! Preliminary data & results ! 15.11.2025
Screening, in duplicate, of a first batch yielded 14 IPDMAs that included at least 1 CRT in their meta-analysis. Total of 43 CRTs across the 14 IPDMAs.
Packages
req_pkgs <- c("readxl",
"tidyverse",
"here",
"ggplot2",
"ggalluvial", # sankey
"gtsummary",
"forcats", # ordering function
"scales" # likert plot
)
install_if_missing <- function(pkgs){
for(p in pkgs){
if(!requireNamespace(p, quietly=TRUE)){
install.packages(p, repos="https://cloud.r-project.org")
}
library(p, character.only=TRUE)
}
}
install_if_missing(req_pkgs)# Import
df <- read_excel(here("IPDMA_CRT_extraction_10112025.xlsx"))
df$`State Of Play Covidence #` <- as.factor(df$`State Of Play Covidence #`)
# Recode the comparator group description levels
df <- df %>%
mutate(
`Comparator category` = case_when(
str_detect(`Describe Comparator`, regex("placebo", ignore_case = TRUE)) ~ "Placebo or usual care",
`Describe Comparator` %in% c("iron and folic acid supplements only", "Saline") ~ "Active Comparator",
TRUE ~ "Usual care"
)
)
# Recode the intervention group description levels
df <- df %>%
mutate(
`Intervention category` = case_when(
`Intervention category` == "Behavioural" ~ "Behavioural (lifestyle, psychological, exercise)",
`Intervention category` == "Rehabilitation" ~ "Behavioural (lifestyle, psychological, exercise)",
TRUE ~ `Intervention category`
)
)
# Rename the medical field level
df <- df %>%
mutate(
`Medical field` = case_when(
`Medical field` == "Cardiovascular/Cardiac" ~ "Cardiovascular",
TRUE ~ `Medical field`
)
)
# Rename variables
df <- df %>%
rename(
`ICC reported` = `ICC or any other measure of between-cluster variability for at least one of the pooled outcomes (incuding CRTs) reported`,
`CRTs labeled as CRTs` = `CRTs flagged as CRT in ‘Characteristics of included studies’ table`,
`Analysis framework` = `Analysis framework for the main IPDMA analysis`,
`Analysis approach` = `Main IPDMA analysis approach`,
`Treatment effect model` = `Main IPDMA analysis model regarding treatment effect`,
`Accounted for two-level clustering` = `Do the authors account for the two-level clustering, i.e., CRT-level (clusters within CRT(s)) and IPDMA-level (pooling of trials)?`,
`CRT-level estimator` = `If two-stage, what estimator do the authors use to account for clustering at the CRT-level`,
`Small sample correction` = `Do the authors report any correction in case of low cluster number in any of the involved CRTs (\"small sample correction\")?`,
`CRT vs IRT analysis` = `Any subgroup analysis/meta-regression performed based on study design (CRTs vs individual RCTs)?`,
`Covariate adjustment` = `Any covariate adjustment applied for the main analysis?`,
`Risk of bias` = `Any risk of bias assessment done?`,
`CRT-specific risk of bias` = `If RoB done, any CRT-specific risk of biases assessed?`,
)
# Recode the CRT-specific risk of bias variable by explicitly showing Unknown
df <- df %>%
mutate(
`CRT-specific risk of bias` = case_when(
is.na(`CRT-specific risk of bias`) ~ "Not applicable",
TRUE ~ `CRT-specific risk of bias`))
# Create long dataset to work with CRT/IPDMA format
df_long <- df %>%
pivot_longer(
cols = matches("^CRT[0-9]+:"),
names_to = c("CRT", ".value"),
names_pattern = "(CRT[0-9]+): (.*)"
)
df_long <- df_long %>%
filter(!is.na(`name/ID`))
# Rename the consent variable
df_long <- df_long %>%
mutate(
`Consent procedure` = case_when(
`Consent procedure` == "Passive consent (Participation occurs unless they actively refuse “opt-out”)" ~ "Passive consent",
`Consent procedure` == "Waiver of consent (No consent sought/communicated to participants)" ~ "Waiver of consent",
`Consent procedure` == "Active consent (Participant must explicitly agree “opt-in”)" ~ "Active consent",
TRUE ~ `Consent procedure`
)
)baseline_vars <- c(
# "Location of corresponding author",
# "Cochrane review?",
"Medical field",
"Intervention category",
"Comparator category"
)
baseline_table <-
df %>%
select(all_of(baseline_vars)) %>%
mutate(across(
everything(),
~ fct_infreq(factor(.))
