How study design is tested on Step 1
Identify the design from how participants were selected and what direction the study looks in time.
| Phase | Participants | Main question |
|---|---|---|
| I | Small number, often healthy volunteers | Is it safe? Pharmacokinetics and dosing |
| II | Patients with the disease | Does it work? Optimal dose, side effects |
| III | Large patient groups, randomized | Is it better than or equal to standard care? |
| IV | Population after approval | Rare or long-term adverse effects (postmarketing) |
Blinding (single, double, or triple) reduces observer and participant bias in trials.
Bias
Bias is systematic error that pushes results in one direction.
| Bias | What happens | Classic setting |
|---|---|---|
| Selection (e.g., Berkson, healthy worker) | Study groups do not represent the population | Hospital-based or employed samples |
| Attrition (loss to follow-up) | People who leave differ from those who stay | Cohort studies, trials |
| Recall | Cases remember exposures differently than controls | Case-control studies |
| Observer-expectancy (Pygmalion) | Investigator expectations influence results | Unblinded studies |
| Hawthorne effect | Participants change behavior because they are observed | Studies of practice or behavior |
| Lead-time | Earlier detection makes survival look longer without changing death date | Screening studies |
| Length-time | Screening finds more slow-growing, less aggressive disease | Screening studies |
Reduce bias with randomization, blinding, standardized measurement, and complete follow-up.
Confounding vs effect modification
Both involve a third variable, but only one is a flaw.
Hypothesis testing
Know the errors, power, and which test fits which data.
- t-test: compares means of two groups.
- ANOVA: compares means of three or more groups.
- Chi-square: compares proportions of categorical outcomes.
- Pearson correlation (r): strength of linear relationship from −1 to +1; correlation does not prove causation.
Worked example
Name the design and its main threat.
Researchers identify 100 infants with a birth defect and 200 without, then ask their mothers about medications taken during pregnancy. Mothers of affected infants report more drug exposures. What is the design and its main bias?
- Participants were chosen by outcome (birth defect present or absent).
- Exposure was assessed by looking back in time: a case-control study.
- Mothers of affected infants may search their memories harder.
- This differential memory is recall bias; the measure of association is the odds ratio.
Answer. Case-control study at risk of recall bias; report the odds ratio.