匯東華統計顧問有限公司

"匯東華-認真作好每件事"
~統計，不再是阻力，而是助力~

BMJ小小統計問題(6)

題目：

以下哪些因素會影響臨床試驗的樣本數估算？ (複選)

Which of the following will affect the sample size calculation for a clinical trial? (There may be more than one correct statement.)

a) Power (檢定力)

b) Significance level (顯著水準)

c) Size of effect to detect (要偵測的效果量)

d) Variability of primary outcome (主要結果的變異性)

Answer：

These are the four pieces of information required for a sample size calculation.

這四項都是計算樣本數所需的資訊。

說明:

Larger sample sizes will result in increased power.

樣本量愈大，檢定力愈高。

Formal “significance” may be set at any P value. The smaller the P value chosen as indicating significance the larger will be the required sample size.

正式的“顯著性”可以設在任何P值。作為顯著性的P值愈小，所需的樣本數就愈大。

Large clinical differences are easier to detect that small ones. The smaller the clinical effect that it is desired to detect, the larger will be the required sample size.

臨床上的差異愈大，被偵測出差異所需要的樣本數就愈小。反之，當需要檢測的臨床效果愈小，所需的樣本數就愈大。

如果結果測量具高度變異性，就會像“噪音”，讓臨床效果的“信號”難以被檢測出來。連續測量中的變異性可以用標準差來評估。該數值愈大，檢測一定的臨床效果量所需要的樣本量就愈大。有時試驗在設計上會使用嚴格的納入標準來減少個案的變異性，以確保試驗中的族群更相似（例如，相似的疾病分期，相似的年齡，相似的種族）。這將減少所需的樣本數，在以降低更廣泛的概化（generalisability）為代價。

Showme補充說明：概化（generalisability）指的是外推效度(外部效度)，亦即當臨床試驗透過嚴格的納入排除標準選擇同質性高的個案來減少所需樣本數時，內部效度（因果推論）雖然會增加，但是其結果可能很難外推到其他族群，就是臨床試驗結果可能沒法應用到其他族群上。

所以答案為：以上皆是

原文：

If the outcome measure is highly variable this will act like “noise,” making it difficult to detect the “signal” of the clinical effect. Variability in a continuous measure may be assessed by the standard deviation. The larger this is, the larger will be the required sample size to detect a certain clinical effect. Trials are sometimes designed to reduce variability by using strict inclusion criteria to ensure that people in the trial are more similar (for example, similar stage of illness, similar age, similar ethnicity). This will reduce the required sample size at the expense of reducing wider generalisability.