# Date: 4 June 2025 #Author: Julia Colombijn; checked by Jan de Leijer #Load packages library("tidyverse") library("meta") library("metafor") library("metaplus") library("readxl") library("robvis") library("spatstat") library("Hmisc") library("dplyr") library("ggplot2") library("devtools") #Load dataset data <- read_excel("Tiab/Analyse/dataset_sglt2i_glp1ra_combination_obs.xlsx") obs_rob <- read_excel("Tiab/RoB/Overview RoB observationele studies.xlsx") # Baseline data baseline <- tibble( study = c("Riley et al. 2023", "Marfella et al. 2024", "Lau et al. 2022", "Luo et al. 2023", "Lopez et al. 2022", "Dave et al. 2021", "Gorgojo-Martínez et al. 2017", "Simms-Williams et al. 2024 (1)", "Simms-Williams et al. 2024 (2)", "Schechter et al. 2023", "Kobayashi et al 2023", "Liu et al. 2025", "Patel et al 2024", "Wright et al 2022 (1)", "Wright et al 2022 (2)", "Wright et al 2022 (3)", "Chaiyakunapruk et al 2025", "Jensen et al 2020", "Jhu et al 2024", "García-Vega et al 2024", "Horiuchi et al 2025"), n = c(311577, 443, 2888, 1674, 686, 25168, 213, 13392, 17884, 6848, 372, 416, 14088, 1064, 15423, 6859, 439995, 16664, 142372, 15549, 131196), n_ma = c(311577, 443, 2888, 1674, 686, 25168, 213, 13392, 17884, 3899, 372, 416, 14088, NA, NA, NA, 439995, 16664, 142372, NA, NA), FU = c(60, 9, 18, NA, 36, 9.85, 12, 9, 8.4, 81, 47, 12, 12, 40, 48, 48, NA, NA, NA, 20, 36), age = c(60, 69, 58, 53, 68, 58, 59, 57, 58, 59, 63, 63, 63, 60, 61, 60, 64, 59, 57, 69, NA), sd_age = c(NA, NA, 11, 11, 8, 11, 11, 10, 10, 10, 12, 10, 11, 9, 9, 9, 11, 11, 11, 11, NA), n_female = c(149274, 180, 1337, 696, 6, 13050, 99, 6097, 8456, 3067, 172, 134, 6175, 728, 9841, 4369, 60370, 6951, 64470, 6433, NA), HbA1c = c(NA, 7.6, 9.0, 8.6, 8.0, 8.4, 7.4, NA, NA, 9, 8.4, 7.4, NA, 9.2, 9.0, 9.4, 8.0, NA, 8.2, NA, NA), sd_HbA1c = c(NA, NA, 1.5, 2, 1.7, 1.8, 1.3, NA, NA, 1.5, 1.6, 3.1, NA, 1.7, 1.7, 1.7, 1.7, NA, 1.9, NA, NA), years_of_diabetes = c(NA, 15, 8.7, 8, NA, NA, 11, 11, 11, 10, NA, 8, NA, 5, 6, 9, NA, 7, NA, NA, NA), sd_years_of_diabetes = c(NA, NA, 6, 5, NA, NA, NA, 6, 6, 5, NA, 3, NA, 5, 5, 4, NA, 5, NA, NA, NA) ) ################################################################################ ########################### Baseline characteristics ########################### ################################################################################ # Total sample size baseline %>% summarise(n = sum(n), # total N n_ma = sum(n_ma, na.rm = TRUE)) # N in meta-analysis # Follow-up baseline %>% summarise(min_fu = min(FU, na.rm = TRUE), max_fu = max(FU, na.rm = TRUE)) # Weighted mean + SD age baseline %>% summarise(mean_age = weighted.mean(age, n, na.rm = TRUE), sd_age = weighted.mean(sd_age, n, na.rm = TRUE)) # % women baseline %>% filter(!is.na(n_female)) %>% summarise(n_female = sum(n_female), n = sum(n), perc = n_female / n * 100) # Weighted mean + SD HbA1c baseline %>% summarise(mean_hba1c = weighted.mean(HbA1c, n, na.rm = TRUE), sd_hba1c = weighted.mean(sd_HbA1c, n, na.rm = TRUE)) # Convert HbA1c to mmol/mol baseline <- baseline %>% mutate(HbA1c_mmol = (HbA1c * 10.93) - 23.5, # Convert HbA1c to mmol/mol HbA1c_perc_f = sd_HbA1c / HbA1c, # determine % of SD to HbA1c sd_HbA1c_mmol = (HbA1c_mmol * HbA1c_perc_f)) # Convert SD to mmol/mol # weighted mean + SD HbA1c in mmol/mol baseline %>% summarise(mean_hba1c = weighted.mean(HbA1c_mmol, n, na.rm = TRUE), sd_hba1c = weighted.mean(sd_HbA1c_mmol, n, na.rm = TRUE)) # Weighted mean + SD DM duration baseline %>% summarise(mean_dm2_years = weighted.mean(years_of_diabetes, n, na.rm = TRUE), sd_dm2_years = weighted.mean(sd_years_of_diabetes, n, na.rm = TRUE)) ################################## Risk of bias################################# # obs_rob <- obs_rob %>% mutate( across(c(D1:D7, Overall), ~ factor(.x, levels = c("Critical", "Low", "Moderate", "Serious", "No information"))) ) # Set up PNG for RoB of observational studies png("supp_fig_2.png", width = 14000, height = 7000, res = 1200) # Risk of bias trafic light observational studies rob_traffic_light(data = obs_rob, tool = "ROBINS-I") # Close PNG device dev.off() ############################### Primary analysis ############################### ################################################################################ ################################### Efficacy ################################### ################################################################################ ## Filter rows for each outcome # Filter the data for MACE mace_data <- data %>% filter(mace == 1) %>% mutate(high_risk = ifelse(high_risk == 1, "Very high risk", "High risk"), outcome = "MACE") # Filter data for all-cause mortality acm_data <- data %>% filter(acm == 1) %>% mutate(events_exp = acm_events_exp, n_exp = acm_n_exp, events_cont = acm_events_cont, n_cont = acm_n_cont, outcome = "ACM") %>% select(study, outcome, events_exp:n_cont) # Filter data for hospitalisation for heart failure hhf_data <- data %>% filter(hhf == 1) %>% mutate(events_exp = hhf_events_exp, n_exp = hhf_n_exp, events_cont = hhf_events_cont, n_cont = hhf_n_cont, outcome = "HHF") %>% select(study, outcome, events_exp:n_cont) # Filter data for cardiovascular mortality cvm_data <- data %>% filter(cvm == 1) %>% mutate(events_exp = cvm_events_exp, n_exp = cvm_n_exp, events_cont = cvm_events_cont, n_cont = cvm_n_cont, outcome = "CVM") %>% select(study, outcome, events_exp:n_cont) # Filter data for myocardial infarction mi_data <- data %>% filter(mi == 1) %>% mutate(events_exp = mi_events_exp, n_exp = mi_n_exp, events_cont = mi_events_cont, n_cont = mi_n_cont, outcome = "MI") %>% select(study, outcome, events_exp:n_cont) # Filter data for stroke stroke_data <- data %>% filter(stroke == 1) %>% mutate(events_exp = stroke_events_exp, n_exp = stroke_n_exp, events_cont = stroke_events_cont, n_cont = stroke_n_cont, outcome = "Stroke") %>% select(study, outcome, events_exp:n_cont) # Filter data for kidney composite endpoint kf_data <- data %>% filter(kf == 1) %>% mutate(events_exp = kf_events_exp, n_exp = kf_n_exp, events_cont = kf_events_cont, n_cont = kf_n_cont, outcome = "KF") %>% select(study, outcome, events_exp:n_cont) # Create function for labels forest plot mlabfun <- function(text, x) { list(bquote(paste(.(text), " (Tau"^2, " = ", .(fmtx(x$tau2, digits=4)), " Q = ", .(fmtx(x$QE, digits=2)), ", df = ", .(x$k - x$p), ", ", .(fmtp(x$QEp, digits=3, pname="p", add0=TRUE, sep=TRUE, equal=TRUE)), "; ", I^2, " = ", .(fmtx(x$I2, digits=1)), "%, ", ")")))} ##################################### MACE ##################################### # Fit random model ma_mace <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data, # Data frame method = "REML" # Method for random-effects model ) # Extract weights mace_weights <- as.data.frame(paste0(fmtx(weights(ma_mace), digits = 1), "%")) %>% rename(weight = `paste0(fmtx(weights(ma_mace), digits = 1), "%")`) %>% mutate(weight_num = as.numeric(gsub("%", "", weight)), weight_scaled = 0.75 + weight_num * 0.06) #Add numeric weights # Add weights to dataframe mace_data <- cbind(mace_data, mace_weights) # Rerun meta-analysis with weights ma_mace <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data, # Data frame method = "REML" # Method for random-effects model ) #Extract rounded estimate and CI bound pred_mace <- predict(ma_mace, trans = exp, digits = 2) pred_mace <- fmtx(c(pred_mace$pred, pred_mace$ci.lb, pred_mace$ci.ub), digits = 2) # Set up PNG for MACE observational studies png("fig_1.png", width = 16000, height = 8000, res = 1200) # Set plot parameters par(mar = c(10, 5, 1.3, 5), mgp = c(3, 1, 0), tcl = -0.2) # Create plot fp_mace_obs <- forest(ma_mace, xlim=c(-10, 4), at=log(c(0.1, 0.5, 1, 2, 10)), atransf=exp, order = weight_num, xlab = "", lty = c(1, 1, 0), ilab = cbind(mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont, weight), ilab.xpos=c(-7, -6, -5, -4, -3), cex = 0.80, ylim=c(-2, 14), ilab.pos = 2, mlab = "", psize = weight_scaled, header = c("Study", "Risk ratio [95% CI]"), showweights = FALSE) # hide non-bold summary estimate and add it back in bold rect(2, -2, 4, -0.5, col="white", border=NA) text(fp_mace_obs$textpos[2], -1, paste0(pred_mace_obs[1], " [", pred_mace_obs[2], ", ", pred_mace_obs[3], "]"), pos=2, font = 2, cex = 0.80) # Add horizontal line at the top segments(-10, 12, 4, 12, lwd = 0.8) # Expand margins par(xpd = NA) # Add labels on top text(c(-7, -6, -5, -4, -3), 13, c("Events", "Total", "Events", "Total", "Weight"), cex = 0.80, font = 2, pos = 2) text(c(-6.5, -4.5, 0), 14, c("Combination therapy", "Monotherapy", "Risk ratio"), cex = 0.80, font = 2) text(0, 13, c("MH Random, 95% CI"), cex = 0.80, font = 2) # Add outcome labels text(c(-10), c(11), c("Major adverse cardiovascular events"), pos = 4, font = 2, cex = 0.80) # Add Total labels: text(-10, -1, pos = 4, "Total (95% CI)", font = 2, cex = 0.80) #Add text for total events text(c(-6, -4), -1, c(sum(mace_obs_data$mace_n_exp), sum(mace_obs_data$mace_n_cont)), pos = 2, font = 2, cex = 0.80) text(c(-7, -5), -1, c(sum(mace_obs_data$mace_events_exp), sum(mace_obs_data$mace_events_cont)), pos = 2, cex = 0.80) # Extract p-value heterogeneity statistics p_mace_het <- fmtp(mace_obs$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) #Add heterogeneity statistics text(-10, -2, pos = 4, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(mace_obs$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(mace_obs$QE, digits = 2)), ", df = ", .(mace_obs$k - mace_obs$p), " (", italic(p), .(p_mace_het), "); ", I^2, " = ", .(round(mace_obs$I2)), "%")), cex = 0.80) # Extract p-value overall effect p_mace_oe <- fmtp(mace_obs$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) text(-10, -3, pos = 4, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(mace_obs$zval, digits = 2)), " (", italic(p), .(p_mace_oe), ")")), cex = 0.80) # Add label for axes text(log(c(0.1, 10)), -5, c("Favours combination of SGLT-2 inhibitor and GLP-1 RA", "Favours SGLT-2 inhibitor or GLP-1 RA monotherapy"), pos = c(4,2), offset = -7, cex = 0.80) # Close PNG device dev.off() ##################### Cardiovascular mortality, MI, Stroke ##################### # Create dataframe for CVM, MI, stroke cvm_mi_stroke_data <- rbind(stroke_data, mi_data, cvm_data) # Create factor of outcome cvm_mi_stroke_data <- cvm_mi_stroke_data %>% mutate(outcome = case_when(outcome == "Stroke" ~ 1, outcome == "MI" ~ 2, outcome == "CVM" ~ 3)) # Fit random effects model for CVM ma_cvm <- rma( measure = "RR", # Risk Ratio ai = events_exp, # Events in experimental group n1i = n_exp, # Total in experimental group ci = events_cont, # Events in control group n2i = n_cont, # Total in control group slab = study, # Study labels data = cvm_data, # Data frame method = "REML" # Method for random-effects model ) # Extract point estimate and confidence intervals CVM pred_cvm <- predict(ma_cvm, transf = exp, digits = 2) pred_cvm <- fmtx(c(pred_cvm$pred, pred_cvm$ci.lb, pred_cvm$ci.ub), digits=2) # Fit random effects model for MI ma_mi <- rma( measure = "RR", # Risk Ratio ai = events_exp, # Events in experimental group n1i = n_exp, # Total in experimental group ci = events_cont, # Events in control group n2i = n_cont, # Total in control group slab = study, # Study labels data = mi_data, # Data frame method = "REML" # Method for random-effects model ) # Extract point estimate and confidence intervals MI pred_mi <- predict(ma_mi, transf = exp, digits = 2) pred_mi <- fmtx(c(pred_mi$pred, pred_mi$ci.lb, pred_mi$ci.ub), digits=2) # Fit random effects model for stroke ma_stroke <- rma( measure = "RR", # Risk Ratio ai = events_exp, # Events in experimental group n1i = n_exp, # Total in experimental group ci = events_cont, # Events in control group n2i = n_cont, # Total in control group slab = study, # Study labels data = stroke_data, # Data frame method = "REML" # Method for random-effects model ) # Extract point estimate and confidence intervals Stroke pred_stroke <- predict(ma_stroke, transf = exp, digits = 2) pred_stroke <- fmtx(c(pred_stroke$pred, pred_stroke$ci.lb, pred_stroke$ci.ub), digits=2) #Extract weights cvm_weights <- as.data.frame(paste0(fmtx(weights(ma_cvm), digits = 1), "%")) %>% #CVM rename(weight = `paste0(fmtx(weights(ma_cvm), digits = 1), "%")`) mi_weights <- as.data.frame(paste0(fmtx(weights(ma_mi), digits = 1), "%")) %>% #MI rename(weight = `paste0(fmtx(weights(ma_mi), digits = 1), "%")`) stroke_weights <- as.data.frame(paste0(fmtx(weights(ma_stroke), digits = 1), "%")) %>% #Stroke rename(weight = `paste0(fmtx(weights(ma_stroke), digits = 1), "%")`) #create dataframe of ordered weights weights_cvm_mi_stroke <- rbind(stroke_weights, mi_weights, cvm_weights) %>% mutate(weight_num = as.numeric(gsub("%", "", weight)), #make weights numeric weight_scaled = 0.7 + 0.02 * weight_num) # Add weights to df cvm_mi_stroke_data <- cbind(cvm_mi_stroke_data, weights_cvm_mi_stroke) # Sort variabels by weight cvm_mi_stroke_data <- cvm_mi_stroke_data %>% arrange(outcome, desc(weight_num)) # Fit random effects model for all outcomes cvm_mi_stroke <- rma( measure = "RR", # Risk Ratio ai = events_exp, # Events in experimental group n1i = n_exp, # Total in experimental group ci = events_cont, # Events in control group n2i = n_cont, # Total in control group slab = study, # Study labels data = cvm_mi_stroke_data, # Data frame method = "REML" # Method for random-effects model ) # Set up png for observational data for CVM, MI, and stroke png("fig_2.png", width = 16000, height = 18000, res = 1200) # Expand margins par(xpd = TRUE) # Create forest plot for CVM, MI, stroke forest(cvm_mi_stroke, xlim = c(-10, 4), at = log(c(0.1, 0.5, 1, 2, 10)), atransf = exp, order = outcome, xlab = "", lty = c(1, 1, 0), ilab = cbind(events_exp, n_exp, events_cont, n_cont, weight), ilab.xpos = c(-7, -6, -5, -4, -3), cex = 0.80, ylim = c(-2, 32), rows = c(0:6, 13:19, 26:28), mlab = "", psize = weight_scaled, header = c("Study", "Risk ratio [95% CI]"), showweights = FALSE, addfit = FALSE) # Add horizontal line at the top segments(-10, 30, 4, 30, lwd = 0.8) # Add labels on top text(c(-7, -6, -5, -4, -3), 31, c("Events", "Total", "Events", "Total", "Weight"), cex = 0.80, font = 2) text(c(-6.5, -4.5, 0), 32, c("Combination therapy", "Monotherapy", "Risk ratio"), cex = 0.80, font = 2) text(0, 31, c("MH Random, 95% CI"), cex = 0.80, font = 2) # Add pooled effect estimates per outcome addpoly(ma_cvm, row = 24, cex = 0.80) addpoly(ma_mi, row = 11, cex = 0.80) addpoly(ma_stroke, row = -2, cex = 0.80) #Hide study labels and model rect(-10, 29, -7.25, -3, col = "White", border = NA) # Add outcome labels text(c(-10), c(29, 20, 7), c("Cardiovascular mortality", "Myocardial infarction", "Stroke"), pos = 4, font = 2, cex = 0.80) # Add study labels text(c(-10), c(0:6, 13:19, 26:28), c("Riley et al (2023) [41]", "Chaiyakunapruk et al (2025) [29]", "Dave et al (2021) [30]", "Simms-Williams et al (2024b) [43]", "Simms-Williams et al (2024a) [43]", "Luo et al (2023) [38]", "Liu et al (2025) [36]", "Riley et al (2023) [41]", "Chaiyakunapruk et al (2025) [29]", "Dave et al (2021) [30]", "Simms-Williams et al (2024b [43]", "Simms-Williams et al (2024a) [43]", "Luo et al (2023) [38]", "Liu et al (2025) [36]", "Simms-Williams et al (2024b) [43]", "Simms-Williams et al (2024a) [43]", "Liu et al (2025) [36]"), cex = 0.80, pos = 4) # Hide point estimates RE rect(1, c(25, 12, -1), 4, c(23, 10, -3), col = "White", border = NA) # Add pooled effect estimates estimates in bold text(4, 24, paste0(pred_cvm[1], " [", pred_cvm[2], ", ", pred_cvm[3], "]"), pos = 2, cex = 0.80, font = 2) # CVM text(4, 11, paste0(pred_mi[1], " [", pred_mi[2], ", ", pred_mi[3], "]"), pos = 2, cex = 0.80, font = 2) # MI text(4, -2, paste0(pred_stroke[1], " [", pred_stroke[2], ", ", pred_stroke[3], "]"), pos = 2, cex = 0.80, font = 2) # Stroke # Add Total labels: text(-10, c(24, 11, -2), pos = 4, "Total (95% CI)", font = 2, cex = 0.80) #Add text for total number of patients CVM text(c(-6, -4), 24, c(sum(cvm_data$n_exp), sum(cvm_data$n_cont)), font = 2, cex = 0.80) # Add text for total number of events CVM text(c(-7, -5), 24, c(sum(cvm_data$events_exp), sum(cvm_data$events_cont)), cex = 0.80) #Add text for total number of patients MI text(c(-6, -4), 11, c(sum(mi_data$n_exp), sum(mi_data$n_cont)), font = 2, cex = 0.80) # Add text for total number of events MI text(c(-7, -5), 11, c(sum(mi_data$events_exp), sum(mi_data$events_cont)), cex = 0.80) #Add text for total number of patients Stroke text(c(-6, -4), -2, c(sum(stroke_data$n_exp), sum(stroke_data$n_cont)), font = 2, cex = 0.80) # Add text for total number of events Stroke text(c(-7, -5), -2, c(sum(stroke_data$events_exp), sum(stroke_data$events_cont)), cex = 0.80) # Cardiovascular mortality # Extract p-values p_cvm_het <- fmtp(ma_cvm$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_cvm_oe <- fmtp(ma_cvm$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Heterogeneity text(-10, 23, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_cvm$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_cvm$QE, digits = 2)), ", df = ", .(ma_cvm$k - ma_cvm$p), " (", italic(p), .(p_cvm_het), "); ", I^2, " = ", .(round(ma_cvm$I2)), "%"))) # Overall effect text(-10, 22, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_cvm$zval, digits = 2)), " (", italic(p), .(p_cvm_oe), ")"))) # Myocardial infarction # Extract p-values p_mi_het <- fmtp(ma_mi$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_mi_oe <- fmtp(ma_mi$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Heterogeneity text(-10, 10, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mi$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mi$QE, digits = 2)), ", df = ", .(ma_mi$k - ma_mi$p), " (", italic(p), .(p_mi_het), "); ", I^2, " = ", .(round(ma_mi$I2)), "%"))) # Overall effect text(-10, 9, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mi$zval, digits = 2)), " (", italic(p), .(p_mi_oe), ")"))) # Stroke # Extract p-values p_stroke_het <- fmtp(ma_stroke$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_stroke_oe <- fmtp(ma_stroke$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Heterogeneity text(-10, -3, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_stroke$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_stroke$QE, digits = 2)), ", df = ", .(ma_stroke$k - ma_stroke$p), " (", italic(p), .(p_stroke_het), "); ", I^2, " = ", .(round(ma_stroke$I2)), "%"))) # Overall effect text(-10, -4, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_stroke$zval, digits = 2)), " (", italic(p), .(p_stroke_oe), ")"))) # Add label for axes text(log(c(0.1, 10)), -5, c("Favours combination of SGLT-2 inhibitor and GLP-1 RA", "Favours SGLT-2 inhibitor or GLP-1 RA monotherapy"), pos = c(4,2), offset = -7, cex = 0.80) # Close PNG device dev.off() ######################### All-cause mortality, HHF, KF ######################### # Create dataframe for all-cause mortality, HHF, KF acm_hhf_kf_data <- rbind(acm_data, kf_data, hhf_data) # Create factor of outcome acm_hhf_kf_data <- acm_hhf_kf_data %>% mutate(outcome = case_when(outcome == "HHF" ~ 1, outcome == "KF" ~ 2, outcome == "ACM" ~ 3)) # Fit random effects model for HHF ma_hhf <- rma( measure = "RR", # Risk Ratio ai = events_exp, # Events in experimental group n1i = n_exp, # Total in experimental group ci = events_cont, # Events in control group n2i = n_cont, # Total in control group slab = study, # Study labels data = hhf_data, # Data frame method = "REML" # Method for random-effects model ) # Extract point estimate and confidence intervals HHF pred_hhf <- predict(ma_hhf, transf = exp, digits = 2) pred_hhf <- fmtx(c(pred_hhf$pred, pred_hhf$ci.lb, pred_hhf$ci.ub), digits=2) # Fit random effects model for KF ma_kf <- rma( measure = "RR", # Risk Ratio ai = events_exp, # Events in experimental group n1i = n_exp, # Total in experimental group ci = events_cont, # Events in control group n2i = n_cont, # Total in control group slab = study, # Study labels data = kf_data, # Data frame method = "REML" # Method for random-effects model ) # Extract point estimate and confidence intervals KF pred_kf <- predict(ma_kf, transf = exp, digits = 2) pred_kf <- fmtx(c(pred_kf$pred, pred_kf$ci.lb, pred_kf$ci.ub), digits=2) # Fit random effects model for ACM ma_acm <- rma( measure = "RR", # Risk Ratio ai = events_exp, # Events in experimental group n1i = n_exp, # Total in experimental group ci = events_cont, # Events in control group n2i = n_cont, # Total in control group slab = study, # Study labels data = acm_data, # Data frame method = "REML" # Method for random-effects model ) # Extract point estimate and confidence intervals ACM pred_acm <- predict(ma_acm, transf = exp, digits = 2) pred_acm <- fmtx(c(pred_acm$pred, pred_acm$ci.lb, pred_acm$ci.ub), digits=2) # Extract weights hhf_weights <- as.data.frame(paste0(fmtx(weights(ma_hhf), digits = 1), "%")) %>% #HHF rename(weight = `paste0(fmtx(weights(ma_hhf), digits = 1), "%")`) kf_weights <- as.data.frame(paste0(fmtx(weights(ma_kf), digits = 1), "%")) %>% #KF rename(weight = `paste0(fmtx(weights(ma_kf), digits = 1), "%")`) acm_weights <- as.data.frame(paste0(fmtx(weights(ma_acm), digits = 1), "%")) %>% #ACM rename(weight = `paste0(fmtx(weights(ma_acm), digits = 1), "%")`) # Create dataframe of ordered weights weights_acm_hhf_kf <- rbind(acm_weights, kf_weights, hhf_weights) %>% mutate(weight_num = as.numeric(gsub("%", "", weight)), #make weights numeric weight_scaled = 0.7 + 0.02 * weight_num) # Add weights to df acm_hhf_kf_data <- cbind(acm_hhf_kf_data, weights_acm_hhf_kf) # Sort variabels by weight acm_hhf_kf_obs_data <- acm_hhf_kf_obs_data %>% arrange(outcome, desc(weight_num)) # Fit random effects model for all outcomes ma_acm_hhf_kf <- rma( measure = "RR", # Risk Ratio ai = events_exp, # Events in experimental group n1i = n_exp, # Total in experimental group ci = events_cont, # Events in control group n2i = n_cont, # Total in control group slab = study, # Study labels data = acm_hhf_kf_obs_data, # Data frame method = "REML" # Method for random-effects model ) # Set up png for observational data for HHF, KF, ACM, CVM, MI, and stroke png("fig_3.png", width = 16000, height = 18000, res = 1200) # Expand margins par(xpd = TRUE) # Create forest plot for HHF, KF, and ACM forest(ma_acm_hhf_kf, xlim = c(-10, 4), at = log(c(0.1, 0.5, 1, 2, 10)), atransf = exp, order = outcome, xlab = "", lty = c(1, 1, 0), ilab = cbind(events_exp, n_exp, events_cont, n_cont, weight), ilab.xpos = c(-7, -6, -5, -4, -3), cex = 0.80, ylim = c(-2, 39), rows = c(0:4, 11:18, 25:35), mlab = "", psize = weight_scaled, header = c("Study", "Risk ratio [95% CI]"), showweights = FALSE, addfit = FALSE) # Add horizontal line at the top segments(-10, 37, 4, 37, lwd = 0.8) # Add labels on top text(c(-7, -6, -5, -4, -3), 38, c("Events", "Total", "Events", "Total", "Weight"), cex = 0.80, font = 2) text(c(-6.5, -4.5, 0), 39, c("Combination therapy", "Monotherapy", "Risk ratio"), cex = 0.80, font = 2) text(0, 38, c("MH Random, 95% CI"), cex = 0.80, font = 2) # Add pooled effect estimates per outcome addpoly(ma_acm, row = 23, cex = 0.80) addpoly(ma_kf, row = 9, cex = 0.80) addpoly(ma_hhf, row = -2, cex = 0.80) #Hide study labels and model rect(-10, 36, -7.25, -3, col = "White", border = NA) # Add outcome labels text(c(-10), c(36, 19, 5), c("All-cause mortality", "Kidney composite endpoint", "Hospitalisation for heart failure"), pos = 4, font = 2, cex = 0.80) # Add study labels text(c(-10), c(0:4, 11:18, 25:35), c("Patel et al (2024) [40]", "Dave et al (2021) [30]", "Lopez et al (2022) [37]", "Lau et al (2022) [35]", "Liu et al (2025) [36]", "Riley et al (2023) [41]", "Jhu et al (2024) [33]", "Patel et al (2024) [40]", "Kobayashi et al (2023) [34]", "Simms-Williams et al (2024a) [43]", "Schechter et al (2023) [42]", "Simms-Williams et al (2024b) [43]", "Lau et al (2022) [35]", "Riley et al (2023) [41]", "Jhu et al (2024) [33]", "Patel et al (2024) [40]", "Simms-Williams et al (2024b) [43]", "Simms-Williams et al (2024a) [43]", "Jensen et al (2020) [32]", "Lau et al (2022) [35]", "Dave et al (2021) [30]", "Luo et al (2023) [38]", "Liu et al (2025) [36]", "Gorgojo-Martínez et al (2017) [31]"), cex = 0.80, pos = 4) # Hide point estimates RE rect(1, c(24, 10, -1), 4, c(22, 8, -3), col = "White", border = NA) # Add pooled effect estimates estimates in bold text(4, 23, paste0(pred_acm[1], " [", pred_acm[2], ", ", pred_acm[3], "]"), pos = 2, cex = 0.80, font = 2) # ACM text(4, 9, paste0(pred_kf[1], " [", pred_kf[2], ", ", pred_kf[3], "]"), pos = 2, cex = 0.80, font = 2) # KF text(4, -2, paste0(pred_hhf[1], " [", pred_hhf[2], ", ", pred_hhf[3], "]"), pos = 2, cex = 0.80, font = 2) # HHF # Add Total labels: text(-10, c(23, 9, -2), pos = 4, "Total (95% CI)", font = 2, cex = 0.80) #Add text for total number of patients ACM text(c(-6, -4), 23, c(sum(acm_data$n_exp), sum(acm_data$n_cont)), font = 2, cex = 0.80) # Add text for total number of events ACM text(c(-7, -5), 23, c(sum(acm_data$events_exp), sum(acm_data$events_cont)), cex = 0.80) #Add text for total number of patients KF text(c(-6, -4), 9, c(sum(kf_data$n_exp), sum(kf_data$n_cont)), font = 2, cex = 0.80) # Add text for total number of events KF text(c(-7, -5), 9, c(sum(kf_data$events_exp), sum(kf_data$events_cont)), cex = 0.80) #Add text for total number of patients HHF text(c(-6, -4), -2, c(sum(hhf_data$n_exp), sum(hhf_data$n_cont)), font = 2, cex = 0.80) # Add text for total number of events HHF text(c(-7, -5), -2, c(sum(hhf_data$events_exp), sum(hhf_data$events_cont)), font = 2, cex = 0.80) # All-cause mortality # Extract p-values p_acm_het <- fmtp(ma_acm$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_acm_oe <- fmtp(ma_acm$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Heterogeneity text(-10, 22, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_acm$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_acm$QE, digits = 2)), ", df = ", .