Choosing a statistical test to compare pre and post intervention outcomes- which are the correct tests to use?

17 January 2023 4 5K Report

Hi all, I have a question along on a similar topic if anyone could help me.

I am writing a research proposal looking at medication review in primary care incorporating an ACB calculation. I will measure ACB scores in patients before and after the medication review. I also have a control of a medication review with no ACB calculation and I will measure the ACB scores before and after review in these patients as well.

I want to use a statistical test first to calculate if there is a statistically significant reduction in ACB in the treatment and then the control arm. I then want to compare both to see if the treatment arm is better than the control arm.

I am thinking of using the Wilcoxon Signed Rank Test and the Z test for two samples. Does this make sense?

Thank in advance,

Ieuan

Godswill Olemgbe

When comparing pre and post-intervention outcomes, there are several statistical tests that can be used depending on the type of data and the research question. Some commonly used tests include:

Paired t-test: This test is used when the data is continuous and normally distributed and the sample size is small (less than 30). It compares the means of the pre and post-intervention outcomes within the same individuals. The paired t-test assumes that the differences between the pre and post intervention outcomes are normally distributed, and that the variances of the two groups are equal. Limitations of this test include the assumption of normality and equal variances.

Wilcoxon signed-rank test: This test is also used when the data is continuous, but the sample size is small and the data is not normally distributed. It compares the medians of the pre and post intervention outcomes within the same individuals. The Wilcoxon test does not assume normality, but it does assume that the differences between the pre and post intervention outcomes are symmetric. Limitations of this test include the assumption of symmetry and the fact that it is less powerful than the paired t-test.

McNemar test: This test is used when the data is binary (e.g. yes/no, pass/fail) and the sample size is small. It compares the proportion of individuals who change from one category to another between pre and post intervention outcomes. The McNemar test does not assume normality, but it does assume that the sample size is large enough to ensure that the proportions are not too close to zero or one. Limitations of this test include the assumption of independence and the fact that it is less powerful than chi-squared test.

Repeated Measures ANOVA: This test can be used when there are multiple measurements taken pre and post-intervention and the sample size is large. It compares the means of the pre and post-intervention outcomes within the same individuals and assumes that the data is normally distributed and the variances of the groups are equal. Limitations of this test include the assumption of normality and equal variances.

Ultimately, the choice of statistical test will depend on the characteristics of the data, the research question, and the sample size. It is important to consider the assumptions of the chosen test and to be aware of its limitations.

Namir G Al-Tawil

I would to like to add to the answer of Dr. Olemgbe.

1. The McNemar-Bowker test can compare a variable with three levels (not binary) before and after the intervention.

2. The non-parametric equivalent to repeated measures ANOVA is the Friedman test.

Namir G Al-Tawil

Regarding the comparison between the two groups. First you have to calculate the difference between the pre and post score in each group, and then compare the difference, so that you can decide, in which group the difference is more. In this way, you can adjust for the possibility that the baseline scores of the two groups are different.

How to compare two groups with only two measurements?

What SPSS analysis should I run with one IV (with two levels) and two DVs?

Does the circuit modeling for planar monopole antennas using Foster canonical forms give good approximation?

Statistics: include groups of respondents made by one person for analysis?

Why am I getting 'Infinity' and 'NaN' in my mixed-mode energy release values?

Sample size determination for non-parametric tests?

Why cannot world and community leaders be gentle? Where did ethics of care go?

Statistical Tests for Quantitative Study?

What is the most prevalent drug related problem in the developed countries?

How to model rupture in geogrid in abaqus?