Contents
On this page
- Architecture of the electronic health record
- Interpretation of EHR adoption statistics
- Meaningful use and the gap between capability and use
- Health information exchange as a distinct problem
- Terminologies, common data models and the limits of coverage
- Changes in the clinical note
- Secondary use of EHR data for research
- Study design using EHR data
- Key points
- Application to your study
- Frequently asked questions
Academy · Research Data Analytics & AI · Electronic health records
The Electronic Health Record: Definition, Function and Implications for Research
An electronic health record (EHR) is a clinical workflow and billing system that accumulates data as a by-product. This origin accounts for many of the difficulties researchers encounter: the most complete fields are those linked to reimbursement, clinically important information is often recorded as free text, and “adoption” statistics measure installed capability rather than clinical use.
EHR adoption is now close to universal in the United States: ONC reports that 95% of office-based physicians had adopted any EHR by 2024 and 91% a certified EHR, up from 42% with any EHR in 2008. This trajectory is one of the fastest technology transitions in the history of clinical medicine, and it was substantially policy-driven. For researchers, however, the relevant question is not how many systems are installed but what is reliably recorded in them, by whom, under what incentive and in what form. The answer varies by field, vendor and site, and it determines which research questions a study can address.
Learning objectives
- Describe the architecture of an EHR and explain how it shapes what the data can support
- Read adoption and meaningful-use statistics for what they measure rather than what they suggest
- State which EHR fields are reliable for research and which are artefacts of billing or documentation workflow
- Explain what the common data models and terminologies do, and where their coverage stops
- Assess the five dimensions of EHR data quality before designing an analysis
- Write a data-source methods paragraph that a reviewer can evaluate
Common misconceptions
- “EHR adoption means that clinicians use electronic records for all functions.” Adoption measures installed function. Among hospitals without a basic EHR, physician notes were the function most often not yet adopted, at 61% of those hospitals — and capability has consistently run ahead of use.
- “Meeting meaningful-use criteria means the system is being used meaningfully.” The criteria were a policy construct with defined thresholds. In 2014, 40.5% of hospitals could meet core Stage 2 criteria, up from 5.8% in 2013 — a jump that reflects capability and attestation, not a transformation of clinical behaviour in one year.
- “The record contains the complete clinical picture.” It contains what workflow and reimbursement required someone to enter. Only 65.5% of nursing flowsheet concepts and 56.0% of flowsheet values could be mapped to SNOMED CT and LOINC target codes in one study — much of what nurses record has no standard representation at all.
- “A common data model solves interoperability.” It standardises representation where mappings exist. The OHDSI Standardized Vocabularies comprise over 10 million concepts drawn from 136 source vocabularies, of which 40.5% of valid concepts are standard concepts — so the majority of concepts are not standard ones, and the mapping decisions form part of the study method.
- “Routinely collected data are inexpensive data.” Extraction is inexpensive; validation is not. The costly component of EHR research is establishing that a variable has the intended meaning in a given system, at a given site and in a given period.
Architecture of the electronic health record
An electronic health record comprises three combined systems: a transactional system that supports clinical workflow (orders, results, scheduling, medication administration), a documentation system that captures narrative, and a revenue cycle system that produces codes for billing. The underlying database serves all three, and when they conflict, the revenue cycle usually takes precedence — because that is the component with a direct financial consequence for incomplete entry.
This has a direct and underappreciated consequence for research. The completeness of a field correlates with whether someone is accountable for it. Diagnosis codes attached to billed encounters are near-complete and may be imprecise. Laboratory results arriving by interface are complete and structured. Vital signs are usually structured and frequently missing at the times a clinician did not consider them necessary — which is informative missingness, not random 30,31. Smoking status, functional status, social circumstances and the reasoning behind a decision reside in free text or in fields for which no one is accountable, and their apparent values are driven by documentation behaviour rather than by patient characteristics.
Hripcsak and Albers, in the Journal of the American Medical Informatics Association, set out what next-generation phenotyping from these records requires, and the framing is useful: a phenotype derived from an EHR is a statement about the record-generating process as much as about the patient 31. Weiskopf and Weng, in the same journal, identified five dimensions of EHR data quality — completeness, correctness, concordance, plausibility and currency — alongside seven broad categories of assessment method, with little consistency or potential generalisability across studies 30. That final observation is the key finding: there is no settled standard method for assessing whether EHR data are fit for a given purpose.
Interpretation of EHR adoption statistics
The trajectory is pronounced, and the figures should be cited precisely, because they are frequently misquoted in introductions.
For office-based physicians, ONC reports that 95% had adopted any EHR by 2024 and 91% a certified EHR, up from 42% with any EHR in 2008 2. For hospitals, the picture is best read as a sequence. In 2013, 59% of US non-federal acute care hospitals had adopted at least a basic EHR and 93% possessed certified EHR technology, with comprehensive EHR adoption at 26% — and state rates for at least a basic system ranged from 26% to 83% 22. By the following measurement, 75% of US hospitals had adopted at least a basic EHR, up from that 59% 20.
The policy effect is measurable. Adler-Milstein and Jha, in Health Affairs, found the annual EHR adoption rate among eligible short-term acute care hospitals rose from 3.2% before the HITECH Act to 14.2% by 2015, across more than 4,200 eligible hospitals 16. That is the cleanest available statement of what the incentive programme did.
