In chjackson/adverse: Data on adverse events of adjuvant breast cancer treatment
load_all("../../gemtc/gemtc")
load_all("..")
library(dplyr)
filter <- dplyr::filter
library(ggplot2)
library(plotly)
library(stringr)
source("theme.r")
clean_aetype <- function(aetype){
aetype <- stringr::str_to_sentence(aetype)
aetype[aetype=="Ggt increased"] <-"GGT increased"
aetype[aetype=="Cpk increased"] <-"CPK increased"
aetype[aetype=="Inr increased"] <-"INR increased"
aetype[aetype=="Withdrawals because of aes"] <-"Withdrawals because of AEs"
aetype
}
Raw data on frequency of each type of event, aggregated over all studies
Proportions of individuals experiencing each type of event, aggregated over all studies and treatments (including placebo/observation only)
knitr::opts_chunk$set(fig.width=15, fig.height=20)
aecount <-
bpaearmtype %>%
filter(!aetype %in% c("ANY ADVERSE EVENT", "SERIOUS ADVERSE EVENTS")) %>%
group_by(aetype) %>%
summarise(N=sum(N), count=sum(count)) %>%
mutate(prop = count/N) %>%
mutate(aetype=factor(aetype, levels=unique(aetype)[order(prop)])) %>%
mutate(aetype = clean_aetype(aetype)) %>%
mutate(pse = N*prop*(1-prop))
p <- ggplot(aecount, aes(x=prop, y=aetype, size=count)) +
geom_point() +
xlab("Proportion having event") +
ylab("") +
paper_theme
# ggplotly(p)
Comparison of heterogeneity in odds ratio and risk difference. or baseline risk?
bpplot <- bpaeriskdiff %>%
mutate(
oddsact = ((count+0.5)/(N+1)) / (1 - (count+0.5)/(N+1)),
oddscon = ((rcon+0.5)/(ncon+1)) / (1 - (rcon+0.5)/(ncon+1)),
est = oddsact / oddscon,
selogor = sqrt(1/(rcon+0.5) + 1/(ncon-rcon+0.5) + 1/(count+0.5) + 1/(N-count+0.5)),
or = est,
orlower = exp(log(est) - qnorm(0.975)*selogor),
orupper = exp(log(est) + qnorm(0.975)*selogor),
a = count, b=N-count, c=rcon, d=ncon-rcon,
serd = sqrt(a*b/(a+b)^3 + c*d/(c+d)^3),
rdlower = riskdiff - qnorm(0.975)*serd,
rdupper = riskdiff + qnorm(0.975)*serd,
brisk = rcon / ncon,
brisklower = qbeta(0.025, 0.5 + rcon, 0.5 + ncon - rcon),
briskupper = qbeta(0.975, 0.5 + rcon, 0.5 + ncon - rcon)
) %>%
filter(aetype=="FEVER") %>%
select(study, or, orlower, orupper, riskdiff, rdlower, rdupper, brisk, brisklower, briskupper, count, N, rcon, ncon)
orres <- metabin(count, N, rcon, ncon, studlab=study, data=bpplot, sm="OR")
rdres <- metabin(count, N, rcon, ncon, studlab=study, data=bpplot, sm="RD")
orres <- as.data.frame(lapply(orres[c("TE.fixed","lower.fixed","upper.fixed")], exp))
names(orres) <- c("or", "orlower", "orupper")
rdres <- as.data.frame(rdres[c("TE.fixed","lower.fixed","upper.fixed")])
names(rdres) <- c("riskdiff", "rdlower", "rdupper")
orres$study <- rdres$study <- "Pooled estimate"
psize <- 3
lsize <- 1.5
wsize <- 0.2
p1 <- ggplot(bpplot, aes(x=study, y=or)) +
coord_flip(ylim=c(0.2, 20)) +
geom_hline(yintercept=1, col="gray") +
geom_point(size=psize) +
geom_errorbar(aes(ymin = orlower, ymax = orupper), size=lsize, width=wsize) +
scale_y_continuous(trans="log", breaks=c(0.2, 0.5, 1, 2, 5, 10, 20)) +
geom_point(data=orres, aes(y=or, x=0), col="red") +
geom_errorbar(data=orres, aes(ymin = orlower, ymax = orupper, x=0), col="red", width=2*wsize, size=lsize) +
geom_text(aes(x=0, y=1), hjust=1, label="Pooled estimate", col="red") +
paper_theme +
ylab("Odds ratio") + xlab("")
p2 <- ggplot(bpplot, aes(x=study, y=riskdiff)) +
coord_flip() +
geom_hline(yintercept=0, col="gray") +
geom_point(size=psize) +
geom_errorbar(aes(ymin = rdlower, ymax = rdupper), size=lsize, width=wsize) +
geom_point(data=rdres, aes(y=riskdiff, x=0), col="red") +
geom_errorbar(data=rdres, aes(ymin = rdlower, ymax = rdupper, x=0),
col="red", width=2*wsize, size=lsize) +
geom_text(aes(x=0, y=0), hjust=1, label="Pooled estimate", col="red") +
ylab("Risk difference") + xlab("") +
paper_theme +
theme(axis.text.y = element_blank())
p2
p3 <- ggplot(bpplot, aes(x=study, y=brisk)) +
coord_flip() +
geom_point(size=psize) +
geom_errorbar(aes(ymin = brisklower, ymax = briskupper), size=lsize, width=wsize) +
ylab("Proportion of controls having event") + xlab("") +
paper_theme +
theme(axis.text.y = element_blank())
gridExtra::grid.arrange(p1, p2, p3, ncol=3, widths=c(3,2,2))
Simple meta-analysis of bisphosphonate effects on risk for each event type
Including only studies where bisphosphonate compared to a control
Standard fixed effects meta analysis
library(meta)
mares <- data.frame(aetype=unique(as.character(aecount$aetype)))
mares$totn <- mares$totr <- mares$totncon <- mares$totrcon <-
mares$or <- mares$orl <- mares$oru <-
mares$rd <- mares$rdl <- mares$rdu <- NA
for (i in seq(along.with=mares$aetype)){
bpma <-
bpaeriskdiff %>%
mutate(aetype = clean_aetype(aetype)) %>%
filter(trtcon %in% c("100","101","103")) %>%
filter(aetype==mares$aetype[i])