(ma_acm$k - ma_acm$p), " (", italic(p), .(p_acm_het), "); ", I^2, " = ", .(round(ma_acm$I2)), "%"))) # Overall effect text(-10, 21, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_acm$zval, digits = 2)), " (", italic(p), .(p_acm_oe), ")"))) # Kidney failure # Extract p-values p_kf_het <- fmtp(ma_kf$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_kf_oe <- fmtp(ma_kf$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Heterogeneity text(-10, 8, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_kf$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_kf$QE, digits = 2)), ", df = ", .(ma_kf$k - ma_kf$p), " (", italic(p), .(p_kf_het), "); ", I^2, " = ", .(round(ma_kf$I2)), "%"))) # Overall effect text(-10, 7, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_kf$zval, digits = 2)), " (", italic(p), .(p_kf_oe), ")"))) # Hospitalisation for heart failure # Extract p-values p_hhf_het <- fmtp(ma_hhf$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_hhf_oe <- fmtp(ma_hhf$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Heterogeneity text(-10, -3, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_hhf$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_hhf$QE, digits = 2)), ", df = ", .(ma_hhf$k - ma_hhf$p), " (", italic(p), .(p_hhf_het), "); ", I^2, " = ", .(round(ma_hhf$I2)), "%"))) # Overall effect text(-10, -4, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_hhf$zval, digits = 2)), " (", italic(p), .(p_hhf_oe), ")"))) # Add label for axes text(log(c(0.1, 10)), -6, c("Favours combination of SGLT-2 inhibitor and GLP-1 RA", "Favours SGLT-2 inhibitor or GLP-1 RA monotherapy"), pos = c(4,2), offset = -7, cex = 0.80) # Close PNG device dev.off() ################################################################################ ########################### Subgroup analyses for MACE ######################### ################################################################################ # Create SGLT2i control arm mace_data_sglt2i <- mace_data %>% filter(sglt2 == 1) %>% # Select rows with SGLT-2i as control select(-weight:-weight_scaled) %>% # Drop weight columns mutate(control = "SGLT-2 inhibitor", mace_n_cont = mace_n_sglt2, mace_events_cont = mace_events_sglt2) # Create GLP-1 RA control arm mace_data_glp1_ra <- mace_data %>% filter(glp_1_ra == 1) %>% # Select rows with GLP-1 RA as control select(-weight:-weight_scaled) %>% # Drop weight columns mutate(control = "GLP-1 receptor agonist", mace_n_cont = mace_n_glp1_ra, mace_events_cont = mace_events_glp1_ra) # Fit random effects model for SGLT-2i ma_mace_sglt2i <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_sglt2, # Events in control group n2i = mace_n_sglt2, # Total in control group slab = study, # Study labels data = mace_data_sglt2i, # Data frame method = "REML" # Method for random-effects model ) # Extract point estimate and confidence intervals SGLT2i pred_mace_sglt2i <- predict(ma_mace_sglt2i, transf = exp, digits = 2) pred_mace_sglt2i <- fmtx(c(pred_mace_sglt2i$pred, pred_mace_sglt2i$ci.lb, pred_mace_sglt2i$ci.ub), digits=2) # Extract weights for SGLT2i mace_sglt2i_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_sglt2i), digits = 1), "%")) %>% rename(weight = `paste0(fmtx(weights(ma_mace_sglt2i), digits = 1), "%")`) %>% mutate(weight_num = as.numeric(gsub("%", "", weight)), weight_scaled = 0.7 + weight_num * 0.02) #Add numeric weights # Add weights to dataframe mace_data_sglt2i <- cbind(mace_data_sglt2i, mace_sglt2i_weights) # Fit random effects model for GLP-1 RA ma_mace_glp1ra <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_glp1_ra, # Events in control group n2i = mace_n_glp1_ra, # Total in control group slab = study, # Study labels data = mace_data_glp1_ra, # Data frame method = "REML" # Method for random-effects model ) # Extract point estimate and confidence intervals GLP-1 RA pred_mace_glp1ra <- predict(ma_mace_glp1ra, transf = exp, digits = 2) pred_mace_glp1ra <- fmtx(c(pred_mace_glp1ra$pred, pred_mace_glp1ra$ci.lb, pred_mace_glp1ra$ci.ub), digits=2) # Extract weights for GLP-1 RA mace_glp1ra_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_glp1ra), digits = 1), "%")) %>% rename(weight = `paste0(fmtx(weights(ma_mace_glp1ra), digits = 1), "%")`) %>% mutate(weight_num = as.numeric(gsub("%", "", weight)), weight_scaled = 0.7 + weight_num * 0.02) #Add numeric weights # Add weights to dataframe mace_data_glp1_ra <- cbind(mace_data_glp1_ra, mace_glp1ra_weights) # bind dataframes together mace_data_control <- rbind(mace_data_sglt2i, mace_data_glp1_ra) mace_data_control <- mace_data_control %>% mutate(control_num = ifelse(control == "SGLT-2 inhibitor", 2, 1)) # order dataframe by weight mace_data_control <- mace_data_control %>% arrange(control, desc(weight_num)) # Run meta-analysis for control intervention ma_mace_control <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_control, # Data frame method = "REML" # Method for random-effects model ) # Set up PNG for forest plot for MACE by control intervention png("fig_4.png", width = 16000, height = 8500, res = 1200) # Create forest plot fp_mace_control <- forest(ma_mace_control, xlim = c(-10, 4), at = log(c(0.1, 0.5, 1, 2, 10)), atransf = exp, order = control_num, efac = 0.5, xlab = "", lty = c(1, 1, 0), ilab = cbind(mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont, weight), ilab.xpos = c(-7, -6, -5, -4, -3), cex = 0.80, ylim = c(-2, 20), rows = c(0:3, 10:16), mlab = "", psize = weight_scaled, header = c("Study", "Risk ratio [95% CI]"), showweights = FALSE, addfit = FALSE) # Add pooled effect of SGLT2-i and GLP-1 RA addpoly(ma_mace_sglt2i, row = 8, cex = 0.80) addpoly(ma_mace_glp1ra, row = -2, cex = 0.80) # hide non-bold summary estimates and study labels rect(-10, 17, -7.25, -3, col = "White", border = NA) # Hide point estimates RE rect(2.4, c(9, -1), 4, c(7, -3), col = "White", border = NA) rect(-10, c(9, -1), -8, c(7, -3), col = "White", border = NA) # Place summary estimates back in bold text(fp_mace_control$textpos[2], 8, paste0(pred_mace_sglt2i[1], " [", pred_mace_sglt2i[2], ", ", pred_mace_sglt2i[3], "]"), pos=2, font = 2, cex = 0.80) text(fp_mace_control$textpos[2], -2, paste0(pred_mace_glp1ra[1], " [", pred_mace_glp1ra[2], ", ", pred_mace_glp1ra[3], "]"), pos=2, font = 2, cex = 0.80) # Add horizontal line at the top segments(-10, 18, 4, 18, lwd = 0.8) # Expand margins par(xpd = NA) # Add labels on top text(c(-7, -6, -5, -4, -3), 19, c("Events", "Total", "Events", "Total", "Weight"), cex = 0.80, font = 2) text(c(-6.5, -4.5, 0), 20, c("Combination therapy", "Monotherapy", "Risk ratio"), cex = 0.80, font = 2) text(0, 19, c("MH Random, 95% CI"), cex = 0.80, font = 2) # Add subgroup labels text(c(-10), c(17, 4), c("SGLT-2 inhibitor", "GLP-1 receptor agonist"), pos = 4, font = 2, cex = 0.80) # Add study labels text(-10, c(0:3, 10:16), c("Simms-Williams et al (2024a) [43]", "Jensen et al (2020) [43]", "Lau et al (2022) [35]", "Marfella et al (2024) [39]", "Chaiyakunapruk et al (2025) [29]", "Jhu et al (2024) [33]", "Simms-Williams et al (2024b) [43]", "Lopez et al (2022) [37]", "Liu et al (2025) [36]", "Jensen et al (2020) [32]", "Marfella et al (2024) [39]"), cex = 0.80, pos = 4) # Add subtotal labels text(-10, c(8, -2), "Total (95% CI)", font = 2, cex = 0.80, pos = 4) #Add text for total events text(c(-6, -4), 8, c(sum(mace_data_sglt2i$mace_n_exp), sum(mace_data_sglt2i$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), 8, c(sum(mace_data_sglt2i$mace_events_exp), sum(mace_data_sglt2i$mace_events_cont)), cex = 0.80) text(c(-6, -4), -2, c(sum(mace_data_glp1_ra$mace_n_exp), sum(mace_data_glp1_ra$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), -2, c(sum(mace_data_glp1_ra$mace_events_exp), sum(mace_data_glp1_ra$mace_events_cont)), cex = 0.80) # Add sum of weights text(-3, c(8, -2), "100%", cex = 0.80, font = 2) # SGLT-2 inhibitors # Extract p-values p_sglt2i_het <- fmtp(ma_mace_sglt2i$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_sglt2i_oe <- fmtp(ma_mace_sglt2i$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Heterogeneity text(-10, 7, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_sglt2i$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_sglt2i$QE, digits = 2)), ", df = ", .(ma_mace_sglt2i$k - ma_mace_sglt2i$p), " (", italic(p), .(p_sglt2i_het), "); ", I^2, " = ", .(round(ma_mace_sglt2i$I2)), "%"))) # Overall effect text(-10, 6, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_sglt2i$zval, digits = 2)), " (", italic(p), .(p_sglt2i_oe), ")"))) # GLP-1 RAs # Extract p-values p_glp1ra_het <- fmtp(ma_mace_glp1ra$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_glp1ra_oe <- fmtp(ma_mace_glp1ra$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Heterogeneity text(-10, -3, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_glp1ra$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_glp1ra$QE, digits = 2)), ", df = ", .(ma_mace_glp1ra$k - ma_mace_glp1ra$p), " (", italic(p), .(p_glp1ra_het), "); ", I^2, " = ", .(round(ma_mace_glp1ra$I2)), "%"))) # Overall effect text(-10, -4, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_glp1ra$zval, digits = 2)), " (", italic(p), .(p_glp1ra_oe), ")"))) # Add label for axes text(log(c(0.1, 10)), -5, c("Favours combination of SGLT-2 inhibitor and GLP-1 RA", "Favours SGLT-2 inhibitor or GLP-1 RA monotherapy"), pos = c(4,2), offset = -7, cex = 0.80) # Close PNG device dev.off() ############################ History of CVD, HF, CKD ########################### # Filter studies with patients at very high risk of CVD mace_very_high_risk_data <- data %>% filter(high_risk_mace == 1 & high_risk == 1) %>% select(study, high_risk, mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont) # Run meta-analysis for patients with history of CVD ma_mace_very_high_risk <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_very_high_risk_data, # Data frame method = "REML" # Method for random-effects model ) # Extract weights mace_very_high_risk_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_very_high_risk), digits = 1), "%")) %>% rename(weight = `paste0(fmtx(weights(ma_mace_very_high_risk), digits = 1), "%")`) %>% mutate(weight_num = as.numeric(gsub("%", "", weight)), weight_scaled = 0.7 + weight_num * 0.02) #Add numeric weights #Extract rounded estimate and CI bound pred_mace_very_high_risk <- predict(ma_mace_very_high_risk, transf = exp, digits = 2) pred_mace_very_high_risk <- fmtx(c(pred_mace_very_high_risk$pred, pred_mace_very_high_risk$ci.lb, pred_mace_very_high_risk$ci.ub), digits=2) # Add weights to dataframe mace_very_high_risk_data <- cbind(mace_very_high_risk_data, mace_very_high_risk_weights) # Filter studies with patients at high risk of CVD mace_high_risk_data <- data %>% filter(high_risk_mace == 1 & high_risk == 0) %>% select(study, high_risk, mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont) # Run meta-analysis for patients without a history of CVD ma_mace_high_risk <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_high_risk_data, # Data frame method = "REML" # Method for random-effects model ) # Extract weights mace_high_risk_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_high_risk), digits = 1), "%")) %>% rename(weight = `paste0(fmtx(weights(ma_mace_high_risk), digits = 1), "%")`) %>% mutate(weight_num = as.numeric(gsub("%", "", weight)), weight_scaled = 0.70 + weight_num * 0.02) #Add numeric weights #Extract rounded estimate and CI bound pred_mace_high_risk <- predict(ma_mace_high_risk, transf = exp, digits = 2) pred_mace_high_risk <- fmtx(c(pred_mace_high_risk$pred, pred_mace_high_risk$ci.lb, pred_mace_high_risk$ci.ub), digits=2) # Add weights to dataframe mace_high_risk_data <- cbind(mace_high_risk_data, mace_high_risk_weights) # Bind dataframes with weights back together mace_cvd_data <- rbind(mace_high_risk_data, mace_very_high_risk_data) # Order dataframe mace_cvd_data <- mace_cvd_data %>% arrange(high_risk, desc(weight_num)) # Run meta-analysis by risk of CVD ma_mace_cvd <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_cvd_data, # Data frame method = "REML" # Method for random-effects model ) mace_cvd_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_cvd), digits = 1), "%")) %>% rename(weight_cvd = `paste0(fmtx(weights(ma_mace_cvd), digits = 1), "%")`) %>% mutate(weight_num_cvd = as.numeric(gsub("%", "", weight_cvd)), weight_scaled_cvd = 0.70 + weight_num_cvd * 0.02) #Add numeric weights #Extract rounded estimate and CI bound pred_mace_cvd <- predict(ma_mace_cvd, transf = exp, digits = 2) pred_mace_cvd <- fmtx(c(pred_mace_cvd$pred, pred_mace_cvd$ci.lb, pred_mace_cvd$ci.ub), digits=2) # Add weights mace_cvd_data <- cbind(mace_cvd_data, mace_cvd_weights) # Rerun meta-analysis by risk of CVD ma_mace_cvd <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_cvd_data, # Data frame method = "REML" # Method for random-effects model ) par(mar = c(10, 5, 1.3, 5), mgp = c(3, 