Three cautions apply to the use of these numbers. The denominators differ — office-based physicians, non-federal acute care hospitals, eligible short-term acute care hospitals are three different populations. “Basic”, “comprehensive” and “certified” are three different constructs: certification is a property of the software, adoption of a defined function set is a property of the site 2,22. Furthermore, the state range of 26% to 83% in a single year should discourage any use of a national figure as a description of a particular setting.
EHR adoption figures, with what each one measures
| Figure | Population | What it measures |
|---|---|---|
| 95% any EHR, 91% certified EHR by 2024, from 42% with any EHR in 2008 | US office-based physicians | Possession of a system, reported by ONC; not a measure of what is recorded in it |
| 59% at least a basic EHR in 2013; 93% possessing certified EHR technology; 26% comprehensive | US non-federal acute care hospitals | Three different constructs. Certification is a property of the software; basic and comprehensive adoption are defined function sets at the site |
| State range 26% to 83% for at least a basic EHR | US non-federal acute care hospitals, same year | The spread, which is why a national figure should never be used to describe a particular setting |
| 75% at least a basic EHR, up from 59% | US hospitals | Adoption of the defined basic function set |
| Annual adoption rate 3.2% pre-HITECH rising to 14.2% by 2015, across more than 4,200 eligible hospitals | Eligible short-term acute care hospitals | Rate of change, which is the cleanest available estimate of the policy effect |
| 42.2% met the Stage 1 meaningful-use proxy in 2012; 5.1% the Stage 2 proxy | 2,796 general acute care hospitals in the 2012 AHA annual survey IT supplement | A proxy measure against policy criteria, not clinical use |
| 40.5% could meet core Stage 2 criteria in 2014, up from 5.8% in 2013 | US hospitals | Capability — note the word “could”. A sevenfold rise in one year reflects configuration and attestation against a deadline |
Office-based physician figures from Barker W, ONC Data Brief No. 84. Hospital 2013 figures from Charles D, Gabriel M, Furukawa MF, ONC. Stage 1 and 2 proxy figures from DesRoches CM, Charles D, Furukawa MF and colleagues, Health Affairs. Basic EHR and Stage 2 capability figures from Adler-Milstein J, DesRoches CM, Kralovec P and colleagues, Health Affairs. Adoption rate change from Adler-Milstein J, Jha AK, Health Affairs.
Office-based physician EHR adoption, from Barker W. Office-based Physician Electronic Health Record Adoption, 2008-2024. ONC Data Brief No. 84, Office of the National Coordinator for Health Information Technology. These figures measure possession of a system, not what is recorded in it.
Meaningful use and the gap between capability and use
The meaningful-use programme defined staged criteria that providers had to meet to receive incentive payments, and the resulting statistics are among the most commonly misread in this field. DesRoches, Charles, Furukawa and colleagues, in Health Affairs, found that 42.2% of US hospitals met the proxy measure for Stage 1 meaningful use in 2012 and 5.1% met the Stage 2 proxy, from the 2012 American Hospital Association annual survey IT supplement covering 2,796 general acute care hospitals 29. Adler-Milstein, DesRoches, Kralovec and colleagues then found 40.5% of hospitals in 2014 could meet core Stage 2 criteria, up from 5.8% in 2013 6,20.
The second pair of figures requires careful reading. The word is “could” — the measure is capability, not performance 20,26. A sevenfold increase in a single year is what a compliance deadline produces, and it describes software configuration and attestation rather than a change in how clinicians work. The same survey found that among hospitals lacking a basic EHR, physician notes were the function most often not yet adopted, at 61% of those hospitals — indicating that the component closest to clinical reasoning was the last to be digitised 20.
How hospitals used the data once they had it is a separate and more informative question. As of 2017, 94% of hospitals used their EHR data to perform hospital processes that inform clinical practice, most often to support quality improvement (82%) and monitor patient safety (81%), with 77% using it to measure organisational performance and 51% to query patient data 13. The gap between the last figure and the first accurately summarises the position of secondary use: nearly all hospitals used the data for institutional purposes, and only about half for querying patient data 13.
What hospitals did with EHR data, as of 2017
| Use | Share of hospitals |
|---|---|
| Any use of EHR data to perform hospital processes that inform clinical practice | 94% |
| Support quality improvement | 82% |
| Monitor patient safety | 81% |
| Measure organisational performance | 77% |
| Query patient data | 51% |
Parasrampuria S, Henry J. Hospitals’ Use of Electronic Health Records Data, 2015-2017. ONC Data Brief No. 46, Office of the National Coordinator for Health Information Technology, U.S. Department of Health and Human Services. The difference between the first and last rows is the informative element: institutional uses are near-universal, patient-data querying roughly half.
Health information exchange as a distinct problem
Having a record is not the same as being able to send it. Operational health information exchange efforts numbered 119 in 2012, up from 75 in 2010, with hospital participation rising from 14% to 30% and ambulatory practice participation from 3% to 10% 6,20,24. More recently, about 70% of non-federal acute care hospitals routinely participate in all four core electronic exchange functions — sending, receiving, finding and integrating patient information.
This last figure must be qualified: participation in an exchange function is not the same as clinicians using outside information in a consultation, and fewer than half of clinicians regularly use outside information during clinical encounters 23,36. That gap between infrastructural capability and clinical use recurs throughout this subject and is the single most reliable pattern in it.
