Getting Started with omophub

Introduction

The omophub package provides an R interface to the OMOPHub API for accessing OHDSI ATHENA standardized medical vocabularies. This vignette demonstrates basic usage patterns.

Installation

Install from GitHub:

# install.packages("devtools")
devtools::install_github("omophub/omophub-R")

Authentication

The package requires an API key from OMOPHub.

Set your API key as an environment variable:

Sys.setenv(OMOPHUB_API_KEY = "your_api_key_here")

Or add it to your .Renviron file for persistence:

OMOPHUB_API_KEY=your_api_key_here

Creating a Client

library(omophub)

# Create client (reads API key from environment)
client <- OMOPHubClient$new()

# Or provide API key explicitly
client <- OMOPHubClient$new(api_key = "your_api_key")

# With additional options
client <- OMOPHubClient$new(
  api_key = "your_api_key",
  timeout = 30,
  max_retries = 3,
  vocab_version = "2025.1"
)

Getting Concepts

Retrieve a concept by its OMOP concept ID:

concept <- client$concepts$get(201826)
print(concept$concept_name)
# [1] "Type 2 diabetes mellitus"

Get a concept by vocabulary-specific code:

concept <- client$concepts$get_by_code("SNOMED", "44054006")
print(concept$concept_name)
# [1] "Type 2 diabetes mellitus"

Batch Operations

Retrieve multiple concepts in a single request:

result <- client$concepts$batch(c(201826, 4329847, 1112807))
for (concept in result$concepts) {
  cat(sprintf("%s: %s\n", concept$concept_id, concept$concept_name))
}

Searching Concepts

Basic search:

results <- client$search$basic("diabetes mellitus", page_size = 10)
for (concept in results$data) {
  cat(sprintf("%s: %s\n", concept$concept_id, concept$concept_name))
}

Search with filters:

results <- client$search$basic(
  "heart attack",
  vocabulary_ids = "SNOMED",
  domain_ids = "Condition",
  page_size = 20
)

Semantic Search

Search using natural language queries powered by neural embeddings:

# Natural language search - understands clinical intent
results <- client$search$semantic("high blood sugar levels")
for (r in results$data$results) {
  cat(sprintf("%s (similarity: %.2f)\n", r$concept_name, r$similarity_score))
}

Filter semantic search results:

results <- client$search$semantic(
  "heart attack",
  vocabulary_ids = "SNOMED",
  domain_ids = "Condition",
  threshold = 0.5
)

Fetch all semantic search results with automatic pagination:

all_results <- client$search$semantic_all(
  "chronic kidney disease",
  page_size = 50,
  max_pages = 5,
  progress = TRUE
)
print(nrow(all_results))

Similarity Search

Find concepts similar to a reference concept:

# Find concepts similar to Type 2 diabetes mellitus
similar <- client$search$similar(concept_id = 201826)
for (s in similar$similar_concepts) {
  cat(sprintf("%s (score: %.2f)\n", s$concept_name, s$similarity_score))
}

Search by natural language query with different algorithms:

# Semantic similarity (neural embeddings)
similar <- client$search$similar(
  query = "high blood pressure",
  algorithm = "semantic"
)

# Lexical similarity (string matching)
similar <- client$search$similar(
  query = "high blood pressure",
  algorithm = "lexical"
)

# Hybrid (combined - default)
similar <- client$search$similar(
  query = "high blood pressure",
  algorithm = "hybrid",
  include_scores = TRUE,
  include_explanations = TRUE
)

Bulk Search

Search for multiple queries in a single API call — much faster than individual requests when you have many terms to look up.

