Module 3- Observational Studies

Observational studies

Research designs where investigators observe exposures and outcomes as they naturally occur, without assigning treatments or interventions.

Why we use observational studies

1. Some exposures are unethical or impractical to assign (e.g., smoking).

2. Often cheaper and faster than randomized trials.

3. Reflect real-world conditions (high external validity).

Types of observational studies

1. Case Report

2. Case Series

3. Case-Control Study

4. Cross-Sectional Study

5. Ecological Study

6. Cohort Studies (Prospective & Retrospective)

Case report 

• A detailed description of an unusual clinical case in a single patient.

• i.e., in 2020: 56-year-old man with fever and non-productive cough after travel from Wuhan.

Purpose of a case report

• Identify new or rare diseases, unexpected symptoms, or side effects.

• Generate hypotheses for future studies.

Advantages of a case report 

• Quick, simple, inexpensive.

• Useful for detecting novel findings (e.g., new infections, mutations).

Limitations of a case report

• Not generalizable (n=1).

• No control group or causation inference.

Case series

• A collection of case reports involving patients with similar clinical presentations or treatment responses.

  • Inclusion/exclusion criteria clearly defined.

  • Collect detailed clinical data (age, sex, comorbidities, treatment, outcomes).

• i.e., Series of all ICU-admitted COVID-19 patients in Vancouver (Feb–Apr 2020).

Advantages of case series

• Easy to conduct; low cost.

• Useful for hypothesis generation and identifying rare conditions.

Limitations of case series

• No control group (cannot assess causality).

• May not represent all patients with that condition (selection bias).

Case-control study

• Compares people with a disease (cases) to people without it (controls) to determine whether prior exposure differs between them.

• Direction:

  • retrospective (start with outcome → look back for exposure).

• Structure (2×2 Table)

• Main Measure:

  • Odds Ratio (OR) = (a×d)/(b×c)

• Interpretation:

  • OR = 1 → no association

  • OR < 1 → exposure protective

  • OR > 1 → exposure increases risk

• Example:

  • Smokers vs non-smokers and heart attacks → OR = 9 → smoking increases odds of heart attack.

Advantages of case control study

• Efficient for rare diseases or diseases with long latency.

• Can examine multiple exposures for one outcome.

• Quicker and cheaper than cohort studies.

Limitations of case-control study

• Cannot calculate risk or incidence.

• Prone to recall bias and selection bias.

• Inefficient for rare exposures.

Cross-sectional study

• Examines exposure and disease at the same time in a defined population (“snapshot”)

• Estimates prevalence of outcomes or associations between exposure and outcome

• Features:

  • Selection independent of disease status.

  • Cannot infer temporality (which came first).

• i.e., Survey at a fertility clinic — stress vs infertility → OR = 3.0 (stressed patients 3× more likely to have ovarian infertility).

Advantages of cross-sectional study

• Quick, low cost, generalizable.

• Useful for generating hypotheses and public health planning

Limitations of cross-sectional study 

• Cannot establish cause and effect.

• Prevalent cases may represent long-duration survivors (survival bias).

Confounding

Refers to a situation where an external variable, or confounder, influences both the independent and dependent variables in a study, leading to a false association or a distorted understanding of the true relationship between the variables of interest

Confounder

An external, often unmeasured, third variable that correlates with both the independent and dependent variables in a study, distorting the observed relationship between them and leading to a potentially false conclusion about a cause-and-effect link

Ecological study

• Examines population-level exposure and disease rates rather than individual-level data.

• Unit of analysis: the population (e.g., countries, cities).

• i.e., Comparing national cabbage consumption vs COVID-19 mortality

Advantages of ecological study

• Low cost, uses existing data.

• Allows exploration of contextual and environmental factors.

Limitations of ecological study

• Ecological fallacy: associations at population level ≠ associations at individual level.

• Confounding likely; lacks detailed personal data.

Cohort studies

• A group of people (cohort) sharing a common characteristic (e.g., exposure) followed over time to see if they develop an outcome.

• 2 types— prospective cohort and retrospective cohort

Prospective cohort

Identify exposure now → follow into future for outcome.

Retrospective cohort

Both exposure and outcome already occurred; use records/databases.

Types of populations

1. Open (dynamic) 

2. Fixed 

3. Closed

Open (dynamic) population

• Members can enter or leave; exposure-defined (e.g., smokers).

• Measures incidence rate 

Fixed population

• Defined by an event (e.g., 9/11 survivors).

• Can lose members but not gain.

• Measures incidence rate 

Closed population

• No gain or loss after baseline (e.g., attendees at event).

• Measures cumulative incidence

Selecting a cohort 

• Well-defined: clear inclusion/exclusion.

• At risk: disease-free at baseline.

• Stable: likely to remain in study.

• Large enough: adequate statistical power.

• Comparable groups (exposed vs unexposed).

Selecting exposed

• Based on hypothesis and exposure frequency.

• May use general (e.g., Nurses’ Health Study) or special populations (e.g., Hiroshima survivors).

Selecting unexposed

“Counterfactual” principle → as similar as possible except for exposure.

Assessing exposure

Questionnaires, interviews, health records, employment data, environmental monitoring, biospecimens.

Assessing outcome

• Health records, physical exams, lab tests, registries, or self-reports.

• Must ensure accurate and consistent measurement.

Cohort retention 

• Loss to follow-up reduces sample size & may bias results if related to exposure/outcome (selection bias).

• Retention strategies: regular contact, incentives, accessible data collection 

Induction period

• Time between exposure and start of disease process.

• Cohort studies are feasible when this periods is short enough to observe within the study duration.

Latent period

• Time between disease onset and clinical detection.

• Cohort studies are feasible when this periods is short enough to observe within the study duration.

Calculations in cohort studies

1. Risk (cumulative incidence)

2. Risk ratio 

3. Risk difference

4. Incidence density

5. Incidence density ratio 

Risk (cumulative incidence)

• Probability of disease among exposed

• Formula: a/(a+b)

Risk ratio (RR)

• Relative risk comparing exposed vs unexposed

• (a/(a+b))/(c/(c+d))

Risk difference (RD)

• Absolute difference in risk

• a/(a+b) - c/(c+d)

Incidence density (IR) 

• Rate of new cases per person-time.

• Formula: New cases/Person-time

Incidence density ratio (IDR)

• Relative rate of disease.

• IRexposed / IRunexposed 

No association

RR/IDR = 1

Exposure increases risk

RR/IDR > 1

Exposure protective

RR/IDR < 1

Harmful

RD > 0

Protective

RD < 0

Relative risk (RR)

• Compares probabilities.

• Can exaggerate small absolute effects

Absolute risk (RD)

Shows real-world impact.

Advantages of prospective cohorts

• Establishes temporality (exposure before outcome).

• Directly measures incidence rates.

• Reduces recall bias.

• Can evaluate multiple outcomes and exposures.

Limitations of prospective cohort studies

• Expensive, time-intensive.

• Potential loss to follow-up or confounding.

Retrospective cohort studies

• Cohort study where both exposure and outcomes have already occurred;

• Data obtained from existing records:

  • Hospital or employment records

  • National registries/databases

  • Insurance or administrative data

Advantages of retrospective cohort study

• Time and cost efficient.

• Can study rare exposures and long-term outcomes.

• Can directly measure risk (unlike case-control).

Limitations of retrospective cohort study

• Limited control over data quality.

• Possible misclassification or incomplete data.

• Confounding from unmeasured variables.