Study designs
Đã giới thiệu| English | Ý nghĩa |
|---|---|
| randomised controlled trial | randomised controlled trial |
| cohort study | cohort study |
| relative risk | relative risk |
| case-control study | case-control study |
| odds ratio | odds ratio |
The right design for the question
- Does a new vaccine work? Did a factory chemical cause a rare cancer? You cannot answer both with the same kind of study.
- Each design answers some questions well and others not at all, and the design decides which statistic you may calculate.
Randomised controlled trial
- A randomised controlled trial 随机对照试验 assigns the exposure by chance, then follows both groups forward.
- Chance balances known and unknown differences between groups, so it is the strongest single test of a treatment. It is only possible when you can ethically assign the exposure.
Cohort study
- A cohort study 队列研究 starts with exposed and unexposed people and follows them forward to count outcomes.
- Because it samples by exposure, it measures risks directly and gives a relative risk 相对危险度: $RR = \dfrac{a/(a+b)}{c/(c+d)}$.
(Synthetic) Exposed 300 with 30 events; unexposed 300 with 10 events. What is the relative risk?
Độ khó 2/5 · 140 xu + phần thưởng nhỏ khi hoàn thành lần đầu tiên hôm nay
0.10 ÷ 0.0333 = 3.0.
Case-control study
- A case-control study 病例对照研究 starts with people who have the outcome (cases) and without it (controls), then looks back at exposure.
- It suits rare diseases. Since the researcher sets the number of cases, risks cannot be found; only the odds ratio 比值比 $OR = ad/bc$ is valid.
A relative risk can be calculated directly from a case-control study.
Độ khó 1/5 · 100 xu + phần thưởng nhỏ khi hoàn thành lần đầu tiên hôm nay
The researcher chose how many cases and controls to include, so only the odds ratio is valid.
(Fictional) Cases: 40 exposed, 20 unexposed. Controls: 20 exposed, 40 unexposed. What is the odds ratio?
Độ khó 2/5 · 140 xu + phần thưởng nhỏ khi hoàn thành lần đầu tiên hôm nay
OR = (40 × 40) ÷ (20 × 20) = 4.0.
Cross-sectional and ecological
- A cross-sectional study takes one snapshot: exposure and outcome are measured together, so which came first is unclear.
- An ecological study compares groups, not individuals, so a link between countries may not hold for people.
Match each question to the best design.
Độ khó 2/5 · 140 xu + phần thưởng nhỏ khi hoàn thành lần đầu tiên hôm nay
The question and the rarity of the outcome choose the design.
Order these designs from strongest to weakest protection against bias.
Độ khó 3/5 · 200 xu + phần thưởng nhỏ khi hoàn thành lần đầu tiên hôm nay
This is the usual ladder; each step adds a source of bias.
Countries that eat more chocolate win more Nobel prizes. What is the main problem with concluding that chocolate makes people clever?
Độ khó 1/5 · 100 xu + phần thưởng nhỏ khi hoàn thành lần đầu tiên hôm nay
An ecological link between groups may not hold for people.
Worked example: a synthetic cohort
- (Synthetic) Exposed 200 with 20 events; unexposed 200 with 10: risks 0.10 and 0.05, so RR = 2.0.
- Attributable fraction in the exposed = (RR − 1)/RR = 1/2 = 50%: half their events are due to the exposure, if the study is unbiased.
The relative risk is 4. What is the attributable fraction among the exposed, in %?
Độ khó 2/5 · 140 xu + phần thưởng nhỏ khi hoàn thành lần đầu tiên hôm nay
(4 − 1) ÷ 4 = 0.75, so 75%.
Worked example: a case-control table
- (Fictional) 30 of 40 cases were exposed and 20 of 60 controls were exposed. So a = 30, b = 20, c = 10, d = 40.
- $OR = (30 \times 40) \div (20 \times 10) = 6.0$. When the outcome is rare, the OR is close to the RR.
Three common slips
- Computing RR from a case-control study. The researcher chose the denominators.
- Reading a cross-sectional association as cause. Exposure and outcome were measured together.
- Thinking "prospective" removes confounding. Only randomisation balances the unknowns.
Recap
- Trial (assigned by chance) → cohort (by exposure, RR) → case-control (by outcome, OR) → cross-sectional → ecological.
- $RR = \dfrac{a/(a+b)}{c/(c+d)}$; $OR = ad/bc$; rare outcome: OR ≈ RR.
- Attributable fraction in the exposed = (RR − 1)/RR.