Introduction to Data Visualization for Meta-Analysis

with tidymeta and ggplot2

Malcolm Barrett
Install R: bit.ly/pm605_r
Handout: bit.ly/pm605_tut

04/23/2018

1 / 73

Data Visualization with Rggplot2 and the tidyverse are friendly and consistent tools for data analysis and visualization2 / 73

Data Visualization with Rggplot2 and the tidyverse are friendly and consistent tools for data analysis and visualizationBetter plots are better communication3 / 73

Data Visualization with Rggplot2 and the tidyverse are friendly and consistent tools for data analysis and visualizationBetter plots are better communicationtidymeta makes it easy to manipulate and plot meta-analysis results4 / 73

Introduction to the DataWhat's the impact of intrauterine device (IUD) use on risk of  cervical cancer?5 / 73

Introduction to the DataWhat's the impact of intrauterine device (IUD) use on risk of  cervical cancer?16 studies: 4,945 cases and 7,537 controls6 / 73

Introduction to the DataWhat's the impact of intrauterine device (IUD) use on risk of  cervical cancer?16 studies: 4,945 cases and 7,537 controlsWomen who used IUDs were at a third less risk than those who didn't (OR 0.64)7 / 73

library(tidymeta)
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>

8 / 73

Five variables from iud_cxca we'll usestudy_namelnesselnesgrouppub_year9 / 73

Five variables from iud_cxca we'll usestudy_name = Author + study yearlnesselnesgrouppub_year10 / 73

Five variables from iud_cxca we'll usestudy_namelnes = ln(Odds Ratio)selnesgrouppub_year11 / 73

Five variables from iud_cxca we'll usestudy_namelnesselnes = SE of ln(OR)grouppub_year12 / 73

Five variables from iud_cxca we'll usestudy_namelnesselnesgroup = Study designpub_year13 / 73

Five variables from iud_cxca we'll usestudy_namelnesselnesgrouppub_year = Publication year14 / 73

Meta-Analysis Plot Types15 / 73

Meta-Analysis Plot TypesForest Plot16 / 73

`forest_plot()`

17 / 73

Meta-Analysis Plot TypesForest PlotFunnel Plot18 / 73

`funnel_plot()`

19 / 73

Meta-Analysis Plot TypesForest PlotFunnel PlotInfluence/Sensitivity Plot20 / 73

`influence_plot()`

21 / 73

Meta-Analysis Plot TypesForest PlotFunnel PlotInfluence/Sensitivity PlotCumulative Plot22 / 73

`cumulative_plot()`

23 / 73

A Crash Course in the Tidyverse24 / 73

ggplot2: Elegant Data Visualizations in R25 / 73

ggplot2: Elegant Data Visualizations in RBased on a Grammar of Graphics26 / 73

ggplot2: Elegant Data Visualizations in RBased on a Grammar of GraphicsData is mapped to aesthetics; Statistics and plot are linked27 / 73

ggplot2: Elegant Data Visualizations in RBased on a Grammar of GraphicsData is mapped to aesthetics; Statistics and plot are linkedSensible defaults; Infinitely extensible28 / 73

library(ggplot2)
p <- ggplot(iud_cxca, aes(case_num + control_num, lnes, color = group))
p

29 / 73

library(ggplot2)
p <- p + geom_point()
p

30 / 73

p <- p + geom_smooth(method = "lm", se = FALSE)
p

31 / 73

p + 
  labs(title = "The Effect of Sample Size on Estimate",
      x = "Sample Size",
      y = "ln(Odds Ratio)") + 
  scale_color_discrete(name = "Study Design") +
  theme_minimal() + 
  theme(text = element_text(size = 16))

32 / 73

Tidy Data is Easier to Plot

33 / 73

Tidy Data is Easier to Plot

Each column is a single variable

34 / 73

Tidy Data is Easier to Plot

Each column is a single variable

Each row is a single observation

35 / 73

Tidy Data is Easier to Plot

Each column is a single variable

Each row is a single observation

Each cell is a value

36 / 73

Our Tidy Tools

`%>%`

`mutate()`

`arrange()`

`group_by()`

`tidy()`

37 / 73

Our Tidy Tools

`%>%`: passes the results of one function to the next

`mutate()`

`arrange()`

`group_by()`

`tidy()`

38 / 73

Our Tidy Tools

`%>%`

`mutate()`: changes or creates a new variable

`arrange()`

`group_by()`

`tidy()`

39 / 73

Our Tidy Tools

`%>%`

`mutate()`

`arrange()`: sorts a data set by a variable

`group_by()`

`tidy()`

40 / 73

Our Tidy Tools

`%>%`

`mutate()`

`arrange()`

`group_by()`: groups a data set by a variable

`tidy()`

41 / 73

Our Tidy Tools

`%>%`

`mutate()`

`arrange()`

`group_by()`

`tidy()`: tidies statistical results

42 / 73

Tidy Meta-Analysismeta_analysis()43 / 73

Tidy Meta-Analysis

`meta_analysis()`

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>

44 / 73

Forest Plotforest_plot()45 / 73

Forest Plot

`forest_plot()`

ma %>% 
  forest_plot(group = group)