Bulk Lexical Search

Execute up to 50 keyword searches at once:

results <- client$search$bulk_basic(list(
  list(search_id = "q1", query = "diabetes mellitus"),
  list(search_id = "q2", query = "hypertension"),
  list(search_id = "q3", query = "aspirin")
), defaults = list(vocabulary_ids = list("SNOMED"), page_size = 5))

# Each result is matched by search_id
for (item in results$results) {
  cat(sprintf("%s: %d results\n", item$search_id, length(item$results)))
}

Bulk Semantic Search

Execute up to 25 natural-language searches using neural embeddings:

results <- client$search$bulk_semantic(list(
  list(search_id = "s1", query = "heart failure treatment options"),
  list(search_id = "s2", query = "type 2 diabetes medication")
), defaults = list(threshold = 0.5, page_size = 10))

for (item in results$results) {
  cat(sprintf("%s: %d results\n", item$search_id,
              item$result_count %||% length(item$results)))
}

Defaults apply to all searches; per-search values override them:

# Different domains per query, shared vocabulary filter
results <- client$search$bulk_basic(list(
  list(search_id = "cond", query = "diabetes", domain_ids = list("Condition")),
  list(search_id = "drug", query = "metformin", domain_ids = list("Drug"))
), defaults = list(vocabulary_ids = list("SNOMED", "RxNorm"), page_size = 5))

Autocomplete

Get suggestions for autocomplete:

suggestions <- client$concepts$suggest("diab", page_size = 5)
for (s in suggestions$suggestions) {
  print(s$suggestion)
}

Pagination

Manual Pagination

# First page
results <- client$search$basic("diabetes", page = 1, page_size = 50)

# Check pagination info
print(results$meta$total_items)
print(results$meta$has_next)

# Get next page if available
if (isTRUE(results$meta$has_next)) {
  page2 <- client$search$basic("diabetes", page = 2, page_size = 50)
}

Automatic Pagination

Fetch all results as a tibble:

all_results <- client$search$basic_all(
  "diabetes",
  page_size = 100,
  max_pages = 5,
  progress = TRUE
)

# Results are a tibble
print(nrow(all_results))
print(names(all_results))

Hierarchy Navigation

Get ancestors (parent concepts):

result <- client$hierarchy$ancestors(201826, max_levels = 3)
for (ancestor in result$ancestors) {
  print(ancestor$concept_name)
}

Get descendants (child concepts):

result <- client$hierarchy$descendants(201826, max_levels = 2)
for (descendant in result$descendants) {
  print(descendant$concept_name)
}

Concept Mappings

Find how a concept maps to other vocabularies:

result <- client$mappings$get(201826)
for (mapping in result$mappings) {
  cat(sprintf("%s: %s\n",
              mapping$target_vocabulary_id,
              mapping$target_concept_name))
}

Map to specific vocabularies:

result <- client$mappings$get(
  201826,
  target_vocabulary = "ICD10CM"
)

Error Handling

Use tryCatch to handle errors:

tryCatch(
  {
    concept <- client$concepts$get(999999999)
  },
  omophub_not_found = function(e) {
    message("Concept not found: ", e$message)
  },
  omophub_api_error = function(e) {
    message("API error: ", e$message)
  }
)

FHIR-to-OMOP Resolution

The FHIR resolver translates FHIR coded values to OMOP standard concepts in a single call — handling URI mapping, code lookup, Maps to traversal, and CDM target table assignment automatically.

Single Coding

result <- client$fhir$resolve(
  system = "http://snomed.info/sct",
  code = "44054006",
  resource_type = "Condition"
)
cat(result$resolution$standard_concept$concept_name)
cat(result$resolution$target_table)   # "condition_occurrence"
cat(result$resolution$mapping_type)   # "direct"

Non-Standard Code (Automatic Maps-to Traversal)

result <- client$fhir$resolve(
  system = "http://hl7.org/fhir/sid/icd-10-cm",
  code = "E11.9"
)
cat(result$resolution$mapping_type)                    # "mapped"
cat(result$resolution$standard_concept$vocabulary_id)  # "SNOMED"

Batch Resolution

batch <- client$fhir$resolve_batch(list(
  list(system = "http://snomed.info/sct", code = "44054006"),
  list(system = "http://loinc.org", code = "2339-0")
))
cat(sprintf("Resolved: %d/%d\n", batch$summary$resolved, batch$summary$total))