The standards picture has improved in a specific way. Mandel, Kreda, Mandl, Kohane and Ramoni, in the Journal of the American Medical Informatics Association, described SMART on FHIR as a standards-based interoperable apps platform for electronic health records, and the HL7 Fast Healthcare Interoperability Resources standard is now the common substrate; Ayaz and colleagues have reviewed its uptake systematically 1,11,18. The 21st Century Cures Act provisions on information blocking and application programming interfaces changed the regulatory position 4,12. What has not changed is semantic equivalence: an interface can move a value without the receiving system knowing what it means, which is covered in detail on our page about the limitations of digital health records 8,35.
Terminologies, common data models and the limits of coverage
Research use of EHR data almost always involves mapping local representations onto a shared model, and the mapping is a methodological decision rather than a technical preliminary.
The OHDSI Standardized Vocabularies comprise over 10 million concepts drawn from 136 source vocabularies, of which 40.5% of valid concepts are standard concepts, as Reich, Ostropolets, Ryan, Rijnbeek, Schuemie and colleagues describe in the Journal of the American Medical Informatics Association 6,20. The scale is impressive and the proportion is the operative number: most concepts are source concepts that map to standard ones, and the mapping is where information is lost or changed 6,9.
Two findings indicate the range to be expected. Oja, Tamm, Mooses and colleagues converted Estonian national health data to the OMOP common data model and achieved an average mapping rate of 95% across more than 100 million transformed entries — a national-scale dataset, structured, with a mature terminology base 2,9. At the other end, Austin, Lalich, Stewart and colleagues found that only 65.5% of nursing flowsheet concepts and 56.0% of flowsheet values could be mapped to SNOMED CT and LOINC target codes. The difference is not quality of effort; it is that laboratory and administrative data have mature standard representations and nursing observation largely does not.
The practical implication is that the mapping rate should be reported as a result. A study that states “data were mapped to the OMOP common data model” without a rate, and without stating what happened to the unmapped remainder, has concealed a selection step. Our pages on challenges in medical data collection and research data services cover the handling, and data management planning covers documenting it.
Changes in the clinical note
The clinical note contains most of the clinically informative content and is where structured extraction is most difficult. Two bodies of evidence are relevant.
First, note quality appears to improve substantially with time on the system rather than immediately. Burke, Sessums, Hoang, Becher, Fontelo and colleagues, in the Journal of the American Medical Informatics Association, rated outpatient notes with the QNOTE instrument and found mean quality rose from 52.0 before EHR adoption to 61.2 at about six months and 80.4 at five years, with p less than 0.0001 21. The shape of that curve matters for study design: notes from the first year or two after an implementation are not comparable with notes from year five at the same site.
Second, documentation is now being generated differently. Ma, Liang, Shah and colleagues, in the same journal, studied 45 ambulatory physicians over three months and found an ambient AI scribe was used in 9,629 of 17,428 encounters, or 55.25% 3. For a researcher this is a provenance problem as much as a quality one: within a single site and period, roughly half the notes in that cohort were machine-drafted and half were not, and nothing in the record structure necessarily indicates which. Any study using note text from 2023 onwards should establish whether scribe-generated text is present and whether it can be identified.
Natural language processing offers partial access to this content — Wieland-Jorna and colleagues have reviewed systems for extracting information from records — but the output is an estimate with its own error rate, and should be reported as such rather than as a variable 5.
Secondary use of EHR data for research
The appeal of these records for research lies in volume and reach: whole populations, long follow-up, no recruitment. The clearest demonstration of what this enables is Rajkomar, Oren, Chen, Dai and colleagues in npj Digital Medicine, who used data from 216,221 adult hospitalisations of at least 24 hours, unrolled into 46,864,534,945 data points including clinical notes 14. The second figure is the significant one: this is a scale of data that no purpose-built study could assemble, and it enabled prediction across several outcomes from the raw record without manual feature engineering.
Scale does not, however, confer validity. Volume reduces random error and does nothing to systematic error, and the systematic errors in EHR data are structural: informative missingness, differential ascertainment between patients who attend frequently and those who do not, diagnoses recorded for billing rather than clinical accuracy, and site-specific documentation conventions that resemble patient characteristics in a pooled analysis 30,31. Hribar, Read-Brown, Goldstein and colleagues’ work on secondary use is instructive on how much local knowledge is required to use a single specialty’s records correctly 15.
Cheng, Shyr, Lewis and colleagues discuss the opportunities, barriers and requirements for using these data in research, and in practical terms the costly component is never extraction but establishing that a variable means the same thing across the sites, periods and patient groups being compared. Our page on significant medical data collection sources covers the alternatives and complements, and biostatistics support covers the analysis.
Study design using EHR data
Four decisions determine whether an EHR study is credible, and all four belong in the protocol rather than in the discussion.
Define each variable operationally, per site. Which field, which codes, which value ranges, which exclusions. If the definition differs between sites, that is a finding to be reported rather than an inconsistency to be smoothed over.
Validate at least the exposure and outcome against a reference. A chart-review validation on a sample, with sensitivity and positive predictive value reported, converts an assumption into a measured property. This is the step most often omitted and the one reviewers increasingly require.
Characterise the missingness rather than imputing silently. Report what proportion is missing, by group and by period, and state the assumed mechanism. Informative missingness is the default in these data, not the exception.
Report the mapping and the data model. Which model, which version, which vocabulary release, what mapping rate, and what happened to unmapped records.
Wilkinson, Dumontier, Aalbersberg and colleagues’ FAIR guiding principles for scientific data management provide the framework for documentation; our study database build and medical data collection pages cover the practicalities, and our page on the future of medical data collection covers where this is heading 19.