)) %>%
tbl_summary(
missing = "ifany",
type = all_categorical() ~ "categorical",
statistic = all_categorical() ~ "{n} ({p}%)"
) %>%
modify_header(label = "**IPDMA Characteristics**") %>%
bold_labels()
baseline_table| IPDMA Characteristics | N = 141 |
|---|---|
| Medical field | |
| Cardiovascular | 4 (29%) |
| Paediatric | 2 (14%) |
| Psychiatry | 2 (14%) |
| Dermatology | 1 (7.1%) |
| Endocrinology | 1 (7.1%) |
| Gynecology / Obstetrics | 1 (7.1%) |
| Intensive Care | 1 (7.1%) |
| Primary Care / Public Health | 1 (7.1%) |
| Rheumatology | 1 (7.1%) |
| Intervention category | |
| Behavioural (lifestyle, psychological, exercise) | 7 (50%) |
| Drug/Product/Device | 7 (50%) |
| Comparator category | |
| Usual care | 9 (64%) |
| Placebo or usual care | 3 (21%) |
| Active Comparator | 2 (14%) |
| 1 n (%) | |
# summarized data
alluvial_df <- df %>%
select(`Intervention category`, `Comparator category`, `Medical field`) %>%
filter(complete.cases(.)) %>%
count(`Medical field`, `Intervention category`, `Comparator category`)
# First order by total frequency
medical_freq <- alluvial_df %>%
group_by(`Medical field`) %>%
summarise(total = sum(n), .groups = "drop") %>%
arrange(desc(total))
# Define custom priority for those with total = 1
custom_order_1 <- c(
"Primary Care / Public Health",
"Rheumatology",
"Endocrinology",
"Intensive Care",
"Gynecology / Obstetrics",
"Dermatology"
)
# Create final ordering vector:
# Psychiatry first among the >=2 group, then custom order for the 1-count group
final_medical_order <- c(
medical_freq$`Medical field`[medical_freq$total > 1],
custom_order_1
)
# reorder Intervention category with Drug/Product/Device on top
intervention_levels <- alluvial_df %>%
distinct(`Intervention category`) %>%
pull()
final_intervention_order <- c("Drug/Product/Device", setdiff(intervention_levels, "Drug/Product/Device"))
# Apply factor releveling
alluvial_df <- alluvial_df %>%
mutate(
`Medical field` = factor(`Medical field`, levels = final_medical_order),
`Intervention category` = factor(`Intervention category`, levels = final_intervention_order))
# determine axis tick range
max_n <- 14
ggplot(alluvial_df,
aes(axis1 = `Medical field`,
axis2 = `Intervention category`,
axis3 = `Comparator category`,
y = n)) +
geom_alluvium(aes(fill = `Intervention category`), width = 0.25) +
geom_stratum(width = 0.25) +
geom_label(stat = "stratum", aes(label = after_stat(stratum))) +
scale_x_discrete(limits = c("Medical Field", "Intervention", "Comparator")) +
scale_y_continuous(breaks = seq(0, max_n, by = 1)) +
theme_minimal(base_size = 14) +
theme(
legend.position = "none",
panel.grid.minor = element_blank(), # remove horizontal helper lines
panel.grid.major.x = element_blank(), # remove vertical lines
panel.grid.major.y = element_line(color = "grey80"),
plot.background = element_blank(),
panel.background = element_blank(),
# axis.title.x = element_blank(),
# axis.text.x = element_text(size = 12),
# axis.text.y = element_text(size = 12),
# force tight plot wrapping (to fix Quarto spacing)
plot.margin = margin(t = 10, r = 5, b = 5, l = 5)
) +
labs(
y = "Number of IPDMAs"
) 
# Select CRT participant columns
crt_part_cols <- grep("CRT[0-9]+: number of participants randomized", names(df), value = TRUE)
# Calculate total participants in CRTs per IPDMA
df <- df %>%
rowwise() %>%
mutate(
total_CRT_participants = sum(c_across(all_of(crt_part_cols)), na.rm = TRUE)
) %>%
ungroup()
# Calculate CRT trial and CRT participant percentages per IPDMA
df <- df %>%
mutate(
pct_trials_CRT = (`Of RCTs with IPD obtained, number of CRTs` /
