统计问题第78问：样本量计算（2）

Question

A randomised controlled trial investigated whether a computerised feedback device modified eating behaviour and resulted in weight loss in obese adolescents. The primary outcome was change in body mass index (BMI) from recruitment to 12 months. BMI was calculated as weight (kg)/[height (m)]²,adjusted for age and sex. The control intervention was standard lifestyle modification therapy.

The optimal sample size to compare the computerised device with standard care was calculated for the primary outcome. Using data from a previous study, the researchers predicted that the mean change in BMI at 12 months with standard therapy would be a reduction of 0.17 (SD=0.267)kg/m². For the computerised device to be considered effective, it should double this mean reduction and achieve a decrease in BMI of 0.34 kg/m² (the smallest effect of clinical interest). A total sample size of 80 children (40 in each treatment arm) at baseline was required to achieve 80% power using a two sided hypothesis test and critical level of significance of 0.05. A total of 106 individuals were randomly allocated to each study arm, but by 12 months some participants in each arm had withdrawn.

Which of the following statements, if any, are true?

a) A decrease in the magnitude of the smallest effect of clinical interest would require a larger sample size

b) The power of the study was increased as a result of the withdrawal of participants

c) The sample size should have been adjusted at baseline for possible withdrawals in order to maintain power of 80%

d) The above trial was a parallel study design

提示：这是一道多选题。

Answer

Answers a, c, and d are true; b is false.

In last week’s question, a sample size calculation for the difference in rates (percentages) of major events between two interventions was presented. The sample size calculation for comparing two interventions on the basis of the difference in the mean primary outcome proceeds in a similar fashion. In addition to calculating the proposed mean change in BMI for standard care and smallest effect of clinical interest, it was also necessary to estimate the standard deviation of the mean changes in BMI. It was necessary to assume the standard deviation was equal in both groups.

The researchers predicted that the mean change in BMI at 12 months in the standard care arm would be a reduction of 0.17 kg/m². The smallest effect of clinical interest for the computerised device was a mean reduction of 0.34 kg/m². If the smallest effect of clinical interest decreased in magnitude, the study would require a larger sample size (a is true). Large samples are more likely to reflect small differences in the population, if they exist, because large samples are more representative of the population. Small samples are less likely to reflect the distribution of individuals and changes in primary outcome in the population.

Power is the probability of detecting the smallest effect of clinical interest if it exists in the population. Last week’s question outlined how power increases as sample size increases (b is false). Generally, as sample size increases and approaches that of the population, the difference between interventions in the sample will become similar to that in the population. Therefore, if the smallest effect of clinical interest exists in the population, we are more likely to observe it as sample size increases.

Sometimes, the proposed sample size in a study is adjusted to account for participants who withdraw from the study or are lost to follow-up during the trial period. On the basis of previous research, the researchers estimated that 26% of those individuals receiving the standard intervention would withdraw or be lost to follow-up during the 12 month study period. The sample size was increased to account for withdrawals and to ensure the optimal sample size was achieved at 12 months, and, therefore, maintain the required power (c is true).

The above trial is a parallel study design (d is true), also known as a between subject design.Participants are randomly allocated to one of two interventions and receive the same intervention for the entire study period. The two groups receive the intervention and are followed up in parallel. Treatment outcomes are then compared between subjects—that is, between independent groups of patients.

所以答案是选择 a c d

每天学习一点，你会更强大！