if (nrow(bpma) > 0){
mres <- metabin(count, N, rcon, ncon, studlab=study, data=bpma, sm="OR")
mresrd <- metabin(count, N, rcon, ncon, studlab=study, data=bpma, sm="RD")
mares$totn[i] <- sum(bpma$N)
mares$totr[i] <- sum(bpma$count)
mares$totncon[i] <- sum(bpma$ncon)
mares$totrcon[i] <- sum(bpma$rcon)
mares[i,c("or","orl","oru")] <- exp(unlist(mres[c("TE.fixed","lower.fixed","upper.fixed")]))
mares[i,c("rd","rdl","rdu")] <- unlist(mresrd[c("TE.fixed","lower.fixed","upper.fixed")])
} else
mares[i,c("totn","totr","totncon","totrcon")] <- 0
}
mares <- mares %>%
mutate(sig = (or > 1.5) | (rd > 0.02)) %>%
mutate(label = sprintf("\n%s/%s treatment\n%s/%s control\nOR = %s\nRisk difference = %s",
totr,totn,totrcon,totncon,
round(or,3), round(rd,3)))
cres <- aecount %>% select(aetype, prop, count) %>% rename(est=prop) %>% mutate(measure="Proportion having event")
rdres <- mares %>% select(aetype, rd, rdl, rdu) %>% rename(est=rd, lower=rdl, upper=rdu) %>% mutate(measure="Risk difference (vs control)") %>% mutate(aetype=as.character(aetype))
orres <- mares %>% select(aetype, or, orl, oru) %>% rename(est=or, lower=orl, upper=oru) %>% mutate(measure="Odds ratio (bisphosphonates / control)") %>% mutate(aetype=as.character(aetype))
directplot <- cres %>%
full_join(rdres) %>% full_join(orres)
p0 <- directplot %>%
filter(measure=="Proportion having event") %>%
mutate(aetype = factor(.data$aetype, levels=cres$aetype[order(cres$est)])) %>%
ggplot(aes(y=est, x=aetype)) +
coord_flip() +
geom_point(position=position_dodge(-0.4)) +
geom_errorbar(aes(ymin=lower, ymax=upper),
position=position_dodge(-0.4)) +
scale_x_discrete(drop=FALSE) +
scale_y_continuous(breaks = seq(0, 0.3, by=0.05)) +
scale_colour_manual(values=c("black","red")) +
theme(legend.position = "none",
axis.text.y = element_blank()) +
ggtitle("Proportion having event") +
paper_theme +
ylab("") + xlab("")
levs <- cres$aetype[order(cres$est)]
p1dat <- directplot %>%
filter(measure=="Risk difference (vs control)") %>%
mutate(aetype = factor(.data$aetype, levels=levs)) %>%
mutate(sig = (est > 0.02 & lower > 0)) %>%
filter(!is.na(est))
levels(p1dat$aetype) <- sprintf("%s (%s)", levels(p1dat$aetype), round(cres$est[order(cres$est)],2))
p2dat <- directplot %>%
filter(measure=="Odds ratio (bisphosphonates / control)") %>%
mutate(aetype = factor(.data$aetype, levels=levs)) %>%
mutate(sig = (est > 1.5 & lower > 1)) %>%
filter(!is.na(est))
levels(p2dat$aetype) <- sprintf("%s (%s)", levels(p2dat$aetype), round(cres$est[order(cres$est)],2))
midblue <- "#9999FF"
p1dat <- droplevels(p1dat)
p2dat <- droplevels(p2dat)
levelssig1 <- p1dat$sig[match(levels(p1dat$aetype), p1dat$aetype)]
levelssig2 <- p2dat$sig[match(levels(p1dat$aetype), p2dat$aetype)]
levels(p2dat$aetype)[levelssig2]
sort(p2dat$aetype[p2dat$sig])
faccols <- ifelse((levelssig1 | levelssig2), "black", midblue)
p1 <- p1dat %>%
droplevels %>%
ggplot(aes(y=est, x=aetype, color=sig)) +
coord_flip(ylim=c(-0.1, 0.2)) +
scale_x_discrete(drop=FALSE) +
geom_hline(yintercept=0.0, col="gray") +
scale_y_continuous(breaks = seq(-0.1, 0.2, by=0.05)) +
geom_point(position=position_dodge(-0.4)) +
geom_errorbar(aes(ymin=lower, ymax=upper),
position=position_dodge(-0.4)) +
scale_colour_manual(values=c(midblue,"black")) +
theme(legend.position = "none",
axis.text.y = element_text(colour=faccols),
plot.title = element_text(hjust = 0)) +
ggtitle("Risk difference (vs control)") +
paper_theme +
ylab("") + xlab("")
p2 <- p2dat %>%
droplevels %>%
ggplot(aes(y=est, x=aetype, color=sig)) +
coord_flip(ylim=c(0.2, 20)) +
scale_x_discrete(drop=FALSE) +
scale_y_continuous(trans="log",
breaks=c(0.2, 0.5, 1, 2, 5, 10, 20)) +
geom_hline(yintercept=1.0, col="gray") +
geom_point(position=position_dodge(-0.4)) +
geom_errorbar(aes(ymin=lower, ymax=upper),
position=position_dodge(-0.4)) +
scale_colour_manual(values=c(midblue,"black")) +
theme(legend.position = "none",
axis.text.y = element_blank(),
plot.title = element_text(hjust = 0)) +
ggtitle("Odds ratio (bisphosphonates / control)") +
paper_theme +
ylab("") + xlab("")
library(gridExtra)
grid.arrange(p1, p2, ncol=2, widths=c(3,2))
#png("direct.png")
#grid.arrange(p1, p2, ncol=2, widths=c(3,2))
#dev.off()
Comparison of effects between subgroups
Meta-regression or subgroup analysis?
metareg function in meta package.
standard fixed effects model estimtes the weighted ave of study-specific effects among included studies
Moderators can be included in the fixed-effects model, yielding a
fixed-effects with moderators model...
the regression coefficients from the fitted model estimate the
weighted relationship (in the least squares sense) between the
true effects/outcomes and the moderator variables ...
