Operational Profiles and Reporting Guidance from a Systematic Meta-Review on Digital Patient Frameworks

1. Introduction

Digital representations of health and disease support physiological simulation, clinical prediction, education, surgical planning, and in silico experimentation. The same broad family is described as digital twins (DTs), patient or human digital twins, digital patients (DPs), virtual patients, synthetic patients, avatars, and in silico patients or trials. These terms have intersected histories, but they do not specify equivalent relationships with a physical person. A virtual patient may be a teaching case; an in silico patient may be one member of a simulated cohort; and a detailed patient-specific model may never be updated after acquisition. Treating these configurations as less complete versions of a twin mistakes difference for deficiency and obscures their intended functions [1,2,3,4,5].

Three concepts are therefore separated throughout this paper. Construct identity is the label an author uses and locates a system within a discourse. An operational profile describes the represented subject and scale, patient specificity, temporality or update pattern, and physical–digital information pathway. Credibility concerns whether evidence supports fitness for a stated purpose. A DT label does not establish a particular operational profile; a particular profile does not establish credibility; and credibility for one purpose does not determine construct identity. In short, labels identify discourse, profiles describe systems, and credibility supports claims for use.

The 2024 National Academies framework is a valuable reference specification. It describes a DT as a virtual representation of a physical counterpart that is dynamically updated with data, makes predictions, informs value-producing decisions, and is connected through bidirectional interaction [6]. We use that specification to make attributes explicit, not as the sole valid definition of a useful medical DP and not as an inclusion threshold. In healthcare, a meaningful update may occur at an image, laboratory panel, or clinical encounter, and a return pathway may be mediated by a clinician rather than an actuator. The framework itself relates update frequency to the decision task [6,7]. A static patient-specific planning model, a one-way monitoring system, and an educational virtual patient may all be useful without satisfying that DT specification.

Existing reviews have organized systems by biological hierarchy, application, data flow, or maturity [1,3,8,9].Those schemes can answer the questions for which they were designed. Used as a universal ladder, however, they can conflate what is represented, how a system is configured, what it is used for, and how well its claims are supported. Medical DPs require a description that keeps these matters separate. The historical Digital Patient program was itself broader than a synchronized replica, joining multiscale models, data resources, interfaces, and decision support [10]. That breadth supports a family-level concept with explicit operational profiles rather than a label-based hierarchy.

This systematic meta-review asks: which labels identify DP constructs in reviews; how the four profile dimensions are reported and combined; how review-level appraisal relates to those descriptions; and what minimum reporting would let authors, reviewers, funders, and standards initiatives compare systems without inferring configuration or quality from a name. The aim is not to decide which systems are or are not DTs or DPs. It is to provide a practical research and reporting framework for stating what a system is and what evidence supports its intended use.

2. Methods

2.1. Evidence-Map Provenance and Protocol Status

The present systematic meta-review draws on the search, screening, eligibility assessment, data extraction, duplicate resolution, and review-level appraisal conducted under a protocol registered in PROSPERO (CRD420261458411). While companion reports evaluate computational pipeline readiness and sociotechnical clinical adoption, this manuscript focuses exclusively on the lexical evolution, taxonomy, and conceptual framing of digital patient representations. It provides a focused synthesis of terminology, operational profiles, and reporting guidance within the protocol-defined scope. The analytical framework was finalized after assembly of the evidence map and before manuscript submission, with its methods documented in Appendix S1. Reporting was guided by PRISMA 2020 [11,12].

The evidence-map plan included English-language reviews published from 2020 onward that addressed patient-linked physiological, biomechanical, biological, or adjacent computational representations. Web of Science, PubMed, IEEE Xplore, and Google Scholar were searched, with reference-list and designated preprint-server supplementation. The search was conducted in October 2025; no later search was incorporated, making October 2025 the evidence cut-off.

Of 247 reports assessed for eligibility, 183 were excluded and 64 entered extraction (Figure 1). Bibliographic reconciliation then removed three duplicate reports before stable identifiers P01–P61 were assigned (Table 7), leaving 61 unique reviews. Records were managed and screened in CADIMA. Data were extracted using a standardized CADIMA form aligned with the protocol-defined research questions. The identifier-to-citation and review-appraisal lookup is provided in Supplementary Appendix S3.

2.2. Analytical Layers

Three distinct components: construct identity, operational profile, and credibility evidence; were analyzed independently, and none was inferred from or used as a proxy for another.

2.2.1. Construct Identity

Primary terminology was assigned by a documented precedence rule using title, related construct, and definition fields. Phrase-aware filters removed null and negated statements, and a small set of contrastive expressions received documented manual overrides. Construct identity records the terminology a review used for its subject; it does not characterize the system’s configuration or quality.

