Instant Patient State Assessment Based on 12-Lead ECG: Codified Terms, Streaming Oscillograms, and Traceable Cardiology Indicators

Authors

1. Prozorov A.A., Director of Technology, RTLAB; Architect, Sbertech; ap@rtlab.ru;
2. Redkin I.V., Director of Science, RTLAB; Cand.Sc. (Med.); Leading Research Scientist, Negovsky Research Institute of General Reanimatology, FSCC for Resuscitation and Emergency Medicine; iredkin@rtlab.ru;

Abstract

Objective. To present a scientific model of instant patient state assessment using 12-lead electrocardiography, in which streaming ECG signals, standardized measurements, and codified clinical terms are interpreted as time-synchronized evidence suitable for physician review.

Materials and Methods. A conceptual literature search and critical analysis of source documents on the HealthOS platform was performed with refinement of cardiology, electrophysiology, terminological, and regulatory vocabulary. The concept was compared with AHA/ACCF/HRS recommendations on ECG standardization and interpretation, LOINC terminology for the 12-lead EKG panel, IEC 60601-2-25 and ANSI/AAMI EC57 standards, and contemporary literature on algorithmic and neural-network ECG interpretation.

Results. An interpretation architecture is proposed in which an authorized ECG knowledge dictionary, measurement rules, lead-level signal evidence, current values of codified terms, mimic-exclusion conditions, and algorithm version form a single evidence chain. The output is not an autonomous diagnosis, but a bounded cardiology indicator reporting that a given rule was satisfied, absent, excluded, unsupported, or unresolved.

Conclusion. The 12-lead ECG model is clinically safest when algorithmic output remains traceable, bounded, and reviewable. Cardio indicators can improve analysis reproducibility and audit quality, but do not replace clinical diagnosis, physician interpretation, regulatory evaluation, or prospective validation.

Keywords: 12-lead ECG; electrocardiography; automated ECG interpretation; codified terms; LOINC; cardio indicator; cardio flag; ST-segment; arrhythmia; traceability; clinical decision support; medical cybernetics

Abbreviations and Terminological Designations

ECG — electrocardiography/electrocardiogram; HR — heart rate; QRS — ventricular complex; PR/PQ — atrioventricular interval; ST — ST segment; QTc — corrected QT interval; LOINC — Logical Observation Identifiers Names and Codes.
Cardio indicator/cardio flag — a bounded machine-readable rule or model output requiring clinical review and not constituting a diagnosis by itself.

1. Introduction

Electrocardiography is one of the most informative and simultaneously methodologically sensitive data sources for instant patient state assessment. The 12-lead ECG combines cardiac electrical activity, waveform and segment morphology, rhythm, conduction, ischemic changes, recording artifacts, and clinical context. Therefore, automated ECG interpretation requires not only signal-shape analysis, but also strict terminological, temporal, and evidential discipline.

In clinical practice, computerized ECG interpretation is often perceived as a diagnostic statement. This creates a risk of over-assertion, especially with artifacts, impaired recording quality, incomplete lead sets, cardiac pacing, bundle branch blocks, electrolyte disorders, drug effects, and mimic ischemic patterns. International ECG standardization recommendations emphasize the need for uniform terminology, correct measurements, and cautious interpretation of computer conclusions [1-5].

2. Terminological Framework

In this article, a codified clinical term is a governed unit of medical meaning with a stable identifier, version, current value, and scope of applicability. A term may denote a PR interval measurement, QRS duration, electrical axis position, rhythm type, P-wave feature, ST elevation or depression, artifact class, exclusion condition, or resulting cardiology indicator.

An ECG oscillogram is continuous or time-discretized electrical signal material obtained from standard leads. It is a source of signal evidence but does not by itself assign clinical semantics. Clinical meaning arises only when a signal fragment is linked to a measurement, term, rule, context, and reviewable result.

A cardiology indicator, or cardio flag, should be treated as a bounded output of a computable rule. It is not equivalent to a diagnosis. It reports that, for a given rule version, admissible signal window, and current term values, a specified condition was satisfied, not satisfied, excluded, unsupported, or left unresolved.

3. ECG Standardization as the Basis for Digital Interpretation

Standardized ECG interpretation requires uniform rules for recording, display, measurement, and terminology. AHA/ACCF/HRS documents describe technological requirements for ECG, the diagnostic lexicon, conduction parameters, ST-T-U changes, and ischemia/infarction criteria [1-5]. They are not a ready-made software algorithm, but define the clinical language without which digital interpretation becomes ambiguous.