also among the included studies [ can't generalize ]
Fixed effects meta-regression [ Viechtbauer, W. (2010). Conducting meta-analyses in R with the "metafor" package. Journal of Statistical Software, 36(3), 1-48. ]
mares <- data.frame(aetype=unique(as.character(aecount$aetype)))
mares[,c("totn","totr","totncon","totrcon",
"or","orl","oru", "rd","rdl","rdu",
"nmeno", "nbone", "nmenopre", "nmenopost", "nbonenone", "nboneall",
"meno.orp","bone.orp","meno.rdp","bone.rdp",
"ormeno","ormenol","ormenou", "orbone","orbonel","orboneu",
"rdmeno","rdmenol","rdmenou", "rdbone","rdbonel","rdboneu",
"orbonenone","orbonenonel","orbonenoneu", "orboneall","orbonealll","orboneallu",
"ormenopre","ormenoprel","ormenopreu", "ormenopost","ormenopostl","ormenopostu",
"rdbonenone","rdbonenonel","rdbonenoneu", "rdboneall","rdbonealll","rdboneallu",
"rdmenopre","rdmenoprel","rdmenopreu", "rdmenopost","rdmenopostl","rdmenopostu",
"totnmenopre", "totnmenopost", "totrmenopre", "totrmenopost",
"totnbonenone", "totnboneall", "totrbonenone", "totrboneall",
"totnconmenopre", "totnconmenopost", "totrconmenopre", "totrconmenopost",
"totnconbonenone", "totnconboneall", "totrconbonenone", "totrconboneall"
)] <- NA
for (i in seq(along.with=mares$aetype)){
bpma <-
bpaeriskdiff %>%
mutate(aetype = clean_aetype(aetype)) %>%
filter(trtcon %in% c("100","101","103")) %>%
filter(aetype==mares$aetype[i]) %>%
mutate(meno2 = ifelse(meno=="mixed", NA, as.character(meno)),
meno2 = factor(meno2, levels=c("pre","post")),
bonemeta2 = ifelse(bonemeta=="mixed", NA, as.character(bonemeta)),
bonemeta2 = factor(bonemeta2, levels = c("none", "all")))
if (nrow(bpma) > 0){
mres <- metabin(count, N, rcon, ncon, studlab=study, data=bpma, sm="OR")
mresrd <- metabin(count, N, rcon, ncon, studlab=study, data=bpma, sm="RD")
mares[i,"nmeno"] <- sum(!is.na(bpma$meno2))
mares[i,"nmenopre"] <- sum(bpma$meno2=="pre", na.rm=TRUE)
mares[i,"nmenopost"] <- sum(bpma$meno2=="post", na.rm=TRUE)
if (mares[i,"nmeno"] > 0){
mrtry <- try(mregmeno <- metareg(mres, ~ meno2))
if (!inherits(mrtry, "try-error") && (length(coef(mrtry))>1)){
orsmeno <- exp(coef(summary(mregmeno)))[c("meno2post"),c("estimate","ci.lb","ci.ub")]
mares$meno.orp[i] <- mregmeno$QMp
mares[i,c("ormeno","ormenol","ormenou")] <- orsmeno
ormenopre <- unlist(unclass(predict(mrtry, newmods = 0))[c("pred","ci.lb","ci.ub")])
ormenopost <- unlist(unclass(predict(mrtry, newmods = 1))[c("pred","ci.lb","ci.ub")])
mares[i,c("ormenopre","ormenoprel","ormenopreu")] <- exp(ormenopre)
mares[i,c("ormenopost","ormenopostl","ormenopostu")] <- exp(ormenopost)
}
mrtryd <- try(mregmeno <- metareg(mresrd, ~ meno2))
if (!inherits(mrtryd, "try-error") && (length(coef(mrtryd))>1)){
rdsmeno <- coef(summary(mregmeno))[c("meno2post"),c("estimate","ci.lb","ci.ub")]
mares$meno.rdp[i] <- mregmeno$QMp
mares[i,c("rdmeno","rdmenol","rdmenou")] <- rdsmeno
rdmenopre <- unlist(unclass(predict(mrtryd, newmods = 0))[c("pred","ci.lb","ci.ub")])
rdmenopost <- unlist(unclass(predict(mrtryd, newmods = 1))[c("pred","ci.lb","ci.ub")])
mares[i,c("rdmenopre","rdmenoprel","rdmenopreu")] <- rdmenopre
mares[i,c("rdmenopost","rdmenopostl","rdmenopostu")] <- rdmenopost
}
}
mares[i,"nbone"] <- sum(!is.na(bpma$bonemeta2))
mares[i,"nboneall"] <- sum(bpma$bonemeta2=="all", na.rm=TRUE)
mares[i,"nbonenone"] <- sum(bpma$bonemeta2=="none", na.rm=TRUE)
if (mares[i,"nbone"] > 0){
mrtry <- try(mregbone <- metareg(mres, ~ bonemeta2))
if (!inherits(mrtry, "try-error") && (length(coef(mrtry))>1)){
orsbone <- exp(coef(summary(mregbone)))[c("bonemeta2all"),c("estimate","ci.lb","ci.ub")]
mares$bone.orp[i] <- mregbone$QMp
mares[i,c("orbone","orbonel","orboneu")] <- orsbone
orbonenone <- unlist(unclass(predict(mrtry, newmods = 0))[c("pred","ci.lb","ci.ub")])
orboneall <- unlist(unclass(predict(mrtry, newmods = 1))[c("pred","ci.lb","ci.ub")])
mares[i,c("orbonenone","orbonenonel","orbonenoneu")] <- exp(orbonenone)
mares[i,c("orboneall","orbonealll","orboneallu")] <- exp(orboneall)
}
mrtryd <- try(mregbone <- metareg(mresrd, ~ bonemeta2))
if (!inherits(mrtryd, "try-error") && (length(coef(mrtryd))>1)){
rdsbone <- coef(summary(mregbone))[c("bonemeta2all"),c("estimate","ci.lb","ci.ub")]
mares$bone.rdp[i] <- mregbone$QMp
mares[i,c("rdbone","rdbonel","rdboneu")] <- rdsbone
rdbonenone <- unlist(unclass(predict(mrtryd, newmods = 0))[c("pred","ci.lb","ci.ub")])
rdboneall <- unlist(unclass(predict(mrtryd, newmods = 1))[c("pred","ci.lb","ci.ub")])
mares[i,c("rdbonenone","rdbonenonel","rdbonenoneu")] <- rdbonenone
mares[i,c("rdboneall","rdbonealll","rdboneallu")] <- rdboneall
}
}
mares$totn[i] <- sum(bpma$N)
mares$totr[i] <- sum(bpma$count)
mares$totncon[i] <- sum(bpma$ncon)
mares$totrcon[i] <- sum(bpma$rcon)
mares$totnmenopre[i] <- sum(bpma$N[bpma$meno2=="pre"],na.rm=TRUE)
mares$totnmenopost[i] <- sum(bpma$N[bpma$meno2=="post"],na.rm=TRUE)
mares$totrmenopre[i] <- sum(bpma$count[bpma$meno2=="pre"],na.rm=TRUE)
mares$totrmenopost[i] <- sum(bpma$count[bpma$meno2=="post"],na.rm=TRUE)
mares$totnbonenone[i] <- sum(bpma$N[bpma$bonemeta2=="none"],na.rm=TRUE)
mares$totnboneall[i] <- sum(bpma$N[bpma$bonemeta2=="all"],na.rm=TRUE)
mares$totrbonenone[i] <- sum(bpma$count[bpma$bonemeta2=="none"],na.rm=TRUE)