46 / 73

Forest Plot

`forest_plot()`

ma %>% 
  forest_plot(group = group)

`text_table()`

47 / 73

Forest Plot

`forest_plot()`

ma %>% 
  forest_plot(group = group)

`text_table()`

ma %>% 
  text_table(group = group, "Weights" = weight)

48 / 73

patchwork: Compose ggplots

49 / 73

patchwork: Compose ggplots

Join ggplots quickly and accurately

50 / 73

patchwork: Compose ggplots

Join ggplots quickly and accurately

library(patchwork)
forest_plot() + text_table()

51 / 73

Funnel Plotfunnel_plot()52 / 73

Funnel Plot

`funnel_plot()`

ma %>% 
  funnel_plot(log_summary = TRUE)

53 / 73

Influence Plotsensitivity()54 / 73

Influence Plot

`sensitivity()`

ma %>% 
  sensitivity(exponentiate = TRUE)

55 / 73

Influence Plot

`sensitivity()`

ma %>% 
  sensitivity(exponentiate = TRUE)

`influence_plot()`

56 / 73

Influence Plot

`sensitivity()`

ma %>% 
  sensitivity(exponentiate = TRUE)

`influence_plot()`

ma %>% 
  sensitivity(exponentiate = TRUE) %>% 
  influence_plot()

57 / 73

Cumulative Plotcumulative()58 / 73

Cumulative Plot

`cumulative()`

ma %>% 
  arrange(desc(weight)) %>% 
  cumulative(exponentiate = TRUE)

59 / 73

Cumulative Plot

`cumulative()`

ma %>% 
  arrange(desc(weight)) %>% 
  cumulative(exponentiate = TRUE)

`cumulative_plot()`

60 / 73

Cumulative Plot

`cumulative()`

ma %>% 
  arrange(desc(weight)) %>% 
  cumulative(exponentiate = TRUE)

`cumulative_plot()`

ma %>% 
  arrange(desc(weight)) %>% 
  cumulative(exponentiate = TRUE) %>% 
  cumulative_plot(sum_lines = FALSE)

61 / 73

Importing Stata data, saving ggplots62 / 73

Importing Stata data, saving ggplotshaven: read_dta()63 / 73

Importing Stata data, saving ggplots

haven: `read_dta()`

library(haven)
data <- read_dta("stata_data.dta")

64 / 73

Importing Stata data, saving ggplots

haven: `read_dta()`

library(haven)
data <- read_dta("stata_data.dta")

ggplot2: `ggsave()`

65 / 73

Importing Stata data, saving ggplots

haven: `read_dta()`

library(haven)
data <- read_dta("stata_data.dta")

ggplot2: `ggsave()`

library(ggplot2)
p <- forest_plot(ma, group = group)
ggsave(p, "forest_plot.png", dpi = 320, height = 8)

66 / 73

tidymeta67 / 73

tidymetameta_analysis()/your_favorite_function() + tidy()68 / 73

tidymetameta_analysis()/your_favorite_function() + tidy()forest_plot()/text_table()69 / 73

tidymetameta_analysis()/your_favorite_function() + tidy()forest_plot()/text_table()sensitivity()/influence_plot()70 / 73

tidymetameta_analysis()/your_favorite_function() + tidy()forest_plot()/text_table()sensitivity()/influence_plot()cumulative()/cumulative_plot()71 / 73

Resources

R for Data Science: A comprehensive but friendly introduction to the tidyverse. Free online.

DataCamp: ggplot2 courses and tidyverse courses

ggplot2: Elegant Graphics for Data Analysis: The official ggplot2 book

72 / 73

github.com/malcolmbarrett/tidymeta

github.com/malcolmbarrett/ma_viz_workshop

Slides created via the R package xaringan.

