Skip to content

DATA SERVICES

See a sample review

A synthetic imaging data flow with known planted errors, the checks we run on it, and what they caught. Nothing on this page is from a client.

Everything here is synthetic: a fictitious resident and facility, synthetic NPIs absent from the NPI Registry, generic payers, a fixed report template. No patient data.

Talk about a reviewThe public repository on GitHub

The flow

One exam, six hops. The review joins them on three keys: the resident's MRN, the accession number, and the placer order number.

  1. 1

    Registration

    ADT^A04: who the resident is and what coverage they hold.

  2. 2

    Order

    OMI^O23 (v2.8) or ORM^O01 (v2.5.1): the exam, its side, the accession, the ordering practitioner and the reason.

  3. 3

    Image

    DICOM DX: the header repeats the accession, the patient, the body part and the side.

  4. 4

    Report

    ORU^R01: exam time, report time, final status, the reader.

  5. 5

    FHIR

    ServiceRequest, ImagingStudy and DiagnosticReport as R4 resources, sharing the accession.

  6. 6

    Claim

    Service lines: the exam's technical component, set-up, and the transport fee prorated by patients at the stop.

The messages

One sample for each HL7 step, in the current v2.8 form. The fields the review reads are marked, and listed below each message.

ADT^A04 (v2.8, structure ADT_A01)

MSH|^~\&|RIS|HARBORLINE|PACS|HARBORLINE|20250304081600||ADT^A04^ADT_A01|ADT0000001|P|2.8
EVN||20250304081600
PID|1||MRN1000001^^^HARBORLINE^MR||ASHFORD^AVERY||19420305|F
PV1|1|O|F01^^^HARBORLINE|||||||||||||||||||||||||||||||||||||||||20240115100000
IN1|1|MAA^Medicare Advantage Plan A^L|MAA|Medicare Advantage Plan A||||||||20241101|20251031||MEDICARE_ADVANTAGE||SEL|||||1||||||||||||||MAA12345676
IN1|2|CPA^Commercial Plan A^L|CPA|Commercial Plan A||||||||20241101|20251031||COMMERCIAL||SEL|||||2||||||||||||||CPA123456781
  • PID-3 — the MRN the review joins everything on
  • IN1-2 — the payer
  • IN1-12 — coverage start
  • IN1-13 — coverage end
  • IN1-22 — primary or secondary
  • IN1-36 — the member ID

OMI^O23 imaging order (v2.8)

MSH|^~\&|RIS|HARBORLINE|PACS|HARBORLINE|20250304081600||OMI^O23^OMI_O23|ORD0000001|P|2.8
PID|1||MRN1000001^^^HARBORLINE^MR||ASHFORD^AVERY||19420305|F
PV1|1|O|F01^^^HARBORLINE
ORC|NW|PLC0000001^RIS|FIL0000001^RIS||SC||||20250304081500
PRT|PLC0000001-OP^RIS|AD||OP^Ordering Provider^HL70912|1234567893^BARLOW^BLAKE^^^^^^NPI&2.16.840.1.113883.4.6&ISO^^^^NPI
TQ1|1||||||20250304100000
OBR|1|PLC0000001^RIS|FIL0000001^RIS|73562^Radiologic examination knee 3 views LT^C4|||||||||||||||||||||||||||^Fall from bed with knee pain and swelling
IPC|ACC00000001^RIS|FIL0000001-RP1^RIS|1.2.826.0.1.3680043.8.498.10000000000000000000000000000001^RIS|FIL0000001-SPS1^RIS|DX^Digital Radiography^DCM
  • ORC-1 — new or cancelled
  • ORC-2 — placer order number, joined to the report
  • OBR-4 — the procedure and its side
  • OBR-31 — the statement of the patient's condition
  • PRT-5 — the ordering NPI, matched to the claim
  • IPC-1 — the accession, matched to the image and the claim

ORU^R01 report (v2.8)

MSH|^~\&|PACS|HARBORLINE|RIS|HARBORLINE|20250304144000||ORU^R01^ORU_R01|RPT0000001|P|2.8
PID|1||MRN1000001^^^HARBORLINE^MR||ASHFORD^AVERY||19420305|F
ORC|NW|PLC0000001^RIS|FIL0000001^RIS||CM||||20250304081500
OBR|1|PLC0000001^RIS|FIL0000001^RIS|73562^Radiologic examination knee 3 views LT^C4|||20250304101200|||||||||||||||20250304144000||RAD|F
PRT|ACC00000001-PP^PACS|AD||PP^Primary Performing Provider^HL70912|2522787128^CALDWELL^CAMERON^^^^^^NPI&2.16.840.1.113883.4.6&ISO^^^^NPI
OBX|1|TX|18748-4^Diagnostic imaging study^LN||SYNTHETIC - NOT A READ||||||F|||20250304101200
  • OBR-7 — when the exam was done
  • OBR-22 — when the report was signed
  • OBR-25 — final
  • PRT-5 — the reading radiologist
  • OBX-5 — the report body, always the synthetic template

The image header

The review reads a DICOM header, never the pixels. These are the tags it compares with the order.