mares$totrboneall[i] <- sum(bpma$count[bpma$bonemeta2=="all"],na.rm=TRUE)
mares$totnconmenopre[i] <- sum(bpma$ncon[bpma$meno2=="pre"],na.rm=TRUE)
mares$totnconmenopost[i] <- sum(bpma$ncon[bpma$meno2=="post"],na.rm=TRUE)
mares$totrconmenopre[i] <- sum(bpma$rcon[bpma$meno2=="pre"],na.rm=TRUE)
mares$totrconmenopost[i] <- sum(bpma$rcon[bpma$meno2=="post"],na.rm=TRUE)
mares$totnconbonenone[i] <- sum(bpma$ncon[bpma$bonemeta2=="none"],na.rm=TRUE)
mares$totnconboneall[i] <- sum(bpma$ncon[bpma$bonemeta2=="all"],na.rm=TRUE)
mares$totrconbonenone[i] <- sum(bpma$rcon[bpma$bonemeta2=="none"],na.rm=TRUE)
mares$totrconboneall[i] <- sum(bpma$rcon[bpma$bonemeta2=="all"],na.rm=TRUE)
mares$totr[i] <- sum(bpma$count)
mares$totncon[i] <- sum(bpma$ncon)
mares$totrcon[i] <- sum(bpma$rcon)
mares[i,c("or","orl","oru")] <- exp(unlist(mres[c("TE.fixed","lower.fixed","upper.fixed")]))
mares[i,c("rd","rdl","rdu")] <- unlist(mresrd[c("TE.fixed","lower.fixed","upper.fixed")])
} else
mares[i,c("totn","totr","totncon","totrcon")] <- 0
}
# Filter to those for which the OR or RD is prac and stat sig for a subgroup
# 1.5 for ORs, and 0.02 for RDs
mares <- mares %>% mutate(
statsigbonerd = bone.rdp < 0.05 & !is.na(bone.rdp),
statsigboneor = bone.orp < 0.05 & !is.na(bone.orp),
statsigmenord = meno.rdp < 0.05 & !is.na(meno.rdp),
statsigmenoor = meno.orp < 0.05 & !is.na(meno.orp),
sigbonenone = (rdbonenone > 0.02 & !is.na(rdbonenone) & statsigbonerd) | (orbonenone > 1.5 & !is.na(orbonenone) & statsigboneor),
sigboneall = (rdboneall > 0.02 & !is.na(rdboneall) & statsigbonerd) | (orboneall > 1.5 & !is.na(orboneall) & statsigboneor),
sigmenopre = (rdmenopre > 0.02 & !is.na(rdmenopre) & statsigmenord) | (ormenopre > 1.5 & !is.na(ormenopre) & statsigmenoor),
sigmenopost = (rdmenopost > 0.02 & !is.na(rdmenopost) & statsigmenord) | (ormenopost > 1.5 & !is.na(ormenopost) & statsigmenoor),
datamenopre = sprintf("%s/%s vs %s/%s", totrmenopre, totnmenopre, totrconmenopre, totnconmenopre),
datamenopost = sprintf("%s/%s vs %s/%s", totrmenopost, totnmenopost, totrconmenopost, totnconmenopost),
databonenone = sprintf("%s/%s vs %s/%s", totrbonenone, totnbonenone, totrconbonenone, totnconbonenone),
databoneall = sprintf("%s/%s vs %s/%s", totrboneall, totnboneall, totrconboneall, totnconboneall),
orfmenopre = sprintf("%s (%s,%s)", round(ormenopre,2), round(ormenoprel,2), round(ormenopreu,2)),
orfmenopost = sprintf("%s (%s,%s)", round(ormenopost,2), round(ormenopostl,2), round(ormenopostu,2)),
rdfmenopre = sprintf("%s (%s,%s)", round(rdmenopre,2), round(rdmenoprel,2), round(rdmenopreu,2)),
rdfmenopost = sprintf("%s (%s,%s)", round(rdmenopost,2), round(rdmenopostl,2), round(rdmenopostu,2)),
orfbonenone = sprintf("%s (%s,%s)", round(orbonenone,2), round(orbonenonel,2), round(orbonenoneu,2)),
orfboneall = sprintf("%s (%s,%s)", round(orboneall,2), round(orbonealll,2), round(orboneallu,2)),
rdfbonenone = sprintf("%s (%s,%s)", round(rdbonenone,2), round(rdbonenonel,2), round(rdbonenoneu,2)),
rdfboneall = sprintf("%s (%s,%s)", round(rdboneall,2), round(rdbonealll,2), round(rdboneallu,2))
)
Need a clean presentation for tables
figure this out with single meno result first
Data in pre: 2/40 vs 9/41
Data in post:
OR in pre, OR in post
RD in pre, RD in post (each estimate and CI)
cardia OR 1.5 in bone none (but other subgroup has opposite effect)
To investigate whether the adverse effects of bisphosphonates were modified by menopausal status or by the presence of bone metastases, the previous meta-analyses of the studies with direct comparisons of a bisphosphonate with a non-bisphosphonate control were extended to meta-regression analyses [ ref Viechtbauer, W. (2010). Conducting meta-analyses in R with the "metafor" package. Journal of Statistical Software, 36(3), 1-48. ]. The studies were categorised according to whether they included pre-menopausal, post-menopausal women or a mixture, and according to whether the participants had bone metastases (no participants, all participants or a mixture). The studies with a mixture of patient types were excluded, then a linear regression term for "all metastatic patients" or "all post-menopausal patients", compared to none, was included on the log odds ratio or the risk difference.
The meta-regression results for all adverse events were filtered to include only those where the regression effect was statistically significant, and the subgroup-specific treatment effect on adverse event risk for the higher-risk subgroup was practically significant (odds ratio > 1.5 or risk difference > 2%). The remaining results are presented in Table ??? and ???
Meno.
only sig one is Insomnia. big OR for post, but it's not stat sig
This is based on small counts
Bone
Note the thing that is significant is the contrast between the two subgroups.
Want the contrast to be stat sig, but also the absolute risk for the higher risk subgroup to be prac sig
- Cardiac - bone none has bigger O, but RD not sig
- Dizziness - same
- Fatigue - bone all has bigger OR and RD
- Flu - bone none has bigger RD, but bone all has bigger OR?
- Hypocalcaemia - same
- Increased bone pain - bone none has bigger OR and RD
- Flu-like - bone none has bigger RD, but small counts.
- Neuralgia - small counts.
- Neutropenia - sig RD but not sig OR.