1, 0), tcl = -0.2) # Set up PNG for forest plot for MACE by history of CVD, CKD, HF png("supp_fig_3.png", width = 16000, height = 8000, res = 1200) fp_mace_cvd <- forest(ma_mace_cvd, xlim=c(-10, 4), at=log(c(0.1, 0.5, 1, 2, 10)), atransf=exp, order = high_risk, xlab = "", lty = c(1, 1, 0), ilab = cbind(mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont, mace_cvd_weights$weight_cvd), ilab.xpos=c(-7, -6, -5, -4, -3), cex=0.80, ylim=c(-1, 26), rows = c(5:10, 17:22), mlab = "", psize = weight_scaled_cvd, header = c("Study", "Risk ratio [95% CI]"), showweights = FALSE, addfit = TRUE) # Add pooled effect of studies with and without patients with CVD addpoly(ma_mace_very_high_risk, row = 15, cex = 0.80) addpoly(ma_mace_high_risk, row = 3, cex = 0.80) addpoly(ma_mace_cvd, row = -1, cex = 0.80) # hide non-bold summary estimates and study labels rect(-10, 23, -7.25, -3, col = "White", border = NA) # Hide point estimates RE rect(2, c(16, 4, 0), 4, c(14, 2, -2), col = "White", border = NA) # Place summary estimates back in bold text(fp_mace_cvd$textpos[2], -1, paste0(pred_mace_cvd[1], " [", pred_mace_cvd[2], ", ", pred_mace_cvd[3], "]"), pos=2, font = 2, cex = 0.80) text(fp_mace_cvd$textpos[2], 15, paste0(pred_mace_very_high_risk[1], " [", pred_mace_very_high_risk[2], ", ", pred_mace_very_high_risk[3], "]"), pos=2, font = 2, cex = 0.80) text(fp_mace_cvd$textpos[2], 3, paste0(pred_mace_high_risk[1], " [", pred_mace_high_risk[2], ", ", pred_mace_high_risk[3], "]"), pos=2, font = 2, cex = 0.80) # Add horizontal line at the top segments(-10, 24, 4, 24, lwd = 0.8) # Expand margins par(xpd = NA) # Add labels on top text(c(-7, -6, -5, -4, -3), 25, c("Events", "Total", "Events", "Total", "Weight"), cex = 0.80, font = 2) text(c(-6.5, -4.5, 0), 26, c("Combination therapy", "Monotherapy", "Risk ratio"), cex = 0.80, font = 2) text(0, 25, c("MH Random, 95% CI"), cex = 0.80, font = 2) # Add outcome labels text(c(-10), c(23, 11), c("History of CVD, CKD, or heart failure", "No history of CVD, CKD, or heart failure"), pos = 4, font = 2, cex = 0.80) # Add study labels text(-10, c(5:10, 17:22), c("Jhu et al (2024) [33]", "Dave et al (2021) [30]", "Jensen et al (2020) [32]", "Lau et al (2022) [35]", "Simms-Williams et al (2024b) [42]", "Simms-Williams et al (2024a) [43]", "Chaiyakunapruk et al (2025) [29]", "Lopez et al (2022) [37]", "Simms-Williams et al (2024b) [43]", "Simms-Williams et al (2024a) [43]", "Marfella et al (2024) [39]"), cex = 0.80, pos = 4) # Add Total labels: text(-10, -1, pos = 4, "Total (95% CI)", font = 2, cex = 0.80) # Add subtotal labels text(-10, c(3, 15), "Subtotal (95% CI)", font = 2, cex = 0.80, pos = 4) #Add text for total events text(c(-6, -4), -1, c(sum(mace_cvd_data$mace_n_exp), sum(mace_cvd_data$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), -1, c(sum(mace_cvd_data$mace_events_exp), sum(mace_cvd_data$mace_events_cont)), cex = 0.80) # Add text for subtotal events text(c(-6, -4), 15, c(sum(mace_very_high_risk_data$mace_n_exp), sum(mace_very_high_risk_data$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), 15, c(sum(mace_very_high_risk_data$mace_events_exp), sum(mace_very_high_risk_data$mace_events_cont)), cex = 0.80) text(c(-6, -4), 3, c(sum(mace_high_risk_data$mace_n_exp), sum(mace_high_risk_data$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), 3, c(sum(mace_high_risk_data$mace_events_exp), sum(mace_high_risk_data$mace_events_cont)), cex = 0.80) # Calculate sum of weights for observational studies mace_cvd_data %>% group_by(high_risk) %>% summarise(sum_weight = sum(weight_num_cvd)) # Add sum of weights text(-3, c(15, 3, -1), c("50.2%", "49.8%", "100%"), cex = 0.80, font = 2) # Extract p-values p_very_high_het <- fmtp(ma_mace_very_high_risk$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_high_het <- fmtp(ma_mace_high_risk$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_pooled_het_hr <- fmtp(ma_mace_cvd$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_very_high_oe <- fmtp(ma_mace_very_high_risk$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_high_oe <- fmtp(ma_mace_high_risk$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_pooled_oe_hr <- fmtp(ma_mace_cvd$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Subgroup difference p-value p_subgroup_diff_hr <- fmtp(0.10, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Heterogeneity statistics # History of CVD, CKD or HF text(-10, 14, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_very_high_risk$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_very_high_risk$QE, digits = 2)), ", df = ", .(ma_mace_very_high_risk$k - ma_mace_very_high_risk$p), " (", italic(p), .(p_very_high_het), "); ", I^2, " = ", .(round(ma_mace_very_high_risk$I2)), "%"))) # No history of CVD, CKD or HF text(-10, 2, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_high_risk$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_high_risk$QE, digits = 2)), ", df = ", .(ma_mace_high_risk$k - ma_mace_high_risk$p), " (", italic(p), .(p_high_het), "); ", I^2, " = ", .(round(ma_mace_high_risk$I2)), "%"))) # Pooled text(-10, -2, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_cvd$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_cvd$QE, digits = 2)), ", df = ", .(ma_mace_cvd$k - ma_mace_cvd$p), " (", italic(p), .(p_pooled_het_hr), "); ", I^2, " = ", .(round(ma_mace_cvd$I2)), "%"))) # Overall effect # History of CVD, CKD or HF text(-10, 13, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_very_high_risk$zval, digits = 2)), " (", italic(p), .(p_very_high_oe), ")"))) # No history of CVD, CKD or HF text(-10, 1, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_high_risk$zval, digits = 2)), " (", italic(p), .(p_high_oe), ")"))) # Pooled text(-10, -3, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_cvd$zval, digits = 2)), " (", italic(p), .(p_pooled_oe_hr), ")"))) # Determine Q values rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_cvd_data, # Data frame method = "REML", # Method for random-effects model mod = high_risk ) # Subgroup differences text(-10, -4, pos = 4, cex = 0.80, bquote(paste("Test for subgroup differences: ", "Chi"^2, " = ", .(fmtx(2.66, digits = 2)), ", df = ", 1, " (", italic(p), .(p_subgroup_diff_hr), ")"))) # Add label for axes text(log(c(0.1, 10)), -5, c("Favours combination of SGLT-2 inhibitor and GLP-1 RA", "Favours SGLT-2 inhibitor or GLP-1 RA monotherapy"), pos = c(4,2), offset = -7, cex = 0.80) # Close PNG device dev.off() ########################## Subgroup by baseline HbA1c ########################## # Filter studies with baseline HbA1c mace_data_hba1c <- mace_data %>% filter(!is.na(hba1c)) %>% mutate(hba1c_cat = ifelse(hba1c >8, "HbA1c >8%", "HbA1c ≤8"), hba1c_num = ifelse(hba1c_cat == "HbA1c >8%", 1, 0)) # Filter studies with high baseline HbA1c mace_data_hba1c_h <- mace_data_hba1c %>% filter(hba1c >8) %>% select(study,hba1c_cat, hba1c_num, mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont) # Run meta-analysis for patients with high HbA1c ma_mace_hba1c_h <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_hba1c_h, # Data frame method = "REML" # Method for random-effects model ) # Extract weights mace_hba1c_h_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_hba1c_h), digits = 1), "%")) %>% rename(weight = `paste0(fmtx(weights(ma_mace_hba1c_h), digits = 1), "%")`) %>% mutate(weight_num = as.numeric(gsub("%", "", weight)), weight_scaled = 0.7 + weight_num * 0.02) #Add numeric weights #Extract rounded estimate and CI bound pred_mace_hba1c_h <- predict(ma_mace_hba1c_h, transf = exp, digits = 2) pred_mace_hba1c_h <- fmtx(c(pred_mace_hba1c_h$pred, pred_mace_hba1c_h$ci.lb, pred_mace_hba1c_h$ci.ub), digits=2) # Add weights to dataframe mace_data_hba1c_h <- cbind(mace_data_hba1c_h, mace_hba1c_h_weights) # Filter studies with low baseline HbA1c mace_data_hba1c_l <- mace_data_hba1c %>% filter(hba1c <= 8) %>% select(study,hba1c_cat, hba1c_num, mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont) # Run meta-analysis for patients with low HbA1c ma_mace_hba1c_l <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_hba1c_l, # Data frame method = "REML" # Method for random-effects model ) # Extract weights mace_hba1c_l_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_hba1c_l), digits = 1), "%")) %>% rename(weight = `paste0(fmtx(weights(ma_mace_hba1c_l), digits = 1), "%")`) %>% mutate(weight_num = as.numeric(gsub("%", "", weight)), weight_scaled = 0.7 + weight_num * 0.02) #Add numeric weights #Extract rounded estimate and CI bound pred_mace_hba1c_l <- predict(ma_mace_hba1c_l, transf = exp, digits = 2) pred_mace_hba1c_l <- fmtx(c(pred_mace_hba1c_l$pred, pred_mace_hba1c_l$ci.lb, pred_mace_hba1c_l$ci.ub), digits=2) # Add weights to dataframe mace_data_hba1c_l <- cbind(mace_data_hba1c_l, mace_hba1c_l_weights) # Bind dataframes with weights back together mace_data_hba1c <- rbind(mace_data_hba1c_h, mace_data_hba1c_l) # Order dataframe by subgroup and weights # Order dataframe mace_data_hba1c <- mace_data_hba1c %>% arrange(hba1c_num, desc(weight_num)) # Run meta-analysis for stratified for baseline HbA1c ma_mace_hba1c <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_hba1c, # Data frame method = "REML" # Method for random-effects model ) #Extract rounded estimate and CI bound pred_mace_hba1c <- predict(ma_mace_hba1c, transf = exp, digits = 2) pred_mace_hba1c <- fmtx(c(pred_mace_hba1c$pred, pred_mace_hba1c$ci.lb, pred_mace_hba1c$ci.ub), digits=2) # Extract weights mace_hba1c_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_hba1c), digits = 1), "%")) %>% rename(weight_hba1c = `paste0(fmtx(weights(ma_mace_hba1c), digits = 1), "%")`) %>% mutate(weight_num_hba1c = as.numeric(gsub("%", "", weight_hba1c)), weight_scaled_hba1c = 0.70 + weight_num_hba1c * 0.02) #Add numeric weights # Add weights to dataframe mace_data_hba1c <- cbind(mace_data_hba1c, mace_hba1c_weights) # Rerun meta-analysis stratified for baseline HbA1c ma_mace_hba1c <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_hba1c, # Data frame method = "REML" # Method for random-effects model ) # Set up PNG for forest plot for MACE by baseline HbA1c png("supp_fig_4.png", width = 16000, height = 7000, res = 1200) fp_mace_hba1c <- forest(ma_mace_hba1c, xlim=c(-10, 4), at=log(c(0.1, 0.5, 1, 2, 10)), atransf=exp, order = hba1c_num, xlab = "", lty = c(1, 1, 0), ilab = cbind(mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont, weight_hba1c), ilab.xpos=c(-7, -6, -5, -4, -3), cex=0.80, ylim=c(-1, 23), rows = c(5:8, 15:19), mlab = "", psize = weight_scaled_hba1c, header = c("Study", "Risk ratio [95% CI]"), showweights = FALSE, addfit = TRUE) # Add pooled estimates of studies with and without high baseline HbA1c addpoly(ma_mace_hba1c_h, row = 13, cex = 0.80) addpoly(ma_mace_hba1c_l, row = 3, cex = 0.80) # hide non-bold summary estimates rect(-10, c(14, 4), -7.25, c(12, 2), col = "White", border = NA) # Hide point estimates RE rect(2.5, c(14, 4), 4, c(12, -4), col = "White", border = NA) # Place summary estimates back in bold text(fp_mace_hba1c$textpos[2], -1, paste0(pred_mace_hba1c[1], " [", pred_mace_hba1c[2], ", ", pred_mace_hba1c[3], "]"), pos=2, font = 2, cex = 0.80) text(fp_mace_hba1c$textpos[2], 13, paste0(pred_mace_hba1c_h[1], " [", pred_mace_hba1c_h[2], ", ", pred_mace_hba1c_h[3], "]"), pos=2, font = 2, cex = 0.80) text(fp_mace_hba1c$textpos[2], 3, paste0(pred_mace_hba1c_l[1], " [", pred_mace_hba1c_l[2], ", ", pred_mace_hba1c_l[3], "]"), pos=2, font = 2, cex = 0.80) # Add horizontal line at the top segments(-10, 21, 4, 21, lwd = 0.8) # Expand margins par(xpd = NA) # Add labels on top text(c(-7, -6, -5, -4, -3), 22, c("Events", "Total", "Events", "Total", "Weight"), cex = 0.80, font = 2) text(c(-6.5, -4.5, 0), 23, c("Combination therapy", "Monotherapy", "Risk ratio"), cex = 0.80, font = 2) text(0, 22, c("MH Random, 95% CI"), cex = 0.80, font = 2) # Add outcome labels text(c(-10), c(20, 9), c("Baseline HbA1c >64 mmol/mol", "Baseline HbA1c ≤64 mmol/mol"), pos = 4, font = 2, cex = 0.80) # Add Total labels: text(-10, -1, pos = 4, "Total (95% CI)", font = 2, cex = 0.80) # Add subtotal labels text(-10, c(3, 13), "Subtotal (95% CI)", font = 2, cex = 0.80, pos = 4) #Add text for total events text(c(-6, -4), -1, c(sum(mace_data_hba1c$mace_n_exp), sum(mace_data_hba1c$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), -1, c(sum(mace_data_hba1c$mace_events_exp), sum(mace_data_hba1c$mace_events_cont)), cex = 0.80) # Add text for subtotal events text(c(-6, -4), 13, c(sum(mace_data_hba1c_h$mace_n_exp), sum(mace_data_hba1c_h$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), 13, c(sum(mace_data_hba1c_h$mace_events_exp), sum(mace_data_hba1c_h$mace_events_cont)), cex = 0.80) text(c(-6, -4), 3, c(sum(mace_data_hba1c_l$mace_n_exp), sum(mace_data_hba1c_l$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), 3, c(sum(mace_data_hba1c_l$mace_events_exp), sum(mace_data_hba1c_l$mace_events_cont)), cex = 0.80) # Calculate sum of weights for studies with an HbA1c > and =< 8% mace_data_hba1c %>% group_by(hba1c_cat) %>% summarise(sum_weight = sum(weight_num_hba1c)) # Add sum of weights text(-3, c(13, 3, -1), c("56.6%", "43.4%", "100%"), cex = 0.80, font = 2) # Extract p-values p_hba1c_h_het <- fmtp(ma_mace_hba1c_h$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_hba1c_l_het <- fmtp(ma_mace_hba1c_l$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_hba1c_pooled_het <- fmtp(ma_mace_hba1c$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_hba1c_h_oe <- fmtp(ma_mace_hba1c_h$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_hba1c_l_oe <- fmtp(ma_mace_hba1c_l$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_hba1c_pooled_oe <- fmtp(ma_mace_hba1c$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Subgroup difference p-value p_subgroup_diff_hba1c <- fmtp(0.30, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Heterogeneity statistics # High HbA1c text(-10, 12, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_hba1c_h$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_hba1c_h$QE, digits = 2)), ", df = ", .