Systematic Review and Meta-Analysis (SRMA) Tutorial

A step-by-step guide to conducting systematic reviews and meta-analyses in public health, with R code examples.

Systematic Review and Meta-Analysis (SRMA) Tutorial

This tutorial provides a comprehensive, step-by-step guide to conducting systematic reviews and meta-analyses (SRMA) in public health research. It follows the methodology outlined in the Cochrane Handbook for Systematic Reviews of Interventions.

Systematic reviews involve a rigorous, transparent process to identify, evaluate, and synthesize all relevant studies on a particular topic. Meta-analysis is the statistical combination of results from multiple studies to provide a more precise estimate of effect.

Prerequisites

Before starting, ensure you have R and RStudio installed. You’ll need the following packages:

install.packages(c("revtools", "dplyr", "stringr", "metagear", "meta", "metafor", "PRISMAstatement"))

================================

1. Define Research Question and Protocol

================================

The first step is to clearly define your research question using the PICO framework: - P: Population - I: Intervention - C: Comparison - O: Outcome

Develop a protocol that outlines your methods in advance to minimize bias.

R Code Example:

# Define PICO elements
pico <- list(
  population = "Adults with type 2 diabetes",
  intervention = "Metformin therapy",
  comparison = "Placebo or no treatment",
  outcome = "HbA1c reduction"
)

# Print protocol summary
cat("Research Question:\n")
cat("In", pico$population, ", does", pico$intervention, "compared to", pico$comparison, "result in", pico$outcome, "?\n")

================================

2. Develop Search Strategy

================================

Create a comprehensive search strategy for multiple databases (PubMed, Cochrane, Web of Science, etc.). Use Boolean operators and controlled vocabulary.

R Code Example:

library(dplyr)

# Define search terms
search_terms <- list(
  population = c("diabetes mellitus, type 2", "type 2 diabetes"),
  intervention = c("metformin", "glucophage"),
  outcome = c("glycated hemoglobin", "HbA1c")
)

# Create search string
create_search_string <- function(terms) {
  paste(sapply(terms, function(x) paste0("(", paste(x, collapse = " OR "), ")")), collapse = " AND ")
}

search_string <- create_search_string(search_terms)
cat("PubMed Search String:\n", search_string, "\n")

================================

3. Import and Deduplicate References

================================

Import references from multiple sources and remove duplicates.

R Code Example:

library(revtools)
library(dplyr)

# Import RIS files from different databases
pubmed_refs <- read_bibliography("pubmed_results.ris") %>% as.data.frame()
cochrane_refs <- read_bibliography("cochrane_results.ris") %>% as.data.frame()

# Combine all references
all_refs <- bind_rows(pubmed_refs, cochrane_refs)

# Find and remove duplicates
deduped_refs <- find_duplicates(all_refs, match_variable = "title", group_variable = "source")

cat("Original references:", nrow(all_refs), "\n")
cat("After deduplication:", nrow(deduped_refs), "\n")

================================

4. Screen Titles and Abstracts

================================

Use screening tools to assess eligibility based on titles and abstracts.

R Code Example:

library(metagear)

# Create screening dataset
screening_data <- data.frame(
  title = deduped_refs$title,
  abstract = deduped_refs$abstract,
  stringsAsFactors = FALSE
)

# Launch abstract screener (interactive)
# Note: This opens an interactive interface
# abstract_screener(screening_data, title_column = "title", abstract_column = "abstract")

# For demonstration, simulate screening results
set.seed(123)
screening_data$include <- sample(c(TRUE, FALSE), nrow(screening_data), prob = c(0.2, 0.8), replace = TRUE)

included_abstracts <- screening_data %>% filter(include)
cat("Studies included after abstract screening:", nrow(included_abstracts), "\n")

================================

5. Full-Text Review

================================

Retrieve and review full texts of potentially eligible studies.

R Code Example:

# Assuming you have full-text PDFs in a directory
full_text_dir <- "full_texts/"

# List available PDFs
pdf_files <- list.files(full_text_dir, pattern = "*.pdf", full.names = TRUE)

# Simulate full-text screening results
set.seed(456)
full_text_data <- included_abstracts
full_text_data$full_text_include <- sample(c(TRUE, FALSE), nrow(full_text_data), prob = c(0.8, 0.2), replace = TRUE)

included_studies <- full_text_data %>% filter(full_text_include)
cat("Studies included after full-text review:", nrow(included_studies), "\n")

================================

6. Data Extraction

================================

Extract relevant data from included studies using standardized forms.

R Code Example:

# Create data extraction template
extraction_template <- data.frame(
  study_id = character(),
  author_year = character(),
  study_design = character(),
  sample_size = numeric(),
  intervention_mean = numeric(),
  intervention_sd = numeric(),
  control_mean = numeric(),
  control_sd = numeric(),
  stringsAsFactors = FALSE
)

# Simulate data extraction for included studies
set.seed(789)
extracted_data <- data.frame(
  study_id = 1:nrow(included_studies),
  author_year = paste("Author", 1:nrow(included_studies), "2020"),
  study_design = sample(c("RCT", "Cohort", "Case-control"), nrow(included_studies), replace = TRUE),
  sample_size = sample(50:200, nrow(included_studies), replace = TRUE),
  intervention_mean = rnorm(nrow(included_studies), mean = -1.5, sd = 0.5),
  intervention_sd = runif(nrow(included_studies), 0.5, 1.5),
  control_mean = rnorm(nrow(included_studies), mean = -0.5, sd = 0.3),
  control_sd = runif(nrow(included_studies), 0.3, 1.0)
)

print(head(extracted_data))

================================

7. Risk of Bias Assessment

================================

Assess the risk of bias in included studies using tools like Cochrane’s RoB 2.0.