CodeableConcept with Vocabulary Preference

result <- client$fhir$resolve_codeable_concept(
  coding = list(
    list(system = "http://snomed.info/sct", code = "44054006"),
    list(system = "http://hl7.org/fhir/sid/icd-10-cm", code = "E11.9")
  ),
  resource_type = "Condition"
)
# SNOMED wins over ICD-10-CM per OHDSI preference
cat(result$best_match$resolution$source_concept$vocabulary_id)  # "SNOMED"

Tibble Output for Batch Resolution

For dplyr / tidyr workflows, pass as_tibble = TRUE to get a flat tibble with one row per input coding and columns for the source and standard concepts, target CDM table, mapping type, and resolution status. This is the most ergonomic shape for ETL pipelines processing many codes:

library(dplyr)

tbl <- client$fhir$resolve_batch(
  list(
    list(system = "http://hl7.org/fhir/sid/icd-10-cm", code = "E11.9"),
    list(system = "http://hl7.org/fhir/sid/icd-10-cm", code = "I10"),
    list(system = "http://hl7.org/fhir/sid/icd-10-cm", code = "J45.909")
  ),
  as_tibble = TRUE
)

tbl |>
  filter(status == "resolved") |>
  select(source_code, standard_concept_name, target_table)

Failed rows are kept in-place with status = "failed" and the error text in status_detail — you can filter them out rather than silently dropping them. The batch summary (total / resolved / failed) is attached as an attribute:

attr(tbl, "summary")

The default as_tibble = FALSE still returns the list-shaped list(results, summary), so existing code keeps working unchanged.

Standalone Wrapper Functions

For pipe-friendly workflows, omophub also exports standalone wrapper functions that take the client as their first argument. Both forms are fully supported — pick whichever reads better for the surrounding code:

# Equivalent to client$fhir$resolve()
client |>
  fhir_resolve(
    system = "http://snomed.info/sct",
    code = "44054006",
    resource_type = "Condition"
  )

# Tibble-shaped batch in a pipe
tbl <- client |>
  fhir_resolve_batch(
    codings = list(
      list(system = "http://snomed.info/sct", code = "44054006"),
      list(system = "http://loinc.org", code = "2339-0")
    ),
    as_tibble = TRUE
  )

client |>
  fhir_resolve_codeable_concept(
    coding = list(
      list(system = "http://snomed.info/sct", code = "44054006"),
      list(system = "http://hl7.org/fhir/sid/icd-10-cm", code = "E11.9")
    ),
    resource_type = "Condition"
  )

FHIR Client Interop with `omophub_fhir_url()`

When you need raw FHIR Parameters / Bundle responses instead of the Concept Resolver envelope, omophub_fhir_url() returns the OMOPHub FHIR Terminology Service base URL so you can talk to it directly with httr2 or fhircrackr. Supports FHIR versions "r4" (default), "r4b", "r5", and "r6":

omophub_fhir_url()
#> "https://fhir.omophub.com/fhir/r4"

omophub_fhir_url("r5")
#> "https://fhir.omophub.com/fhir/r5"

Example: call CodeSystem/$lookup directly with httr2:

library(httr2)

resp <- request(omophub_fhir_url()) |>
  req_url_path_append("CodeSystem/$lookup") |>
  req_url_query(
    system = "http://snomed.info/sct",
    code = "44054006"
  ) |>
  req_headers(Authorization = paste("Bearer", Sys.getenv("OMOPHUB_API_KEY"))) |>
  req_perform()

params <- resp_body_json(resp)
# Raw FHIR Parameters resource with the concept display and designations.

Use client$fhir$resolve() (or fhir_resolve()) when you want OMOP-enriched answers (standard concept, CDM target table, mapping quality). Use omophub_fhir_url() + httr2 when you need FHIR-native responses for FHIR-native tooling.