`Number of eligible trials for which IPD were obtained and included in the MA`) * 100,
pct_participants_CRT = total_CRT_participants /
`Number of eligible participants for which IPD were obtained and included in the MA` * 100
) %>%
mutate(IPDMA_index = 1:n())
# Calculate overall percentages
overall_pct_trials <- sum(df$`Of RCTs with IPD obtained, number of CRTs`) /
sum(df$`Number of eligible trials for which IPD were obtained and included in the MA`) * 100
overall_pct_participants <- sum(df$total_CRT_participants, na.rm = TRUE) /
sum(df$`Number of eligible participants for which IPD were obtained and included in the MA`) * 100
# Create data for stacked bars
crt_stacked <- df %>%
mutate(
n_nonCRT = `Number of eligible trials for which IPD were obtained and included in the MA` - `Of RCTs with IPD obtained, number of CRTs`,
pct_nonCRT = 100 - pct_trials_CRT,
n_nonCRT_participants = `Number of eligible participants for which IPD were obtained and included in the MA` - total_CRT_participants,
pct_nonCRT_participants = 100 - pct_participants_CRT
)
# Stacked plot for trials
trial_stack_df <- crt_stacked %>%
select(IPDMA_index, `Of RCTs with IPD obtained, number of CRTs`, n_nonCRT, pct_trials_CRT, pct_nonCRT) %>%
pivot_longer(cols = c(`Of RCTs with IPD obtained, number of CRTs`, n_nonCRT), names_to = "Type", values_to = "Count") %>%
mutate(
pct = ifelse(Type == "Of RCTs with IPD obtained, number of CRTs", pct_trials_CRT, pct_nonCRT),
Type = factor(Type, levels = c("n_nonCRT", "Of RCTs with IPD obtained, number of CRTs"), labels = c("Non-CRT", "CRT"))
)
ggplot(trial_stack_df, aes(x = IPDMA_index, y = pct, fill = Type)) +
geom_col(color = "white", linewidth = 0.2) +
geom_hline(yintercept = overall_pct_trials, linetype = "dashed", color = "darkgrey", linewidth = 1) +
geom_text(aes(label = Count),
position = position_stack(vjust = 0.5), size = 3, color = "white") +
scale_fill_manual(values = c("Non-CRT" = "#b0c4de", "CRT" = "#2E8B57")) +
scale_x_continuous(breaks = 1:14) +
scale_y_continuous(limits = c(0, 100)) +
labs(
x = "IPDMA",
y = "Percentage of trials",
fill = NULL,
title = "CRT vs Non-CRT trials per IPDMA",
caption = "Counts in stacked bars, percentage on y-axis. Grey line: Average across all."
) +
theme_minimal() +
theme(
axis.text.x = element_text(angle = 0, hjust = 0.5),
plot.caption = element_text(hjust = 0)
)
# Stacked plot for participants
participant_stack_df <- crt_stacked %>%
select(IPDMA_index, total_CRT_participants, n_nonCRT_participants, pct_participants_CRT, pct_nonCRT_participants) %>%
pivot_longer(cols = c(total_CRT_participants, n_nonCRT_participants), names_to = "Type", values_to = "Count") %>%
mutate(
pct = ifelse(Type == "total_CRT_participants", pct_participants_CRT, pct_nonCRT_participants),
Type = factor(Type, levels = c("n_nonCRT_participants", "total_CRT_participants"), labels = c("Non-CRT participants", "CRT participants"))
)
ggplot(participant_stack_df, aes(x = IPDMA_index, y = pct, fill = Type)) +
geom_col(color = "white", linewidth = 0.2) +
geom_hline(yintercept = overall_pct_participants, linetype = "dashed", color = "darkgrey", linewidth = 1) +
geom_text(aes(label = Count),
position = position_stack(vjust = 0.5), size = 3, color = "white") +
scale_fill_manual(values = c("Non-CRT participants" = "#b0c4de", "CRT participants" = "#2E8B57")) +
scale_x_continuous(breaks = 1:14) +
scale_y_continuous(limits = c(0, 100)) +
labs(
x = "IPDMA",
y = "Percentage of participants",
fill = NULL,
title = "CRT vs Non-CRT participants per IPDMA",
caption = "Counts in stacked bars, percentage on y-axis. Grey line: Average across all."
