Clinical data abstraction

Clinical data abstraction that starts from the coded record.

Clinical Extract fills every coded field from the EHR first. Your abstractors spend their hours on the chart review that needs a person.

Diagnosis codes, dates, results, medications and procedures, filled from the EHR with the FHIR source next to each value.

Request a demo See the split

1 Definition

What is clinical data abstraction?

Clinical data abstraction is the work of reading a patient's record and recording defined data elements in a structured form, for a registry, a quality measure or a study. Medical chart abstraction, medical record abstraction and chart abstraction name the same work. Every layout has two kinds of element: the coded ones (diagnosis codes, dates, laboratory results, medications, procedures, staging) that the EHR holds as structured data, and the judgment calls (a date that is a reading of a note, an ambiguous stage, a cause) that take a person.

Coded elements make up most of the fields and most of the lookup time. Clinical Extract exports them from the EHR's certified bulk FHIR API, one column per element with its FHIR source. The abstraction form opens filled, and the abstractor starts at the fields that need a person.

2 The split

Coded fields and judgment calls in one breast case.

Take one breast case abstracted to the NAACCR layout. The diagnosis and its date, the histology, the biomarkers with their LOINC codes, the stage group, the first course of treatment with its RxNorm codes and dates, the procedures. Clinical Extract fills all of these from coded EHR data. The six fields that take a reading of the notes (a date of first contact that depends on which note counts, a stage the pathology and the imaging disagree on) stay with the abstractor, who now starts there.

