#Required packages: install.packages("statmod") install.packages("metafor") install.packages("meta") install.packages("lifecycle") install.packages("readr") install.packages("tidyverse") install.packages('questionr') install.packages('dplyr') install.packages("pacman") # Cite when using orchard plots. Full reference here: https://doi.org/10.1111/2041-210X.14152 install.packages("devtools") install.packages("orchaRd") # for overall pooled effect and meta-regression install.packages("ggplot2") install.packages("grid") # for forest plot titles/subtitles install.packages("broom") # orchaRd installation rm(list = ls()) devtools::install_github("daniel1noble/orchaRd", ref = "main", force = TRUE) pacman::p_load(devtools, tidyverse, metafor, patchwork, R.rsp, orchaRd, emmeans, ape, phytools, flextable) #Loading packages library(statmod) library(meta) library(metafor) library(lifecycle) library(readr) library(tidyverse) library(questionr) library(readxl) library(dplyr) library(pacman) library(orchaRd) library(ggplot2) library(grid) library(broom) ######################### OUTCOME: OPIOID CONSUMPTION ################################ # Read and view data dataopicons <- read.csv("dataopioidcons.csv") # Overall analysis # Calculate Mean Difference (MD = yi) as effect size and its Variance (MDV = vi) opiconsMD <- escalc(measure = "MD", #Specify Effect Size Measure m1i = meanexp, m2i = meanctrl, sd1i = sdexp, sd2i = sdctrl, n1i = nexp, n2i = nctrl, data = dataopicons, append = TRUE) dataopicons <- opiconsMD #add MD (yi) and MDV (vi) to the original data set head(dataopicons) #check if yi and vi are now included # Perform multilevel (ML) MA of Mean Difference (MD) in opioid consumption # Calculate variance-covariance (VCV) matrix to account for correlated/non-independent errors opiconsMDVCV <- vcalc(vi = vi, cluster = studyID, obs = compID, data = dataopicons, rho = 0.5) # Perform multilevel meta-analytic model using the constructed VCV matrix (VCV) opiconsMD_ML <- rma.mv(yi = yi, V = opiconsMDVCV, #VCV matrix random = list(~1 | studyID/compID), #compID is nested within studyID, not a separate source of random variation method = "REML", data = dataopicons) summary(opiconsMD_ML) #show MA results # Assess I2 levels and extract between-study I2 opiconsi2 <- i2_ml(opiconsMD_ML) #breaks down the I2 into the ML levels opiconsi2 #show the I2 per level # Extract Q-statistic and p-value opicons_overallp <- opiconsMD_ML$pval opiconsQ_value <- opiconsMD_ML$QE opiconsQ_pvalue <- opiconsMD_ML$QEp # Display the heterogeneity statistics print(paste("I² = ", round(opiconsi2[1],0), "%, Q = ", round(opiconsQ_value, 1), ", p = ", round(opiconsQ_pvalue, 7))) # Forest plot of multilevel meta-analysis model with VCV png(filename="FPopiconsMD_ML.png", res=200, width=1600, height=1800) #set up a png file FPopiconsMD_ML<- forest(opiconsMD_ML, annotate=TRUE, showweights=TRUE, slab = dataopicons$compID, predstyle = "bar", #adds prediction interval as bar order=dataopicons$compID, #effect sizes sorted by compid main = "MLMA Opioid (equivalent) consumption 24h post-op", header=c("Comparison", " Weights MD 95% CI "), digits = c(1,2,1), #number of decimals for ESCI, SE and weights xlim=c(-80,65), #horizontal limits of the plot region at=seq(-50,25, by=5), #position of the x-axis tick marks and corresponding labels efac=c(0.1,1), #length of ticks on error bars xlab = " Favours experimental <-- 0 --> Favours control", #x-axis label ) # Add heterogeneity statistics to the plot mtext(side = 1, line = 4, text = paste("I² = ", round(opiconsi2[1],0), "%, Q = ", round(opiconsQ_value, 2), ", p = ", round(opiconsQ_pvalue, 3)), cex = 0.8) dev.off() #------- Opioid consumption: exploring heterogeneity (post-hoc)-----------------------# #-------- Moderator: added multimodal component----------------------------# freq(dataopicons$mmc_added) # n % val% #clonidine 1 4.2 4.2 #dexa 3 12.5 12.5 #dexmed 3 12.5 12.5 #ketamine 4 16.7 16.7 #lidocaine 3 