The data source paragraph in the methods of an EHR-based study
Unevaluable by a reviewer
Data were extracted from the electronic health record for all patients seen between the study dates. Diagnoses were identified using relevant ICD codes and laboratory values were obtained from the clinical data warehouse. Patients with missing data were excluded. Data were mapped to a common data model prior to analysis.
“Relevant ICD codes” is not a definition. Without the code list and the fields it was applied to, nobody can reproduce the cohort or judge whether it is the right one.
No validation of exposure or outcome. A coded diagnosis is treated as a clinical fact, when its sensitivity and positive predictive value against chart review are measurable properties that vary by condition.
“Patients with missing data were excluded” selects on care intensity. In EHR data, having a value recorded means someone decided the test was needed; excluding incomplete records therefore selects sicker or more closely followed patients.
“Mapped to a common data model” with no rate. The mapping is a selection step, and the proportion that failed to map and what became of it is part of the result.
No site-level detail in a multi-site study. With different vendors, the same named field can carry different content, which appears in a pooled analysis as a patient characteristic.
States what the variables mean and how well
Data were extracted from the electronic health records of the three participating sites, which run two different vendor systems; variable definitions are given per site in Appendix 1 because the structured fields are not equivalent across them. The exposure was defined by the code list in Appendix 2, applied to coded problem-list and encounter-diagnosis fields but not to free text; we validated this definition against blinded chart review on a random sample, and report its sensitivity and positive predictive value in Table 2, because a coded diagnosis is a billing artefact as much as a clinical statement. The outcome was taken from interfaced laboratory results, which are structured and effectively complete for tests that were performed — the limitation being that a test is performed when a clinician considers it necessary, so missingness here is informative and we model it as such rather than excluding incomplete records, which would select on care intensity. Missingness is reported by variable, by site and by study period in Table 3. Data were mapped to the OMOP common data model, with the model version and vocabulary release stated; the achieved mapping rate and the disposition of unmapped records are reported, and the nursing-recorded variables were excluded in advance because flowsheet content has poor standard-terminology coverage. Notes were not used, in part because the study period spans the introduction of ambient documentation tools at one site, and we could not reliably identify which notes were machine-drafted.
Variables are defined per site, with the non-equivalence treated as a reportable feature rather than smoothed over.
The exposure definition is validated against chart review with sensitivity and predictive value reported, converting an assumption into a measured property.
Missingness is characterised and modelled rather than removed, with the informative mechanism stated explicitly.
The mapping rate and the disposition of unmapped records are reported, so the selection introduced by the data model is visible.
A variable class with known poor terminology coverage is excluded in advance and the reason given, which is stronger than discovering it in sensitivity analysis.
The note-provenance problem created by ambient documentation tools is identified and handled by exclusion, with the reasoning stated.
A reviewer assessing EHR research asks one question: whether these variables mean what the authors assume, consistently across the comparison being made. The first version provides no means of answering it; the second answers it in advance and declares where it could not.
Key points
Key points
An EHR is a clinical workflow and billing system that accumulates data, and the completeness of any field reflects whether someone was accountable for entering it. Adoption statistics should be read for what they measure: ONC reports 95% of US office-based physicians with any EHR by 2024 and 91% with a certified EHR, up from 42% in 2008, while hospital figures describe different constructs for different populations — 59% with at least a basic EHR in 2013 against 93% possessing certified technology, state rates ranging from 26% to 83%, and an annual adoption rate that rose from 3.2% before HITECH to 14.2% by 2015. Meaningful-use figures measure capability, not behaviour: 40.5% of hospitals could meet core Stage 2 criteria in 2014 against 5.8% a year earlier. The recurring pattern is capability outpacing use — 94% of hospitals used EHR data for institutional processes in 2017 but only 51% to query patient data, and about 70% of hospitals now participate in all four core exchange functions while fewer than half of clinicians regularly use outside information in a consultation. For research, variable definitions should be reported per site, exposure and outcome validated against chart review, missingness characterised as informative rather than excluded, and the mapping rate stated.
Application to your study
Application to your study
- Researchers should verify which construct an adoption figure measures. Certified, basic, comprehensive and meaningful-use capability are four different constructs with four different denominators. The figure cited should match the claim being made.
- Variables should be ranked by accountability for data entry. Billed codes and interfaced results are near-complete; nurse-recorded and social variables reflect documentation behaviour. Studies should be built on the former, and the latter treated as uncertain.
- Exposure and outcome should be validated on a sample. Blinded chart review with sensitivity and positive predictive value reported. This is the step reviewers increasingly require and the one most protocols omit.
- Records should not be excluded on the basis of missingness without a stated missingness mechanism. A recorded value usually means a clinician thought the test was needed, so complete-case analysis selects on care intensity.
- The mapping rate and the handling of unmapped records should be reported. The common data model version and vocabulary release should be named, since both affect what maps.
- In multi-site studies, variables should be defined per site. Different vendors produce different content in identically named fields, and in a pooled analysis that difference masquerades as a patient characteristic.
- Researchers should establish whether ambient documentation tools were in use during the period from which notes were drawn. If notes form part of the data and machine-drafted text cannot be identified, this should be stated.