(ma_mace_hba1c_h$k - ma_mace_hba1c_h$p), " (", italic(p), .(p_hba1c_h_het), "); ", I^2, " = ", .(round(ma_mace_hba1c_h$I2)), "%"))) # Low HbA1c text(-10, 2, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_hba1c_l$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_hba1c_l$QE, digits = 2)), ", df = ", .(ma_mace_hba1c_l$k - ma_mace_hba1c_l$p), " (", italic(p), .(p_hba1c_l_het), "); ", I^2, " = ", .(round(ma_mace_hba1c_l$I2)), "%"))) # Pooled text(-10, -2, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_hba1c$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_hba1c$QE, digits = 2)), ", df = ", .(ma_mace_hba1c$k - ma_mace_hba1c$p), " (", italic(p), .(p_hba1c_pooled_het), "); ", I^2, " = ", .(round(ma_mace_hba1c$I2)), "%"))) # Overall effect # High HbA1c text(-10, 11, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_hba1c_h$zval, digits = 2)), " (", italic(p), .(p_hba1c_h_oe), ")"))) # Low HbA1c text(-10, 1, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_hba1c_l$zval, digits = 2)), " (", italic(p), .(p_hba1c_l_oe), ")"))) # Pooled text(-10, -3, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_hba1c$zval, digits = 2)), " (", italic(p), .(p_hba1c_pooled_oe), ")"))) # Determine subgroup differences rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_hba1c, # Data frame method = "REML", # Method for random-effects model mod = hba1c_num ) # Subgroup differences text(-10, -4, pos = 4, cex = 0.80, bquote(paste("Test for subgroup differences: ", "Chi"^2, " = ", .(fmtx(1.09, digits = 2)), ", df = ", 1, " (", italic(p), .(p_subgroup_diff_hba1c), ")"))) # Add label for axes text(log(c(0.1, 10)), -5, c("Favours combination of SGLT-2 inhibitor and GLP-1 RA", "Favours SGLT-2 inhibitor or GLP-1 RA monotherapy"), pos = c(4,2), offset = -7, cex = 0.80) # Close PNG device dev.off() ######################### Subgroup by diabetes duration ######################## # Filter studies with diabetes duration mace_data_dm_years <- mace_data %>% filter(!is.na(dm_years)) %>% mutate(dm_years_cat = ifelse(dm_years >10, "Diabetes duration > 10 years", "Diabetes duration ≤10 years"), dm_years_num = ifelse(dm_years_cat == "Diabetes duration > 10 years", 1, 0)) # Filter studies with long diabetes duration mace_data_dm_years_l <- mace_data_dm_years %>% filter(dm_years >10) %>% select(study,dm_years_cat, dm_years_num, mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont) # Run meta-analysis for patients with long diabetes duration ma_mace_dm_years_l <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_dm_years_l, # Data frame method = "REML" # Method for random-effects model ) # Extract weights mace_dm_years_l_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_dm_years_l), digits = 1), "%")) %>% rename(weight = `paste0(fmtx(weights(ma_mace_dm_years_l), digits = 1), "%")`) %>% mutate(weight_num = as.numeric(gsub("%", "", weight)), weight_scaled = 0.7 + weight_num * 0.02) #Add numeric weights #Extract rounded estimate and CI bound pred_mace_dm_years_l <- predict(ma_mace_dm_years_l, transf = exp, digits = 2) pred_mace_dm_years_l <- fmtx(c(pred_mace_dm_years_l$pred, pred_mace_dm_years_l$ci.lb, pred_mace_dm_years_l$ci.ub), digits=2) # Add weights to dataframe mace_data_dm_years_l <- cbind(mace_data_dm_years_l, mace_dm_years_l_weights) # Filter studies with short diabetes duration mace_data_dm_years_s <- mace_data_dm_years %>% filter(dm_years <=10) %>% select(study,dm_years_cat, dm_years_num, mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont) # Run meta-analysis for patients with short diabetes duration ma_mace_dm_years_s <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_dm_years_s, # Data frame method = "REML" # Method for random-effects model ) # Extract weights mace_dm_years_s_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_dm_years_s), digits = 1), "%")) %>% rename(weight = `paste0(fmtx(weights(ma_mace_dm_years_s), digits = 1), "%")`) %>% mutate(weight_num = as.numeric(gsub("%", "", weight)), weight_scaled = 0.7 + weight_num * 0.02) #Add numeric weights #Extract rounded estimate and CI bound pred_mace_dm_years_s <- predict(ma_mace_dm_years_s, transf = exp, digits = 2) pred_mace_dm_years_s <- fmtx(c(pred_mace_dm_years_s$pred, pred_mace_dm_years_s$ci.lb, pred_mace_dm_years_s$ci.ub), digits=2) # Rerun meta-analysis stratified for baseline HbA1c ma_mace_dm_years <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_dm_years, # Data frame method = "REML" # Method for random-effects model ) # Extract weights mace_dm_years_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_dm_years), digits = 1), "%")) %>% rename(weight_dm_years = `paste0(fmtx(weights(ma_mace_dm_years), digits = 1), "%")`) %>% mutate(weight_num_dm_years = as.numeric(gsub("%", "", weight_dm_years)), weight_scaled_dm_years = 0.70 + weight_num_dm_years * 0.02) #Add numeric weights # Add weights to dataframe mace_data_dm_years <- cbind(mace_data_dm_years, mace_dm_years_weights) mace_data_dm_years <- mace_data_dm_years %>% arrange(dm_years_cat, desc(weight_num_dm_years)) # Rerun meta-analysis stratified for diabetes duration ma_mace_dm_years <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_dm_years, # Data frame method = "REML" # Method for random-effects model ) #Extract rounded estimate and CI bound pred_mace_dm_years <- predict(ma_mace_dm_years, transf = exp, digits = 2) pred_mace_dm_years <- fmtx(c(pred_mace_dm_years$pred, pred_mace_dm_years$ci.lb, pred_mace_dm_years$ci.ub), digits=2) # Set up PNG for forest plot for MACE by baseline HbA1c png("supp_fig_5.png", width = 16000, height = 7000, res = 1200) fp_mace_dm_years <- forest(ma_mace_dm_years, xlim=c(-10, 4), at=log(c(0.1, 0.5, 1, 2, 10)), atransf=exp, order = dm_years_num, xlab = "", lty = c(1, 1, 0), ilab = cbind(mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont, weight_dm_years), ilab.xpos=c(-7, -6, -5, -4, -3), cex=0.80, ylim=c(-1, 20), rows = c(5:7, 14:16), mlab = "", psize = weight_scaled_dm_years, header = c("Study", "Risk ratio [95% CI]"), showweights = FALSE, addfit = TRUE) # Add pooled estimates of studies with and without high baseline HbA1c addpoly(ma_mace_dm_years_l, row = 12, cex = 0.80) addpoly(ma_mace_dm_years_s, row = 3, cex = 0.80) # hide non-bold summary estimates rect(-10, c(13, 4), -7.25, c(11, 2), col = "White", border = NA) # Hide point estimates RE rect(2.5, c(13, 4), 4, c(11, -4), col = "White", border = NA) # Place summary estimates back in bold text(fp_mace_dm_years$textpos[2], -1, paste0(pred_mace_dm_years[1], " [", pred_mace_dm_years[2], ", ", pred_mace_dm_years[3], "]"), pos=2, font = 2, cex = 0.80) text(fp_mace_dm_years$textpos[2], 12, paste0(pred_mace_dm_years_l[1], " [", pred_mace_dm_years_l[2], ", ", pred_mace_dm_years_l[3], "]"), pos=2, font = 2, cex = 0.80) text(fp_mace_dm_years$textpos[2], 3, paste0(pred_mace_dm_years_s[1], " [", pred_mace_dm_years_s[2], ", ", pred_mace_dm_years_s[3], "]"), pos=2, font = 2, cex = 0.80) # Add horizontal line at the top segments(-10, 18, 4, 18, lwd = 0.8) # Expand margins par(xpd = NA) # Add labels on top text(c(-7, -6, -5, -4, -3), 19, c("Events", "Total", "Events", "Total", "Weight"), cex = 0.80, font = 2) text(c(-6.5, -4.5, 0), 20, c("Combination therapy", "Monotherapy", "Risk ratio"), cex = 0.80, font = 2) text(0, 19, c("MH Random, 95% CI"), cex = 0.80, font = 2) # Add outcome labels text(c(-10), c(17, 8), c("Baseline diabetes duration >10 years", "Baseline diabetes duration ≤10 years"), pos = 4, font = 2, cex = 0.80) # Add Total labels: text(-10, -1, pos = 4, "Total (95% CI)", font = 2, cex = 0.80) # Add subtotal labels text(-10, c(3, 12), "Subtotal (95% CI)", font = 2, cex = 0.80, pos = 4) #Add text for total events text(c(-6, -4), -1, c(sum(mace_data_dm_years$mace_n_exp), sum(mace_data_dm_years$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), -1, c(sum(mace_data_dm_years$mace_events_exp), sum(mace_data_dm_years$mace_events_cont)), cex = 0.80) # Add text for subtotal events text(c(-6, -4), 12, c(sum(mace_data_dm_years_l$mace_n_exp), sum(mace_data_dm_years_l$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), 12, c(sum(mace_data_dm_years_l$mace_events_exp), sum(mace_data_dm_years_l$mace_events_cont)), cex = 0.80) text(c(-6, -4), 3, c(sum(mace_data_dm_years_s$mace_n_exp), sum(mace_data_dm_years_s$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), 3, c(sum(mace_data_dm_years_s$mace_events_exp), sum(mace_data_dm_years_s$mace_events_cont)), cex = 0.80) # Calculate sum of weights for studies with a diabetes duration > and =< 10 years mace_data_dm_years %>% group_by(dm_years_cat) %>% summarise(sum_weight = sum(weight_num_dm_years)) # Add sum of weights text(-3, c(12, 3, -1), c("49.9%", "50.2%", "100%"), cex = 0.80, font = 2) # Extract p-values p_dm_long_het <- fmtp(ma_mace_dm_years_l$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_dm_short_het <- fmtp(ma_mace_dm_years_s$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_dm_pooled_het <- fmtp(ma_mace_dm_years$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_dm_long_oe <- fmtp(ma_mace_dm_years_l$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_dm_short_oe <- fmtp(ma_mace_dm_years_s$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_dm_pooled_oe <- fmtp(ma_mace_dm_years$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Subgroup difference p-value p_subgroup_diff_dm <- fmtp(0.01, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Heterogeneity statistics # Long diabetes duration text(-10, 11, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_dm_years_l$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_dm_years_l$QE, digits = 2)), ", df = ", .(ma_mace_dm_years_l$k - ma_mace_dm_years_l$p), " (", italic(p), .(p_dm_long_het), "); ", I^2, " = ", .(round(ma_mace_dm_years_l$I2)), "%"))) # Short diabetes duration text(-10, 2, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_dm_years_s$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_dm_years_s$QE, digits = 2)), ", df = ", .(ma_mace_dm_years_s$k - ma_mace_dm_years_s$p), " (", italic(p), .(p_dm_short_het), "); ", I^2, " = ", .(round(ma_mace_dm_years_s$I2)), "%"))) # Pooled text(-10, -2, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_dm_years$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_dm_years$QE, digits = 2)), ", df = ", .(ma_mace_dm_years$k - ma_mace_dm_years$p), " (", italic(p), .(p_dm_pooled_het), "); ", I^2, " = ", .(round(ma_mace_dm_years$I2)), "%"))) # Overall effect # Long diabetes duration text(-10, 10, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_dm_years_l$zval, digits = 2)), " (", italic(p), .(p_dm_long_oe), ")"))) # Short diabetes duration text(-10, 1, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_dm_years_s$zval, digits = 2)), " (", italic(p), .(p_dm_short_oe), ")"))) # Pooled text(-10, -3, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_dm_years$zval, digits = 2)), " (", italic(p), .(p_dm_pooled_oe), ")"))) # Determine subgroup differences rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_dm_years, # Data frame method = "REML", # Method for random-effects model mod = dm_years_num ) # Subgroup differences text(-10, -4, pos = 4, cex = 0.80, bquote(paste("Test for subgroup differences: ", "Chi"^2, " = ", .(fmtx(7.67, digits = 2)), ", df = ", 1, " (", italic(p), .