Coded fields and judgment callsA cancer abstraction form with three columns: the field, the value Clinical Extract pre-fills from coded EHR data (highlighted), and the fields the abstractor decides. Case: Date of 1st Contact [580], abstractor decides which visit came first for this tumor, a biopsy, scan or consult; Class of Case [610], abstractor decides whether the tumor was diagnosed here, treated here, or both. Diagnosis: Date of Diagnosis [390] pre-filled 20250314 from Condition.onsetDateTime; Primary Site [400] C504 and Laterality [410] 2 from ICD-10-CM C50.412; Histologic Type ICD-O-3 [522] 8500 from SNOMED CT 82711006; Sequence Number--Hospital [560], abstractor decides new primary or recurrence by the Solid Tumor Rules. Biomarkers: Estrogen Receptor Summary [3827] 1 from LOINC 16112-5; Progesterone Receptor Summary [3915] 1 from LOINC 16113-3; HER2 Overall Summary [3855] 0 from LOINC 48676-1. Stage: AJCC TNM Path Stage Group [1014] 1A from LOINC 21902-2; AJCC TNM Clin T [1001], abstractor decides when exam and imaging disagree on tumor size. Treatment: RX Date Surgery [1200] 20250402 from Procedure CPT 19301; RX Summ--Hormone [1400] 01 from RxNorm 72965 letrozole; Reason for No Radiation [1430], abstractor reads the radiation oncology consult. Follow-up: Date of Last Contact [1750] 20260811 from Encounter.period.end; Vital Status [1760] 1 from Patient.deceased; Cancer Status [1770], abstractor reads the last oncology note. Twelve fields are pre-filled and six are judgment calls.FIELDNAACCR itemPRE-FILLED BY CLINICAL EXTRACT12 fields, from coded EHR dataABSTRACTOR DECIDES6 judgment calls, from the notesCASEDate of 1st Contact [580]which visit came first for this tumor:biopsy, scan or consultClass of Case [610]diagnosed here, treated here, or bothDIAGNOSISDate of Diagnosis [390]20250314Condition.onsetDateTimePrimary Site [400]C504ICD-10-CM C50.412Laterality [410]2ICD-10-CM C50.412Histologic Type ICD-O-3 [522]8500SNOMED CT 82711006Sequence Number--Hospital [560]new primary or recurrence, by the SolidTumor RulesBIOMARKERSEstrogen Receptor Summary [3827]1LOINC 16112-5Progesterone Receptor Summary [3915]1LOINC 16113-3HER2 Overall Summary [3855]0LOINC 48676-1STAGEAJCC TNM Path Stage Group [1014]1ALOINC 21902-2AJCC TNM Clin T [1001]exam and imaging disagree on tumor sizeTREATMENTRX Date Surgery [1200]20250402Procedure, CPT 19301RX Summ--Hormone [1400]01RxNorm 72965 letrozoleReason for No Radiation [1430]read from the radiation oncologyconsultFOLLOW-UPDate of Last Contact [1750]20260811Encounter.period.endVital Status [1760]1Patient.deceased[x]Cancer Status [1770]read from the last oncology note
Coded fields and judgment callsA cancer abstraction form with three columns: the field, the value Clinical Extract pre-fills from coded EHR data (highlighted), and the fields the abstractor decides. Case: Date of 1st Contact [580], abstractor decides which visit came first for this tumor, a biopsy, scan or consult; Class of Case [610], abstractor decides whether the tumor was diagnosed here, treated here, or both. Diagnosis: Date of Diagnosis [390] pre-filled 20250314 from Condition.onsetDateTime; Primary Site [400] C504 and Laterality [410] 2 from ICD-10-CM C50.412; Histologic Type ICD-O-3 [522] 8500 from SNOMED CT 82711006; Sequence Number--Hospital [560], abstractor decides new primary or recurrence by the Solid Tumor Rules. Biomarkers: Estrogen Receptor Summary [3827] 1 from LOINC 16112-5; Progesterone Receptor Summary [3915] 1 from LOINC 16113-3; HER2 Overall Summary [3855] 0 from LOINC 48676-1. Stage: AJCC TNM Path Stage Group [1014] 1A from LOINC 21902-2; AJCC TNM Clin T [1001], abstractor decides when exam and imaging disagree on tumor size. Treatment: RX Date Surgery [1200] 20250402 from Procedure CPT 19301; RX Summ--Hormone [1400] 01 from RxNorm 72965 letrozole; Reason for No Radiation [1430], abstractor reads the radiation oncology consult. Follow-up: Date of Last Contact [1750] 20260811 from Encounter.period.end; Vital Status [1760] 1 from Patient.deceased; Cancer Status [1770], abstractor reads the last oncology note. Twelve fields are pre-filled and six are judgment calls.PRE-FILLED BY CLINICAL EXTRACT12 fields, from coded EHR dataDate of Diagnosis [390]20250314Condition.onsetDateTimePrimary Site [400]C504ICD-10-CM C50.412Laterality [410]2ICD-10-CM C50.412Histologic Type ICD-O-3 [522]8500SNOMED CT 82711006Estrogen Receptor Summary [3827]1LOINC 16112-5Progesterone Receptor Summary [3915]1LOINC 16113-3HER2 Overall Summary [3855]0LOINC 48676-1AJCC TNM Path Stage Group [1014]1ALOINC 21902-2RX Date Surgery [1200]20250402Procedure, CPT 19301RX Summ--Hormone [1400]01RxNorm 72965 letrozoleDate of Last Contact [1750]20260811Encounter.period.endVital Status [1760]1Patient.deceased[x]ABSTRACTOR DECIDES6 judgment calls, from the notesDate of 1st Contact [580]which visit came first for this tumor:biopsy, scan or consultClass of Case [610]diagnosed here, treated here, or bothSequence Number--Hospital [560]new primary or recurrence, by the SolidTumor RulesAJCC TNM Clin T [1001]exam and imaging disagree on tumor sizeReason for No Radiation [1430]read from the radiation oncology consultCancer Status [1770]read from the last oncology note
Figure 1. Coded fields and judgment calls. Clinical Extract fills every field that coded EHR data answers, with its FHIR source next to the value, and the abstractor starts at the six fields that take a reading of the notes. NAACCR version 25 items; the breast case and its values are illustrative.

3 Minutes per case

Medical chart abstraction with the coded fields pre-filled.

An abstractor's minutes on a case go to finding values, checking them and deciding the calls. With the coded fields pre-filled, there is nothing to find: the values are in the form, each with the FHIR element it came from, and the abstractor's time goes to checking them and making the calls. Figure 2 draws the minutes for the same breast case, before and after.