) +
theme_minimal() +
theme(
axis.text.x = element_text(angle = 0, hjust = 0.5),
plot.caption = element_text(hjust = 0)
)
# Cluster design
# df_long %>%
# count(`cluster design`, sort = TRUE)# Summarize CRT-level characteristics
crt_sankey <- df_long %>%
filter(!is.na(`Order of recruitment and randomization`),
!is.na(`Consent procedure`)) %>%
count(`Order of recruitment and randomization`,
`Consent procedure`)
crt_sankey <- crt_sankey %>%
mutate(
`Order of recruitment and randomization` =
fct_infreq(`Order of recruitment and randomization`),
`Consent procedure` =
fct_infreq(`Consent procedure`)
)
# determine axis tick range
max_n <- 43
ggplot(crt_sankey,
aes(axis1 = `Order of recruitment and randomization`,
axis2 = `Consent procedure`,
y = n)) +
geom_alluvium(aes(fill = `Order of recruitment and randomization`),
width = 0.18, alpha = 0.85) +
geom_stratum(width = 0.18, color = "black", fill = "grey90") +
geom_label(stat = "stratum", aes(label = after_stat(stratum)),
size = 4, label.padding = unit(0.15, "lines")) +
scale_x_discrete(
limits = c("Order of recruitment & randomization", "Consent procedure"),
expand = c(0.3, 0.3) # adjust extra space on x-axis!!
) +
scale_y_continuous(
name = "Number of CRTs",
breaks = seq(0, max_n, by = 1),
expand = c(0, 0) # no padding above/below
) +
theme_minimal(base_size = 14) +
theme(
legend.position = "none",
panel.grid.minor = element_blank(),
panel.grid.major.x = element_blank(), # remove vertical lines
panel.grid.major.y = element_line(color = "grey80"),
plot.background = element_blank(),
panel.background = element_blank(),
axis.title.x = element_blank(),
axis.text.x = element_text(size = 12),
axis.text.y = element_text(size = 12),
# force tight plot wrapping (fixes Quarto spacing)
plot.margin = margin(t = 10, r = 5, b = 5, l = 5)
)
vars_reporting <- c(
"ICC reported",
"CRTs labeled as CRTs",
"Risk of bias",
"CRT-specific risk of bias"
)
vars_exist <- vars_reporting %in% names(df)
if(!all(vars_exist)) {
message("Warning: some variable names were not found. Missing: ",
paste(vars_reporting[!vars_exist], collapse = "; "))
}
# create and display gtsummary table
tbl_reporting <- df %>%
select(any_of(vars_reporting)) %>%
mutate(across(everything(), ~ fct_infreq(factor(.)))) %>%
tbl_summary(
missing = "ifany",
type = all_categorical() ~ "categorical",
statistic = all_categorical() ~ "{n} ({p}%)"
) %>%
modify_header(label = "**IPDMA reporting characteristics**") %>%
bold_labels()
tbl_reporting| IPDMA reporting characteristics | N = 141 |
|---|---|
| ICC reported | |
| No | 13 (93%) |
| Yes | 1 (7.1%) |
| CRTs labeled as CRTs | |
| No | 9 (64%) |
| Yes | 5 (36%) |
| Risk of bias | |
| Yes | 11 (79%) |
| No | 3 (21%) |
| CRT-specific risk of bias | |
| No | 9 (64%) |
| Not applicable | 3 (21%) |
| Yes | 2 (14%) |
| 1 n (%) | |
# Reshape and calculate percentages
likert_df <- df %>%
select(any_of(vars_reporting)) %>%
pivot_longer(everything(), names_to = "Variable", values_to = "Category") %>%
filter(!is.na(Category)) %>%
mutate(
# enforce left to right ordering of stack
Category = factor(Category, levels = c("Yes", "No", "Not applicable")),
# control order (top-down)
Variable = factor(Variable, levels = rev(vars_reporting))
) %>%
count(Variable, Category) %>%
group_by(Variable) %>%
mutate(percent = n / sum(n) * 100) %>%
ungroup()
category_colors <- c(
"Yes" = "#2E8B57",
"No" = "#D95F02",
"Not applicable" = "#E0E0E0"
)
# Plot
ggplot(likert_df, aes(x = Variable, y = percent, fill = Category)) +
