Study Design in Pharmacoepidemiology

What is Pharmacoepidemiology?

Overview: The Study Design Framework

Step 1 - Define the Research Question

• The exposure (a drug, drug class, dose, or duration)
• The outcome (adverse drug reaction, therapeutic benefit, death, hospitalization)
• The target population (e.g., elderly, pregnant women, specific disease group)
• The time frame (prospective vs. retrospective)
• The causal question (association or causation?)

Step 2 - Descriptive Study Designs (Hypothesis Generating)

2a. Case Reports and Case Series

• A case report describes a single patient experiencing a drug-related event.
• A case series describes a group of patients with a similar experience.
• Strength: Fast, inexpensive, useful for signaling rare or unexpected adverse events.
• Limitation: No denominator, cannot rule out chance or bias, cannot determine incidence.
• Example: First reports linking thalidomide to phocomelia were case reports.

2b. Ecological (Secular Trend) Studies

• Compare population-level exposure data (e.g., aggregate drug sales) with population-level outcomes (e.g., disease rates) across geographic regions or time periods.
• Strength: Quick support for or against a hypothesis; useful for drug utilization patterns.
• Limitation: Ecological fallacy - associations at the group level may not apply to individuals.

2c. Cross-Sectional Studies

• Measure drug exposure and health outcomes at a single point in time in a defined population.
• Provides a "snapshot" - useful for measuring prevalence of drug use and associated outcomes.
• Strength: Can establish prevalence; relatively quick.
• Limitation: Cannot establish a temporal relationship between exposure and outcome; susceptible to prevalent user bias.

2d. Drug Utilization Studies (DUS)

• A form of descriptive study specific to pharmacoepidemiology.
• Examines how drugs are prescribed, dispensed, and used in populations, including patterns, trends, and deviations from guidelines.
• Uses the Defined Daily Dose (DDD) as the unit of measurement (WHO standard).
• Subdivided into: prescribing studies, dispensing studies, and patient-use studies.

Step 3 - Analytic Observational Study Designs (Hypothesis Testing)

3a. Cohort Studies

• A group (cohort) of individuals exposed to a drug and a comparison group not exposed are followed over time to see who develops the outcome.
• Prospective: Exposure is identified and participants followed forward in time.
• Retrospective (Historical): Uses existing records; exposure and outcomes already occurred.

1. Define the study cohort (new users vs. prevalent users - "new user design" is preferred to avoid prevalent user bias)
2. Define and measure drug exposure (start, stop, dose)
3. Define the outcome (e.g., myocardial infarction, hospitalisation)
4. Follow participants until the outcome, censoring, or end of study
5. Calculate Relative Risk (RR) or Incidence Rate Ratio (IRR)

3b. Case-Control Studies

• Cases (patients who developed the outcome) are identified and compared to controls (patients who did not), and prior drug exposure is measured in both groups.

1. Define the case (outcome definition, e.g., upper GI bleed)
2. Select controls from the same source population (population-based, hospital-based)
3. Measure past drug exposure in both cases and controls (e.g., via prescriptions, pharmacy records)
4. Match or adjust for confounders
5. Calculate the Odds Ratio (OR) as an estimate of relative risk

3c. Nested Case-Control Study

• Cases and controls are both drawn from within a pre-defined cohort (e.g., an insurance database or registry).
• All cases occurring in the cohort are identified; controls are randomly sampled from the cohort risk set at the time each case occurs (risk-set sampling).
• Advantages: More efficient than a full cohort analysis; eliminates confounding factors outside the source cohort; allows use of electronic health records databases efficiently.
• Calculates an Incidence Density Ratio (equivalent to RR under risk-set sampling).

3d. Case-Cohort Study

• Similar to nested case-control but the control group (sub-cohort) is randomly selected from the full cohort at baseline rather than matched to individual cases.
• Allows study of multiple outcomes using a single sub-cohort.
• The sub-cohort may include some future cases (overlap is handled statistically).

3e. Case-Only Designs (Self-Controlled Studies)

• Case-Crossover Study: Compares a patient's drug exposure during a "hazard window" (just before the adverse event) to exposure during "control windows" (earlier time periods). Ideal for transient exposures and acute outcomes (e.g., does a sleeping pill taken last night cause a fall today?).
• Self-Controlled Case Series (SCCS): Compares the rate of outcomes during risk periods (when the person is exposed to the drug) vs. control periods (unexposed periods) within the same individual.
• Limitation: These designs are biased if the exposure is long-term/chronic or if the outcome affects future exposure (e.g., a side effect causes discontinuation of the drug).

Step 4 - Experimental Designs

4a. Randomized Controlled Trial (RCT)

• The gold standard for establishing causation. Participants are randomly allocated to the drug (treatment arm) or placebo/comparator (control arm).
• Parallel design: Each participant receives only one treatment.
• Crossover design: Each participant receives both treatments in sequence (each serves as their own control). Requires a washout period.
• Factorial design: Two or more interventions are tested simultaneously (e.g., drug A vs. no drug A, AND drug B vs. no drug B - in a 2×2 arrangement).

• Short duration - miss long-term adverse events
• Small sample sizes - miss rare adverse events (one in 10,000 or rarer)
• Strict inclusion/exclusion criteria - not representative of real-world populations (elderly, pregnant, multi-morbid patients excluded)
• Ethical constraints - cannot randomize to harmful exposures

4b. Cluster Randomized Trial

• Groups (clusters) such as hospitals, clinics, or communities are randomized rather than individuals.
• Used when intervention is delivered at the group level (e.g., prescriber education programs, formulary changes).

Step 5 - Emerging Design: Target Trial Emulation

1. Step 1: Design the "target trial" - specify eligibility criteria, treatment strategies, assignment procedure, follow-up, outcomes, causal contrast, and analysis plan.
2. Step 2: Use available observational data (e.g., electronic health records, claims databases) to emulate each element of that target trial.
3. This approach minimizes biases (e.g., immortal time bias, prevalent user bias) that are common in naive observational analyses.

Step 6 - Data Sources Used in Pharmacoepidemiology

Step 7 - Addressing Bias and Confounding

• Confounding by indication: Drug users are sicker than non-users - addressed by active comparator, new user design, or propensity score methods.
• Selection bias: Who gets into the study; includes prevalent user bias (resolved by new user designs).
• Information/Misclassification bias: Incorrect exposure or outcome measurement.
• Immortal time bias: Time during which the patient is alive but incorrectly assigned to an exposure group.
• Methods: Propensity score matching/weighting, instrumental variable analysis, negative control outcomes, restriction strategies.

Summary Hierarchy of Evidence

Meta-analysis / Systematic Reviews (highest)
         ↓
Randomized Controlled Trials
         ↓
Cohort Studies
         ↓
Case-Control Studies (Nested > Traditional)
         ↓
Cross-sectional Studies
         ↓
Case Series / Case Reports (lowest, hypothesis-generating)