Frequently asked questions
Architecturally an EHR is three systems sharing a database: a transactional system supporting clinical workflow such as orders, results and medication administration; a documentation system holding narrative; and a revenue cycle system producing billing codes. The distinction from an electronic medical record is usually framed in terms of scope — a record intended to follow a patient across organisations rather than within one — but for research purposes the more useful distinction is the one above, because it predicts data quality. Fields with a financial or workflow consequence for incomplete entry are near-complete; fields with no accountable owner reflect documentation behaviour rather than patient characteristics.
ONC reports that 95% of US office-based physicians had adopted any EHR by 2024 and 91% a certified EHR, against 42% with any EHR in 2008. For hospitals the figures depend on the construct: in 2013, 59% of US non-federal acute care hospitals had adopted at least a basic EHR while 93% possessed certified EHR technology and 26% had a comprehensive system, with state rates for at least a basic system ranging from 26% to 83%. A later measurement put at least basic adoption at 75%. Adler-Milstein and Jha found the annual adoption rate among eligible short-term acute care hospitals rose from 3.2% before HITECH to 14.2% by 2015 across more than 4,200 hospitals.
No. The meaningful-use measures record capability against policy criteria, and the increase in those figures tracks compliance deadlines rather than clinical behaviour: 40.5% of hospitals could meet core Stage 2 criteria in 2014, up from 5.8% in 2013, and DesRoches and colleagues found 42.2% meeting a Stage 1 proxy and 5.1% a Stage 2 proxy in 2012 across 2,796 hospitals in the AHA IT supplement. A sevenfold single-year change in capability reflects software configuration and attestation. It indicates that certain functions exist and certain fields are populated to a threshold, which is weak evidence that any particular variable is accurate for a given research question.
Fields should be ranked by accountability. Interfaced laboratory results are structured and effectively complete for tests that were performed. Billed diagnosis and procedure codes are near-complete and may be clinically imprecise, since they serve reimbursement. Medication orders and administrations are generally reliable within the system that issued them. Vital signs are structured and missing non-randomly. Smoking, functional and social variables, and the reasoning behind decisions, reside in free text or unowned fields and reflect documentation practice. One concrete indicator of the limits: only 65.5% of nursing flowsheet concepts and 56.0% of flowsheet values could be mapped to SNOMED CT and LOINC target codes in one study, so much of nursing observation has no standard representation at all.
It standardises representation where mappings exist, which is a real and substantial benefit but not the same as solving interoperability. The OHDSI Standardized Vocabularies comprise over 10 million concepts from 136 source vocabularies, of which 40.5% of valid concepts are standard concepts — so most concepts are source concepts requiring a mapping, and each mapping is a decision that can lose or change meaning. Achievable rates vary enormously with data type: Oja and colleagues reported an average 95% mapping rate converting Estonian national health data to OMOP across more than 100 million entries, while nursing flowsheet content maps at around two-thirds of concepts. The mapping rate and the disposition of the remainder should be reported.
The available evidence suggests improvement over years rather than immediately. Burke, Sessums, Hoang, Becher, Fontelo and colleagues rated outpatient notes using the QNOTE instrument and found mean quality rose from 52.0 before EHR adoption to 61.2 at about six months and 80.4 at five years, with p less than 0.0001. For study design the shape of that curve matters more than the endpoint: notes from the first year or two after implementation are not comparable with notes from year five at the same site, so a study spanning an implementation has a documentation-quality trend that can masquerade as a clinical one.
Yes, as a provenance problem. Ma, Liang, Shah and colleagues studied 45 ambulatory physicians over three months and found an ambient AI scribe used in 9,629 of 17,428 encounters, or 55.25%. In that cohort roughly half the notes were machine-drafted and half were not, within the same site and period, and nothing in the record structure necessarily distinguishes them. Any study using note text from recent years should establish whether such tools were in use, whether affected notes can be identified, and whether the mix changed over the study period — because a change in drafting method will alter the measured properties of text without any change in clinical practice.
Treating extraction as the difficult part. Extraction is inexpensive; establishing that a variable means the same thing across the sites, periods and patient groups being compared is costly, and it is the step that determines whether the result is meaningful. Three specific failures follow from omitting it: using a coded diagnosis as a clinical fact without validating it against chart review, excluding records with missing values when a recorded value implies a clinician thought the test necessary, and pooling multi-site data without site-level variable definitions when different vendor systems put different content in identically named fields. Weiskopf and Weng identified five data quality dimensions — completeness, correctness, concordance, plausibility and currency — and found little consistency in how any of them are assessed.