geom_col(position = "fill", color = "white", linewidth = 0.2) +
geom_text(aes(label = n),
position = position_fill(vjust = 0.5),
color = "white", size = 3) +
coord_flip() +
scale_fill_manual(values = category_colors, drop = FALSE) +
scale_y_continuous(labels = scales::percent_format(scale = 100)) +
labs(x = NULL, y = NULL, fill = NULL) +
theme_minimal(base_size = 13) +
theme(
axis.text.x = element_text(size = 11),
axis.text.y = element_text(size = 11),
legend.position = "bottom",
panel.grid.major.y = element_blank()
)
vars_analysis <- c(
"Type of Primary Outcome(s)",
"Analysis framework",
"Treatment effect model",
"CRT vs IRT analysis"
# "CRT-level estimator",
# "Small sample correction",
# "Covariate adjustment",
)
vars_exist <- vars_analysis %in% names(df)
if(!all(vars_exist)) {
message("Warning: some variable names were not found. Missing: ",
paste(vars_analysis[!vars_exist], collapse = "; "))
}
# create and display gtsummary table
tbl_analysis <- df %>%
select(any_of(vars_analysis)) %>%
mutate(across(everything(), ~ fct_infreq(factor(.)))) %>%
tbl_summary(
missing = "ifany",
type = all_categorical() ~ "categorical",
statistic = all_categorical() ~ "{n} ({p}%)"
) %>%
modify_header(label = "**IPDMA analysis characteristics**") %>%
bold_labels()
tbl_analysis| IPDMA analysis characteristics | N = 141 |
|---|---|
| Type of Primary Outcome(s) | |
| Continuous | 6 (43%) |
| Binary | 3 (21%) |
| Time-to-event | 3 (21%) |
| Combination | 1 (7.1%) |
| Ordinal | 1 (7.1%) |
| Analysis framework | |
| Frequentist | 13 (93%) |
| Bayesian | 1 (7.1%) |
| Treatment effect model | |
| Fixed | 8 (57%) |
| Random | 6 (43%) |
| CRT vs IRT analysis | |
| No | 13 (93%) |
| Yes | 1 (7.1%) |
| 1 n (%) | |
vars_analysis2 <- c(
"Analysis approach",
"Accounted for two-level clustering"
)
vars_exist <- vars_analysis2 %in% names(df)
if(!all(vars_exist)) {
message("Warning: some variable names were not found. Missing: ",
paste(vars_analysis2[!vars_exist], collapse = "; "))
}
# create and display gtsummary table
tbl_analysis2 <- df %>%
select(any_of(vars_analysis2)) %>%
mutate(across(everything(), ~ fct_infreq(factor(.)))) %>%
tbl_summary(
missing = "ifany",
type = all_categorical() ~ "categorical",
statistic = all_categorical() ~ "{n} ({p}%)"
) %>%
modify_header(label = "**IPDMA analysis characteristics**") %>%
bold_labels()
tbl_analysis2| IPDMA analysis characteristics | N = 141 |
|---|---|
| Analysis approach | |
| One-stage | 7 (50%) |
| Two-stage | 7 (50%) |
| Accounted for two-level clustering | |
| Yes | 10 (71%) |
| No | 4 (29%) |
| 1 n (%) | |
# Summarize CRT-level characteristics
sankey_analysis2 <- df %>%
count(`Accounted for two-level clustering`,
`Analysis approach`,)
sankey_analysis2 <- sankey_analysis2 %>%
mutate(
`Accounted for two-level clustering` =
fct_infreq(`Accounted for two-level clustering`),
`Analysis approach` =
fct_infreq(`Analysis approach`)
)
# determine axis tick range
max_n <- 14
ggplot(sankey_analysis2,
aes(axis1 = `Accounted for two-level clustering`,
axis2 = `Analysis approach`,
y = n)) +
geom_alluvium(aes(fill = `Accounted for two-level clustering`),
width = 0.18, alpha = 0.85) +
geom_stratum(width = 0.18, color = "black", fill = "grey90") +
geom_label(stat = "stratum", aes(label = after_stat(stratum)),
size = 4, label.padding = unit(0.15, "lines")) +
scale_x_discrete(
limits = c("Accounted for two-level clustering", "Analysis approach"),
expand = c(0.3, 0.3) # adjust extra space on x-axis!!