LOINC terminology complements this framework at the observation level: the 12-lead EKG panel contains codes for basic elements of the ECG study, including intervals, QRS duration, and other structured parameters [6]. For a digital system this matters because inputs and outputs must be comparable with external medical dictionaries and electronic health records.

IEC 60601-2-25 and ANSI/AAMI EC57 standards are important as technical and methodological context. The first relates to safety and essential performance requirements for electrocardiographs; the second to approaches for testing and reporting results of rhythm and ST-segment measurement algorithms [7,8]. However, citing a standard does not mean that a specific algorithm is validated; validity must be demonstrated in a separate study with predefined metrics.

4. Evidence Architecture of Interpretation

The proposed architecture begins not with an analyzer, but with an authorized ECG knowledge dictionary. It records terms, relations, rules, evidence requirements, admissible signal windows, exclusion conditions, and source references. Compilation translates this knowledge into executable rules but does not add new clinical meaning. The runtime analyzer only applies rules to data, recording version, inputs, outputs, and an evidence trace.

At the moment of interpretation, the system must combine two streams. The first is current values of codified terms: for example, presence of a pacemaker, potassium level, therapy mode, lead quality, or previously established rhythm context. The second is the 12-lead ECG stream: waveform shape, intervals, segments, RR regularity, QRS morphology, P-QRS relationship, signal quality, and inter-lead coherence. Both streams must be time-aligned, because a term value is current only until the next update.

If a rule produces a result without reference to a term, signal window, data quality, or algorithm version, that result must not be considered suitable for clinical review. It may be a technical signal of a traceability defect, but not full clinical evidence.

Table 1. Evidence layers in 12-lead ECG interpretation

Layer	Content	Function	Boundary
Codified terms	intervals, rhythms, morphology, exclusions, patient context	assign governed clinical semantics	do not prove the signal without oscillograms
ECG oscillograms	lead signals, time windows, morphology, quality	supply signal evidence	do not create clinical meaning by themselves
Rules/model	algorithm version, criteria, exclusion conditions	apply authorized interpretation	do not clinically validate themselves
Cardio indicator	output status and evidence references	report a bounded result	not an autonomous diagnosis
Physician reviewer	clinical context and decision	assess plausibility and applicability	must not silently repair semantics during correction

5. Named Cardiology Indicators

A cardio indicator answers a narrow question: is a specific condition supported by the data in the current context? For example, “sinus rhythm supported” means that, in an admissible recording window, sinus-origin P waves are detected before most QRS complexes, PR intervals are within expected limits, RR is sufficiently regular, QRS is not prolonged by rule criteria, and exclusion conditions are not active. Such an indicator does not claim absence of all other abnormalities and does not replace a physician’s conclusion.

For ischemic and morphological indicators, limitations are even greater. ST elevation may reflect acute myocardial infarction, but also pericarditis, early repolarization, left ventricular aneurysm, bundle branch block, ventricular rhythm, hyperkalemia, artifacts, or incorrect electrode placement. Therefore, each target indicator must be accompanied by explicit mimic conditions, quality criteria, and indication of which alternative explanations were checked.

Table 2. Statuses of a cardiology indicator

Status	Meaning	Key physician question
Supported	data and current context satisfy the rule	are references to signal, terms, and rule version sufficient?
Not supported	rule executed but target condition not met	is the input valid enough to trust a negative result?
Excluded	a better explanation is active: artifact, mimic, or other context	is the exclusion rationale shown?
Data-unsupported	required lead, feature, window, or term is missing	is this a capture gap or a terminology gap?
Unresolved	evidence conflicts or policy blocks output	what refinement is needed for re-analysis?

6. Mimic Exclusion and Artifact Control

ECG mimic exclusion is a governed rule recording why an observed pattern must not be interpreted as the target indicator. The reason may be artifact, poor electrode contact, muscle tremor, baseline drift, cardiac pacing, conduction block, ventricular rhythm, electrolyte disorder, drug effect, lead mismatch, or conflicting evidence.

Clinically, exclusion must not be silent. If a target indicator is blocked, the system must show the basis for blocking, corresponding signal segments, terms used, and rule version. Otherwise the physician cannot judge whether a negative result is reliable, unsupported, or due to missing data.

