fhir-inconsistencies.md

Finding inconsistencies in FHIR data

A useful usage for FHIR unified schema is search for inconsistencies. As suggested by Qi Li, one example could be find patients with records for diabetes medication, however without diabetes condition.

Another example is search conditions with findings (like diabetes, for instance) and a record for resources denoting "no known problems", e.g., SNOMED CT: "160245001 |No current problems or disability (situation)|", as discussed in HL7 FHIR documentation.

So, in order to deal with them, I setup two SQL KPIs for querying FHIR SQL schema and return resources which match to these kinds of inconsistencies. For input, I manually changed some files which are imported when the container is created.

For diabetes medication inconsistence, I choose a patient without diabetes conditions and insert manually a diabetes medication (file Mikel238_Dickinson688_62644b68-f18b-46e0-86f9-56b3cb2f6737.json). I found the medication code by searching for "snomed ct insulin".

"medicationCodeableConcept": {
  "coding": [
    {
      "system": "http://snomed.info/sct",
      "code": "325072002",
      "display": "Insulin Aspart (substance)"
    }
  ],
  "text": "Lantus 100 unit/ml injectable solution"

For "no known problems" inconsistencies, I changed the file Adina377_Corkery305_cb12851a-2ebd-4c15-88a9-5bee0f308afc.json, and add a "160245001 |No current problems or disability|" record in an pre-existing "190905008" |Cystic Fibrosis|" condition.

"coding": [
  {
    "system": "http://snomed.info/sct",
    "code": "160245001",
    "display": "No current problems or disability"
  },
  {
    "system": "http://snomed.info/sct",
    "code": "190905008",
    "display": "Cystic Fibrosis"
  }
],

Then, the two SQL shown below were setup. The first one gets diabetes medication inconsistencies; the second, "no known problems" inconsistencies:

-- patients with diabetes medication but no diabetes condition
SELECT 
	(
	SELECT GetProp(GetJSON(GetAtJSON(GetJSON(GetJSON(ResourceString, 'medicationCodeableConcept'), 'coding'),0), 'display'), 'display') FROM HSFHIR_I0001_R.Rsrc r where r.Key = m.Key
	) "Medication (ingredient)",
	encounter "Encounter",
	authoredon "Encounter date"
FROM HSFHIR_I0001_S.MedicationRequest m
WHERE 
	-- data inserted for last 7 days
	_lastUpdated BETWEEN DATEADD('dd', -7, CURRENT_TIMESTAMP) and CURRENT_DATE
	-- diabetes medication
	AND $LISTGET(m.code, 1) %inlist $LISTBUILD('96367001','420837001','421367005','420609005','325072002','427292001','414515005','411529005','411530000','412210000','426313003','422346007')
	-- no diabetes condition
	AND 0 = (
		SELECT count(*) FROM HSFHIR_I0001_S.Condition c
		WHERE 
		-- diabetes conditions
		$LISTGET(c.code, 1) %inlist $LISTBUILD('73211009','733072002','530558861000132104','609568004','609569007','105401000119101','199223000','703136005','46635009','44054006','111552007','716362006','123763000','722206009','8801005') 
		AND m.patient = c.patient
	)
  
-- conditions with "no known problems" no properly 
SELECT 
	'',
	encounter "Encounter",
	onsetDate "Encounter date"
FROM HSFHIR_I0001_S.Condition 
WHERE 
	-- data inserted for last 7 days
	_lastUpdated BETWEEN DATEADD('dd', -7, CURRENT_TIMESTAMP) and CURRENT_DATE
	-- 160245001 = no known problems
	AND $FIND(code, 160245001) > 0
	-- more than 1 condition code recorded (each code recorded generates 2 entries into code list - code and its description)
	AND $LISTLENGTH(code) > 2

I found codes for diabetes conditions and medications in the CDC site, here and here.

These SQL feed up the KPIs DiabetesMedicationInconsistence and NoKnowProblemsInconsistence, respectively.

Finally, I setup a dashboard where this information is diplayed. Note that two inconsistencies issue were detected - one for each of them.

You can access this dashboard by these links:

http://localhost:32783/csp/healthshare/fhirserver/_DeepSee.UserPortal.DashboardViewer.zen?DASHBOARD=User/Inconsistencies.dashboard

http://localhost:32783/dsw/index.html#!/d/User/Inconsistencies.dashboard?ns=FHIRSERVER