Minutes per case, before and afterTwo stacked bars on a scale of 0 to 60 minutes, illustrative. Manual lookup, 58 minutes per case: judgment calls 11, finding the case records 9, diagnosis and histology 8, biomarkers 6, stage 7, treatment 12, follow-up 5. With Clinical Extract, 18 minutes per case: the same 11 minutes of judgment calls, then 7 minutes checking the pre-filled fields. The difference, 40 minutes back per case, is highlighted.Manual lookup  58 min per casejudgment calls 11find the records 9diagnosis, histology 8biomarkers 6stage 7treatment 12follow-up 5With Clinical Extract  18 min per casejudgment calls 11check pre-filled fields 740 min back per case0102030405060 min
Minutes per case, before and afterTwo stacked bars on a scale of 0 to 60 minutes, illustrative. Manual lookup, 58 minutes per case: judgment calls 11, finding the case records 9, diagnosis and histology 8, biomarkers 6, stage 7, treatment 12, follow-up 5. With Clinical Extract, 18 minutes per case: the same 11 minutes of judgment calls, then 7 minutes checking the pre-filled fields. The difference, 40 minutes back per case, is highlighted.Manual lookup  58 min per caseWith Clinical Extract  18 min per case40 min back per caseTASKBEFOREAFTERjudgment calls1111find the records90diagnosis, histology80biomarkers60stage70treatment120follow-up50check pre-filled fields07minutes per case58180204060 min
Figure 2 as a table
TaskManual lookup, minWith Clinical Extract, min
Judgment calls1111
Finding the case records90
Diagnosis and histology lookup80
Biomarker lookup60
Stage lookup70
Treatment lookup120
Follow-up lookup50
Checking the pre-filled fields07
Minutes per case5818
Figure 2. Minutes per case, before and after. Clinical Extract pre-fills the coded fields, so the record lookups drop out and the abstractor’s time goes to checking values and making the judgment calls. The minutes are illustrative, for one breast case abstracted to the NAACCR layout.

4 Software, services, or both

Clinical data abstraction software and services.

Clinical data abstraction services send abstractors; clinical data abstraction software gives them a form and a workflow. Clinical Extract feeds both. A service that starts from a filled case bills fewer hours per case; software that imports study.csv opens each case pre-populated, with the dictionary as the field map. Either way the coded fields come from the EHR as recorded, and the abstractor keeps the decisions that need the notes.

5 Quality and registry abstractors

Filling SEP-1, NAACCR and NCDR elements from the EHR.

A sepsis bundle review (SEP-1) turns on times: the lactate result, the antibiotic order, the fluid start, each a timestamped coded element in the EHR, and one judgment call, time zero, which is a reading of the note. A cancer registry case is the NAACCR items; a cardiac registry case is the NCDR elements. For every one, the coded elements arrive as columns with their sources and the abstractor's review starts at the calls. The registry page shows the crosswalk from the EHR to the NAACCR items; running a chart review study shows the same split for a research team.

6 Questions

Questions about abstraction

What does a clinical data abstractor do?

Reads a patient’s record against a defined layout (a registry’s data items, a quality measure’s elements, a study’s variables) and records each element in a structured form: dates, codes, results, and the judgments the layout asks for. With the coded fields filled from the EHR, the abstractor’s hours go to checking values and deciding the ambiguous fields.

What is data abstraction in healthcare?

The step between the record and the report: turning what the chart holds into the defined elements a registry, a measure or a study needs. Most of those elements are already coded in the EHR; the rest take a reading of the notes.

What is medical record abstraction?

Medical record abstraction is reading a patient’s chart and copying the facts a registry or study needs into set fields. Clinical Extract fills the coded fields from the EHR first, so the abstractor reviews them and adds the judgment calls.

Is there an AI tool that reads medical records?

AI tools that read clinical notes exist. Coded data does not need one: diagnosis codes, dates, results, medications and procedures come out of the EHR as they are. Clinical Extract delivers those values as the EHR holds them, with the FHIR source next to each value, and leaves the notes to your abstractors and whatever reads them.

How much time does it save per case?

The record lookups drop out. Figure 2 shows the minutes for one breast case abstracted to the NAACCR layout, before and after; the minutes that remain go to checking values and reading the notes.

Does it replace abstractors?

It fills the coded fields. Abstractors keep the judgment calls, the note review and the sign-off.

Which layouts does it fill?

Any layout whose elements are coded in the EHR: NAACCR data items for cancer registries, NCDR elements for cardiac registries, the timed elements of a quality measure such as SEP-1, a study’s variable list. The data dictionary maps each column to the element it fills.

Next

See Clinical Extract run on one of your studies.

Tell us which EHR you run and what the study or registry needs. We reply within one business day to set a meeting time.

Request a demo