R Code Example:

# Simulate risk of bias assessment
bias_domains <- c("randomization", "allocation_concealment", "blinding", "incomplete_outcome", "selective_reporting")

set.seed(101)
bias_assessment <- data.frame(
  study_id = extracted_data$study_id,
  matrix(sample(c("Low", "High", "Unclear"), 
                nrow(extracted_data) * length(bias_domains), 
                replace = TRUE), 
         nrow = nrow(extracted_data), 
         dimnames = list(NULL, bias_domains))
)

# Calculate overall risk of bias
bias_assessment$overall_bias <- apply(bias_assessment[, bias_domains], 1, 
                                      function(x) ifelse(any(x == "High"), "High", 
                                                         ifelse(any(x == "Unclear"), "Unclear", "Low")))

print(table(bias_assessment$overall_bias))

================================

8. Meta-Analysis

================================

Perform statistical synthesis of the results.

R Code Example:

library(meta)
library(metafor)

# Calculate effect sizes (mean difference)
es_data <- escalc(measure = "MD", 
                  m1i = extracted_data$intervention_mean, 
                  sd1i = extracted_data$intervention_sd, 
                  n1i = extracted_data$sample_size / 2,  # Assuming equal groups
                  m2i = extracted_data$control_mean, 
                  sd2i = extracted_data$control_sd, 
                  n2i = extracted_data$sample_size / 2)

# Fit random-effects model
meta_model <- rma(yi = es_data$yi, vi = es_data$vi, method = "REML")

# Print results
summary(meta_model)

# Create forest plot
forest(meta_model, 
       xlab = "Mean Difference in HbA1c", 
       main = "Meta-Analysis of Metformin vs Control")

================================

9. Assess Heterogeneity and Subgroup Analyses

================================

Investigate sources of heterogeneity and perform subgroup analyses.

R Code Example:

# Test for heterogeneity
cat("Heterogeneity test:\n")
cat("Q =", meta_model$QE, ", p =", meta_model$QEp, "\n")
cat("I² =", meta_model$I2, "%\n")

# Subgroup analysis by study design
subgroup_model <- rma(yi = es_data$yi, vi = es_data$vi, 
                      mods = ~ extracted_data$study_design, method = "REML")

summary(subgroup_model)

================================

10. Assess Publication Bias

================================

Check for publication bias using funnel plots and statistical tests.

R Code Example:

# Funnel plot
funnel(meta_model, main = "Funnel Plot")

# Egger's test for asymmetry
regtest(meta_model)

================================

11. Generate PRISMA Flow Diagram

================================

Create a PRISMA flow diagram to document the review process.

R Code Example:

library(PRISMAstatement)

# Define PRISMA data
prisma_data <- data.frame(
  identification = c(records_identified = nrow(all_refs)),
  screening = c(records_screened = nrow(deduped_refs), 
                records_excluded = nrow(deduped_refs) - nrow(included_abstracts)),
  eligibility = c(full_text_assessed = nrow(included_abstracts),
                  full_text_excluded = nrow(included_abstracts) - nrow(included_studies),
                  reasons_excluded = "Not meeting inclusion criteria"),
  included = c(studies_included = nrow(included_studies))
)

# Generate PRISMA diagram
prisma_diagram <- prisma(prisma_data)
plot(prisma_diagram)

================================

12. Report Results and Conclusions

================================

Compile all findings into a comprehensive report following PRISMA guidelines.

R Code Example:

# Generate summary statistics
summary_stats <- data.frame(
  total_studies = nrow(included_studies),
  total_participants = sum(extracted_data$sample_size),
  pooled_effect = meta_model$beta[,1],
  ci_lower = meta_model$ci.lb,
  ci_upper = meta_model$ci.ub,
  heterogeneity_i2 = meta_model$I2,
  publication_bias_p = regtest(meta_model)$pval
)

print(summary_stats)

# Save results
write.csv(extracted_data, "extracted_data.csv", row.names = FALSE)
write.csv(bias_assessment, "bias_assessment.csv", row.names = FALSE)
save(meta_model, file = "meta_analysis_model.RData")

Additional Resources

• Cochrane Handbook for Systematic Reviews of Interventions
• PRISMA Statement
• ROBIS Tool for assessing risk of bias in systematic reviews
• R packages: meta, metafor, revtools, metagear, PRISMAstatement

Notes

This tutorial provides a simplified overview. In practice, each step requires careful consideration and expertise. Always consult with methodologists and follow the latest guidelines. The R code examples are illustrative and may need adaptation for your specific research question and data.