Clinical trials and systematic reviews
導入済み| English | 意味 |
|---|---|
| systematic review | systematic review |
| allocation concealment | allocation concealment |
| blinding/ˈblaɪndɪŋ/ | blinding |
| intention-to-treat analysis | intention-to-treat analysis |
| number needed to treat/ˈnʌmbə ˈniːdɪd tə triːt/ | number needed to treat |
| heterogeneity | heterogeneity |
How many must take it for one to benefit?
- A drug cuts heart attacks from 20% to 12%. That sounds impressive, but 13 people must take it for one of them to benefit.
- Trials produce the evidence and systematic reviews combine it. Their vocabulary appears in every abstract.
Randomisation and concealment
- Randomisation assigns treatment by chance, which balances known and unknown factors between the groups.
- Allocation concealment 分配隐藏 keeps the next assignment hidden from recruiters, so they cannot steer patients into a group.
Blinding
- Blinding 盲法 hides the assignment from participants, clinicians and outcome assessors, so expectations do not shape the results.
- Randomisation balances groups at the start; blinding protects how they are treated and measured. A trial can have one without the other.
A randomised trial is always blinded.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
Randomisation and blinding are separate; a surgical trial may be randomised but not blinded.
Intention to treat
- Intention-to-treat analysis 意向性分析 counts everyone in the group they were randomised to, even those who stopped treatment.
- It keeps the protection of randomisation and answers the real-world policy question. Per-protocol analysis breaks that protection.
Put the parts of a trial in order.
Difficulty 3/5 · 200 coins + a small bonus on first completion today
Each step protects against a different bias.
In an intention-to-treat analysis, where is a patient counted who stopped the drug after a week?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
Everyone is analysed as randomised.
ARR, RRR and NNT
- Absolute risk reduction (ARR) = control risk − treated risk. Relative risk reduction (RRR) = ARR ÷ control risk.
- Number needed to treat 需治疗人数 (NNT) = 1/ARR, rounded up. A lower baseline risk gives a smaller ARR and a larger NNT.
(Fictional) Control risk 15%, treated risk 10%. What is the NNT?
Difficulty 2/5 · 140 coins + a small bonus on first completion today
ARR = 5% = 0.05, so NNT = 1 ÷ 0.05 = 20.
(Fictional) Control risk 15%, treated risk 10%. What is the RRR in %? Give one decimal place.
Difficulty 2/5 · 140 coins + a small bonus on first completion today
5 ÷ 15 ≈ 0.333, so 33.3%.
Reading a forest plot
- In a systematic review 系统综述, each trial is a square (size = weight) with a line (its CI). The diamond is the pooled estimate.
- Heterogeneity 异质性 ($I^2$) measures how much the trials disagree; a high $I^2$ means asking why before trusting the diamond.
Match each part of a forest plot to its meaning.
Difficulty 2/5 · 140 coins + a small bonus on first completion today
Read the squares, then the diamond, then $I^2$.
Worked example: the NNT
- (Fictional trial) Control 20%, treated 12%: ARR = 8%, RRR = 8/20 = 40%, NNT = 1/0.08 = 12.5, rounded up to 13.
- In a clinic with a 5% baseline and the same RRR: ARR = 2%, NNT = 50.
A treatment keeps the same RRR in a lower-risk clinic. What happens to the NNT?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
A smaller baseline gives a smaller ARR, so more people must be treated.
Worked example: reading the diamond
- (Fictional) Five trials with RRs from 1.1 to 2.4; pooled RR 1.7 (1.3–2.1), $I^2$ = 22%. The diamond lies right of 1 and excludes it.
- Low $I^2$ means the trials broadly agree. With $I^2$ = 78%, you would first ask why they disagree.
Three common slips
- Equating randomisation with blinding. Chance balances groups; blinding balances expectations.
- Trusting per-protocol as the honest analysis. It breaks randomisation.
- Trusting the diamond despite high heterogeneity.
Recap
- Randomise (concealed) → blind → analyse by intention to treat.
- ARR = control − treated; RRR = ARR/control; NNT = 1/ARR, rounded up.
- Forest plot: weighted squares, CI lines, pooled diamond, $I^2$ for disagreement.