Tidy and plot meta-analyses from popular meta-analytic tools in R. Currently in early development.
tidymeta requires the development version of several packages,
including ggplot2, to function correctly. You can install the required
packages for this vignette with the following code:
install.packages(c("devtools", "yaml", "ggrepel"))
library(devtools)
install_github("r-lib/rlang")
install_github("malcolmbarrett/tidymeta")
install_github("malcolmbarrett/mbmisc")
install_github("tidyverse/ggplot2")tidymeta is a toolkit for working with meta-analyses in R. Currently,
it includes a data set, called iud_cxca, for a meta-analysis of the
relationship between IUD use and cervical cancer.
library(tidymeta)
library(ggplot2)
library(dplyr)
library(broom)
iud_cxca
#> # A tibble: 16 x 26
#> study_id study_name author es l95 u95 lnes lnl95 lnu95
#> <int> <chr> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 1 Roura, 2016 Roura 0.600 0.300 1.20 -0.511 -1.20 0.182
#> 2 2 Lassise, 1991 Lassi… 0.800 0.500 1.20 -0.223 -0.693 0.182
#> 3 3 Li, 2000 Li 0.890 0.730 1.08 -0.117 -0.315 0.0770
#> 4 4 Shields, 2004 Shiel… 0.500 0.300 0.820 -0.693 -1.20 -0.198
#> 5 5 Castellsague… Caste… 0.630 0.380 1.06 -0.462 -0.968 0.0583
#> 6 6 Castellsague… Caste… 0.450 0.300 0.670 -0.799 -1.20 -0.400
#> 7 7 Brinton, 1990 Brint… 0.690 0.500 0.900 -0.371 -0.693 -0.105
#> 8 8 Parazzini, 1… Paraz… 0.600 0.300 1.10 -0.511 -1.20 0.0953
#> 9 9 Williams, 19… Willi… 1.00 0.600 1.60 0. -0.511 0.470
#> 10 10 Hammouda, 20… Hammo… 0.300 0.100 0.500 -1.20 -2.30 -0.693
#> 11 11 Castellsague… Caste… 1.08 0.370 3.20 0.0770 -0.994 1.16
#> 12 12 Castellsague… Caste… 0.340 0.0500 2.56 -1.08 -3.00 0.940
#> 13 13 Castellsague… Caste… 0.870 0.340 2.23 -0.139 -1.08 0.802
#> 14 14 Castellsague… Caste… 0.490 0.190 1.23 -0.713 -1.66 0.207
#> 15 15 Castellsague… Caste… 0.240 0.0900 0.660 -1.43 -2.41 -0.416
#> 16 16 Celentano, 1… Celen… 0.500 0.170 1.47 -0.693 -1.77 0.385
#> # ... with 17 more variables: selnes <dbl>, group <fct>, case_num <dbl>,
#> # control_num <dbl>, start_recruit <dbl>, stop_recruit <dbl>,
#> # pub_year <dbl>, numpap <dbl>, ses <dbl>, gravidity <dbl>,
#> # lifetimepart <dbl>, coitarche <dbl>, hpvstatus <dbl>, smoking <dbl>,
#> # location <chr>, aair <dbl>, hpvrate <dbl>tidymeta includes broom methods for cleaning meta-analysis results,
although it currently only supports the metafor package. The tidy()
function in broom puts results into a tidy data frame.
library(metafor)
#> Warning: package 'Matrix' was built under R version 3.4.4
meta4 <- rma(yi = lnes, sei = selnes, data = iud_cxca)
tidy(meta4) %>%
as_tibble() # for space
#> # A tibble: 17 x 8
#> study type estimate std.error statistic p.value conf.low conf.high
#> <chr> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 1 study -0.511 0.354 -1.44 NA -1.20 0.182
#> 2 2 study -0.223 0.223 -0.999 NA -0.661 0.215
#> 3 3 study -0.117 0.0999 -1.17 NA -0.312 0.0793
#> 4 4 study -0.693 0.257 -2.70 NA -1.20 -0.190
#> 5 5 study -0.462 0.262 -1.77 NA -0.975 0.0509
#> 6 6 study -0.799 0.205 -3.90 NA -1.20 -0.397
#> 7 7 study -0.371 0.150 -2.47 NA -0.665 -0.0772
#> 8 8 study -0.511 0.331 -1.54 NA -1.16 0.139
#> 9 9 study 0. 0.250 0. NA -0.490 0.490
#> 10 10 study -1.20 0.411 -2.93 NA -2.01 -0.399
#> 11 11 study 0.0770 0.550 0.140 NA -1.00 1.16
#> 12 12 study -1.08 1.00 -1.07 NA -3.05 0.889
#> 13 13 study -0.139 0.480 -0.290 NA -1.08 0.801
#> 14 14 study -0.713 0.476 -1.50 NA -1.65 0.221
#> 15 15 study -1.43 0.508 -2.81 NA -2.42 -0.431
#> 16 16 study -0.693 0.550 -1.26 NA -1.77 0.385
#> 17 Overall summa… -0.449 0.0941 -4.77 1.83e-6 -0.634 -0.265tidymeta also includes wrapper functions for working with
meta-analysis packages in the context of the tidyverse. The main
function for this is meta_analysis(), which models and tidies the
object, as well as storing the results in the meta column to
facilitate other analysis.