12.5 12.5 #magnesium 7 29.2 29.2 #pregabalin 3 12.5 12.5 # N.B. limited power due to many small strata # Include moderator into the model, regardless of categorical or continuous opicons_mmcomp <- rma.mv(yi = yi, V = opiconsMDVCV, random = list(~1 | studyID/compID), #compID is nested within studyID, not a separate source of random variation mods = ~ mmc_added, #column of the tested moderator test = "t", method = "REML", data = dataopicons) summary(opicons_mmcomp) opiconsi2_mmcomp <- i2_ml(opicons_mmcomp) #breaks down the I2 into the ML levels opiconsi2_mmcomp #show the I2 per level # Orchard plot visualizing the regression analysis OPopicons_mmcomp<- orchard_plot(opicons_mmcomp, mod = "mmc_added", group = "studyID", angle = 0, xlab = "Mean Difference in opioid (equivalent) consumption 24h post-op", # X-axis label alpha = 0.5, # CI fill transparency col = "darkblue" # point and CI color ) + annotate(geom="text", x= 1, y= -40, label= paste0("italic(I)^{2} == ", round(opiconsi2_mmcomp[1],0), "*\"%\""), color="black", parse = TRUE, size = 3.5) ggsave("OPopicons_mmcomp.png", width = 7, height = 5, dpi = 300) #######Sensitivity analysis: remove non-parametric data (means+SDs recalculated from medians+IQR)###### sens_dataopicons <- dataopicons %>% filter(orgdataformat != "medianiqr_recalc") sens_opiconsMDVCV <- vcalc(vi = vi, cluster = studyID, obs = compID, data = sens_dataopicons, rho = 0.5) sens_opiconsMD_ML <- rma.mv(yi = yi, V = sens_opiconsMDVCV, #VCV matrix random = list(~1 | studyID/compID), #compID is nested within studyID, not a separate source of random variation method = "REML", data = sens_dataopicons) summary(sens_opiconsMD_ML) #show MA results sens_opiconsi2 <- i2_ml(sens_opiconsMD_ML) #breaks down the I2 into the ML levels sens_opiconsi2 #show the I2 per level ######################## sensitivity analysis - applied cluster-robust variance estimation at the study level############# opiconsMD_ML_robust <- robust(opiconsMD_ML, cluster = studyID) summary(opiconsMD_ML_robust) ############################## OUTCOME: PAIN SCALES ############################## #read and view data datapain <- read.csv("datapain.csv") #----------------------------------------------------------------------------- # Overall analysis # Calculate Mean Difference (MD = yi) as effect size and its Variance (MDV = vi) painMD <- escalc(measure = "MD", m1i = meanexp, m2i = meanctrl, sd1i = sdexp, sd2i = sdctrl, n1i = nexp, n2i = nctrl, data = datapain, append = TRUE) # add MD and MDV to the original data set datapain <- painMD head(datapain) #check if yi and vi are now included # Subset for 4h timepoint datapain4h <- subset(datapain, timepoint == 4) # Perform multilevel (ML) MA of Mean Difference (MD) in pain # Calculate variance-covariance (VCV) matrix to account for correlated/non-independent errors pain4hMDVCV <- vcalc(vi = vi, cluster = studyID, obs = compID, data = datapain4h, rho = 0.5) # multilevel meta-analytic model using the constructed VCV matrix (VCV) pain4hMD_ML <- rma.mv(yi = yi, V = pain4hMDVCV, # VCV matrix random = list(~1 | studyID/compID), #compID is nested within studyID, not a separate source of random variation method = "REML", data = datapain4h) summary(pain4hMD_ML) # Investigating I2 and extract between-study I2 pain4hi2 <- i2_ml(pain4hMD_ML) #breaks down the I2 into the ML levels pain4hi2 #show the I2 per level # Extract Q-statistic and p-value pain4h_overallp <- pain4hMD_ML$pval pain4hQ_value <- pain4hMD_ML$QE pain4hQ_pvalue <- pain4hMD_ML$QEp # Display the heterogeneity statistics print(paste("I² = ",round(pain4hi2[1],0), "%, Q = ", round(pain4hQ_value, 2), ", p = ", round(pain4hQ_pvalue, 3))) # Forest plot of multilevel meta-analysis model with VCV png(filename="FPpain4hMD_ML.png", res=200, width=1600, height=1800) #set up a png file FPpain4hMD_ML<- forest(pain4hMD_ML, annotate=TRUE, showweights=TRUE, slab = datapain4h$compID, predstyle = "bar", #adds prediction interval as bar #order=datapain$compID, #effect sizes sorted by compid main = "MLMA pain scale score 4h post-op", header=c("Comparison", " Weights MD 95% CI "), digits = c(1,2,1), #number of decimals for ESCI, SE and weights xlim=c(-7,7), #horizontal limits of the plot region at=seq(-3,3, by=0.5), #position of the x-axis tick marks and corresponding labels efac=c(0.1,1), #length of ticks on error bars xlab = " Favours experimental <-- 0 --> Favours control", #x-axis label ) # Add heterogeneity statistics to the plot mtext(side = 1, line = 4, text = paste("I² = ", round(pain4hi2[1],0), "%, Q = ", round(pain4hQ_value, 2), ", p = ", round(pain4hQ_pvalue, 3)), cex = 0.8) dev.off() #------- Pain at 4 hours post-op: exploring heterogeneity (post-hoc)-----------------------# #-------- Moderator: added multimodal component----------------------------# freq(datapain4h$mmc_added) # n % val% #dexa 7 24.1 24.1 #dexmed 4 13.8 13.8 #ketamine 6 20.7 20.7 #lidocaine 4 13.8 13.8 #magnesium 3 10.3 10.3 #pregabalin 5 17.2 17.2 # N.B. limited power due to many small strata # Code to include moderator into the model, regardless of categorical or continuous pain4h_mmcomp <- rma.mv(yi = yi, V = pain4hMDVCV, random = list(~1 | studyID/compID), #compID is nested within studyID, not a separate source of random variation mods = ~ mmc_added, #column of the tested moderator test = "t", method = "REML", data = datapain4h) summary(pain4h_mmcomp) # Investigating I2 and extract between-study I2 pain4hi2_mmcomp <- i2_ml(pain4h_mmcomp) #breaks down the I2 into the ML levels pain4hi2_mmcomp #show the I2 per level ####### create orchardplot of added multimodal components ############# OPpain4h_mmcomp<- orchard_plot( mod = "mmc_added", # match moderator name exactly object = pain4h_mmcomp, xlab = "Mean Difference in pain scale score", transfm = "none", # no transformation needed for MD angle = 0, # rotate x-axis labels if needed alpha = 0.5, # transparency for raw data dots col = "darkblue", # point and CI color group = "studyID" ) + annotate(geom="text", x= 1, y= -2, label= paste0("italic(I)^{2} == ", round(pain4hi2_mmcomp[1],0), "*\"%\""), color="black", parse = TRUE, size = 3.5) ggsave("OPpain4h_mmcomp.png", width = 7, height = 5, dpi = 300) #---------Sensitivity analysis: remove non-parametric data (means+SDs recalculated from medians+IQR)---------# ########### sens_datapain4h <- datapain4h %>% filter(orgdataformat != "medianiqr_recalc") sens_pain4hMDVCV <- vcalc(vi = vi, cluster = studyID, obs = compID, data = sens_datapain4h, rho = 0.5) sens_pain4hMD_ML <- rma.mv(yi = yi, V = sens_pain4hMDVCV, #VCV matrix random = list(~1 | studyID/compID), method = "REML", data = sens_datapain4h) summary(sens_pain4hMD_ML) #show MA results sens_pain4hi2 <- i2_ml(sens_pain4hMD_ML) #breaks down the I2 into the ML levels sens_pain4hi2 #show the I2 per level ######################## sensitivity analysis - applied cluster-robust variance estimation at the study level############# pain4hMD_ML_robust <- robust(pain4hMD_ML, cluster = studyID) summary(pain4hMD_ML_robust) #---------Sensitivity analysis: change MD to SMD ---------# ########### #read and view data datapainSMD <- read.csv("datapain.csv") painSMD <- escalc(measure = "SMD", m1i = meanexp, m2i = meanctrl, sd1i = sdexp, sd2i = sdctrl, n1i = nexp, n2i = nctrl, data = datapain, append = TRUE) # add SMD and SMDV to the original data set datapainSMD <- painSMD head(datapainSMD) #check if yi and vi are now included # Subset for 4h timepoint datapain4hSMD <- subset(datapainSMD, timepoint == 4) # Perform multilevel (ML) MA of Standardized Mean Difference (SMD) pain # Calculate variance-covariance (VCV) matrix to account for correlated/non-independent errors pain4hSMDVCV <- vcalc(vi = vi, cluster = studyID, obs = compID, data = datapain4hSMD, rho = 0.5) # multilevel meta-analytic model using the constructed VCV matrix (VCV) pain4hSMD_ML <- rma.mv(yi = yi, V = pain4hSMDVCV, # VCV matrix random = list(~1 | studyID/compID), #compID