In patelm9/chariot: chariot Package

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

library(chariot)
library(tidyverse)

conn <- connectAthena()

Pivoting the Concept Ancestor table can provide a rich array representing the class hierarchy an OMOP Concept may have.

To exemplify, a test Drug concept is retrieved. Using more than 1 concept can lead to messy results since the classifications between randomly acquired concepts can be vastly different.

test_data <- 
  get_test_drug_concepts(conn = conn,
                         limit = 1)

test_data

The ancestors of this concept are derived.

test_data_ancestors <-
  join_for_ancestors(data = test_data,
                     descendant_id_column = "concept_id",
                     conn = conn)

test_data_ancestors

The descendants are also derived.

test_data_descendants <-
  join_for_descendants(
    data = test_data, 
    ancestor_id_column = "concept_id",
    conn = conn
  )

test_data_descendants

The construct a full concept lineage, the ancestors and descendants are harmonized. Harmonization is achieved by:

1. Renaming the ancestor_ and descendant_ concepts
   to relative_ while maintaining the original identity in a separate column
2. Arranging the ancestors in descending order by levels of separation and the descendants in ascending order
3. Combining the two datasets to create the complete lineage from the topmost ancestor to the bottommost descendant

test_data_ancestors2 <- 
  test_data_ancestors %>%
  rename_all(str_replace_all, "ancestor_", "relative_") %>% 
  mutate(relative_type = "ancestor") %>%
  arrange(desc(min_levels_of_separation))

test_data_descendants2 <-
  test_data_descendants %>%
  rename_all(str_replace_all, "descendant_", "relative_") %>% 
  mutate(relative_type = "descendant") %>%
  arrange(min_levels_of_separation)

test_data_relatives <-
  bind_rows(test_data_ancestors2,
            test_data_descendants2)

test_data_relatives

The preserve the ordering of the lineage, a levels_of_separation field of the factor datatype is created by combining a marker between relative_type and the min_levels_of_separation.

test_data_relatives2 <-
  test_data_relatives %>%
  unite(col = levels_of_separation, 
        relative_type,
        min_levels_of_separation) %>%
  mutate(levels_of_separation = factor(levels_of_separation,
                                       levels = c("ancestor_3", "ancestor_2",
                                                  "ancestor_1", "ancestor_0", 
                                                  "descendant_0", "descendant_1")))
test_data_relatives2$levels_of_separation

To properly pivot the resultset, the source concept and relative concepts require transformation into a single cell per concept. This is achieved by converting concepts into the strip format.

test_data_relatives3 <-
  test_data_relatives2 %>%
  merge_strip(into = "concept") %>%
  merge_strip(into = "relative", 
              prefix = "relative_")

test_data_relatives3

The dataset can now be pivoted by the level of separation, hinging on the source concept.

output <-
  test_data_relatives3 %>%
  pivot_wider(id_col = concept,
              names_from = levels_of_separation, 
              values_from = relative)
output

Since concept has more than 1 relative at any given levels_of_separation, list-cols are returned in the output. To create a more readable output, an aggregate function on relative must be supplied to convert multiple relative values into a string of length 1.

output <-
  test_data_relatives3 %>%
  pivot_wider(id_col = concept,
              names_from = levels_of_separation, 
              values_from = relative,
              #Add aggregate 
              values_fn = list(relative = ~ paste(unique(.), collapse = "|")))
output

The resulting dataset contains a single row for the source concept provided with each relative cast across in the order from topmost ancestor to bottommost descendant aggregated into a pipe-separated string.

dcAthena(conn = conn)

patelm9/chariot documentation built on Feb. 19, 2022, 11:29 a.m.