Cohort Sample Size Calculations

Summary

This video explains sample size calculations for cohort studies, focusing on binary and continuous outcomes. For binary outcomes, it covers key parameters like prevalence, relative risk, and the ratio of exposed to unexposed individuals. The example used involves physical activity and blood pressure. For continuous outcomes, the video shifts to comparing means, with an example of fiber intake and BMI. The video also touches on handling fixed sample sizes, confounders, and person-time in cohort data, offering guidance on calculating the necessary sample size for different study designs and scenarios.

Takeaways

• 😀 Binary outcome cohort studies require specific parameters like confidence level (α), power (β), prevalence of the outcome (p), and the ratio of exposed to unexposed individuals (r).
• 😀 In binary outcome studies, the magnitude of difference (d) refers to the risk difference you want to detect between exposed and unexposed groups.
• 😀 The total sample size in a cohort study is determined by the number of exposed individuals and the ratio of unexposed to exposed (r), not just by the number of exposed individuals alone.
• 😀 For binary outcomes, you need to calculate the risk of disease in both the exposed and unexposed groups (p1 and p0) using a weighted average formula.
• 😀 In continuous outcome studies, parameters like variance (not proportion) are used to calculate the sample size, with the goal being a detectable difference in means.
• 😀 Continuous outcomes require calculations using variance and absolute differences in means (e.g., BMI), instead of proportions like in binary outcome studies.
• 😀 When calculating sample size for a continuous outcome, be sure to convert any given standard deviation into variance if necessary.
• 😀 To calculate sample size for continuous outcomes, you must also consider the prevalence of exposure and the difference in means you aim to detect.
• 😀 If confounders are present, it's recommended to increase the sample size by 10% for each strong confounder to maintain accurate results.
• 😀 When working with multiple exposure groups (e.g., low, medium, high exposure), sample sizes must be adjusted to account for the full population, not just the comparison between two extreme groups.

Q & A

What are the two types of cohort studies discussed in the video?

The two types of cohort studies discussed are those with binary outcomes (e.g., having lung cancer vs not) and continuous outcomes (e.g., cholesterol concentration or BMI).

How do you calculate the total sample size for a cohort study with a binary outcome?

To calculate the total sample size for a binary outcome, you first determine the number of exposed individuals using the relevant formula, then adjust based on the ratio (r) of unexposed to exposed individuals in the population.

What is the significance of the parameters z_alpha and z_beta in the sample size calculation?

The parameters z_alpha and z_beta represent the critical values for the confidence level (typically 1.96 for 95% confidence) and power (typically 0.84 for 80% power) respectively, which are essential for calculating the sample size in a cohort study.

What does the parameter 'r' represent in the formula?

The parameter 'r' represents the ratio of unexposed to exposed individuals in the population, which is important for determining how many unexposed individuals are needed in the sample.

In the given example, what is the prevalence of high blood pressure and the proportion of people engaging in low physical activity?

In the example, the prevalence of high blood pressure is 5%, and 40% of people engage in low physical activity.

How do you calculate the risk of disease in the exposed and unexposed groups (p1 and p0)?

The risk of disease in the exposed group (p1) is the proportion of exposed individuals who are diseased. Similarly, p0 is the risk of disease in the unexposed group, calculated as the proportion of unexposed individuals who are diseased.

What is the purpose of calculating the 'magnitude of difference' (d) in the sample size calculation?

The 'magnitude of difference' (d) represents the difference in proportions or means that you want to detect between the exposed and unexposed groups, such as a risk difference or difference in BMI.

How does the calculation change when dealing with a continuous outcome, such as BMI?

For continuous outcomes, the sample size calculation focuses on the difference in means between groups rather than proportions. You use the variance of the outcome instead of prevalence to determine the sample size.

In the continuous outcome example, how is the variance of BMI used in the sample size calculation?

The variance of BMI in the population is used to determine how much variability exists in the outcome. This variance is then incorporated into the formula to calculate the sample size required to detect the desired difference in means.

What adjustments should be made if there are confounders or if the study uses person-time data?

If there are strong confounders, the sample size should be inflated by 10% for each confounder. For studies using person-time data, an estimate of the prevalence can be obtained by multiplying by the mean follow-up time, as time cancels out in the denominator.