is nested within studyID, not a separate source of random variation method = "REML", data = datapain4hSMD) summary(pain4hSMD_ML) # Investigating I2 and extract between-study I2 pain4hSMDi2 <- i2_ml(pain4hSMD_ML) #breaks down the I2 into the ML levels pain4hSMDi2 #show the I2 per level # Extract Q-statistic and p-value pain4hSMD_overallp <- pain4hSMD_ML$pval pain4hSMDQ_value <- pain4hSMD_ML$QE pain4hSMDQ_pvalue <- pain4hSMD_ML$QEp # Display the heterogeneity statistics print(paste("I² = ", round(pain4hSMDi2[1],0) , "%, Q = ", round(pain4hSMDQ_value, 2), ", p = ", round(pain4hSMDQ_pvalue, 3))) # Forest plot of multilevel meta-analysis model with VCV png(filename="FPpain4hSMD_ML.png", res=200, width=1600, height=1800) #set up a png file FPpain4hSMD_ML<- forest(pain4hSMD_ML, annotate=TRUE, showweights=TRUE, slab = datapain4hSMD$compID, predstyle = "bar", #adds prediction interval as bar #order=datapainSMD$compID, #effect sizes sorted by compid main = "MLMA pain scale score 4h post-op", header=c("Comparison", " Weights SMD 95% CI "), digits = c(1,2,1), #number of decimals for ESCI, SE and weights xlim=c(-7,7), #horizontal limits of the plot region at=seq(-3,3, by=0.5), #position of the x-axis tick marks and corresponding labels efac=c(0.1,1), #length of ticks on error bars xlab = " Favours experimental <-- 0 --> Favours control", #x-axis label ) # Add heterogeneity statistics to the plot mtext(side = 1, line = 4, text = paste("I² = ", round(pain4hSMDi2[1],0), "%, Q = ", round(pain4hSMDQ_value, 2), ", p = ", round(pain4hSMDQ_pvalue, 3)), cex = 0.8) dev.off() ######################### PAIN over timepoints ################################ # the escalc aleady existis in painMD for all timepoints painallMDVCV <- vcalc(vi = vi, cluster = studyID, obs = compID, data = datapain, subgroup = timepoint, #only comparisons within the same study and same timepoint are assumed correlated rho = 0.8) #assumed strong correlation between timepoints of the same study (standard is 0.5) # multilevel meta-analytic model using the constructed VCV matrix (VCV) painallMD_ML <- rma.mv(yi = yi, V = painallMDVCV, # VCV matrix random = list(~1 | studyID/compID), #compID is nested within studyID, not a separate source of random variation method = "REML", data = datapain) summary(painallMD_ML) # test timepoint as moderator datapain$timepoint <- factor(datapain$timepoint) # do this before model fitting painallMD_ML_timepoint <- rma.mv(yi = yi, V = painallMDVCV, # VCV matrix random = list(~1 | studyID/compID), #compID is nested within studyID, not a separate source of random variation method = "REML", mods = ~ factor(timepoint), #enable to test timepoint as modifier data = datapain) summary(painallMD_ML_timepoint) ####### create orchardplot of timepoints ############# orchard_plot( object = painallMD_ML_timepoint, mod = "timepoint", # match moderator name exactly xlab = "Mean Difference (MD)", transfm = "none", # no transformation needed for MD angle = 45, # rotate x-axis labels if needed alpha = 0.5, # transparency for raw data dots group = TRUE, # show study-level points g = "studyID" # grouping variable from your random effects ) ggsave("OPpainall_timepoints.png", width = 7, height = 5, dpi = 400) #----------------------------------------------------------------------------- #Publication bias - funnel plots # Funnel plot for opioid consumption png(filename="FUNopiconsMD_ML.png", res=200, width=1600, height=1600) #set up a png file funnel(opiconsMD_ML, main = "Funnel plot - Opioid Consumption") dev.off() # Funnel plot for pain at 4h png(filename="FUNpain4hMD_ML.png", res=200, width=1600, height=1600) #set up a png file funnel(pain4hMD_ML, main = "Funnel plot - Pain Scores at 4h") dev.off() #run regtests on unilevel MA (not available for multilevel) regtest(rma(yi, vi, data = dataopicons), model = "rma", predictor = "sei") regtest(rma(yi, vi, data = datapain4h), model = "rma", predictor = "sei") #END OF ANALYSIS#