) +
scale_y_continuous(
name = "Number of IPDMAs",
breaks = seq(0, max_n, by = 1),
expand = c(0, 0) # no padding above/below
) +
theme_minimal(base_size = 14) +
theme(
legend.position = "none",
panel.grid.minor = element_blank(),
panel.grid.major.x = element_blank(), # remove vertical lines
panel.grid.major.y = element_line(color = "grey80"),
plot.background = element_blank(),
panel.background = element_blank(),
axis.title.x = element_blank(),
axis.text.x = element_text(size = 12),
axis.text.y = element_text(size = 12),
# force tight plot wrapping (fixes Quarto spacing)
plot.margin = margin(t = 10, r = 5, b = 5, l = 5)
)
crt_stacked_no_clustering <- crt_stacked %>%
filter(`Accounted for two-level clustering` == "No") %>%
mutate(IPDMA_index = 1:n())
# Stacked plot for trials
trial_stack_df <- crt_stacked_no_clustering %>%
select(IPDMA_index, `Of RCTs with IPD obtained, number of CRTs`, n_nonCRT, pct_trials_CRT, pct_nonCRT) %>%
pivot_longer(cols = c(`Of RCTs with IPD obtained, number of CRTs`, n_nonCRT), names_to = "Type", values_to = "Count") %>%
mutate(
pct = ifelse(Type == "Of RCTs with IPD obtained, number of CRTs", pct_trials_CRT, pct_nonCRT),
Type = factor(Type, levels = c("n_nonCRT", "Of RCTs with IPD obtained, number of CRTs"), labels = c("Non-CRT", "CRT"))
)
ggplot(trial_stack_df, aes(x = IPDMA_index, y = pct, fill = Type)) +
geom_col(color = "white", linewidth = 0.2) +
geom_text(aes(label = Count),
position = position_stack(vjust = 0.5), size = 3, color = "white") +
scale_fill_manual(values = c("Non-CRT" = "#b0c4de", "CRT" = "#2E8B57")) +
# scale_x_continuous(breaks = 1:14) +
scale_y_continuous(limits = c(0, 100)) +
labs(
x = "IPDMA",
y = "Percentage of trials",
fill = NULL,
title = "CRT vs Non-CRT trials per IPDMA",
caption = "Counts in stacked bars, percentage on y-axis"
) +
theme_minimal() +
theme(
axis.text.x = element_text(angle = 0, hjust = 0.5),
plot.caption = element_text(hjust = 0)
)
# Stacked plot for participants
participant_stack_df <- crt_stacked_no_clustering %>%
select(IPDMA_index, total_CRT_participants, n_nonCRT_participants, pct_participants_CRT, pct_nonCRT_participants) %>%
pivot_longer(cols = c(total_CRT_participants, n_nonCRT_participants), names_to = "Type", values_to = "Count") %>%
mutate(
pct = ifelse(Type == "total_CRT_participants", pct_participants_CRT, pct_nonCRT_participants),
Type = factor(Type, levels = c("n_nonCRT_participants", "total_CRT_participants"), labels = c("Non-CRT participants", "CRT participants"))
)
ggplot(participant_stack_df, aes(x = IPDMA_index, y = pct, fill = Type)) +
geom_col(color = "white", linewidth = 0.2) +
geom_text(aes(label = Count),
position = position_stack(vjust = 0.5), size = 3, color = "white") +
scale_fill_manual(values = c("Non-CRT participants" = "#b0c4de", "CRT participants" = "#2E8B57")) +
# scale_x_continuous(breaks = 1:14) +
scale_y_continuous(limits = c(0, 100)) +
labs(
x = "IPDMA",
y = "Percentage of participants",
fill = NULL,
title = "CRT vs Non-CRT participants per IPDMA",
caption = "Counts in stacked bars, percentage on y-axis"
) +
theme_minimal() +
theme(
axis.text.x = element_text(angle = 0, hjust = 0.5),
plot.caption = element_text(hjust = 0)
)
crt_stacked_onestage <- crt_stacked %>%
filter(`Accounted for two-level clustering` == "Yes" & `Analysis approach` == "One-stage") %>%
mutate(IPDMA_index = 1:n())
df_onestage <- df %>%
filter(`Accounted for two-level clustering` == "Yes" & `Analysis approach` == "One-stage")IPDMA #1:
Binary outcome
Bayesian model with two hierarchical layers:
trial-level intercept
cluster-level intercepts nested within trials (or ICUs in individually randomized trials)
Intervention as fixed effect
IPDMA #2 & #3:
Time-to-event outcome
Cox PH shared frailty model with random effects for clusters (or sites in individually randomized trials)
Intervention as fixed effect
Unclear if a single baseline hazard across all trials was assumed
IPDMA #4:
Continuous outcome
Generalized Estimating Equations to account for clustering within trials (if exists) and trial as fixed effect (separate intercepts)
Intervention as fixed effect