2.2.2. Operational Profile

Four dimensions were coded non-exclusively from the relevant standardized extraction fields. They answer four independent descriptive questions:

□

Represented subject and biological scale: What is represented? Molecular or cellular, tissue, organ or physiological system, person or whole body, population, and system or other. Broader biological coverage was not treated as better than narrower coverage.

□

Patient specificity: Whose model is it? Generic, archetypal, or population-based representation versus a model initialized or calibrated with data from one person. A review could contain both, and individual specificity was not treated as inherently preferable.

□

Temporality or update pattern: How does it change? Explicitly static versus dynamic, longitudinal, episodic, continuous, or real time. Simulating change was not assumed to mean that new patient observations were assimilated during operation, and faster updating was not treated as progress.

□

Coupling or physical–digital information pathway: How does information cross the boundary? None or manual transfer, one-way transfer from person to model, or review-described bidirectional exchange. A recommendation returned through a clinician counted as a return pathway but not as automated control. Bidirectionality was not treated as proof of usefulness or credibility.

Initial term and profile assignments were generated using documented, phrase-aware rules applied to the standardized extraction fields. The rules removed null and negated statements, and documented overrides addressed explicit contrastive expressions. All record-level assignments for construct identity and the four operational-profile dimensions were then manually verified across the 61-review cohort against the extracted evidence, with ambiguous cases adjudicated. Only the verified classifications were used in the reported counts, combinations, tables, and figures.

For descriptive comparison only, we examined review-level co-reporting of individual specificity, dynamic or continuous orientation, and bidirectional coupling. This exploratory three-facet overlap is not a DT definition, a DP threshold, a maturity category, or evidence that one primary implementation contained all three attributes. The extraction unit was the review, and a review could describe several system classes.

Intended decision was not retrofitted as a fifth profile dimension. We distinguish clinical decision-making (CDM), concerning direct care such as diagnosis, testing, or treatment, from operational decision-making (ODM), concerning delivery-system decisions such as waiting times, staffing, beds, or revenue [13]. The extraction form did not support reliable retrospective counts for this distinction. Purpose therefore appears as a reporting requirement, not a numerical result. Credibility evidence is likewise required for interpretation but remains outside the four-dimension profile.

2.2.3. Credibility

AMSTAR 2 was used where appropriate for systematic reviews and SANRA for structured narrative reviews; CADIMA-aligned criteria provided additional review-level context [14,15]. Review-method appraisal and whether source reviews appraised their included evidence were displayed separately. These measures inform confidence in the evidence provided in the review and do not establish the credibility of every system described in the record.

2.3. Synthesis Safeguards

Our synthesis was strictly descriptive. Counts reflect reviews rather than primary studies; categories are non-exclusive, and primary-study overlap was not estimated. Co-reporting within a single review cannot justify system-level inferences. Rather than forcing reviews into mutually exclusive classes, the radial evidence matrix shown in Figure 2 retained all positive values across categories.

3. Results

3.1. Construct Identity: One Dominant Label, Variable Descriptions

Digital twin was the primary term in 46 reviews (75.4%). Patient digital twin, human digital twin, and virtual patient each led four reviews (6.6% apiece); in silico patient or trial led three (4.9%). The numerical dominance of one label did not produce one operational definition.

Boundary cases show what the labels leave unstated. John et al [8]. (P17) separated digital models, shadows, and twins according to data flow. Pradíes et al [16]. (P21) described a patient-specific dental virtual patient with captured jaw motion but did not report longitudinal synchronization with the person. Karaduman et al [4]. (P57) examined interactive virtual-patient simulations designed for learning rather than for mirroring one person’s changing state [17]. Tudor et al [2]. (P31) applied personalization, dynamic updating, and clinical decision support criteria to 149 primary studies; 18 (12.1%) met all three criteria, 17 through a human-mediated return and one through automatic action on the physical system. These examples describe different configurations and purposes, not successively better instances of one object.

3.2. Operational Profile: Four Dimensions That Did Not Move Together

Represented subject and biological scale. Organ or physiological-system scale appeared in 44 reviews, person or body scale in 43, population scale in 17, molecular or cellular scale in 14, tissue scale in 12, and system or other scale in six.

Patient specificity. Individual specificity appeared in 55 reviews (90.2%), while generic, archetypal, or population representations appeared in 29 (47.5%).

Temporality or update pattern. Dynamic or continuous orientation appeared in 53 reviews (86.9%), and explicitly static constructs appeared in 15 (24.6%).

Coupling or physical–digital information pathway. One-way coupling was reported in 39 reviews (63.9%), bidirectional coupling in 26 (42.6%), and none or manual exchange in 13 (21.3%).

Characteristics and safeguards of the included evidence base are summarized in Table 1, and the construct-identity and operational-profile findings are summarized in Table 2.