Table 3. Risks of automated ECG interpretation and control measures

Risk	Potential failure	Control measure
Clinical over-assertion	indicator read as diagnosis	explicit limitation of result status and physician review
Signal artifact	false arrhythmia or false ST shift	lead quality assessment, baseline drift, and noise evaluation
Data asynchrony	term not current for ECG window	temporal binding and term-value validity period
Missing exclusion	mimic does not block target indicator	governed exclusion base and diagnostic refinement
Non-reproducibility	output cannot be repeated	rule version, result hash, log of inputs and outputs

7. Example: Sinus Rhythm Determination

Consider a bounded example: evaluation of a sinus-rhythm candidate. Before rule execution, the system checks recording quality, lead identity, window duration, cardiac-cycle regularity, and presence of a sufficient number of QRS complexes. P-wave features, P-QRS linkage, PR intervals, RR regularity, and QRS duration are then analyzed. Lead II may be the primary source, but final assessment must account for inter-lead coherence and signal quality.

When criteria are met, an indicator “sinus rhythm supported” is issued. With a pacemaker, atrial arrhythmia, significant artifact, incomplete lead set, or conflicting features, the indicator may be excluded, unsupported, or unresolved. Such output is clinically useful precisely because it does not hide uncertainty and shows which evidence was used.

8. Relation to Modern Algorithms and Artificial Intelligence

Modern neural-network models demonstrate high classification performance for arrhythmias and several ECG abnormalities, including single-lead ambulatory ECG and 12-lead ECG [11,12]. However, high metrics on a test set do not remove the need for clinical traceability, artifact control, data-shift analysis, independent validation, and understandable presentation of results to the physician. For decision-support systems it is essential that the user can evaluate the basis of the output, not only receive a class label.

Clinical studies show that automated AI conclusions and recommendations amplify automation bias: physicians over-trust model outputs and under-check borderline cases [13,14]. In 12-lead ECG interpretation, incorrect automated interpretation reduces diagnostic accuracy (approximately 43% among cardiologists and 59% among non-cardiologists) and affects less experienced interpreters more strongly [13]; in a randomized clinical trial with large language model recommendations, similar dependence on erroneous output persisted even among physicians trained in AI [14]. Rule-based systems make uncertainty explicit — a rule either fires or it does not; the physician sees exactly which rule fired and can challenge it if needed.

Under these conditions, model determinism becomes an engineering requirement. It must be reproducible, validatable, resistant to drift, and yield the same results on repeated runs with the same rule package and component versions.

In the HealthOS platform, AI is appropriate for source search and analysis, material structuring, knowledge-graph draft preparation, and analytics of clinical practice results to create candidates for knowledge-graph refinement. It may help reveal a possible new rule, rule correction, or borderline case, but the result remains within the responsibility of the author of knowledge-graph changes and expert review.

The ultimate clinical output of the system preserves linkage to data, terms, exclusion conditions, signal quality, model version, and assertion boundaries. This is especially important in the regulatory logic of SaMD/CDS [9,10].

9. Evidence, Audit, and Physician Review

A reviewable ECG result must include rule identifier, knowledge-base version, references to source leads, time window, measured features, current term values, activated exclusions, indicator status, and a result hash or other integrity-control mechanism. Such structure allows the physician to understand why the system formed a specific output and to distinguish clinical uncertainty from a technical defect.

If a missing term, obsolete rule, undescribed exclusion, or unsuitable signal is detected, the problem must be passed to a governed knowledge-refinement contour, not corrected arbitrarily in the runtime layer. The analyzer must not become a source of new clinical semantics.

10. Limitations

This article does not claim clinical effectiveness, safety, regulatory approval, conformity to a specific standard, or complete coverage of all rhythm, conduction, ischemia, infarction, electrolyte, and drug-related abnormalities. The model presented is conceptual and terminological, not a clinical trial report. Practical deployment requires technical verification, independent clinical labeling, external validation, assessment of impact on physician decisions, and regulatory qualification.

11. Conclusion

Instant patient state assessment based on 12-lead ECG must combine standardized measurements, codified terms, streaming oscillograms, exclusion conditions, and traceable results. Algorithmic output must remain a bounded cardio indicator, not an autonomous diagnosis. Such a framework improves reproducibility, facilitates audit and review, but preserves the central role of the physician, clinical context, and prospective validation.

Conflict of Interest

The manuscript describes implementation of a specific aspect of the HealthOS platform, namely: instant patient state assessment based on interpretation of codified term values and a real-time 12-lead ECG stream.