TAGNAMEVALUEWHY IT IS READ
(0008,0050)AccessionNumberACC00000001matched to the order (E01, E04)
(0010,0020)PatientIDMRN1000001the MRN
(0010,0010)PatientNameASHFORD^AVERYmust match registration
(0020,000D)StudyInstanceUID1.2.826.0.1.3680043.8.498.10000000000000000000000000000001the study's identity
(0008,0020)StudyDate20250304the date of service
(0008,0060)ModalityDXDX
(0018,0015)BodyPartExaminedKNEEmust be a defined term (E03)
(0020,0060)LateralityLfallback side (E02)
(0020,0062)ImageLateralityLthe side the review compares (E02)
(0018,5101)ViewPositionAPone per instance; the count feeds E11
(0008,0090)ReferringPhysicianNameBARLOW^BLAKEthe orderer
(0008,0070)ManufacturerSYNTHETICSYNTHETIC

The checks

Each check, the error it catches, and why that error costs money.

E01The image carries a different accession number from the order

Catches: A typo or a swapped study number between the order and the image.

Costs: The claim cannot be tied to a valid order, so it is deniable on review.

E02The image's side disagrees with the order

Catches: A left knee ordered, a right knee in the header, or no side at all.

Costs: The LT/RT modifier on the claim no longer matches the record of what was done.

E03The body part is missing or not a defined term

Catches: An empty or non-standard BodyPartExamined tag.

Costs: The exam cannot be matched to its procedure code in a documentation review.

E04An image with no order

Catches: A study that no order explains.

Costs: Work that cannot be billed, or a billed exam with no order behind it.

E05The member ID is wrong

Catches: A wrong payer prefix, a failed check digit, another patient's number.

Costs: The payer rejects the claim.

E06Coverage was not in force on the date of service

Catches: A claim sent to a plan whose period ended before, or started after, the exam.

Costs: The payer denies it; the time to rebill is the cost.

E07The wrong payer

Catches: Fee-for-service billed for an Advantage member; Medicaid billed first for a dual-eligible.

Costs: A denial, and sometimes a timely-filing loss on the correct payer.

E08The ordering NPI is missing or invalid

Catches: A blank, a failed check digit, an organization where an individual is required.

Costs: Medicare requires a valid ordering practitioner on the claim.

E09The report is late, early or missing

Catches: Turnaround past the service level, a report time before the exam, no report at all.

Costs: An exam with no final report is not complete; a late one is a service failure.

E11Views billed differ from views taken

Catches: A two-view code on a one-image study, or the reverse.

Costs: Over-coding is an overpayment to return; under-coding is money left on the table.

E12The transport line disagrees with the stop

Catches: A solo-trip fee at a shared stop, a shared-stop code for one patient, transport on a day with no exam.

Costs: The transport fee is prorated by patients served; the wrong code pays the wrong amount.

What the checks caught

These numbers measure our checks on synthetic data with known planted errors. They are not results from a client.

Seed 20261101, 24 months, 10,108 performed exams, 1,128 planted errors. Recall is the share of planted errors a check caught, written as caught of planted. Hold-out variants were written after the checks were frozen and are counted apart. The clean run, with nothing planted, produced 0 flags.

A planted error often reaches more than one record: shorten a coverage period and every later claim for that payer is outside it; send a claim to the wrong payer and every line of it went there. Flags on those records are counted as cascade flags, not as planted errors. A flag on a record planted with a different error is a cross-flag. A flag on nothing anyone planted is unexplained, and the target for that column is zero.

CHECKCAUGHT OF PLANTEDRECALLHOLD-OUTCASCADE / CROSS FLAGSUNEXPLAINED
E01114 of 1141.00 (0.97 to 1.00)38 of 380 / 10
E0230 of 301.00 (0.89 to 1.00)0 of 90 / 140
E0375 of 751.00 (0.95 to 1.00)24 of 240 / 00
E0459 of 600.98 (0.91 to 1.00)20 of 200 / 00
E0545 of 451.00 (0.92 to 1.00)15 of 151133 / 250
E06103 of 1070.96 (0.91 to 0.98)36 of 36669 / 510
E0775 of 751.00 (0.95 to 1.00)24 of 25198 / 00
E08114 of 1141.00 (0.97 to 1.00)38 of 380 / 00
E0976 of 761.00 (0.95 to 1.00)25 of 250 / 00
E1167 of 680.98 (0.92 to 1.00)33 of 3436 / 20
E1275 of 751.00 (0.95 to 1.00)25 of 250 / 00

These numbers measure our checks on synthetic data with known planted errors. They are not results from a client. The interval is a 95% Wilson interval on the recall.

A sample findings report

The report a review ends with, built from the synthetic run above and marked synthetic on every page. It lists each issue, the records affected, and the fix, and says what could not be checked.

Open the sample findings report (PDF)Read the code and the data on GitHub

All data services