Overlaps within each dimension were expected because one review could include several model classes. Across the three attributes used in the exploratory overlap, the largest combinations were individual-specific plus dynamic or continuous plus bidirectional (25 reviews) and individual-specific plus dynamic or continuous without bidirectional coupling (23). Six reviews described individual-specific constructs without either dynamic orientation or bidirectionality; five described dynamic orientation without individual specificity or bidirectionality. One review combined individual specificity and bidirectional coupling without a dynamic or continuous orientation; one reported none of the three attributes. No review reported bidirectional coupling without individual specificity. These combinations describe review-level reporting patterns. Figure 2 displays operational-profile information and review appraisal together for comparison while keeping them conceptually separate.

3.3. Exploratory Overlap and Evidence Appraisal

Twenty-five reviews (41.0%) co-reported individual specificity, dynamic or continuous orientation, and bidirectional coupling. Their primary terms were digital twin in 20, patient digital twin in three, and human digital twin in two. No virtual-patient or in silico primary-term record overlapped. That pattern describes discourse review only. It does not estimate the number of implemented DTs, show that all three features occurred together in one system, or establish that the overlap defines a DT.

More granular reviews reinforce the need for separate dimensions. Drummond et al [1]. (P12) identified 80 claimed patient digital twins and reported one-way patient-to-model flow for 58, two-way flow for nine, and, among 64 patient-specific models, 44 static and 20 dynamic configurations. The effect of definitional boundaries was especially visible in dermatology: Akbarialiabad et al [18]. (P03) screened 157 records but found no implementation that satisfied their bounded definition. Vallée et al [19]. (P44) used individual initialization, updating, and counterfactual simulation to distinguish included from borderline systems. These reviews were used to expose differences among definitions and dimensions, not to endorse a particular threshold. Usable review-method appraisal was available for 39/61 reviews (63.9%) and 17/25 reviews (68.0%) in the exploratory overlap. By primary term, appraisal was available for 30/46 digital-twin reviews, 2/4 patient-digital-twin reviews, 3/4 human-digital-twin reviews, 3/4 virtual-patient reviews, and 1/3 in silico reviews. Only three of the 25 reviews in the exploratory overlap formally appraised their included primary evidence. Construct identity did not determine a single operational profile, and neither construct identity nor operational profile certified credibility.

4. A Six-Question Reporting Grammar

The practical response to unstable terminology is not to ban labels but to reduce the information they are expected to carry. The six-question grammar has two explicitly different parts. Questions 1–4 report on the four dimensions of the operational profile. Questions 5–6 state the purpose of the decision and credibility evidence needed to interpret that profile. The latter are reporting requirements, not additional profile dimensions. The complete grammar and the inferences each question is intended to prevent are summarized in Table 3. The grammar is informed by DT and model-credibility guidance [6,20,21].

4.1. Part A: Four Operational-Profile Questions

• What is represented? Name the physical or conceptual subject and its biological or service-system scale.
• Whose model is it? State whether it is generic, population-based, stratified, or initialized or calibrated to one person.
• How does it change? Report whether it is static or updated episodically, longitudinally, continuously, or in real time, including the trigger and cadence.
• How does information cross the boundary? Describe manual, one-way, or bidirectional flow and whether any return path is clinician-mediated or automated.

4.2. Part B: Two Requirements for Interpreting the Profile

5.

What decision does it serve? Name the intended user, action, and whether the purpose is CDM, ODM, both, or neither.

6.

Why should it be trusted for that purpose? Report the evidence of fitness for the stated use, including uncertainty, limitations, and performance monitoring.

For the fifth question, a model that forecasts tumor growth to compare treatments serves CDM; one that forecasts patient flow to allocate beds or staff serves ODM. A platform may support both, but each decision and user should be stated because predictive accuracy alone does not establish clinical benefit or operational value [13]. The sixth question is purpose-dependent: evidence supporting education, one-time planning, or cohort experimentation may differ from evidence required for a patient-specific clinical recommendation.

The grammar is deliberately label-agnostic. A static patient-specific anatomical model can be appropriate for one-time planning. A generic educational virtual patient can suit a defined learning objective. A one-way monitoring representation can support surveillance, and a clinician-mediated return pathway can support care without automatic actuation. A virtual cohort can support an in silico experiment without maintaining a persistent counterpart. Conversely, bidirectional real-time coupling does not by itself show that a system is accurate, safe, useful, or credible. Each system should be described by its four-dimensional profile and judged against its own stated purpose.

5. Discussion

5.1. Labels Identify Discourse; Profiles Describe Systems; Credibility Supports Claims

The principal finding is not merely that terminology is inconsistent. Rather, labels, configurations, and evidentiary support are different analytical objects. Represented subject and scale, patient specificity, temporality, and coupling answered different questions and did not move together across reviews. Purpose then specifies what the system is meant to do, while credibility concerns whether the evidence supports that claim for use. Keeping these layers separate prevents identity from being mistaken for quality and prevents a technically elaborate configuration from being mistaken for demonstrated value.