References
- Office of the National Coordinator for Health Information Technology. Health Level 7 (HL7) Fast Healthcare Interoperability Resources (FHIR). HealthIT.gov, Office of the National Coordinator for Health Information Technology. 2026. read 9 October 2026. healthit.gov
- Barker W. Office-based Physician Electronic Health Record Adoption, 2008-2024 (ONC Data Brief No. 84). Office of the National Coordinator for Health Information Technology. 2026. read 9 October 2026. healthit.gov
- Ma SP, Liang AS, Shah SJ, Smith M, Jeong Y, Devon-Sand A, Crowell T, Delahaie C, Hsia C, Lin S, Shanafelt T, Pfeffer MA, Sharp C, Garcia P. Ambient artificial intelligence scribes: utilization and impact on documentation time. Journal of the American Medical Informatics Association. 2025;32(2):381-385. doi:10.1093/jamia/ocae304 read 9 October 2026. doi.org/10.1093/jamia/ocae304
- Cheng AC, Shyr C, Lewis A, Delacqua F, Bosler T, Banasiewicz MK, Taylor R, Lindsell CJ, Harris PA. Opportunities, barriers, and remedies for implementing REDCap integration with electronic health records via Fast Healthcare Interoperability Resources (FHIR). JAMIA Open. 2025;8(5):ooaf111. doi:10.1093/jamiaopen/ooaf111 read 9 October 2026. doi.org/10.1093/jamiaopen/ooaf111
- Wieland-Jorna Y, van Kooten D, Verheij RA, de Man Y, Francke AL, Oosterveld-Vlug MG. Natural language processing systems for extracting information from electronic health records about activities of daily living. A systematic review. JAMIA Open. 2024;7(2):ooae044. doi:10.1093/jamiaopen/ooae044 read 9 October 2026. doi.org/10.1093/jamiaopen/ooae044
- Reich C, Ostropolets A, Ryan P, Rijnbeek P, Schuemie M, Davydov A, Dymshyts D, Hripcsak G. OHDSI Standardized Vocabularies-a large-scale centralized reference ontology for international data harmonization. Journal of the American Medical Informatics Association. 2024;31(3):583-590. doi:10.1093/jamia/ocad247 read 9 October 2026. doi.org/10.1093/jamia/ocad247
- Parker G. Future of Electronic Health Records: A Challenge to Maximize Their Utility. SSRN Electronic Journal. 2023. doi:10.2139/ssrn.4457214 read 9 October 2026. doi.org/10.2139/ssrn.4457214
- Walker DM, Tarver WL, Jonnalagadda P, Ranbom L, Ford EW, Rahurkar S. Perspectives on Challenges and Opportunities for Interoperability: Findings From Key Informant Interviews With Stakeholders in Ohio. JMIR Medical Informatics. 2023;11(1):e43848. doi:10.2196/43848 read 9 October 2026. doi.org/10.2196/43848
- Oja M, Tamm S, Mooses K, Pajusalu M, Talvik HA, Ott A, Laht M, Malk M, et al. Transforming Estonian health data to the Observational Medical Outcomes Partnership (OMOP) Common Data Model: lessons learned. JAMIA Open. 2023;6(4):ooad100. doi:10.1093/jamiaopen/ooad100 read 9 October 2026. doi.org/10.1093/jamiaopen/ooad100
- Li E, Clarke J, Ashrafian H, Darzi A, Neves AL. The Impact of Electronic Health Record Interoperability on Safety and Quality of Care in High-Income Countries: Systematic Review. Journal of Medical Internet Research. 2022;24(9):e38144. doi:10.2196/38144 read 9 October 2026. doi.org/10.2196/38144
- Ayaz M, Pasha MF, Alzahrani MY, Budiarto R, Stiawan D. The Fast Health Interoperability Resources (FHIR) Standard: Systematic Literature Review of Implementations, Applications, Challenges and Opportunities. JMIR Medical Informatics. 2021;9(7):e21929. doi:10.2196/21929 read 9 October 2026. doi.org/10.2196/21929
- Office of the National Coordinator for Health Information Technology, U.S. Department of Health and Human Services. 21st Century Cures Act: Interoperability, Information Blocking, and the ONC Health IT Certification Program (final rule). Federal Register. 2020;85:25642-25961. read 9 October 2026. federalregister.gov
- Parasrampuria S, Henry J. Hospitals' Use of Electronic Health Records Data, 2015-2017 (ONC Data Brief No. 46). Office of the National Coordinator for Health Information Technology (ONC), U.S. Department of Health and Human Services. 2019. read 9 October 2026. healthit.gov
- Rajkomar A, Oren E, Chen K, Dai AM, Hajaj N, Hardt M, Liu PJ, Liu X, Marcus J, Sun M, Sundberg P, Yee H, Zhang K, Zhang Y, Flores G, Duggan GE, Irvine J, Le Q, Litsch K, Mossin A, Tansuwan J, Wang D, Wexler J, Wilson J, Ludwig D, Volchenboum SL, Chou K, Pearson M, Madabushi S, Shah NH, Butte AJ, Howell MD, Cui C, Corrado GS, Dean J. Scalable and accurate deep learning with electronic health records. npj Digital Medicine. 2018;1(1):18. doi:10.1038/s41746-018-0029-1 read 9 October 2026. doi.org/10.1038/s41746-018-0029-1
- Hribar MR, Read-Brown S, Goldstein IH, Reznick LG, Lombardi L, Parikh M, Chamberlain W, Chiang MF. Secondary use of electronic health record data for clinical workflow analysis. Journal of the American Medical Informatics Association. 2018;25(1):40-46. doi:10.1093/jamia/ocx098 read 9 October 2026. doi.org/10.1093/jamia/ocx098
- Adler-Milstein J, Jha AK. HITECH Act Drove Large Gains In Hospital Electronic Health Record Adoption. Health Affairs. 2017;36(8):1416-1422. doi:10.1377/hlthaff.2016.1651 read 9 October 2026. doi.org/10.1377/hlthaff.2016.1651
- Sinsky CA, Colligan L, Li L, et al. Allocation of Physician Time in Ambulatory Practice: A Time and Motion Study in 4 Specialties. Annals of Internal Medicine. 2016;165(11):753-760. doi:10.7326/M16-0961 read 9 October 2026. doi.org/10.7326/M16-0961