- Osteo - tiny counts
do we want it prac sig for both OR and RD? clearly OR doesn't matter if RD is low.
but does RD matter if OR is low? e.g. 70%vs60% is less exciting than 20%vs10%
maybe
options(width=150)
mares %>%
filter(sigmenopre | sigmenopost) %>%
select(aetype, nmenopre, nmenopost, datamenopre, datamenopost, orfmenopre, rdfmenopre, orfmenopost, rdfmenopost, meno.orp, meno.rdp)
mares %>%
filter(sigbonenone | sigboneall) %>%
mutate(bone.orp = round(bone.orp,2), bone.rdp=round(bone.rdp,2)) %>%
select(aetype, nbonenone, nboneall, databonenone, databoneall, orfbonenone, rdfbonenone, orfboneall, rdfboneall, bone.orp, bone.rdp)
chjackson/adverse documentation built on June 16, 2022, 4:53 p.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
load_all("../../gemtc/gemtc") load_all("..") library(dplyr) filter <- dplyr::filter library(ggplot2) library(plotly) library(stringr) source("theme.r") clean_aetype <- function(aetype){ aetype <- stringr::str_to_sentence(aetype) aetype[aetype=="Ggt increased"] <-"GGT increased" aetype[aetype=="Cpk increased"] <-"CPK increased" aetype[aetype=="Inr increased"] <-"INR increased" aetype[aetype=="Withdrawals because of aes"] <-"Withdrawals because of AEs" aetype }
Raw data on frequency of each type of event, aggregated over all studies
Proportions of individuals experiencing each type of event, aggregated over all studies and treatments (including placebo/observation only)
knitr::opts_chunk$set(fig.width=15, fig.height=20)
aecount <- bpaearmtype %>% filter(!aetype %in% c("ANY ADVERSE EVENT", "SERIOUS ADVERSE EVENTS")) %>% group_by(aetype) %>% summarise(N=sum(N), count=sum(count)) %>% mutate(prop = count/N) %>% mutate(aetype=factor(aetype, levels=unique(aetype)[order(prop)])) %>% mutate(aetype = clean_aetype(aetype)) %>% mutate(pse = N*prop*(1-prop)) p <- ggplot(aecount, aes(x=prop, y=aetype, size=count)) + geom_point() + xlab("Proportion having event") + ylab("") + paper_theme # ggplotly(p)
Comparison of heterogeneity in odds ratio and risk difference. or baseline risk?
bpplot <- bpaeriskdiff %>% mutate( oddsact = ((count+0.5)/(N+1)) / (1 - (count+0.5)/(N+1)), oddscon = ((rcon+0.5)/(ncon+1)) / (1 - (rcon+0.5)/(ncon+1)), est = oddsact / oddscon, selogor = sqrt(1/(rcon+0.5) + 1/(ncon-rcon+0.5) + 1/(count+0.5) + 1/(N-count+0.5)), or = est, orlower = exp(log(est) - qnorm(0.975)*selogor), orupper = exp(log(est) + qnorm(0.975)*selogor), a = count, b=N-count, c=rcon, d=ncon-rcon, serd = sqrt(a*b/(a+b)^3 + c*d/(c+d)^3), rdlower = riskdiff - qnorm(0.975)*serd, rdupper = riskdiff + qnorm(0.975)*serd, brisk = rcon / ncon, brisklower = qbeta(0.025, 0.5 + rcon, 0.5 + ncon - rcon), briskupper = qbeta(0.975, 0.5 + rcon, 0.5 + ncon - rcon) ) %>% filter(aetype=="FEVER") %>% select(study, or, orlower, orupper, riskdiff, rdlower, rdupper, brisk, brisklower, briskupper, count, N, rcon, ncon) orres <- metabin(count, N, rcon, ncon, studlab=study, data=bpplot, sm="OR") rdres <- metabin(count, N, rcon, ncon, studlab=study, data=bpplot, sm="RD") orres <- as.data.frame(lapply(orres[c("TE.fixed","lower.fixed","upper.fixed")], exp)) names(orres) <- c("or", "orlower", "orupper") rdres <- as.data.frame(rdres[c("TE.fixed","lower.fixed","upper.fixed")]) names(rdres) <- c("riskdiff", "rdlower", "rdupper") orres$study <- rdres$study <- "Pooled estimate" psize <- 3 lsize <- 1.5 wsize <- 0.2 p1 <- ggplot(bpplot, aes(x=study, y=or)) + coord_flip(ylim=c(0.2, 20)) + geom_hline(yintercept=1, col="gray") + geom_point(size=psize) + geom_errorbar(aes(ymin = orlower, ymax = orupper), size=lsize, width=wsize) + scale_y_continuous(trans="log", breaks=c(0.2, 0.5, 1, 2, 5, 10, 20)) + geom_point(data=orres, aes(y=or, x=0), col="red") + geom_errorbar(data=orres, aes(ymin = orlower, ymax = orupper, x=0), col="red", width=2*wsize, size=lsize) + geom_text(aes(x=0, y=1), hjust=1, label="Pooled estimate", col="red") + paper_theme + ylab("Odds ratio") + xlab("") p2 <- ggplot(bpplot, aes(x=study, y=riskdiff)) + coord_flip() + geom_hline(yintercept=0, col="gray") + geom_point(size=psize) + geom_errorbar(aes(ymin = rdlower, ymax = rdupper), size=lsize, width=wsize) + geom_point(data=rdres, aes(y=riskdiff, x=0), col="red") + geom_errorbar(data=rdres, aes(ymin = rdlower, ymax = rdupper, x=0), col="red", width=2*wsize, size=lsize) + geom_text(aes(x=0, y=0), hjust=1, label="Pooled estimate", col="red") + ylab("Risk difference") + xlab("") + paper_theme + theme(axis.text.y = element_blank()) p2 p3 <- ggplot(bpplot, aes(x=study, y=brisk)) + coord_flip() + geom_point(size=psize) + geom_errorbar(aes(ymin = brisklower, ymax = briskupper), size=lsize, width=wsize) + ylab("Proportion of controls having event") + xlab("") + paper_theme + theme(axis.text.y = element_blank()) gridExtra::grid.arrange(p1, p2, p3, ncol=3, widths=c(3,2,2))
Simple meta-analysis of bisphosphonate effects on risk for each event type