Funding

Work funded by RTLAB.

References

1. Kligfield P, Gettes LS, Bailey JJ, et al. Recommendations for the Standardization and Interpretation of the Electrocardiogram: Part I: The Electrocardiogram and Its Technology. J Am Coll Cardiol. 2007;49(10):1109-1127. doi:10.1016/j.jacc.2007.01.024.
2. Mason JW, Hancock EW, Gettes LS, et al. Recommendations for the Standardization and Interpretation of the Electrocardiogram: Part II: Electrocardiography Diagnostic Statement List. J Am Coll Cardiol. 2007;49(10):1128-1135. doi:10.1016/j.jacc.2007.01.025.
3. Surawicz B, Childers R, Deal BJ, et al. AHA/ACCF/HRS Recommendations for the Standardization and Interpretation of the Electrocardiogram: Part III: Intraventricular Conduction Disturbances. J Am Coll Cardiol. 2009;53(11):976-981. doi:10.1016/j.jacc.2008.12.013.
4. Rautaharju PM, Surawicz B, Gettes LS, et al. AHA/ACCF/HRS Recommendations for the Standardization and Interpretation of the Electrocardiogram: Part IV: The ST Segment, T and U Waves, and the QT Interval. J Am Coll Cardiol. 2009;53(11):982-991. doi:10.1016/j.jacc.2008.12.014.
5. Wagner GS, Macfarlane P, Wellens H, et al. AHA/ACCF/HRS Recommendations for the Standardization and Interpretation of the Electrocardiogram: Part VI: Acute Ischemia/Infarction. J Am Coll Cardiol. 2009;53(11):1003-1011. doi:10.1016/j.jacc.2008.12.016.
6. Regenstrief Institute. LOINC 34534-8: 12 lead EKG panel. Indianapolis: Regenstrief Institute; current online terminology resource.
7. International Electrotechnical Commission. IEC 60601-2-25:2011. Medical electrical equipment - Part 2-25: Particular requirements for the basic safety and essential performance of electrocardiographs. Geneva: IEC; 2011.
8. Association for the Advancement of Medical Instrumentation. ANSI/AAMI EC57:2012/(R2020). Testing and reporting performance results of cardiac rhythm and ST segment measurement algorithms. Arlington: AAMI; 2020 reaffirmation.
9. U.S. Food and Drug Administration. Clinical Decision Support Software: Guidance for Industry and Food and Drug Administration Staff. Silver Spring: FDA; 2026.
10. International Medical Device Regulators Forum. Software as a Medical Device (SaMD): possible framework for risk categorization and corresponding considerations. IMDRF/SaMD WG/N12FINAL:2014. 2014.
11. Hannun AY, Rajpurkar P, Haghpanahi M, et al. Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network. Nat Med. 2019;25(1):65-69. doi:10.1038/s41591-018-0268-3.
12. Ribeiro AH, Ribeiro MH, Paixao GMM, et al. Automatic diagnosis of the 12-lead ECG using a deep neural network. Nat Commun. 2020;11:1760. doi:10.1038/s41467-020-15432-4.
13. Bond RR, Novotny T, Andrsova I, et al. Automation bias in medicine: The influence of automated diagnoses on interpreter accuracy and uncertainty when reading electrocardiograms. J Electrocardiol. 2018;51(6 Suppl):S6-S11. doi:10.1016/j.jelectrocard.2018.08.007.
14. Qazi IA, Ali A, Khawaja AU, et al. Automation bias in large language model-assisted diagnostic reasoning among physicians trained in AI literacy: a randomized clinical trial. NEJM AI. 2026;3(5). doi:10.1056/AIoa2501001.
15. Berikashvili LB, Kuzovlev AN, Yadgarov MYa, et al. Prognostic ability of the MEWS nomogram after cardiac surgery in the ICU. Bulletin of Anesthesiology and Reanimatology. 2022;19(2):6-13. doi:10.21292/2078-5658-2022-19-2-6-13.
16. Moroz VV, Borisov KYu, Grebenchikov OA, et al. Anesthetic myocardial preconditioning and biochemical markers of cardiac and coronary insufficiency. General Reanimatology. 2013;9(5):29-35.
17. Likhvantsev VV, Moroz VV, Grebenchikov OA, et al. Ischemic and pharmacological preconditioning. General Reanimatology. 2011;7(6):59-65.