- Mandel JC, Kreda DA, Mandl KD, Kohane IS, Ramoni RB. SMART on FHIR: a standards-based, interoperable apps platform for electronic health records. Journal of the American Medical Informatics Association. 2016;23(5):899-908. doi:10.1093/jamia/ocv189 read 9 October 2026. doi.org/10.1093/jamia/ocv189
- Wilkinson MD, Dumontier M, Aalbersberg IJ, et al. (53 authors; Mons B last). The FAIR Guiding Principles for scientific data management and stewardship. Scientific Data. 2016;3:160018. doi:10.1038/sdata.2016.18 read 9 October 2026. doi.org/10.1038/sdata.2016.18
- Adler-Milstein J, DesRoches CM, Kralovec P, Foster G, Worzala C, Charles D, Searcy T, Jha AK. Electronic Health Record Adoption in US Hospitals: Progress Continues, But Challenges Persist. Health Affairs. 2015. doi:10.1377/hlthaff.2015.0992 read 9 October 2026. doi.org/10.1377/hlthaff.2015.0992
- Burke HB, Sessums LL, Hoang A, Becher DA, Fontelo P, Liu F, Stephens M, Pangaro LN, et al. Electronic health records improve clinical note quality. Journal of the American Medical Informatics Association. 2015;22(1):199-205. doi:10.1136/amiajnl-2014-002726 read 9 October 2026. doi.org/10.1136/amiajnl-2014-002726
- Charles D, Gabriel M, Furukawa MF. Adoption of Electronic Health Record Systems among U.S. Non-federal Acute Care Hospitals: 2008-2013 (ONC Data Brief No. 16). Office of the National Coordinator for Health Information Technology (ONC), U.S. Department of Health and Human Services. 2014. read 9 October 2026. healthit.gov
- Furukawa MF, King J, Patel V, Hsiao CJ, Adler-Milstein J, Jha AK. Despite Substantial Progress In EHR Adoption, Health Information Exchange And Patient Engagement Remain Low In Office Settings. Health Affairs (Millwood). 2014;33(9):1672-1679. read 9 October 2026. help.senate.gov
- Adler-Milstein J, Jha A. Health information exchange among U.S. hospitals: who's in, who's out and why? Healthcare. 2014;2(1):26-32. read 9 October 2026. help.senate.gov
- Adler-Milstein J, Scott KW, Jha AK. Leveraging EHRs to Improve Hospital Performance: The Role of Management. American Journal of Managed Care. 2014;20(11 Spec No. 17):SP511-SP519. read 9 October 2026. help.senate.gov
- Adler-Milstein J, DesRoches CM, Furukawa MF, Worzala C, Charles D, Kralovec P, Stalley S, Jha A. More than half of US hospitals have at least a basic EHR, but stage 2 criteria remain challenging for most. Health Affairs (Millwood). 2014;33(9):1664-71. read 9 October 2026. help.senate.gov
- Adler-Milstein J, Jha AK. No Evidence That Hospitals Are Using New Electronic Health Records to Increase Medicare Reimbursements. Health Affairs. 2014;33(7):1271. doi:10.1377/hlthaff.2014.0023 read 9 October 2026. doi.org/10.1377/hlthaff.2014.0023
- Adler-Milstein J, Green CE, Bates DW. A Survey Analysis Suggests That Electronic Health Records Will Yield Revenue Gains For Some Practices And Losses For Many. Health Affairs (Millwood). 2013;32(3):562-70. read 9 October 2026. help.senate.gov
- DesRoches CM, Charles D, Furukawa MF, et al. Adoption of Electronic Health Records Grows Rapidly, But Fewer Than Half of US Hospitals Had At Least A Basic System in 2012. Health Affairs (Millwood). 2013;32(8). read 9 October 2026. healthaffairs.org
- Weiskopf NG, Weng C. Methods and dimensions of electronic health record data quality assessment: enabling reuse for clinical research. Journal of the American Medical Informatics Association. 2013;20(1):144-151. doi:10.1136/amiajnl-2011-000681 read 9 October 2026. doi.org/10.1136/amiajnl-2011-000681
- Hripcsak G, Albers DJ. Next-generation phenotyping of electronic health records. Journal of the American Medical Informatics Association. 2013;20(1):117-121. doi:10.1136/amiajnl-2012-001145 read 9 October 2026. doi.org/10.1136/amiajnl-2012-001145
- DesRoches CM, Campbell EG, Rao SR, Donelan K, Ferris T, Jha A, Kaushal R, Levy D, Rosenbaum S, Shields A, Blumenthal D. Electronic health records in ambulatory care - a national survey of physicians. New England Journal of Medicine. 2008;359(1):50-60. read 9 October 2026. help.senate.gov
- Ammenwerth E, Schnell-Inderst P, Machan C, Siebert U. The effect of electronic prescribing on medication errors and adverse drug events: a systematic review. Journal of the American Medical Informatics Association. 2008;15(5):585-600. read 9 October 2026. help.senate.gov
- Kuraitis V, McNicoll I. read 9 October 2026. e-caremanagement.com
- Journal of the American Medical Informatics Association. doi:10.1093/jamia/ocab289 read 9 October 2026. doi.org/10.1093/jamia/ocab289
- Assistant Secretary for Technology Policy (ASTP). Interoperable Exchange of Patient Health Information Among US Hospitals, 2023. read 9 October 2026. techtarget.com
Every source above is a peer-reviewed article, a publisher’s record of one, or a standard issued by the body that owns it. Each carries the date it was last read. Where a standard moves, this article is revised rather than left up with a note.
How to cite this article
Pick your referencing style and copy it. Your university or target journal may vary the punctuation — check its guide if it has one.