Including only studies where bisphosphonate compared to a control
Standard fixed effects meta analysis
library(meta) mares <- data.frame(aetype=unique(as.character(aecount$aetype))) mares$totn <- mares$totr <- mares$totncon <- mares$totrcon <- mares$or <- mares$orl <- mares$oru <- mares$rd <- mares$rdl <- mares$rdu <- NA for (i in seq(along.with=mares$aetype)){ bpma <- bpaeriskdiff %>% mutate(aetype = clean_aetype(aetype)) %>% filter(trtcon %in% c("100","101","103")) %>% filter(aetype==mares$aetype[i]) if (nrow(bpma) > 0){ mres <- metabin(count, N, rcon, ncon, studlab=study, data=bpma, sm="OR") mresrd <- metabin(count, N, rcon, ncon, studlab=study, data=bpma, sm="RD") mares$totn[i] <- sum(bpma$N) mares$totr[i] <- sum(bpma$count) mares$totncon[i] <- sum(bpma$ncon) mares$totrcon[i] <- sum(bpma$rcon) mares[i,c("or","orl","oru")] <- exp(unlist(mres[c("TE.fixed","lower.fixed","upper.fixed")])) mares[i,c("rd","rdl","rdu")] <- unlist(mresrd[c("TE.fixed","lower.fixed","upper.fixed")]) } else mares[i,c("totn","totr","totncon","totrcon")] <- 0 } mares <- mares %>% mutate(sig = (or > 1.5) | (rd > 0.02)) %>% mutate(label = sprintf("\n%s/%s treatment\n%s/%s control\nOR = %s\nRisk difference = %s", totr,totn,totrcon,totncon, round(or,3), round(rd,3)))
cres <- aecount %>% select(aetype, prop, count) %>% rename(est=prop) %>% mutate(measure="Proportion having event") rdres <- mares %>% select(aetype, rd, rdl, rdu) %>% rename(est=rd, lower=rdl, upper=rdu) %>% mutate(measure="Risk difference (vs control)") %>% mutate(aetype=as.character(aetype)) orres <- mares %>% select(aetype, or, orl, oru) %>% rename(est=or, lower=orl, upper=oru) %>% mutate(measure="Odds ratio (bisphosphonates / control)") %>% mutate(aetype=as.character(aetype)) directplot <- cres %>% full_join(rdres) %>% full_join(orres) p0 <- directplot %>% filter(measure=="Proportion having event") %>% mutate(aetype = factor(.data$aetype, levels=cres$aetype[order(cres$est)])) %>% ggplot(aes(y=est, x=aetype)) + coord_flip() + geom_point(position=position_dodge(-0.4)) + geom_errorbar(aes(ymin=lower, ymax=upper), position=position_dodge(-0.4)) + scale_x_discrete(drop=FALSE) + scale_y_continuous(breaks = seq(0, 0.3, by=0.05)) + scale_colour_manual(values=c("black","red")) + theme(legend.position = "none", axis.text.y = element_blank()) + ggtitle("Proportion having event") + paper_theme + ylab("") + xlab("") levs <- cres$aetype[order(cres$est)] p1dat <- directplot %>% filter(measure=="Risk difference (vs control)") %>% mutate(aetype = factor(.data$aetype, levels=levs)) %>% mutate(sig = (est > 0.02 & lower > 0)) %>% filter(!is.na(est)) levels(p1dat$aetype) <- sprintf("%s (%s)", levels(p1dat$aetype), round(cres$est[order(cres$est)],2)) p2dat <- directplot %>% filter(measure=="Odds ratio (bisphosphonates / control)") %>% mutate(aetype = factor(.data$aetype, levels=levs)) %>% mutate(sig = (est > 1.5 & lower > 1)) %>% filter(!is.na(est)) levels(p2dat$aetype) <- sprintf("%s (%s)", levels(p2dat$aetype), round(cres$est[order(cres$est)],2)) midblue <- "#9999FF" p1dat <- droplevels(p1dat) p2dat <- droplevels(p2dat) levelssig1 <- p1dat$sig[match(levels(p1dat$aetype), p1dat$aetype)] levelssig2 <- p2dat$sig[match(levels(p1dat$aetype), p2dat$aetype)] levels(p2dat$aetype)[levelssig2] sort(p2dat$aetype[p2dat$sig]) faccols <- ifelse((levelssig1 | levelssig2), "black", midblue) p1 <- p1dat %>% droplevels %>% ggplot(aes(y=est, x=aetype, color=sig)) + coord_flip(ylim=c(-0.1, 0.2)) + scale_x_discrete(drop=FALSE) + geom_hline(yintercept=0.0, col="gray") + scale_y_continuous(breaks = seq(-0.1, 0.2, by=0.05)) + geom_point(position=position_dodge(-0.4)) + geom_errorbar(aes(ymin=lower, ymax=upper), position=position_dodge(-0.4)) + scale_colour_manual(values=c(midblue,"black")) + theme(legend.position = "none", axis.text.y = element_text(colour=faccols), plot.title = element_text(hjust = 0)) + ggtitle("Risk difference (vs control)") + paper_theme + ylab("") + xlab("") p2 <- p2dat %>% droplevels %>% ggplot(aes(y=est, x=aetype, color=sig)) + coord_flip(ylim=c(0.2, 20)) + scale_x_discrete(drop=FALSE) + scale_y_continuous(trans="log", breaks=c(0.2, 0.5, 1, 2, 5, 10, 20)) + geom_hline(yintercept=1.0, col="gray") + geom_point(position=position_dodge(-0.4)) + geom_errorbar(aes(ymin=lower, ymax=upper), position=position_dodge(-0.4)) + scale_colour_manual(values=c(midblue,"black")) + theme(legend.position = "none", axis.text.y = element_blank(), plot.title = element_text(hjust = 0)) + ggtitle("Odds ratio (bisphosphonates / control)") + paper_theme + ylab("") + xlab("") library(gridExtra) grid.arrange(p1, p2, ncol=2, widths=c(3,2)) #png("direct.png") #grid.arrange(p1, p2, ncol=2, widths=c(3,2)) #dev.off()
Comparison of effects between subgroups
Meta-regression or subgroup analysis?
metareg function in meta package.