73 / 73

Help

Keyboard shortcuts

↑, ←, Pg Up, k

Go to previous slide

↓, →, Pg Dn, Space, j

Go to next slide

Home

Go to first slide

End

Go to last slide

Number + Return

Go to specific slide

b / m / f

Toggle blackout / mirrored / fullscreen mode

Clone slideshow

Toggle presenter mode

Restart the presentation timer

?, h

Toggle this help

Introduction to Data Visualization for Meta-Analysis

with tidymeta and ggplot2

Malcolm Barrett Install R: bit.ly/pm605_r Handout: bit.ly/pm605_tut

04/23/2018

Data Visualization with R

ggplot2 and the tidyverse are friendly and consistent tools for data analysis and visualization

Data Visualization with R

ggplot2 and the tidyverse are friendly and consistent tools for data analysis and visualization

Better plots are better communication

Data Visualization with R

ggplot2 and the tidyverse are friendly and consistent tools for data analysis and visualization

Better plots are better communication

tidymeta makes it easy to manipulate and plot meta-analysis results

Introduction to the Data

What's the impact of intrauterine device (IUD) use on risk of cervical cancer?

Introduction to the Data

What's the impact of intrauterine device (IUD) use on risk of cervical cancer?

16 studies: 4,945 cases and 7,537 controls

Introduction to the Data

What's the impact of intrauterine device (IUD) use on risk of cervical cancer?

16 studies: 4,945 cases and 7,537 controls

Women who used IUDs were at a third less risk than those who didn't (OR 0.64)

Five variables from iud_cxca we'll use

study_name

lnes

selnes

group

pub_year

Five variables from iud_cxca we'll use

study_name = Author + study year

lnes

selnes

group

pub_year

Five variables from iud_cxca we'll use

study_name

lnes = ln(Odds Ratio)

selnes

group

pub_year

Five variables from iud_cxca we'll use

study_name

lnes

selnes = SE of ln(OR)

group

pub_year

Five variables from iud_cxca we'll use

study_name

lnes

selnes

group = Study design

pub_year

Five variables from iud_cxca we'll use

study_name

lnes

selnes

group

pub_year = Publication year

Meta-Analysis Plot Types

Meta-Analysis Plot Types

Forest Plot

forest_plot()

Meta-Analysis Plot Types

Forest Plot

Funnel Plot

funnel_plot()

Meta-Analysis Plot Types

Forest Plot

Funnel Plot

Influence/Sensitivity Plot

influence_plot()

Meta-Analysis Plot Types

Forest Plot

Funnel Plot

Influence/Sensitivity Plot

Cumulative Plot

cumulative_plot()

A Crash Course in the Tidyverse

ggplot2: Elegant Data Visualizations in R

ggplot2: Elegant Data Visualizations in R

Malcolm Barrett
Install R: bit.ly/pm605_r
Handout: bit.ly/pm605_tut

Five variables from `iud_cxca` we'll use

`study_name`

`lnes`

`selnes`

`group`

`pub_year`

Five variables from `iud_cxca` we'll use

`study_name` = Author + study year

`lnes`

`selnes`

`group`

`pub_year`

Five variables from `iud_cxca` we'll use

`study_name`

`lnes` = ln(Odds Ratio)

`selnes`

`group`

`pub_year`

Five variables from `iud_cxca` we'll use

`study_name`

`lnes`

`selnes` = SE of ln(OR)

`group`

`pub_year`

Five variables from `iud_cxca` we'll use

`study_name`

`lnes`

`selnes`

`group` = Study design

`pub_year`

Five variables from `iud_cxca` we'll use

`study_name`

`lnes`

`selnes`

`group`

`pub_year` = Publication year

`forest_plot()`

`funnel_plot()`

`influence_plot()`

`cumulative_plot()`

`%>%`

`mutate()`

`arrange()`

`group_by()`

`tidy()`

`%>%`: passes the results of one function to the next

`mutate()`

`arrange()`

`group_by()`

`tidy()`

`%>%`

`mutate()`: changes or creates a new variable

`arrange()`

`group_by()`

`tidy()`

`%>%`

`mutate()`

`arrange()`: sorts a data set by a variable

`group_by()`

`tidy()`

`%>%`

`mutate()`

`arrange()`

`group_by()`: groups a data set by a variable

`tidy()`

`%>%`

`mutate()`

`arrange()`

`group_by()`

`tidy()`: tidies statistical results

`meta_analysis()`

`meta_analysis()`

`forest_plot()`

`forest_plot()`

`forest_plot()`

`text_table()`

`forest_plot()`

`text_table()`