University:
University of Alberta
Course:
NURS 211 | Statistics and Knowledge Management
Academic year:

2025
Views:

355

Pages:

5
Author:

Alya A

Lecture 12: Systematic Reviews and Meta-analysis Anila’s Slides on Systematic Reviews → TBH I don’t think she gave us any real info Knowledge Synthesis Reviews Systematic Reviews - A properly conducted systematic review faithfully summarizes the evidence from all relevant studies on the topic of interest and does so concisely and transparently - Steps for a systematic review: 1. Develop a research question 2. Create a protocol a. Inclusion and exclusion criteria 3. Literature searching 4. Study selection 5. Evaluate and appraise 6. Extract and meta-analysis 7. Report results Back to Simon’s Slides Systematic Reviews and Meta-Analysis - A systematic review allows a more reliable and clear picture to emerge - Required sample size is often hard to achieve for systematic reviews - In order to detect such an effect with 90% certainty over 10,000 patients in each treatment group would be needed - Critical appraisal is an important step in systematic reviews - - Need to formally assess quality of primary studies Inadequate quality of studies may distort the results from systematic reviews and meta-analysis What is a Meta-Analysis - Meta-analysis refers to the analysis of analyses - Aggregating outcomes from other studies through the use of statistics to then give us a summary statistic - Odds ratio - Mean difference between interventions - Relative risk - Meta-analysis helps identify patterns among study results, increases the power of the study, and can help see the overall effectiveness of an intervention by comparing various studies Steps in a Meta-Analysis 1. Conduct a systematic review 2. Determine the effect size (magnitude of observed effect) 3. Determine the heterogeneity of the included studies 4. Determine the type of meta-analysis (we don’t need to know the different types tho) Test for Heterogeneity - Tested by calculating the Chi-square (X²) or an index called I-squared (I²) - Cochran’s Q (Chi-square test) whether the outcomes are significantly different from each other - Want no significant difference between study outcomes - P is set at 0.1 (because hard to detect differences) - If P > 0.1 studies should not be combined - I² is based on Chi-square, gives a 0-100% (= % variance between studies) - 0-30% is okay - >30% indicates increasing heterogeneity (may then want to do a subgroup analysis to exclude certain studies) - 100% extreme differences between the studies (very bad) - Rule of thumb: >30% need to conduct sub-analysis Interpreting Forest Plots (Example 1) - Note the I² percentage and p-value in blue. The p-value is significant but the I² shows lots of heterogeneity which is not good - The diamond is highlighted in green. It is called the aggregate diamond and shows us the aggregate outcomes of all the studies included in the meta-analysis (or - is it a systematic review?) The top and bottom values of the diamond show the point estimate (1.46), and the width of the diamond shows the confidence interval (1.22, 1.75). You can see that the diamond does not cross the 1 line so thus it is significant. Because the diamond is on the right, it indicates an increased risk. Each of the studies are indicated by the different lines on the plot. The lines with a larger width indicate a larger confidence interval, and are less certain of the outcome of that individual study. Some studies cross the 1 line but thats okay I think Forest Plot Example 2 (Showing Mean Difference) - Tbh this one makes very little sense to me - The p-value is 0.04 which is barely significant - The heterogeneity is low (1.4%) which I believe is a good thing - The diamond confidence interval does not contain 1, but it is very close to 1 so it is also barely significant, and since the diamond is on the left it shows a decreased risk - Mean difference = intervention group (exposed group) value - control group (non exposed group) value - Null = line of no difference/no effect equals 0 when we are referring to mean difference forest plots (instead of 1 in the above example) REMAINING SLIDES ARE REVIEW FOR FINAL Determining Which Statistical Test to Use 1. What is the purpose of the study? 2. What level of data do I have? 3. How many groups? 4. Are the groups related? 5. What's the sample size? 6. Is the data normally distributed? Central Limit Theorem - If the population is normal, the distribution of the sample means is also normal - If the population is not normal, the distribution of sample means approaches a normal distribution as the sample gets larger (need 30+ in sample) Parametric and Non-Parametric Tests - Parametric - Data must be interval or ratio - - Data normally distributed - Using the average - Has the most power (ability to detect a difference when one truly exists) Non-parametric - Data not normally distributed - Using median - Can be used for small sample sizes - Has less power Independent T-test Purpose: to compare means of 2 groups Mann-Whitney U Test Purpose: to compare distribution (not means) of 2 independent samples Assumptions: - Independent variable is dichotomous - The 2 groups are independent - Dependent variable is continuous - Dependent variable is normally distributed Assumptions: - Independent variable is dichotomous - The 2 groups are independent - Dependent variable is at least ordinal - Total sample size at least 8 (or more) Paired T-test Purpose: to compare the means of 2 related/dependent groups (repeated measures tests) Wilcoxon Matched-Pairs Signed Rank Test Purpose: to compare the medians of 2 related/dependent groups when the dependent variable is not normally distributed Assumptions: - The independent variable is dichotomous - The 2 groups are dependent - The dependent variable is continuous - The dependent variable is normally distributed Assumptions: - The independent variable is dichotomous - The 2 groups are dependent - The dependent variable is ordinal, interval, or ratio - There are at least 5 pairs of measurements ANOVA Purpose: compare the means of 3 or more independent samples Kruskal-Wallis H-test Purpose: compare ranked scores of 3 or more independent samples that are not normally distributed Assumptions: - The independent variable has 3 or more categories - BUT: One-way ANOVA only has 2 groups - The groups are independent - Dependent variable is normally distributed - Dependent variable is continuous Assumptions: - The independent variable has 3 or more categories - The groups are independent - Dependent variable is not normally distributed - Dependent variable is at least ordinal Chi-Square Test Purpose: compares expected to observed to see associations between 2 things Assumptions: - The data is frequency data - The measures are both nominal - There is an adequate sample size (>3) - The measures are independent Correlations - Pearson’s r (parametric) or Spearman’s rho (nonparametric) - Once you know the correlation coefficient you can determine the strength of associations - Can be positive or negative association - 0.10-0.30 = small - 0.31-0.50 = medium - 0.51-1.0 = large - You need the p-value to determine statistical significance - Variance is calculated by squaring the correlation coefficient (called coefficient of determination) = r² also expressed as a % (percentage of explained variance)