standard fixed effects model estimtes the weighted ave of study-specific effects among included studies Moderators can be included in the fixed-effects model, yielding a fixed-effects with moderators model... the regression coefficients from the fitted model estimate the weighted relationship (in the least squares sense) between the true effects/outcomes and the moderator variables ... also among the included studies [ can't generalize ]
Fixed effects meta-regression [ Viechtbauer, W. (2010). Conducting meta-analyses in R with the "metafor" package. Journal of Statistical Software, 36(3), 1-48. ]
mares <- data.frame(aetype=unique(as.character(aecount$aetype))) mares[,c("totn","totr","totncon","totrcon", "or","orl","oru", "rd","rdl","rdu", "nmeno", "nbone", "nmenopre", "nmenopost", "nbonenone", "nboneall", "meno.orp","bone.orp","meno.rdp","bone.rdp", "ormeno","ormenol","ormenou", "orbone","orbonel","orboneu", "rdmeno","rdmenol","rdmenou", "rdbone","rdbonel","rdboneu", "orbonenone","orbonenonel","orbonenoneu", "orboneall","orbonealll","orboneallu", "ormenopre","ormenoprel","ormenopreu", "ormenopost","ormenopostl","ormenopostu", "rdbonenone","rdbonenonel","rdbonenoneu", "rdboneall","rdbonealll","rdboneallu", "rdmenopre","rdmenoprel","rdmenopreu", "rdmenopost","rdmenopostl","rdmenopostu", "totnmenopre", "totnmenopost", "totrmenopre", "totrmenopost", "totnbonenone", "totnboneall", "totrbonenone", "totrboneall", "totnconmenopre", "totnconmenopost", "totrconmenopre", "totrconmenopost", "totnconbonenone", "totnconboneall", "totrconbonenone", "totrconboneall" )] <- NA for (i in seq(along.with=mares$aetype)){ bpma <- bpaeriskdiff %>% mutate(aetype = clean_aetype(aetype)) %>% filter(trtcon %in% c("100","101","103")) %>% filter(aetype==mares$aetype[i]) %>% mutate(meno2 = ifelse(meno=="mixed", NA, as.character(meno)), meno2 = factor(meno2, levels=c("pre","post")), bonemeta2 = ifelse(bonemeta=="mixed", NA, as.character(bonemeta)), bonemeta2 = factor(bonemeta2, levels = c("none", "all"))) if (nrow(bpma) > 0){ mres <- metabin(count, N, rcon, ncon, studlab=study, data=bpma, sm="OR") mresrd <- metabin(count, N, rcon, ncon, studlab=study, data=bpma, sm="RD") mares[i,"nmeno"] <- sum(!is.na(bpma$meno2)) mares[i,"nmenopre"] <- sum(bpma$meno2=="pre", na.rm=TRUE) mares[i,"nmenopost"] <- sum(bpma$meno2=="post", na.rm=TRUE) if (mares[i,"nmeno"] > 0){ mrtry <- try(mregmeno <- metareg(mres, ~ meno2)) if (!inherits(mrtry, "try-error") && (length(coef(mrtry))>1)){ orsmeno <- exp(coef(summary(mregmeno)))[c("meno2post"),c("estimate","ci.lb","ci.ub")] mares$meno.orp[i] <- mregmeno$QMp mares[i,c("ormeno","ormenol","ormenou")] <- orsmeno ormenopre <- unlist(unclass(predict(mrtry, newmods = 0))[c("pred","ci.lb","ci.ub")]) ormenopost <- unlist(unclass(predict(mrtry, newmods = 1))[c("pred","ci.lb","ci.ub")]) mares[i,c("ormenopre","ormenoprel","ormenopreu")] <- exp(ormenopre) mares[i,c("ormenopost","ormenopostl","ormenopostu")] <- exp(ormenopost) } mrtryd <- try(mregmeno <- metareg(mresrd, ~ meno2)) if (!inherits(mrtryd, "try-error") && (length(coef(mrtryd))>1)){ rdsmeno <- coef(summary(mregmeno))[c("meno2post"),c("estimate","ci.lb","ci.ub")] mares$meno.rdp[i] <- mregmeno$QMp mares[i,c("rdmeno","rdmenol","rdmenou")] <- rdsmeno rdmenopre <- unlist(unclass(predict(mrtryd, newmods = 0))[c("pred","ci.lb","ci.ub")]) rdmenopost <- unlist(unclass(predict(mrtryd, newmods = 1))[c("pred","ci.lb","ci.ub")]) mares[i,c("rdmenopre","rdmenoprel","rdmenopreu")] <- rdmenopre mares[i,c("rdmenopost","rdmenopostl","rdmenopostu")] <- rdmenopost } } mares[i,"nbone"] <- sum(!is.na(bpma$bonemeta2)) mares[i,"nboneall"] <- sum(bpma$bonemeta2=="all", na.rm=TRUE) mares[i,"nbonenone"] <- sum(bpma$bonemeta2=="none", na.rm=TRUE) if (mares[i,"nbone"] > 0){ mrtry <- try(mregbone <- metareg(mres, ~ bonemeta2)) if (!inherits(mrtry, "try-error") && (length(coef(mrtry))>1)){ orsbone <- exp(coef(summary(mregbone)))[c("bonemeta2all"),c("estimate","ci.lb","ci.ub")] mares$bone.orp[i] <- mregbone$QMp mares[i,c("orbone","orbonel","orboneu")] <- orsbone orbonenone <- unlist(unclass(predict(mrtry, newmods = 0))[c("pred","ci.lb","ci.ub")]) orboneall <- unlist(unclass(predict(mrtry, newmods = 1))[c("pred","ci.lb","ci.ub")]) mares[i,c("orbonenone","orbonenonel","orbonenoneu")] <- exp(orbonenone) mares[i,c("orboneall","orbonealll","orboneallu")] <- exp(orboneall) } mrtryd <- try(mregbone <- metareg(mresrd, ~ bonemeta2)) if (!inherits(mrtryd, "try-error") && (length(coef(mrtryd))>1)){ rdsbone <- coef(summary(mregbone))[c("bonemeta2all"),c("estimate","ci.lb","ci.ub")] mares$bone.rdp[i] <- mregbone$QMp mares[i,c("rdbone","rdbonel","rdboneu")] <- rdsbone rdbonenone <- unlist(unclass(predict(mrtryd, newmods = 0))[c("pred","ci.lb","ci.ub")]) rdboneall <- unlist(unclass(predict(mrtryd, newmods = 1))[c("pred","ci.lb","ci.ub")]) mares[i,c("rdbonenone","rdbonenonel","rdbonenoneu")] <- rdbonenone mares[i,c("rdboneall","rdbonealll","rdboneallu")] <- rdboneall } } mares$totn[i] <- sum(bpma$N) mares$totr[i] <- sum(bpma$count) mares$totncon[i] <- sum(bpma$ncon) mares$totrcon[i] <- sum(bpma$rcon) mares$totnmenopre[i] <- sum(bpma$N[bpma$meno2=="pre"],na.rm=TRUE) mares$totnmenopost[i] <- sum(bpma$N[bpma$meno2=="post"],na.rm=TRUE) mares$totrmenopre[i] <- sum(bpma$count[bpma$meno2=="pre"],na.rm=TRUE) mares$totrmenopost[i] <- sum(bpma$count[bpma$meno2=="post"],na.rm=TRUE) mares$totnbonenone[i] <- sum(bpma$N[bpma$bonemeta2=="none"],na.rm=TRUE) mares$totnboneall[i] <- sum(bpma$N[bpma$bonemeta2=="all"],na.rm=TRUE) mares$totrbonenone[i] <- sum(bpma$count[bpma$bonemeta2=="none"],na.rm=TRUE) mares$totrboneall[i] <- sum(bpma$count[bpma$bonemeta2=="all"],na.rm=TRUE) mares$totnconmenopre[i] <- sum(bpma$ncon[bpma$meno2=="pre"],na.rm=TRUE) mares$totnconmenopost[i] <- sum(bpma$ncon[bpma$meno2=="post"],na.rm=TRUE) mares$totrconmenopre[i] <- sum(bpma$rcon[bpma$meno2=="pre"],na.rm=TRUE) mares$totrconmenopost[i] <- sum(bpma$rcon[bpma$meno2=="post"],na.rm=TRUE) mares$totnconbonenone[i] <- sum(bpma$ncon[bpma$bonemeta2=="none"],na.rm=TRUE) mares$totnconboneall[i] <- sum(bpma$ncon[bpma$bonemeta2=="all"],na.rm=TRUE) mares$totrconbonenone[i] <- sum(bpma$rcon[bpma$bonemeta2=="none"],na.rm=TRUE) mares$totrconboneall[i] <- sum(bpma$rcon[bpma$bonemeta2=="all"],na.rm=TRUE) mares$totr[i] <- sum(bpma$count) mares$totncon[i] <- sum(bpma$ncon) mares$totrcon[i] <- sum(bpma$rcon) mares[i,c("or","orl","oru")] <- exp(unlist(mres[c("TE.fixed","lower.fixed","upper.fixed")])) mares[i,c("rd","rdl","rdu")] <- unlist(mresrd[c("TE.fixed","lower.fixed","upper.fixed")]) } else