Pubrica Academy. (2026, October 9). The Electronic Health Record: Definition, Function and Implications for Research. Pubrica. https://pubrica.com/academy/concepts-definitions/electronic-health-records/
Pubrica Academy. “The Electronic Health Record: Definition, Function and Implications for Research.” Pubrica, 9 Oct. 2026, pubrica.com/academy/concepts-definitions/electronic-health-records/.
Pubrica Academy (2026) The Electronic Health Record: Definition, Function and Implications for Research. Pubrica. Available at: https://pubrica.com/academy/concepts-definitions/electronic-health-records/ (Accessed: 15 October 2026).
Pubrica Academy. “The Electronic Health Record: Definition, Function and Implications for Research.” Pubrica. October 9, 2026. https://pubrica.com/academy/concepts-definitions/electronic-health-records/
Pubrica Academy, ‘The Electronic Health Record: Definition, Function and Implications for Research’ (Pubrica, 9 October 2026) <https://pubrica.com/academy/concepts-definitions/electronic-health-records/> accessed 15 October 2026.
Pubrica Academy. The Electronic Health Record: Definition, Function and Implications for Research. [Internet]. Pubrica; 2026 Oct 9 [cited 2026 Oct 15]. Available from: https://pubrica.com/academy/concepts-definitions/electronic-health-records/
Pubrica Academy, “The Electronic Health Record: Definition, Function and Implications for Research,” Pubrica, Oct. 9, 2026. [Online]. Available: https://pubrica.com/academy/concepts-definitions/electronic-health-records/
@misc{electronic_health_records_2026,
author = {{Pubrica Academy}},
title = {The Electronic Health Record: Definition, Function and Implications for Research},
howpublished = {Pubrica},
year = {2026},
month = {oct},
url = {https://pubrica.com/academy/concepts-definitions/electronic-health-records/},
note = {Accessed: <span data-cite-acc="iso">2026-10-15</span>}
}This article is written to be learned from. Cite it if you are referring to what it says; if you are using the evidence behind it, cite the sources listed above instead.
EHR data source specification for a protocol
This specification should be completed before extraction, and the completed version incorporated into the methods section. A methods reviewer is likely to examine the validation and missingness sections first.
EHR DATA SOURCE SPECIFICATION
1. SYSTEMS AND SITES
Sites included ...................: _______________________
Vendor system at each site .......: _______________________
[ ] Structured fields are equivalent across sites
[ ] They are NOT - variables defined per site in Appendix __
Study period .....................: _______________________
Extraction date ..................: _______________________
Any system upgrade or migration within the period: ________
2. VARIABLE DEFINITIONS (one block per key variable)
Variable .........................: _______________________
Field(s) read ..................: _______________________
Code list / value set ..........: Appendix ____
Free text used? [ ] no [ ] yes - NLP method: _________
Accountability tier:
[ ] billed / coded (near-complete, may be imprecise)
[ ] interfaced result (complete where performed)
[ ] clinician-entered structured (variable)
[ ] nurse-recorded flowsheet (poor terminology coverage)
[ ] free text / unowned field (documentation behaviour)
3. VALIDATION
Exposure validated? [ ] yes [ ] no
Method ........................: _______________________
Sample size ...................: ______
Sensitivity ___% PPV ___% Specificity ___% NPV ___%
Outcome validated? [ ] yes [ ] no
Method ........................: _______________________
Sensitivity ___% PPV ___%
[ ] Reviewers blinded to the coded status
4. MISSINGNESS (do NOT silently exclude)
Proportion missing, per key variable: ______________________
Reported by: [ ] variable [ ] site [ ] period [ ] patient group
Assumed mechanism:
[ ] missing completely at random (justify): ____________
[ ] missing at random (justify): ______________________
[ ] INFORMATIVE - recorded only when clinically indicated
Handling ........................: _______________________
[ ] Complete-case analysis avoided, or its selection on
care intensity is stated as a limitation
5. MAPPING AND DATA MODEL
Common data model and version ....: _______________________
Vocabulary release ...............: _______________________
Mapping rate achieved ............: ______ %
Disposition of unmapped records ..: _______________________
Variable classes excluded in advance for poor coverage: ____
6. NOTES AND PROVENANCE
Are clinical notes used? [ ] no [ ] yes
If yes:
[ ] Ambient or AI documentation tools in use in the period
[ ] Machine-drafted notes can be identified
[ ] They cannot - stated as a limitation
Note-quality trend over the period considered? [ ] yes
7. GOVERNANCE
Approval reference ...............: _______________________
De-identification method .........: _______________________
Data sharing position ............: _______________________
SENTENCES TO DELETE FROM THE METHODS IF PRESENT
[ ] "relevant ICD codes were used"
[ ] "patients with missing data were excluded" with no mechanism
[ ] "data were mapped to a common data model" with no rate
[ ] "data were obtained from the clinical data warehouse" as the
entire description of provenance Pubrica AcademyWritten and maintained by Pubrica’s subject matter experts. Articles are revised when the underlying standards change, and each article states the date on which it was last reviewed.
How Pubrica can help
Pubrica reviews the elements of an EHR study that reviewers scrutinise: whether each variable is defined operationally and per site, whether exposure and outcome have been validated against a reference standard rather than assumed, whether missingness is characterised as informative, and whether the mapping rate and data model version are reported. Pubrica also checks adoption and policy statistics cited in introductions against what they measure, since the certified, basic, comprehensive and meaningful-use figures are frequently used interchangeably.