The distinction also preserves the diversity of constructs across the DP spectrum of definitions. A virtual patient can be assessed for learning validity when education is the purpose; a virtual cohort for representativeness and experimental purpose; a static surgical model for anatomical and decision fitness; and a one-way or clinician-mediated system for the task and safeguards actually intended. None is a failed or lesser DP because it lacks a feature required by one DT definition. Nor is a system more legitimate merely because it is individualized, continuously updated, or bidirectionally coupled.

Primary studies outside the review corpus illustrate how the grammar could sharpen interpretation of implemented systems. For a patient-specific pulmonary-artery DT and a simulation-trained coronary surrogate, explicit reporting would distinguish calibration and update cadence from computationally real-time prediction and clarify whether a physical–digital return pathway was implemented in the study or proposed as a future capability [22,23]. A physics-generated virtual aneurysm cohort further illustrates why simulated temporal evolution, cohort generation, and subsequent individualization should be reported separately rather than inferred from the word “virtual.”[24]

5.2. Why an Operational Profile Is Preferable to a Universal Hierarchy

Hierarchy-, application-, data-flow-, and maturity-based schemes in the cited literature make important differences visible [1,3,8,9]. The problem arises when their categories are interpreted across a single continuum of completeness. Biological scale concerns the represented subject, not sophistication. Application concerns purpose, not configuration. Data flow concerns coupling, not usefulness. A maturity designation may bundle several such judgments, but it cannot substitute for purpose-specific credibility evidence.

The operational-profile approach is therefore better suited to the heterogeneous medical DP literature. It permits more precise comparison without requiring a universal boundary around DT or DP. A molecular model and a whole-body representation can be compared on scale without treating one as superior; a static and a continuously updated system can be compared on temporality without assuming the latter is more useful; and clinician-mediated and automated return pathways can be reported without treating automation as having greater legitimacy. The National Academies definition remains a clear specification for systems intended to meet it, but it is neither the sole vocabulary for medical DPs nor a criterion for whether other configurations can provide value [6].

5.3. Practical Use and Future Development

This framework is intended as a foundation, not a gate. Authors can place the six answers in abstracts, methods, or structured model cards. Reviewers can use them to identify claims that depend on an unstated update cadence, personalization method, or return pathway. Funding agencies can require applicants to state the intended decision and purpose-matched evidence without demanding a preferred label or configuration. Standards initiatives can turn the questions into controlled fields, examples, and conformance checks for reporting completeness while explicitly avoiding a feature-count score. Future DP toolkits can represent the four profile dimensions as separate metadata fields and link each intended use to its own credibility record.

Prospective work should test inter-rater reliability, refine response options with primary-study authors and users, and evaluate whether the grammar improves study selection, reproducibility, and claim appraisal. Such testing may support extensions, but purpose and credibility should remain outside the four-dimension profile so that descriptive configuration is not converted into a ranking system.

5.4. Limitations

The current study included 61 unique reviews after bibliographic duplicate resolution, with stable identifiers assigned after deduplication. The record-level constructs classifications were generated using documented rules and manually verified across the whole cohort, with ambiguous cases adjudicated against the extracted evidence. Categories were coded non-exclusively, and review appraisal was kept separate from construct identity and operational profile. The radial display (Figure 2) retains ambiguity rather than hiding it in a forced class. The unit of analysis, however, was a review, and overlap among underlying primary studies was not quantified. A review-level combination may aggregate attributes from different systems; co-reporting does not warrant system-level inference. Manual verification established what each source review reported, but the source descriptions did not always distinguish simulated dynamics, episodic recalibration, online state estimation, and continuous operation. Similarly, review-described bidirectionality could include either a human-mediated recommendation or an automatic actuation. Intended decision was not a separate countable extraction field. Appraisal methods were heterogeneous and unavailable for many reviews, and only 11/61 source reviews formally assessed their included evidence.

The reporting grammar is a practical conceptual framework, not a validated ontology, taxonomy score, maturity model, readiness level, or classification threshold. Its categories describe what was reported; they do not decide which systems deserve the DT or DP label.

6. Conclusions

This systematic meta-review proposes a practical way to describe medical digital representations without turning terminology into a hierarchy. Construct identity records the label used. The operational profile reports four separate dimensions: represented subject and biological scale, patient specificity, temporality or update pattern, and coupling or physical–digital information pathway. Intended decision states the claim for use, and credibility evidence supports or limits that claim. The six-question grammar reports these elements but does not score them. A static, generic, one-way, or clinician-mediated system can be the right system for its purpose, just as a continuously updated bidirectional system can lack adequate evidence for its own. Labels identify discourse; profiles describe systems; credibility supports claims for use.