mares[i,c("totn","totr","totncon","totrcon")] <- 0 } # Filter to those for which the OR or RD is prac and stat sig for a subgroup # 1.5 for ORs, and 0.02 for RDs mares <- mares %>% mutate( statsigbonerd = bone.rdp < 0.05 & !is.na(bone.rdp), statsigboneor = bone.orp < 0.05 & !is.na(bone.orp), statsigmenord = meno.rdp < 0.05 & !is.na(meno.rdp), statsigmenoor = meno.orp < 0.05 & !is.na(meno.orp), sigbonenone = (rdbonenone > 0.02 & !is.na(rdbonenone) & statsigbonerd) | (orbonenone > 1.5 & !is.na(orbonenone) & statsigboneor), sigboneall = (rdboneall > 0.02 & !is.na(rdboneall) & statsigbonerd) | (orboneall > 1.5 & !is.na(orboneall) & statsigboneor), sigmenopre = (rdmenopre > 0.02 & !is.na(rdmenopre) & statsigmenord) | (ormenopre > 1.5 & !is.na(ormenopre) & statsigmenoor), sigmenopost = (rdmenopost > 0.02 & !is.na(rdmenopost) & statsigmenord) | (ormenopost > 1.5 & !is.na(ormenopost) & statsigmenoor), datamenopre = sprintf("%s/%s vs %s/%s", totrmenopre, totnmenopre, totrconmenopre, totnconmenopre), datamenopost = sprintf("%s/%s vs %s/%s", totrmenopost, totnmenopost, totrconmenopost, totnconmenopost), databonenone = sprintf("%s/%s vs %s/%s", totrbonenone, totnbonenone, totrconbonenone, totnconbonenone), databoneall = sprintf("%s/%s vs %s/%s", totrboneall, totnboneall, totrconboneall, totnconboneall), orfmenopre = sprintf("%s (%s,%s)", round(ormenopre,2), round(ormenoprel,2), round(ormenopreu,2)), orfmenopost = sprintf("%s (%s,%s)", round(ormenopost,2), round(ormenopostl,2), round(ormenopostu,2)), rdfmenopre = sprintf("%s (%s,%s)", round(rdmenopre,2), round(rdmenoprel,2), round(rdmenopreu,2)), rdfmenopost = sprintf("%s (%s,%s)", round(rdmenopost,2), round(rdmenopostl,2), round(rdmenopostu,2)), orfbonenone = sprintf("%s (%s,%s)", round(orbonenone,2), round(orbonenonel,2), round(orbonenoneu,2)), orfboneall = sprintf("%s (%s,%s)", round(orboneall,2), round(orbonealll,2), round(orboneallu,2)), rdfbonenone = sprintf("%s (%s,%s)", round(rdbonenone,2), round(rdbonenonel,2), round(rdbonenoneu,2)), rdfboneall = sprintf("%s (%s,%s)", round(rdboneall,2), round(rdbonealll,2), round(rdboneallu,2)) )
Need a clean presentation for tables figure this out with single meno result first Data in pre: 2/40 vs 9/41 Data in post: OR in pre, OR in post RD in pre, RD in post (each estimate and CI)
cardia OR 1.5 in bone none (but other subgroup has opposite effect)
To investigate whether the adverse effects of bisphosphonates were modified by menopausal status or by the presence of bone metastases, the previous meta-analyses of the studies with direct comparisons of a bisphosphonate with a non-bisphosphonate control were extended to meta-regression analyses [ ref Viechtbauer, W. (2010). Conducting meta-analyses in R with the "metafor" package. Journal of Statistical Software, 36(3), 1-48. ]. The studies were categorised according to whether they included pre-menopausal, post-menopausal women or a mixture, and according to whether the participants had bone metastases (no participants, all participants or a mixture). The studies with a mixture of patient types were excluded, then a linear regression term for "all metastatic patients" or "all post-menopausal patients", compared to none, was included on the log odds ratio or the risk difference.
The meta-regression results for all adverse events were filtered to include only those where the regression effect was statistically significant, and the subgroup-specific treatment effect on adverse event risk for the higher-risk subgroup was practically significant (odds ratio > 1.5 or risk difference > 2%). The remaining results are presented in Table ??? and ???
Meno.
only sig one is Insomnia. big OR for post, but it's not stat sig This is based on small counts
Bone
Note the thing that is significant is the contrast between the two subgroups. Want the contrast to be stat sig, but also the absolute risk for the higher risk subgroup to be prac sig
- Cardiac - bone none has bigger O, but RD not sig
- Dizziness - same
- Fatigue - bone all has bigger OR and RD
- Flu - bone none has bigger RD, but bone all has bigger OR?
- Hypocalcaemia - same
- Increased bone pain - bone none has bigger OR and RD
- Flu-like - bone none has bigger RD, but small counts.
- Neuralgia - small counts.
- Neutropenia - sig RD but not sig OR.
- Osteo - tiny counts
do we want it prac sig for both OR and RD? clearly OR doesn't matter if RD is low. but does RD matter if OR is low? e.g. 70%vs60% is less exciting than 20%vs10% maybe
options(width=150) mares %>% filter(sigmenopre | sigmenopost) %>% select(aetype, nmenopre, nmenopost, datamenopre, datamenopost, orfmenopre, rdfmenopre, orfmenopost, rdfmenopost, meno.orp, meno.rdp) mares %>% filter(sigbonenone | sigboneall) %>% mutate(bone.orp = round(bone.orp,2), bone.rdp=round(bone.rdp,2)) %>% select(aetype, nbonenone, nboneall, databonenone, databoneall, orfbonenone, rdfbonenone, orfboneall, rdfboneall, bone.orp, bone.rdp)
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.