Study*: Evaluate 1,000 adults; Exclude pts who have HTN at start of study; Administer food questionnaire to assess frequency of nut consumption; Follow over 5 years for development of New Onset HTN *2 Qs*: 1.) Is lower nut consumption *associated with* HTN? 2.) Does lower nut consumption *cause* HTN?

Relative Risk

*Relative Risk = Incidence in Exposed/Incidence in Unexposed* Ratio of two incidences in a cohort study - preferably incidence rates, but incidence proportion works too

Relative Risk vs Attributable Risk

*Relative Risk*: applies nicely to an individual - e.g. you can tell a pt, "consuming nuts <1x per week is associated w/ an 80% higher chance of YOU developing HTN" - causality not determined *Attributable Risk and Population Risk*: addresses impact of exposure on a population level - e.g. HTN is a big problem in the US, how can we reducee it

Observational Study Designs

- Case reports / case series - Cross-sectional studies - Cohort studies - Case-control studies Observational studies detect associations NOT causes of disease

Relative Risk Example Interpretations 1

1.) Interpretations Consuming nuts <1 time per week is associated with *1.8-fold greater incidence* of HTN compared to consuming nuts ≥1 time per week 2.) Consuming nuts <1 time per week is associated with *80% greater incidence* of HTN compared to consuming nuts ≥1 time per week

Measures of Excess Risk

1.) Relative Risk (RR) 2.) Attributable Risk 3.) Population attributable risk 4.) Number needed to treat (NNT) Typically applied to clinical trials or situations for which there is a strong body of supportive evidence - Risk measures aside from RR require some certainty that exposure CAUSES disease

Population Attributable Risk

= *Incidence Total - Incidence Unexposed* - In the population, there are X additional cases of disease per person-years attributable to exposure

Cohort Studies 2x2 Tables

Pic

Diversity in Cohorts

Random assignment to cohorts often ignores other factors (e.g. people who eat less nuts are generally older) - Can equalize people (e.g. no one eats nuts for 2 weeks) and assign people to specific treatment groups/cohorts to balance characteristics

Cohort Study Populations - Exclusions

Study population May further exclude people who have "subclinical" disease *Pros*: - Refines disease-free status at baseline - Cases that develop during follow up are truly new *Cons*: - Fewer eligible people - Results less applicable