In softloud/sysrevdata: Methods for designing multifunctional, machine readable systematic review and map databases

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)
library(tippy)

# packages used in this walkthrough
library(sysrevdata)
library(tidyverse)
library(leaflet)

# this avoids tidyverse conflicts with the base function filter
conflicted::conflict_prefer("filter", "dplyr")

complex visualisations and analysis

Along with a narrative synthesis, systematic reviews and maps typically require complex visualisations. In addition, it is common for review authors to upgrade a systematic map into one or more focused systematic reviews, which may involve r tippy("quantitative synthesis", "<span style='font-size:14px;'>e.g. meta-analysis or meta-regression</span>"). Before this can happen, review authors need to convert their database of studies from a condensed or wide format into a format that is ready for analysis. The easiest way to do this is to produce a long or tidy format, one variable per study per row.

This is the ideal way to share data for future syntheses, as it is machine readable for multiple analyses. It also explicitly separates connected data, rather than compressing multiple methods and outcomes onto a single line, losing links between columns. In this long version, each linkage between populations, interventions, comparators, outcomes, study methods, etc. is preserved explicitly on a separate row - a row of independent data.

In this walkthrough, we'll consider the bufferstrips dataset, which starts of as wide-formatted (each level of each variable is presented as a separate column).

# spatial variables
buffer_example %>% 
  select(short_title, contains("spatial"))

wide to long

We want to convert these wide data to long-form data, dropping the r tippy("NA","<span style='font-size:14px;'>'Not applicable, displayed when a cell is empty or blank'</span>") elements and producing a table where each row contains a unique r tippy("level", "<span style='font-size:14px;'>Also referred to as 'code' or 'subcategory''</span>") of each r tippy("factor", "<span style='font-size:14px;'>Also referred to as 'variable' or 'category'</span>") for each study.

We'll use the collection of variables we obtained in the creating a narrative synthesis table vignette.

buffer_variables

Here's one way of transforming the data to long format.

buffer_example_long <-
  buffer_example %>%
  # this function pivots longer
  pivot_longer(
    # see the narrative vignette for how this vector was created
    cols = contains(buffer_variables),
    # name of column we will put the column names of the wide data
    names_to = "category_type",
    # name of column we will put the values of those columns in
    values_to = "subcategory_value",
    # drop the NA values
    values_drop_na = TRUE
  ) %>%

  mutate(
      subcategory_type = map_chr(
    category_type,
    .f = function(x){ifelse(
    str_detect(x, "_"),
    str_match(x, "_(\\w+)") %>% pluck(2),
    NA
  )}),
    category_type = if_else(
    # extract the prefix of the column names with _
    str_detect(category_type, "_"),
    str_extract(category_type, "[a-z]*"),
    category_type
  )
  )

# newly created columns
buffer_example_long %>%
  select(short_title, category_type, subcategory_type, subcategory_value)

usethis::use_data(buffer_example_long)

condensed to long

Suppose, however, that we began with condensed data, as we created in the creating a narrative synthesis table vignette.

condensed_buffer_example

We want to take these data and create the same long-format we have above.

condensed_buffer_example %>% 
  pivot_longer(
    contains(buffer_variables),
    names_to = "category_type",
    values_to = "subcategory_value"
  ) %>% 
  separate(subcategory_value,
           # there are a maximum of 8 different subcategories
           into = letters[1:8],
           sep = "; ") %>% 
  pivot_longer(letters[1:8],
               values_to = "subcategory_value") %>%
  # get rid of nas
  filter(!is.na(subcategory_value)) %>%
  # drop redundant column
  select(-name) %>% 
  # from here is just for display
  select(short_title, category_type, subcategory_value)

summarising

Now we have our data in long-form, we can perform various analyses, including (if we wanted to) meta-analysis on extracted full quantitative data from each study.

We might be interested in the number of countries in the systematic review, in which case we can use the original data where each row is a study (and studies are the independent data needed when we look at countries: each study was conducted in a specific country, so long data aren't necessary yet).

bufferstrips %>% 
  count(study_country)

But to see the number of, say, observations in the farming production system, we will use our long data.

buffer_example_long %>%
  filter(category_type == "farmingproductionsystem") %>%
  count(subcategory_value)

creating evidence atlases

For creating r tippy("evidence atlases", "<span style='font-size:14px;'>Evidence atlases are interactive geographical representations of spatially explicit information, and a common visualisation for systematic maps</span>") we need the data in the wide-format with one row per study. Our bufferstrips dataset is in the wide format already so lets try and reconfigure a wide database from the long formatted data that we created above.

back to wide

back_to_wide <-
  buffer_example_long %>% 
  pivot_wider(
    id_cols = -contains("category"),
    names_from = c(category_type, subcategory_type),
    names_sep = "_",
    values_from = subcategory_value
  ) 

We can use a wide formatted dataset to plot a cartographic map of the study locations for example.

back_to_wide %>% 
  select(short_title,latitute, longitude, google_scholar_link) %>% 
  mutate(lat=as.numeric(latitute)) %>% 
  mutate(lng=as.numeric(longitude)) %>% 
  mutate(tag = paste0("Scholar_link: <a href=", google_scholar_link,">", google_scholar_link, "</a>")) %>% 
  leaflet(width = "100%") %>% 
  addTiles() %>%  # Add default OpenStreetMap map tiles
  addMarkers(lng=~lng, lat=~lat, popup=~tag, clusterOptions = markerClusterOptions())

The code above is just applied to a subset of the data but we can apply the same code to our bufferstrips dataframe.

map<-sysrevdata::bufferstrips %>% 
  select(short_title,latitute, longitude, google_scholar_link) %>% 
  mutate(lat=as.numeric(latitute)) %>% 
  mutate(lng=as.numeric(longitude)) %>% 
  mutate(tag = paste0("Scholar_link: <a href=", google_scholar_link,">", google_scholar_link, "</a>"))

# you might need to tidy up the encoding in the dataframe to get it to work with leaflet 
Encoding(x = map$tag) <- "UTF-8"

# replace all non UTF-8 character strings with an empty space
map$tag <-
  iconv( x = map$tag,
          from = "UTF-8"
         , to = "UTF-8"
         , sub = "" )

map %>% leaflet(width = "100%") %>% 
  addTiles() %>%  # Add default OpenStreetMap map tiles
  addMarkers(lng=~lng, lat=~lat, popup=~tag, clusterOptions = markerClusterOptions())

softloud/sysrevdata documentation built on June 7, 2022, 1:21 p.m.