cohort study

**Definition & design:** Participants are selected based on exposure status (exposed vs. unexposed) and are followed forward in time (or via existing records) to see who develops the outcome. Direction of inquiry: exposure → disease.

**Types:** - *Prospective cohort*: Exposure assessed at baseline; participants followed into the future (e.g., Framingham Heart Study, Nurses' Health Study). Less recall bias but expensive and slow. - *Retrospective (historical) cohort*: Exposure and outcome have both already occurred; investigators use existing records to reconstruct exposure and follow-up. Faster and cheaper, but dependent on record quality.

**Key measures:** - Incidence in exposed and unexposed groups - **Relative risk (RR)** = incidence in exposed / incidence in unexposed = [a/(a+b)] / [c/(c+d)] - **Attributable risk (AR)** = incidence exposed − incidence unexposed - Attributable risk percent = (RR − 1)/RR - Number needed to harm (NNH) = 1/AR - Population attributable risk also derivable - RR > 1 = increased risk; RR < 1 = protective; RR = 1 = no association

**Strengths:** - Establishes temporality (exposure precedes outcome), which is supportive of, but does not prove, causation - Can measure incidence and RR directly - Can examine multiple outcomes from a single exposure - Good for rare exposures (e.g., occupational or toxic exposures) - Less recall bias than case-control (prospective)

**Weaknesses:** - Inefficient for rare diseases (need large samples) - Long follow-up for diseases with long latency; expensive - **Loss to follow-up** (attrition bias) is a major threat - Confounding (no randomization), selection bias (e.g., healthy worker effect) - Cannot establish causation as definitively as an RCT

**Comparison with other designs (high-yield):** - *Case-control*: Starts with disease status (cases vs. controls) and looks back at exposure; yields **odds ratio (OR)**; best for rare diseases; prone to recall and selection bias; cheaper/quicker. - *Cross-sectional*: Exposure and disease measured at one point in time; yields **prevalence**; cannot establish temporality. - *RCT*: Investigator assigns exposure by randomization; gold standard for causality; controls confounding. - Mnemonic: Cohort = "Risk" (RR); Case-control = "Odds" (OR).

**Example/vignette clue:** "Researchers followed 5,000 smokers and 5,000 non-smokers for 20 years and compared the incidence of lung cancer" → cohort study; calculate RR.

**Bias to know:** Loss to follow-up, observer/detection bias (lack of blinding), Berkson/selection bias, healthy worker effect, and confounding. Ways to address confounding: restriction, matching, stratification, multivariable regression (all of which can be used in the analysis phase for cohort studies).

**Step 2 relevance:** Interpreting RR and confidence intervals (if the 95% CI for RR includes 1, the result is not statistically significant) and choosing the appropriate study design for a given research question.