Skip to content
Scintria
ToolsGrant writerFunding
HomeLiteratureClinical trialsEvidence LabProtocol packData analysisLearnProjectsLibraryReferencesApplications
Start free
ScintriaStart free

Learn / Epidemiology / Study designs

Cohort studies

14 minute read · 6-question quiz

Download this lesson as a colour Word or PDF study guide with the figures and answers. Included in paid plans; reading and quizzes stay free.Sign in

What you will learn

  • Describe prospective, retrospective and ambispective cohort designs

  • Calculate relative risk, rate ratio and attributable risk from cohort data

  • Identify the main threats: loss to follow-up, misclassification and confounding

In a cohort study, you start with exposure: you follow groups of exposed and unexposed people over time to see who develops the outcome. It follows the natural direction of cause and effect.

1Free ofdiseaseDefine thepopulation2MeasureexposureExposed vsunexposed3Follow upMonths todecades4Count newcasesIncidence ineach group5CompareRR, rate ratio,risk difference
The cohort design follows the arrow of time.

Types

  • Prospective: the cohort is assembled now; follow-up occurs in the future. Best data quality, expensive and slow. (Framingham Heart Study, Rakai Community Cohort Study.)
  • Retrospective (historical): exposure and outcome have already happened; the researcher uses past records (occupational records linked to cancer registry). Fast, relies on record quality.
  • Ambispective: a mix.

Strengths

  • Clear temporality: exposure precedes outcome.
  • Direct measures of incidence and risk in both groups.
  • Can study many outcomes from one exposure; good for rare exposures (a cohort of workers exposed to a chemical).
  • Less recall bias than case-control studies.

Weaknesses

  • Expensive and slow, especially for rare or long-latency diseases.
  • Loss to follow-up can bias results if those lost differ.
  • Exposure may change over time (smoking cessation); need updated measures.
  • Confounding is still present (not randomised).

Worked example: exposure to cooking smoke

A cohort of 22,000 person-years: 10,000 person-years in children from smoky kitchens, with 40 pneumonia cases, and 12,000 person-years from smoke-free homes, with 15 cases.

  • Incidence rate (exposed) = 40 / 10,000 = 4.0 per 1,000 person-years.
  • Incidence rate (unexposed) = 15 / 12,000 = 1.25 per 1,000 person-years.
  • Rate ratio = 4.0 / 1.25 = 3.2 (95% CI 1.8 to 5.8).
  • Rate difference (attributable rate) = 4.0 − 1.25 = 2.75 per 1,000 person-years among exposed children.
  • Attributable fraction among exposed = (3.2 − 1) / 3.2 = 69%: if the exposure is causal, about 69% of pneumonia in exposed children is due to it.

Interpretation: exposed children had 3.2 times the rate of pneumonia. Removing smoke might prevent about 2.75 cases per 1,000 child-years.

Measures from a cohort

MeasureFormulaQuestion answered
Risk ratio (RR)Risk exposed ÷ risk unexposedHow many times as likely?
Rate ratioRate exposed ÷ rate unexposedSame, using person-time
Risk differenceRisk exposed − risk unexposedHow many extra cases?
Attributable fraction (exposed)(RR − 1) ÷ RRWhat share of disease in the exposed is due to exposure?
Population attributable fractionp(RR − 1) ÷ [1 + p(RR − 1)]What share of disease in the whole population is due to exposure?

(Here p is the proportion of the population exposed.)

Dealing with loss to follow-up

Retention is the cohort's lifeblood. Collect multiple contact details, tracking visits, regular communication, incentives within ethical limits. Compare those lost with those retained; use inverse probability weights or sensitivity analyses to see whether losses could change the result. Aim for 80% or more follow-up, and treat below 60% as seriously threatening.

Famous cohorts

Framingham (cardiovascular risk), Nurses' Health Study, British Doctors Study (smoking and lung cancer), Rakai Community Cohort (HIV), Birth to Twenty (South Africa), Mbarara Adolescent Cohorts. Large cohorts increasingly use biobanks and linked electronic records.

Comparing designs

Cohort: from exposure to outcome, gives incidence and risk, good for rare exposures and many outcomes, slow and costly. Case-control: from outcome to exposure, gives odds ratios, good for rare outcomes and many exposures, faster and cheaper, more prone to bias. Cross-sectional: snapshot, gives prevalence, quick, no temporality.

Healthy-worker and healthy-volunteer effects

Workers are healthier than the general population (those too ill do not work), so comparing a workforce with the general population understates occupational harm. Volunteers for cohorts are healthier too, which limits generalisation but not necessarily internal comparisons.

Summary

  • A cohort follows exposed and unexposed groups forward to compare incidence.

  • It establishes temporality and gives RR, rate ratio, risk difference and attributable fractions.

  • It suits rare exposures and multiple outcomes but is slow, costly and vulnerable to losses.

  • PAF = p(RR − 1) / [1 + p(RR − 1)] gives the public-health impact.

  • Retention and exposure updating are central to quality.

Check your understanding

Question 1
1

A cohort study of 5 years' follow-up most clearly demonstrates:

Question 2
2

Incidence 4 per 1,000 py (exposed) and 1 per 1,000 py (unexposed). The rate difference is:

Question 3
3

With RR = 4, the attributable fraction in the exposed is:

Question 4
4

Which design is best for a rare exposure such as an industrial chemical?

Question 5
5

Many participants with worse health drop out of a cohort. The estimate of incidence is likely:

Question 6
6

PAF depends on both the RR and:

Draft study guide. Spotted an error or something unclear? Tell us at hello@scintria.com. This is educational material, not clinical or regulatory advice.

Case-control studiesAttributable risk and population impact