In chjackson/adverse: Data on adverse events of adjuvant breast cancer treatment

knitr::opts_chunk$set(echo=FALSE,
                      fig.width=10,
                      fig.height=12)
load_all("..")
library(dplyr)
library(ggplot2)
library(plotly)
library(meta)

Meta-analysis of baseline event rates

Include all studies, not just those with complete reporting.

Control groups only, either placebo or observation-only.

Include 2 studies from healthy women (Control arms of the HORIZON, FIT studies) and compare between healthy and cancer

events <- unique(nmadata$aetype)
notevents <- c("ANY ADVERSE EVENT","SERIOUS ADVERSE EVENTS", "Withdrawals because of AEs")
events <- setdiff(events, notevents)
nev <- length(events)

mares <- data.frame(aetype=events)
mares$totncon <- mares$totrcon <-
  mares$p <- mares$pl <- mares$pu <- NA

for (i in 1:nev){ 
    bpma <-
      bpaeriskdiff %>%
      filter(trtcon %in% c("100","101","103")) %>%
      filter(aetype==events[i])
    if (nrow(bpma) > 0){
        mres <- metaprop(rcon, ncon, studlab=`Trial name`, data=bpma, sm="PLN")
        mares[i,"totncon"] <- sum(bpma$ncon) 
        mares[i,"totrcon"] <- sum(bpma$rcon)
        mares[i,c("p","pl","pu")] <- exp(unlist(mres[c("TE.fixed","lower.fixed","upper.fixed")]))
    } else
        mares[i,c("totncon","totrcon")] <- 0
}

hmares <- data.frame(aetype=events)
hmares$totncon <- hmares$totrcon <-
  hmares$p <- hmares$pl <- hmares$pu <- NA

for (i in 1:nev){ 
    bpma <-
      bpaeprevma %>%
      filter(`Trial name` %in% c("FIT","HORIZON")) %>% 
      filter(aetype==events[i])
    if (nrow(bpma) > 0){
        mres <- metaprop(rcon, ncon, studlab=`Trial name`, data=bpma, sm="PLN")
        hmares[i,"totncon"] <- sum(bpma$ncon) 
        hmares[i,"totrcon"] <- sum(bpma$rcon)
        hmares[i,c("p","pl","pu")] <- exp(unlist(mres[c("TE.fixed","lower.fixed","upper.fixed")]))
    } else
        hmares[i,c("totncon","totrcon")] <- 0
}

mares$patient <- "Cancer"
hmares$patient <- "Healthy"

baseres <- rbind(mares, hmares)
baseres <- baseres %>%
  mutate(label = sprintf("\nPooled baseline rate = %s (from total %s/%s)\n",
                         round(p,3), totrcon,totncon))

## Put healthy result and cancer result side by side in long-form data
cp <- baseres$p[baseres$patient=="Cancer"]
names(cp) <- baseres$aetype[baseres$patient=="Cancer"]
baseres$cp <- cp[as.character(baseres$aetype)]
hp <- baseres$p[baseres$patient=="Healthy"]
names(hp) <- baseres$aetype[baseres$patient=="Healthy"]
baseres$hp <- hp[as.character(baseres$aetype)]

p <- baseres %>%
  mutate(aetype = factor(baseres$aetype,
                         levels=unique(baseres$aetype[order(baseres$p)]))) %>%
  filter(!is.na(p)) %>% 
  ggplot(aes(y=p, x=aetype, col=patient, label=label)) + 
  coord_flip(ylim=c(0, 1)) +
  geom_hline(aes(yintercept=1), col="gray") + 
  geom_point(aes(size=totrcon), position=position_dodge(-0.4)) +
  geom_errorbar(aes(ymin=pl, ymax=pu),
                width=0, position=position_dodge(-0.4)) +
  scale_y_continuous(breaks=seq(0, 1, by=0.05)) +
  ylab("Baseline event rate: pooled estimate from meta-analysis") +
  xlab("") 

ggplotly(p)

Events with baseline estimates for both cancer patients and healthy women.

Pooled estimates of baseline event rates from meta-analysis.

pc <- with(mares, sprintf("%s (%s,%s)", round(p,2), round(pl,2), round(pu,2)))
ph <- with(hmares, sprintf("%s (%s,%s)", round(p,2), round(pl,2), round(pu,2)))
pc <- ifelse(is.na(mares$p), NA, pc)
ph <- ifelse(is.na(hmares$p), NA, ph)
dat <- data.frame(Event=as.character(mares$aetype), Cancer=pc, Healthy=ph) 
dat[!is.na(dat$Healthy),]

Seven events where no studies report rates from control group. These shouldn't be relevant, they are rare and there's no evidence of an effect of bisphosphonates.

as.character(mares$aetype[is.na(mares$p)])

chjackson/adverse documentation built on June 16, 2022, 4:53 p.m.