The basic structure of FHIR R4 JSON

We use jsonlite::fromJSON to import a randomly selected FHIR document from a collection simulated by the MITRE corporation. See the associated site for details.

We’ll drill down through the hierarchy of elements collected in a FHIR document with some base R commands, after importing the JSON.

testf = dir(system.file("json", package="BiocFHIR"), full=TRUE)
tt = fromJSON(testf)
names(tt)

## [1] "resourceType" "type"         "entry"

tt[1:2]

## $resourceType
## [1] "Bundle"
## 
## $type
## [1] "transaction"

tte = tt$entry
class(tte)

## [1] "data.frame"

dim(tte)

## [1] 301   3

head(names(tte))

## [1] "fullUrl"  "resource" "request"

tter = tte$resource
dim(tter)

## [1] 301  72

head(names(tter))

## [1] "resourceType" "id"           "text"         "extension"    "identifier"  
## [6] "name"

table(tter$resourceType)

## 
##   AllergyIntolerance             CarePlan             CareTeam 
##                    8                    3                    3 
##                Claim            Condition     DiagnosticReport 
##                   46                   15                    3 
##            Encounter ExplanationOfBenefit         Immunization 
##                   37                   37                   10 
##    MedicationRequest          Observation         Organization 
##                    9                  114                    3 
##              Patient         Practitioner            Procedure 
##                    1                    3                    9

It is by filtering the data frame tter that we acquire information that may be useful in data analysis. The data frame is sparse: many fields are not used in many records. Code in this package attempts to produce useful tables from the sparse information.

As a prologue to table extraction, we do some basic decomposition of tter using process_fhir_bundle.

bu1 = process_fhir_bundle(testf) # just give file path
bu1

## BiocFHIR FHIR.bundle instance.
##   resource types are:
##    AllergyIntolerance CarePlan ... Patient Procedure

bu1 is just a list of data.frames, but with considerable nesting of data.frames and lists within the basic data.frames corresponding to the major FHIR concepts. “Flattening” of such structures is not fully automatic.

Example: a table on Conditions recorded on the patient.

We use process_Condition to extract information.

cond1 = process_Condition(bu1$Condition)
datatable(cond1)

A family of documents

We have collected 50 documents from the synthea resource. These were obtained using random draws from the 1180 records provided. A temporary folder holding them can be produced as follows:

tset = make_test_json_set()
tset[1]

## [1] "/tmp/RtmpZlfiuD/jsontest/Angel97_Swift555_c072e6ad-b03f-4eee-abe0-2dbc93bbadfe.json"

We import ten documents into a list.

myl = lapply(tset[1:10], process_fhir_bundle)
myl[1:2]

## [[1]]
## BiocFHIR FHIR.bundle instance.
##   resource types are:
##    AllergyIntolerance CarePlan ... Patient Procedure
## 
## [[2]]
## BiocFHIR FHIR.bundle instance.
##   resource types are:
##    CarePlan Claim ... Patient Procedure

sapply(myl,length)

##  [1] 10  9  7  9  9  9  9  9  9 10

We see with the last command that documents can have different numbers of components present.