# same as above but stores the meta-analysis object
iud_cxca %>%
meta_analysis(yi = lnes, sei = selnes, slab = study_name)
#> # A tibble: 17 x 10
#> study type estimate std.error statistic p.value conf.low conf.high
#> <chr> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 Roura, … study -0.511 0.354 -1.44 NA -1.20 0.182
#> 2 Lassise… study -0.223 0.223 -0.999 NA -0.661 0.215
#> 3 Li, 2000 study -0.117 0.0999 -1.17 NA -0.312 0.0793
#> 4 Shields… study -0.693 0.257 -2.70 NA -1.20 -0.190
#> 5 Castell… study -0.462 0.262 -1.77 NA -0.975 0.0509
#> 6 Castell… study -0.799 0.205 -3.90 NA -1.20 -0.397
#> 7 Brinton… study -0.371 0.150 -2.47 NA -0.665 -0.0772
#> 8 Parazzi… study -0.511 0.331 -1.54 NA -1.16 0.139
#> 9 William… study 0. 0.250 0. NA -0.490 0.490
#> 10 Hammoud… study -1.20 0.411 -2.93 NA -2.01 -0.399
#> 11 Castell… study 0.0770 0.550 0.140 NA -1.00 1.16
#> 12 Castell… study -1.08 1.00 -1.07 NA -3.05 0.889
#> 13 Castell… study -0.139 0.480 -0.290 NA -1.08 0.801
#> 14 Castell… study -0.713 0.476 -1.50 NA -1.65 0.221
#> 15 Castell… study -1.43 0.508 -2.81 NA -2.42 -0.431
#> 16 Celenta… study -0.693 0.550 -1.26 NA -1.77 0.385
#> 17 Overall summ… -0.449 0.0941 -4.77 1.83e-6 -0.634 -0.265
#> # ... with 2 more variables: meta <list>, weight <dbl>tidymeta doesn’t actually do meta-analysis; it unifies existing tools
with the tidyverse. The benefit of this approach is that you can do
meta-analyses with tidy tools in mind. For example, if you want to
conduct a sub-group analysis, you can use the group_by() function from
dplyr. Here, group is a variable with information about study
design.
ma <- iud_cxca %>%
group_by(group) %>%
meta_analysis(yi = lnes, sei = selnes, slab = study_name, exponentiate = TRUE)
ma
#> # A tibble: 21 x 11
#> group study type estimate std.error statistic p.value conf.low
#> <fct> <chr> <chr> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 Nested … Roura, 2… study 0.600 0.354 -1.44 NA 0.300
#> 2 Nested … Subgroup… summ… 0.600 0.354 -1.44 0.149 0.300
#> 3 Populat… Lassise,… study 0.800 0.223 -0.999 NA 0.516
#> 4 Populat… Li, 2000 study 0.890 0.0999 -1.17 NA 0.732
#> 5 Populat… Shields,… study 0.500 0.257 -2.70 NA 0.302
#> 6 Populat… Castells… study 0.630 0.262 -1.77 NA 0.377
#> 7 Populat… Castells… study 0.450 0.205 -3.90 NA 0.301
#> 8 Populat… Subgroup… summ… 0.655 0.146 -2.90 0.00374 0.492
#> 9 Clinic-… Brinton,… study 0.690 0.150 -2.47 NA 0.514
#> 10 Clinic-… Parazzin… study 0.600 0.331 -1.54 NA 0.313
#> # ... with 11 more rows, and 3 more variables: conf.high <dbl>,
#> # meta <list>, weight <dbl>You can also do sensitivy analyses and cumulative analyses with
sensitivity() and cumulative().
tidymeta includes functionality for working with results in ggplot2,
including meta-analysis specific geoms (such as geom_funnel()) and
quick plots for common visualizations.
With tidy data, most data visualizations for meta-analyses are easy to
build from the ground up. Nevertheless, tidymeta has several quick
plot functions to make the process easier. forest_plot() takes a
tidied meta-analysis and plots the effect sizes.
fp <- ma %>%
forest_plot(group = group)
fp
Because the results are still ggplot2 objects, it’s easy to make
changes to the plot to your liking.
fp <- fp +
scale_x_log() +
theme(axis.text.y = element_text(face = c("bold", rep("plain", 21))))
fp
tidymeta currently supports forest plots, funnel plots, influence
plots, and cumulative plots.