(p_subgroup_diff_dm), ")"))) # Add label for axes text(log(c(0.1, 10)), -5, c("Favours combination of SGLT-2 inhibitor and GLP-1 RA", "Favours SGLT-2 inhibitor or GLP-1 RA monotherapy"), pos = c(4,2), offset = -7, cex = 0.80) # Close PNG device dev.off() ####################### Subgroup analyses by risk of bias ###################### # Make risk of bias categorical and numeric mace_data_rob <- mace_data %>% mutate(rob_cat = ifelse(rob == "low", "Low", "Moderate/serious/critical"), rob_num = ifelse(rob_cat == "Low", 1, 0)) # Filter studies with low risk of bias mace_data_low_rob <- mace_data_rob %>% filter(rob_cat == "Low") %>% select(study,rob_cat, rob_num, mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont) # Run meta-analysis for studies with low RoB ma_mace_low_rob <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_low_rob, # Data frame method = "REML" # Method for random-effects model ) # Extract weights mace_low_rob_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_low_rob), digits = 1), "%")) %>% rename(weight = `paste0(fmtx(weights(ma_mace_low_rob), digits = 1), "%")`) %>% mutate(weight_num = as.numeric(gsub("%", "", weight)), weight_scaled = 0.75 + weight_num * 0.06) #Add numeric weights #Extract rounded estimate and CI bound pred_mace_low_rob <- predict(ma_mace_low_rob, transf = exp, digits = 2) pred_mace_low_rob <- fmtx(c(pred_mace_low_rob$pred, pred_mace_low_rob$ci.lb, pred_mace_low_rob$ci.ub), digits=2) # Add weights to dataframe mace_data_low_rob <- cbind(mace_data_low_rob, mace_low_rob_weights) # Filter studies with high risk of bias mace_data_high_rob <- mace_data_rob %>% filter(rob_cat == "Moderate/serious/critical") %>% select(study,rob_cat, rob_num, mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont) # Run meta-analysis for studies with some concerns/serious/critical RoB ma_mace_high_rob <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_high_rob, # Data frame method = "REML" # Method for random-effects model ) # Extract weights mace_high_rob_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_high_rob), digits = 1), "%")) %>% rename(weight = `paste0(fmtx(weights(ma_mace_high_rob), digits = 1), "%")`) %>% mutate(weight_num = as.numeric(gsub("%", "", weight)), weight_scaled = 0.70 + weight_num * 0.02) #Add numeric weights #Extract rounded estimate and CI bound pred_mace_high_rob <- predict(ma_mace_high_rob, transf = exp, digits = 2) pred_mace_high_rob <- fmtx(c(pred_mace_high_rob$pred, pred_mace_high_rob$ci.lb, pred_mace_high_rob$ci.ub), digits=2) # Add weights to dataframe mace_data_high_rob <- cbind(mace_data_high_rob, mace_high_rob_weights) # Bind dataframes with weights back together mace_data_rob <- rbind(mace_data_low_rob, mace_data_high_rob) # Order dataframe by subgroup and weights # Order dataframe mace_data_rob <- mace_data_rob %>% arrange(rob_num, desc(weight_num)) # Run meta-analysis for stratified for risk of bias ma_mace_rob <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_rob, # Data frame method = "REML" # Method for random-effects model ) #Extract rounded estimate and CI bound pred_mace_rob <- predict(ma_mace_rob, transf = exp, digits = 2) pred_mace_rob <- fmtx(c(pred_mace_rob$pred, pred_mace_rob$ci.lb, pred_mace_rob$ci.ub), digits=2) # Extract weights mace_rob_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_rob), digits = 1), "%")) %>% rename(weight_rob = `paste0(fmtx(weights(ma_mace_rob), digits = 1), "%")`) %>% mutate(weight_num_rob = as.numeric(gsub("%", "", weight_rob)), weight_scaled_rob = 0.70 + weight_num_rob * 0.02) #Add numeric weights # Add weights to dataframe mace_data_rob <- cbind(mace_data_rob, mace_rob_weights) # Run meta-analysis for stratified for risk of bias ma_mace_rob <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_rob, # Data frame method = "REML" # Method for random-effects model ) # Set up PNG for forest plot for MACE by risk of bias png("supp_fig_6.png", width = 16000, height = 8000, res = 1200) fp_mace_rob <- forest(ma_mace_rob, xlim=c(-10, 4), at=log(c(0.1, 0.5, 1, 2, 10)), atransf=exp, order = rob_num, xlab = "", lty = c(1, 1, 0), ilab = cbind(mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont, weight_rob), ilab.xpos=c(-7, -6, -5, -4, -3), cex=0.80, ylim=c(-1, 24), rows = c(5:13, 20), mlab = "", psize = weight_scaled_rob, header = c("Study", "Risk ratio [95% CI]"), showweights = FALSE, addfit = TRUE) # Add pooled estimates of studies with and without high baseline HbA1c addpoly(ma_mace_low_rob, row = 18, cex = 0.80) addpoly(ma_mace_high_rob, row = 3, cex = 0.80) # hide non-bold summary estimates and study labels rect(-10, c(19, 4), -7.25, c(17, 2), col = "White", border = NA) # Hide point estimates RE rect(2.5, c(19, 4), 4, c(17, -4), col = "White", border = NA) # Place summary estimates back in bold text(fp_mace_rob$textpos[2], -1, paste0(pred_mace_rob[1], " [", pred_mace_rob[2], ", ", pred_mace_rob[3], "]"), pos=2, font = 2, cex = 0.80) text(fp_mace_rob$textpos[2], 18, paste0(pred_mace_low_rob[1], " [", pred_mace_low_rob[2], ", ", pred_mace_low_rob[3], "]"), pos=2, font = 2, cex = 0.80) text(fp_mace_rob$textpos[2], 3, paste0(pred_mace_high_rob[1], " [", pred_mace_high_rob[2], ", ", pred_mace_high_rob[3], "]"), pos=2, font = 2, cex = 0.80) # Add horizontal line at the top segments(-10, 22, 4, 22, lwd = 0.8) # Expand margins par(xpd = NA) # Add labels on top text(c(-7, -6, -5, -4, -3), 23, c("Events", "Total", "Events", "Total", "Weight"), cex = 0.80, font = 2) text(c(-6.5, -4.5, 0), 24, c("Combination therapy", "Monotherapy", "Risk ratio"), cex = 0.80, font = 2) text(0, 23, c("MH Random, 95% CI"), cex = 0.80, font = 2) # Add outcome labels text(c(-10), c(21, 14), c("Low risk of bias", "Moderate/serious/critical risk of bias"), pos = 4, font = 2, cex = 0.80) # Add Total labels: text(-10, -1, pos = 4, "Total (95% CI)", font = 2, cex = 0.80) # Add subtotal labels text(-10, c(3, 18), "Subtotal (95% CI)", font = 2, cex = 0.80, pos = 4) #Add text for total events text(c(-6, -4), -1, c(sum(mace_data_rob$mace_n_exp), sum(mace_data_rob$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), -1, c(sum(mace_data_rob$mace_events_exp), sum(mace_data_rob$mace_events_cont)), cex = 0.80) # Add text for subtotal events text(c(-6, -4), 18, c(sum(mace_data_low_rob$mace_n_exp), sum(mace_data_low_rob$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), 18, c(sum(mace_data_low_rob$mace_events_exp), sum(mace_data_low_rob$mace_events_cont)), cex = 0.80) text(c(-6, -4), 3, c(sum(mace_data_high_rob$mace_n_exp), sum(mace_data_high_rob$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), 3, c(sum(mace_data_high_rob$mace_events_exp), sum(mace_data_high_rob$mace_events_cont)), cex = 0.80) # Calculate sum of weights for studies with a low and high risk of bias mace_data_rob %>% group_by(rob_cat) %>% summarise(sum_weight = sum(weight_num_rob)) # Add sum of weights text(-3, c(18, 3, -1), c("8.7%", "91.2%", "100%"), cex = 0.80, font = 2) # Extract p-values p_rob_low_het <- fmtp(ma_mace_low_rob$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_rob_high_het <- fmtp(ma_mace_high_rob$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_rob_pooled_het <- fmtp(ma_mace_rob$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_rob_low_oe <- fmtp(ma_mace_low_rob$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_rob_high_oe <- fmtp(ma_mace_high_rob$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_rob_pooled_oe <- fmtp(ma_mace_rob$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Subgroup difference p-value p_subgroup_diff_rob <- fmtp(0.14, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Heterogeneity statistics # Low risk of bias text(-10, 17, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_low_rob$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_low_rob$QE, digits = 2)), ", df = ", .(ma_mace_low_rob$k - ma_mace_low_rob$p), " (", italic(p), .(p_rob_low_het), "); ", I^2, " = ", .(round(ma_mace_low_rob$I2)), "%"))) # High risk of bias text(-10, 2, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_high_rob$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_high_rob$QE, digits = 2)), ", df = ", .(ma_mace_high_rob$k - ma_mace_high_rob$p), " (", italic(p), .(p_rob_high_het), "); ", I^2, " = ", .(round(ma_mace_high_rob$I2)), "%"))) # Pooled text(-10, -2, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_rob$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_rob$QE, digits = 2)), ", df = ", .(ma_mace_rob$k - ma_mace_rob$p), " (", italic(p), .(p_rob_pooled_het), "); ", I^2, " = ", .(round(ma_mace_rob$I2)), "%"))) # Overall effect # Low risk of bias text(-10, 16, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_low_rob$zval, digits = 2)), " (", italic(p), .(p_rob_low_oe), ")"))) # High risk of bias text(-10, 1, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_high_rob$zval, digits = 2)), " (", italic(p), .(p_rob_high_oe), ")"))) # Pooled text(-10, -3, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_rob$zval, digits = 2)), " (", italic(p), .(p_rob_pooled_oe), ")"))) # Determine subgroup differences rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_rob, # Data frame method = "REML", # Method for random-effects model mod = rob_num ) # Subgroup differences text(-10, -4, pos = 4, cex = 0.80, bquote(paste("Test for subgroup differences: ", "Chi"^2, " = ", .(fmtx(2.23, digits = 2)), ", df = ", 1, " (", italic(p), .(p_subgroup_diff_rob), ")"))) # Add label for axes text(log(c(0.1, 10)), -5, c("Favours combination of SGLT-2 inhibitor and GLP-1 RA", "Favours SGLT-2 inhibitor or GLP-1 RA monotherapy"), pos = c(4,2), offset = -7, cex = 0.80) # Close PNG device dev.off() ########################## Subgroup by MACE definition ######################### # Make risk of MACE definition categorical and numeric mace_data_mace_def <- mace_data %>% mutate(tp_mace_cat = ifelse(ascvd_mace == 1, "Three point MACE", "Expanded MACE"), tp_mace_num = ifelse(tp_mace_cat == "Three point MACE", 1, 0)) # Filter studies with three point MACE mace_data_tp_mace <- mace_data_mace_def %>% filter(tp_mace_cat == "Three point MACE") %>% select(study,tp_mace_cat, tp_mace_num, mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont) # Run meta-analysis for studies with a three point MACE endpoint ma_mace_tp_mace <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_tp_mace, # Data frame method = "REML" # Method for random-effects model ) # Extract weights mace_tp_mace_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_tp_mace), digits = 1), "%")) %>% rename(weight = `paste0(fmtx(weights(ma_mace_tp_mace), digits = 1), "%")`) %>% mutate(weight_num = as.numeric(gsub("%", "", weight)), weight_scaled = 0.80 + weight_num * 0.06) #Add numeric weights #Extract rounded estimate and CI bound pred_mace_tp_mace <- predict(ma_mace_tp_mace, transf = exp, digits = 2) pred_mace_tp_mace <- fmtx(c(pred_mace_tp_mace$pred, pred_mace_tp_mace$ci.lb, pred_mace_tp_mace$ci.ub), digits=2) # Add weights to dataframe mace_data_tp_mace <- cbind(mace_data_tp_mace, mace_tp_mace_weights) # Filter studies with expanded MACE mace_data_exp_mace <- mace_data_mace_def %>% filter(tp_mace_cat == "Expanded MACE") %>% select(study,tp_mace_cat, tp_mace_num, mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont) # Run meta-analysis for studies with some concerns/serious/critical RoB ma_mace_exp_mace <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_exp_mace, # Data frame method = "REML" # Method for random-effects model ) # Extract weights mace_exp_mace_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_exp_mace), digits = 1), "%")) %>% rename(weight = `paste0(fmtx(weights(ma_mace_exp_mace), digits = 1), "%")`) %>% mutate(weight_num = as.numeric(gsub("%", "", weight)), weight_scaled = 0.7 + weight_num * 0.02) #Add numeric weights #Extract rounded estimate and CI bound pred_mace_exp_mace <- predict(ma_mace_exp_mace, transf = exp, digits = 2) pred_mace_exp_mace <- fmtx(c(pred_mace_exp_mace$pred, pred_mace_exp_mace$ci.lb, pred_mace_exp_mace$ci.ub), digits=2) # Add weights to dataframe mace_data_exp_mace <- cbind(mace_data_exp_mace, mace_exp_mace_weights) # Bind dataframes with weights back together mace_data_mace_def <- rbind(mace_data_tp_mace, mace_data_exp_mace) # Order dataframe by subgroup and weights # Order dataframe mace_data_mace_def <- mace_data_mace_def %>% arrange(tp_mace_cat, desc(weight_num)) # Run meta-analysis for stratified by MACE definition ma_mace_mace_def <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_mace_def, # Data frame method = "REML" # Method for random-effects model ) #Extract rounded estimate and CI bound pred_mace_def <- predict(ma_mace_mace_def, transf = exp, digits = 2) pred_mace_def <- fmtx(c(pred_mace_def$pred, pred_mace_def$ci.lb, pred_mace_def$ci.ub), digits=2) # Extract weights mace_mace_def_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_mace_def), digits = 1), "%")) %>% rename(weight_tp_mace = `paste0(fmtx(weights(ma_mace_mace_def), digits = 1), "%")`) %>% mutate(weight_num_tp_mace = as.numeric(gsub("%", "", weight_tp_mace)), weight_scaled_tp_mace = 0.70 + weight_num_tp_mace * 0.02) #Add numeric weights # Add weights to dataframe mace_data_mace_def <- cbind(mace_data_mace_def, mace_mace_def_weights) # Run meta-analysis for stratified for MACE definition ma_mace_mace_def <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_mace_def, # Data frame method = "REML" # Method for random-effects model ) # Set up PNG for forest plot for MACE by MACE definition png("supp_fig_7.png", width = 16000, height = 8000, res = 1200) fp_mace_mace_def <- forest(ma_mace_mace_def, xlim=c(-10, 4), at=log(c(0.1, 0.5, 1, 2, 10)), atransf=exp, order = tp_mace_num, xlab = "", lty = c(1, 1, 0), ilab = cbind(mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont, weight_tp_mace), ilab.xpos=c(-7, -6, -5, -4, -3), cex=0.80, ylim=c(-1, 24), rows = c(5:8, 15:20), mlab = "", psize = weight_scaled_tp_mace, header = c("Study", "Risk ratio [95% CI]"), showweights = FALSE, addfit = TRUE) # Add pooled estimates of studies with three point MACE and other MACE addpoly(ma_mace_tp_mace, row = 13, cex = 0.80) addpoly(ma_mace_exp_mace, row = 3, cex = 0.80) # hide non-bold summary estimates and study labels rect(-10, c(14, 4), -7.25, c(12, 2), col = "White", border = NA) # Hide point estimates RE rect(2.5, c(14, 4), 4, c(12, -4), col = "White", border = NA) # Place summary estimates back in bold text(fp_mace_mace_def$textpos[2], -1, paste0(pred_mace_def[1], " [", pred_mace_def[2], ", ", pred_mace_def[3], "]"), pos=2, font = 2, cex = 0.80) text(fp_mace_mace_def$textpos[2], 13, paste0(pred_mace_tp_mace[1], " [", pred_mace_tp_mace[2], ", ", pred_mace_tp_mace[3], "]"), pos=2, font = 2, cex = 0.80) text(fp_mace_mace_def$textpos[2], 3, paste0(pred_mace_exp_mace[1], " [", pred_mace_exp_mace[2], ", ", pred_mace_exp_mace[3], "]"), pos=2, font = 2, cex = 0.80) # Add horizontal line at the top segments(-10, 22, 4, 22, lwd = 0.8) # Expand margins par(xpd = NA) # Add labels on top text(c(-7, -6, -5, -4, -3), 23, c("Events", "Total", "Events", "Total", "Weight"), cex = 0.80, font = 2) text(c(-6.5, -4.5, 0), 24, c("Combination therapy", "Monotherapy", "Risk ratio"), cex = 0.80, font = 2) text(0, 23, c("MH Random, 95% CI"), cex = 0.80, font = 2) # Add outcome labels text(c(-10), c(21, 9), c("Three point MACE", "Other MACE"), pos = 4, font = 2, cex = 0.80) # Add Total labels: text(-10, -1, pos = 4, "Total (95% CI)", font = 2, cex = 0.80) # Add subtotal labels text(-10, c(3, 13), "Subtotal (95% CI)", font = 2, cex = 0.80, pos = 4) #Add text for total events text(c(-6, -4), -1, c(sum(mace_data_mace_def$mace_n_exp), sum(mace_data_mace_def$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), -1, c(sum(mace_data_mace_def$mace_events_exp), sum(mace_data_mace_def$mace_events_cont)), cex = 0.80) # Add text for subtotal events text(c(-6, -4), 13, c(sum(mace_data_tp_mace$mace_n_exp), sum(mace_data_tp_mace$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), 13, c(sum(mace_data_tp_mace$mace_events_exp), sum(mace_data_tp_mace$mace_events_cont)), cex = 0.80) text(c(-6, -4), 3, c(sum(mace_data_exp_mace$mace_n_exp), sum(mace_data_exp_mace$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), 3, c(sum(mace_data_exp_mace$mace_events_exp), sum(mace_data_exp_mace$mace_events_cont)), cex = 0.80) # Calculate sum of weights for studies with a three-point and expanded MACE mace_data_mace_def %>% group_by(tp_mace_cat) %>% summarise(sum_weight = sum(weight_num_tp_mace)) # Add sum of weights text(-3, c(13, 3, -1), c("39.3%", "60.6%", "100%"), cex = 0.80, font = 2) # Extract p-values p_tp_mace_het <- fmtp(ma_mace_tp_mace$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_exp_mace_het <- fmtp(ma_mace_exp_mace$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_mace_def_het <- fmtp(ma_mace_mace_def$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_tp_mace_oe <- fmtp(ma_mace_tp_mace$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_exp_mace_oe <- fmtp(ma_mace_exp_mace$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_mace_def_oe <- fmtp(ma_mace_mace_def$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Subgroup difference p-value p_subgroup_diff_mace_def <- fmtp(0.09, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Add heterogeneity statistics # Three point MACE text(-10, 12, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_tp_mace$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_tp_mace$QE, digits = 2)), ", df = ", .(ma_mace_tp_mace$k - ma_mace_tp_mace$p), " (", italic(p), .(p_tp_mace_het), "); ", I^2, " = ", .(round(ma_mace_tp_mace$I2)), "%"))) # Other MACE text(-10, 2, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_exp_mace$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_exp_mace$QE, digits = 2)), ", df = ", .(ma_mace_exp_mace$k - ma_mace_exp_mace$p), " (", italic(p), .(p_exp_mace_het), "); ", I^2, " = ", .(round(ma_mace_exp_mace$I2)), "%"))) # Pooled text(-10, -2, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_mace_def$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_mace_def$QE, digits = 2)), ", df = ", .(ma_mace_mace_def$k - ma_mace_mace_def$p), " (", italic(p), .(p_mace_def_het), "); ", I^2, " = ", .(round(ma_mace_mace_def$I2)), "%"))) # Add overall effect # Three point MACE text(-10, 11, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_tp_mace$zval, digits = 2)), " (", italic(p), .(p_tp_mace_oe), ")"))) # Other MACE text(-10, 1, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_exp_mace$zval, digits = 2)), " (", italic(p), .(p_exp_mace_oe), ")"))) # Pooled text(-10, -3, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_mace_def$zval, digits = 2)), " (", italic(p), .(p_mace_def_oe), ")"))) # Determine subgroup differences rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_mace_def, # Data frame method = "REML", # Method for random-effects model mod = tp_mace_num ) # Add subgroup differences text(-10, -4, pos = 4, cex = 0.80, bquote(paste("Test for subgroup differences: ", "Chi"^2, " = ", .(fmtx(2.81, digits = 2)), ", df = ", 1, " (", italic(p), .(p_subgroup_diff_mace_def), ")"))) # Add label for axes text(log(c(0.1, 10)), -5, c("Favours combination of SGLT-2 inhibitor and GLP-1 RA", "Favours SGLT-2 inhibitor or GLP-1 RA monotherapy"), pos = c(4,2), offset = -7, cex = 0.80) # Close PNG device dev.off() ############################ Confounding adjustment ############################ # Filter studies for confounding analysis mace_data_confounding <- mace_data %>% filter(!is.na(confounding)) # Filter studies with propensity score/IPTW mace_data_confounding_ps <- mace_data_confounding %>% filter(confounding == 1) %>% select(study,confounding, mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont) # Run meta-analysis for studies with PS/IPTW ma_mace_confounding_ps <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_confounding_ps, # Data frame method = "REML" # Method for random-effects model ) # Extract weights mace_confounding_ps_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_confounding_ps), digits = 1), "%")) %>% rename(weight = `paste0(fmtx(weights(ma_mace_confounding_ps), digits = 1), "%")`) %>% mutate(weight_num = as.numeric(gsub("%", "", weight)), weight_scaled = 0.7 + weight_num * 0.02) #Add numeric weights #Extract rounded estimate and CI bound pred_mace_confounding_ps <- predict(ma_mace_confounding_ps, transf = exp, digits = 2) pred_mace_confounding_ps <- fmtx(c(pred_mace_confounding_ps$pred, pred_mace_confounding_ps$ci.lb, pred_mace_confounding_ps$ci.ub), digits=2) # Add weights to dataframe mace_data_confounding_ps <- cbind(mace_data_confounding_ps, mace_confounding_ps_weights) # Filter studies with other methods for adjustment mace_data_confounding_mv <- mace_data_confounding %>% filter(confounding == 0) %>% select(study,confounding, mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont) # Run meta-analysis for studies with multivariable adjustment ma_mace_confounding_mv <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_confounding_mv, # Data frame method = "REML" # Method for random-effects model ) # Extract weights mace_confounding_mv_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_confounding_mv), digits = 1), "%")) %>% rename(weight = `paste0(fmtx(weights(ma_mace_confounding_mv), digits = 1), "%")`) %>% mutate(weight_num = as.numeric(gsub("%", "", weight)), weight_scaled = 0.7 + weight_num * 0.02) #Add numeric weights #Extract rounded estimate and CI bound pred_mace_confounding_mv <- predict(ma_mace_confounding_mv, transf = exp, digits = 2) pred_mace_confounding_mv <- fmtx(c(pred_mace_confounding_mv$pred, pred_mace_confounding_mv$ci.lb, pred_mace_confounding_mv$ci.ub), digits=2) # Add weights to dataframe mace_data_confounding_mv <- cbind(mace_data_confounding_mv, mace_confounding_mv_weights) # Bind dataframes with weights back together mace_data_confounding <- rbind(mace_data_confounding_ps, mace_data_confounding_mv) # Order dataframe by subgroup and weights # Order dataframe mace_data_confounding <- mace_data_confounding %>% arrange(confounding, desc(weight_num)) # Run meta-analysis for stratified by method to adjust for confounding ma_mace_confounding <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_confounding, # Data frame method = "REML" # Method for random-effects model ) #Extract rounded estimate and CI bound pred_mace_confounding <- predict(ma_mace_confounding, transf = exp, digits = 2) pred_mace_confounding <- fmtx(c(pred_mace_confounding$pred, pred_mace_confounding$ci.lb, pred_mace_confounding$ci.ub), digits=2) # Extract weights mace_confounding_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_confounding), digits = 1), "%")) %>% rename(weight_confounding = `paste0(fmtx(weights(ma_mace_confounding), digits = 1), "%")`) %>% mutate(weight_num_confounding = as.numeric(gsub("%", "", weight_confounding)), weight_scaled_confounding = 0.70 + weight_num_confounding * 0.02) #Add numeric weights # Add weights to dataframe mace_data_confounding <- cbind(mace_data_confounding, mace_confounding_weights) # Run meta-analysis for stratified for method to adjust for confounding ma_mace_confounding <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_confounding, # Data frame method = "REML" # Method for random-effects model ) # Set up PNG for forest plot for MACE by method to adjust for confounding png("supp_fig_8.png", width = 16000, height = 8000, res = 1200) fp_mace_confounding <- forest(ma_mace_confounding, xlim=c(-10, 4), at=log(c(0.1, 0.5, 1, 2, 10)), atransf=exp, order = confounding, xlab = "", lty = c(1, 1, 0), ilab = cbind(mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont, weight_confounding), ilab.xpos=c(-7, -6, -5, -4, -3), cex=0.80, ylim=c(-1, 24), rows = c(5:6, 13:20), mlab = "", psize = weight_scaled_confounding, header = c("Study", "Risk ratio [95% CI]"), showweights = FALSE, addfit = TRUE) # Add pooled estimates of studies PS/IPTW adjustment and other adjustment addpoly(ma_mace_confounding_ps, row = 11, cex = 0.80) addpoly(ma_mace_confounding_mv, row = 3, cex = 0.80) # hide non-bold summary estimates and study labels rect(-10, c(12, 4), -7.25, c(10, 2), col = "White", border = NA) # Hide point estimates RE rect(2.5, c(12, 4), 4, c(10, -4), col = "White", border = NA) # Place summary estimates back in bold text(fp_mace_confounding$textpos[2], -1, paste0(pred_mace_confounding[1], " [", pred_mace_confounding[2], ", ", pred_mace_confounding[3], "]"), pos=2, font = 2, cex = 0.80) text(fp_mace_confounding$textpos[2], 11, paste0(pred_mace_confounding_ps[1], " [", pred_mace_confounding_ps[2], ", ", pred_mace_confounding_ps[3], "]"), pos=2, font = 2, cex = 0.80) text(fp_mace_confounding$textpos[2], 3, paste0(pred_mace_confounding_mv[1], " [", pred_mace_confounding_mv[2], ", ", pred_mace_confounding_mv[3], "]"), pos=2, font = 2, cex = 0.80) # Add horizontal line at the top segments(-10, 22, 4, 22, lwd = 0.8) # Expand margins par(xpd = NA) # Add labels on top text(c(-7, -6, -5, -4, -3), 23, c("Events", "Total", "Events", "Total", "Weight"), cex = 0.80, font = 2) text(c(-6.5, -4.5, 0), 24, c("Combination therapy", "Monotherapy", "Risk ratio"), cex = 0.80, font = 2) text(0, 23, c("MH Random, 95% CI"), cex = 0.80, font = 2) # Add outcome labels text(c(-10), c(21, 7), c("Propensity scores/IPTW", "Other method to adjust for confounding"), pos = 4, font = 2, cex = 0.80) # Add Total labels: text(-10, -1, pos = 4, "Total (95% CI)", font = 2, cex = 0.80) # Add subtotal labels text(-10, c(3, 11), "Subtotal (95% CI)", font = 2, cex = 0.80, pos = 4) #Add text for total events text(c(-6, -4), -1, c(sum(mace_data_confounding$mace_n_exp), sum(mace_data_confounding$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), -1, c(sum(mace_data_confounding$mace_events_exp), sum(mace_data_confounding$mace_events_cont)), cex = 0.80) # Add text for subtotal events text(c(-6, -4), 11, c(sum(mace_data_confounding_ps$mace_n_exp), sum(mace_data_confounding_ps$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), 11, c(sum(mace_data_confounding_ps$mace_events_exp), sum(mace_data_confounding_ps$mace_events_cont)), cex = 0.80) text(c(-6, -4), 3, c(sum(mace_data_confounding_mv$mace_n_exp), sum(mace_data_confounding_mv$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), 3, c(sum(mace_data_confounding_mv$mace_events_exp), sum(mace_data_confounding_mv$mace_events_cont)), cex = 0.80) # Calculate sum of weights for studies with a low and high risk of bias mace_data_confounding %>% group_by(confounding) %>% summarise(sum_weight = sum(weight_num_confounding)) # Add sum of weights text(-3, c(11, 3, -1), c("83.5%", "16.4%", "100%"), cex = 0.80, font = 2) # Extract p-values p_confounding_ps_het <- fmtp(ma_mace_confounding_ps$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_confounding_mv_het <- fmtp(ma_mace_confounding_mv$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_confounding_pooled_het <- fmtp(ma_mace_confounding$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_confounding_ps_oe <- fmtp(ma_mace_confounding_ps$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_confounding_mv_oe <- fmtp(ma_mace_confounding_mv$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_confounding_pooled_oe <- fmtp(ma_mace_confounding$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Subgroup difference p-value p_subgroup_diff_confounding <- fmtp(0.07, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) #Add heterogeneity statistics # Propensity scores/IPTW text(-10, 10, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_confounding_ps$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_confounding_ps$QE, digits = 2)), ", df = ", .