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

knitr::opts_chunk$set(echo=FALSE,
                      fig.width=15,
                      fig.height=15)
library(knitr)
library(dplyr)
filter <- dplyr::filter
library(flextable)
library(rlang)
library(ggplot2)
library(gridExtra)
library(plotly)
setwd("~/work/winton/adverse/vignettes")
load_all("..")
events <- names(models_all)
nev <- length(events)
load(file="ors.rda")

Alkaline phosphatase

arrfn("ALKALINE PHOSPHATASE")

Network geometry summary

Proportion of potential comparisons with direct data: 0.38

Proportion of treatments compared to more than one other: 0.71

Median number of studies per direct comparison: 1 (interquartile range 1-1)

Proportion of direct comparisons informed by more than one study: 0

Meta analysis results

Estimates (and 95% credible intervals) of the odds ratio of each bisphosphonate treatment, versus observation only, under

(green) the best-fitting network meta-analysis model (if feasible)
(blue) an equivalent meta-analysis of direct comparisons of the same treatment versus an observation-only control, and
(red) all study-specific direct comparisons odds ratios corresponding to direct comparisons of the treatment with a no-treatment control. Placebo controls (dotted lines) are distinguished from observation-only controls (solid lines).

The panels distinguish the treatment categories under the best-fitting network meta-analysis model.

Note:

there is one red line for each active study arm in each study containing a placebo or observation-only control. Note that this excludes comparisons of one bisphosphonate against another, which contribute to the network-meta analysis pooled estimates along with the direct comparisons.
the blue line (direct-data pooled estimate) is the weighted average of all solid red lines (study-specific comparisons against observation-only). If there is no blue line on a particular panel, there were no studies that directly compared that treatment against an observation-only control.
the green line (network meta-analysis pooled estimate) will not necessarily appear to be the average of the red lines, since the network meta-analysis also includes indirect data from the bisphosphonate-bisphosponate comparisons that are not illustrated
there is no green line if no network meta-analysis was feasible.

plotarr[[3]]$ests

Comparison of different treatment groupings

Network meta-analysis models defined on different groupings of treatments are compared.

The treatment groupings compared for this symptom are those for which there is a connected network of trials reporting that symptom.
Lines indicated pooled estimates of the odds ratio for each treatment against observation only, under the network meta-analysis model on that treatment grouping.
The statistical fit of the models on different treatment groupings is compared using the deviance information criterion (DIC). DIC are presented relative to the best-fitting model (DIC = 0), and higher DIC indicate worse fit.

For the event ALKALINE PHOSPHATASE, observation-only and placebo are considered separately.