(ma_mace_confounding_ps$k - ma_mace_confounding_ps$p), " (", italic(p), .(p_confounding_ps_het), "); ", I^2, " = ", .(round(ma_mace_confounding_ps$I2)), "%"))) # Other method to adjust for confounding text(-10, 2, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_confounding_mv$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_confounding_mv$QE, digits = 2)), ", df = ", .(ma_mace_confounding_mv$k - ma_mace_confounding_mv$p), " (", italic(p), .(p_confounding_mv_het), "); ", I^2, " = ", .(round(ma_mace_confounding_mv$I2)), "%"))) # Pooled text(-10, -2, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_confounding$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_confounding$QE, digits = 2)), ", df = ", .(ma_mace_confounding$k - ma_mace_confounding$p), " (", italic(p), .(p_confounding_pooled_het), "); ", I^2, " = ", .(round(ma_mace_confounding$I2)), "%"))) # Add overall effect # Propensity scores/IPTW text(-10, 9, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_confounding_ps$zval, digits = 2)), " (", italic(p), .(p_confounding_ps_oe), ")"))) # Other method to adjust for confounding text(-10, 1, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_confounding_mv$zval, digits = 2)), " (", italic(p), .(p_confounding_mv_oe), ")"))) # Pooled text(-10, -3, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_confounding$zval, digits = 2)), " (", italic(p), .(p_confounding_pooled_oe), ")"))) # Determine subgroup differences rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_confounding, # Data frame method = "REML", # Method for random-effects model mod = confounding ) # Add subgroup differences text(-10, -4, pos = 4, cex = 0.80, bquote(paste("Test for subgroup differences: ", "Chi"^2, " = ", .(fmtx(3.28, digits = 2)), ", df = ", 1, " (", italic(p), .(p_subgroup_diff_confounding), ")"))) # Add label for axes text(log(c(0.1, 10)), -5, c("Favours combination of SGLT-2 inhibitor and GLP-1 RA", "Favours SGLT-2 inhibitor or GLP-1 RA monotherapy"), pos = c(4,2), offset = -7, cex = 0.80) # Close PNG device dev.off() ################################## Data source ################################# mace_data_source <- mace_data %>% filter(!is.na(data_source)) mace_data_ehr <- mace_data_source %>% filter(data_source == 1) %>% select(study,data_source, mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont) # Run meta-analysis for studies with EHR data ma_mace_ehr <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_ehr, # Data frame method = "REML" # Method for random-effects model ) # Extract weights mace_ehr_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_ehr), digits = 1), "%")) %>% rename(weight = `paste0(fmtx(weights(ma_mace_ehr), digits = 1), "%")`) %>% mutate(weight_num = as.numeric(gsub("%", "", weight)), weight_scaled = 0.7 + weight_num * 0.02) #Add numeric weights #Extract rounded estimate and CI bound pred_mace_ehr <- predict(ma_mace_ehr, transf = exp, digits = 2) pred_mace_ehr <- fmtx(c(pred_mace_ehr$pred, pred_mace_ehr$ci.lb, pred_mace_ehr$ci.ub), digits=2) # Add weights to dataframe mace_data_ehr <- cbind(mace_data_ehr, mace_ehr_weights) # Filter studies with other methods for adjustment mace_data_other <- mace_data_source %>% filter(data_source == 0) %>% select(study,data_source, mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont) # Run meta-analysis for studies with other data sources ma_mace_other <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_other, # Data frame method = "REML" # Method for random-effects model ) # Extract weights mace_other_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_other), digits = 1), "%")) %>% rename(weight = `paste0(fmtx(weights(ma_mace_other), digits = 1), "%")`) %>% mutate(weight_num = as.numeric(gsub("%", "", weight)), weight_scaled = 0.7 + weight_num * 0.02) #Add numeric weights #Extract rounded estimate and CI bound pred_mace_other <- predict(ma_mace_other, transf = exp, digits = 2) pred_mace_other <- fmtx(c(pred_mace_other$pred, pred_mace_other$ci.lb, pred_mace_other$ci.ub), digits=2) # Add weights to dataframe mace_data_other <- cbind(mace_data_other, mace_other_weights) # Bind dataframes with weights back together mace_data_source <- rbind(mace_data_ehr, mace_data_other) # Order dataframe by subgroup and weights # Order dataframe mace_data_source <- mace_data_source %>% arrange(data_source, desc(weight_num)) # Run meta-analysis for stratified by data source ma_mace_data_source <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_source, # Data frame method = "REML" # Method for random-effects model ) #Extract rounded estimate and CI bound pred_mace_data_source <- predict(ma_mace_data_source, transf = exp, digits = 2) pred_mace_data_source <- fmtx(c(pred_mace_data_source$pred, pred_mace_data_source$ci.lb, pred_mace_data_source$ci.ub), digits=2) # Extract weights mace_source_weights <- as.data.frame(paste0(fmtx(weights(ma_mace_data_source), digits = 1), "%")) %>% rename(weight_source = `paste0(fmtx(weights(ma_mace_data_source), digits = 1), "%")`) %>% mutate(weight_num_source = as.numeric(gsub("%", "", weight_source)), weight_scaled_source = 0.70 + weight_num_source * 0.02) #Add numeric weights # Add weights to dataframe mace_data_source <- cbind(mace_data_source, mace_source_weights) # Run meta-analysis for stratified data source ma_mace_data_source <- rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_source, # Data frame method = "REML" # Method for random-effects model ) # Set up PNG for forest plot for MACE by data source png("supp_fig_9.png", width = 16000, height = 8000, res = 1200) fp_mace_data_source <- forest(ma_mace_data_source, xlim=c(-10, 4), at=log(c(0.1, 0.5, 1, 2, 10)), atransf=exp, order = data_source, xlab = "", lty = c(1, 1, 0), ilab = cbind(mace_events_exp, mace_n_exp, mace_events_cont, mace_n_cont, weight_source), ilab.xpos=c(-7, -6, -5, -4, -3), cex=0.80, ylim=c(-1, 24), rows = c(5:8, 15:20), mlab = "", psize = weight_scaled_source, header = c("Study", "Risk ratio [95% CI]"), showweights = FALSE, addfit = TRUE) # Add pooled estimates of studies EHR data and outher sources addpoly(ma_mace_ehr, row = 13, cex = 0.80) addpoly(ma_mace_other, row = 3, cex = 0.80) # hide non-bold summary estimates and study labels rect(-10, c(14, 4), -7.25, c(12, 2), col = "White", border = NA) # Hide point estimates RE rect(2.5, c(14, 4), 4, c(12, -4), col = "White", border = NA) # Place summary estimates back in bold text(fp_mace_data_source$textpos[2], -1, paste0(pred_mace_data_source[1], " [", pred_mace_data_source[2], ", ", pred_mace_data_source[3], "]"), pos=2, font = 2, cex = 0.80) text(fp_mace_data_source$textpos[2], 13, paste0(pred_mace_ehr[1], " [", pred_mace_ehr[2], ", ", pred_mace_ehr[3], "]"), pos=2, font = 2, cex = 0.80) text(fp_mace_data_source$textpos[2], 3, paste0(pred_mace_other[1], " [", pred_mace_other[2], ", ", pred_mace_other[3], "]"), pos=2, font = 2, cex = 0.80) # Add horizontal line at the top segments(-10, 22, 4, 22, lwd = 0.8) # Expand margins par(xpd = NA) # Add labels on top text(c(-7, -6, -5, -4, -3), 23, c("Events", "Total", "Events", "Total", "Weight"), cex = 0.80, font = 2) text(c(-6.5, -4.5, 0), 24, c("Combination therapy", "Monotherapy", "Risk ratio"), cex = 0.80, font = 2) text(0, 23, c("MH Random, 95% CI"), cex = 0.80, font = 2) # Add outcome labels text(c(-10), c(21, 9), c("Electronic healthcare records", "Other data sources"), pos = 4, font = 2, cex = 0.80) # Add Total labels: text(-10, -1, pos = 4, "Total (95% CI)", font = 2, cex = 0.80) # Add subtotal labels text(-10, c(3, 13), "Subtotal (95% CI)", font = 2, cex = 0.80, pos = 4) #Add text for total events text(c(-6, -4), -1, c(sum(mace_data_source$mace_n_exp), sum(mace_data_source$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), -1, c(sum(mace_data_source$mace_events_exp), sum(mace_data_source$mace_events_cont)), cex = 0.80) # Add text for subtotal events text(c(-6, -4), 13, c(sum(mace_data_ehr$mace_n_exp), sum(mace_data_ehr$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), 13, c(sum(mace_data_ehr$mace_events_exp), sum(mace_data_ehr$mace_events_cont)), cex = 0.80) text(c(-6, -4), 3, c(sum(mace_data_other$mace_n_exp), sum(mace_data_other$mace_n_cont)), font = 2, cex = 0.80) text(c(-7, -5), 3, c(sum(mace_data_other$mace_events_exp), sum(mace_data_other$mace_events_cont)), cex = 0.80) # Calculate sum of weights for studies with a low and high risk of bias mace_data_source %>% group_by(data_source) %>% summarise(sum_weight = sum(weight_num_source)) # Add sum of weights text(-3, c(13, 3, -1), c("59.6%", "40.3%", "100%"), cex = 0.80, font = 2) # Extract p-values p_ehr_het <- fmtp(ma_mace_ehr$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_other_het <- fmtp(ma_mace_other$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_data_het <- fmtp(ma_mace_data_source$QEp, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_ehr_oe <- fmtp(ma_mace_ehr$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_other_oe <- fmtp(ma_mace_other$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) p_data_oe <- fmtp(ma_mace_data_source$pval, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) # Subgroup difference p-value p_subgroup_diff_data <- fmtp(0.61, digits = 2, pname = "", add0 = TRUE, sep = TRUE, equal = TRUE) #Add heterogeneity statistics # EHR data text(-10, 12, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_ehr$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_ehr$QE, digits = 2)), ", df = ", .(ma_mace_ehr$k - ma_mace_ehr$p), " (", italic(p), .(p_ehr_het), "); ", I^2, " = ", .(round(ma_mace_ehr$I2)), "%"))) # Other data sources text(-10, 2, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_other$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_other$QE, digits = 2)), ", df = ", .(ma_mace_other$k - ma_mace_other$p), " (", italic(p), .(p_other_het), "); ", I^2, " = ", .(round(ma_mace_other$I2)), "%"))) # Pooled text(-10, -2, pos = 4, cex = 0.80, bquote(paste("Heterogeneity: ", "Tau"^2, " = ", .(fmtx(ma_mace_data_source$tau2, digits = 2)), "; ", "Chi"^2, " = ", .(fmtx(ma_mace_data_source$QE, digits = 2)), ", df = ", .(ma_mace_data_source$k - ma_mace_data_source$p), " (", italic(p), .(p_data_het), "); ", I^2, " = ", .(round(ma_mace_data_source$I2)), "%"))) # Add overall effect # EHR data text(-10, 11, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_ehr$zval, digits = 2)), " (", italic(p), .(p_ehr_oe), ")"))) # Other data sources text(-10, 1, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_other$zval, digits = 2)), " (", italic(p), .(p_other_oe), ")"))) # Pooled text(-10, -3, pos = 4, cex = 0.80, bquote(paste("Test for overall effect: ", "Z = ", .(fmtx(ma_mace_data_source$zval, digits = 2)), " (", italic(p), .(p_data_oe), ")"))) # Determine subgroup differences rma( measure = "RR", # Risk Ratio ai = mace_events_exp, # Events in experimental group n1i = mace_n_exp, # Total in experimental group ci = mace_events_cont, # Events in control group n2i = mace_n_cont, # Total in control group slab = study, # Study labels data = mace_data_source, # Data frame method = "REML", # Method for random-effects model mod = data_source ) # Add subgroup differences text(-10, -4, pos = 4, cex = 0.80, bquote(paste("Test for subgroup differences: ", "Chi"^2, " = ", .(fmtx(0.27, digits = 2)), ", df = ", 1, " (", italic(p), .(p_subgroup_diff_data), ")"))) # Add label for axes text(log(c(0.1, 10)), -5, c("Favours combination of SGLT-2 inhibitor and GLP-1 RA", "Favours SGLT-2 inhibitor or GLP-1 RA monotherapy"), pos = c(4,2), offset = -7, cex = 0.80) # Close PNG device dev.off()