Accessibility settings

This paper is in the following e-collection/theme issue:

Viewpoints and Opinions on Human Factors (36) Clinical Informatics (2273) E-Health Policy and Health Systems Innovation (529) Theoretical Frameworks and Concepts (550) Safety and Error Prevention in Health (321) Ontologies, Classifications, and Coding (442) Digital Health Reporting Standards, Quality and Transparency in e-Research (224)

Published on in Vol 13 (2026)

This is a member publication of Bibsam Consortium

Preprints (earlier versions) of this paper are available at https://preprints.jmir.org/preprint/91783, first published .
Doctor in white coat reviewing medical data on a tablet.

Understanding the Limits of Patient Safety Classification Systems for Health Information Technology–Related Incidents

Understanding the Limits of Patient Safety Classification Systems for Health Information Technology–Related Incidents

Authors of this article:

Md Shafiqur Rahman Jabin1, 2 Author Orcid Image
Article Authors Cited by Tweetations Metrics

    Viewpoint

    1Department of Medicine and Optometry, Linnaeus University, Kalmar, Sweden

    2Faculty of Health and Social Care, University of Bradford, Bradford, West Yorkshire, United Kingdom

    Corresponding Author:

    Md Shafiqur Rahman Jabin, PhD

    Department of Medicine and Optometry

    Linnaeus University

    Universitetsplatsen 1

    Kalmar, SE-391 82

    Sweden

    Phone: 44 7915 673612

    Email: mdshafiqur.rahmanjabin@lnu.se


    Patient safety classification systems are fundamental to surveillance, organizational learning, research, and governance because they enable adverse events and near misses to be organized into standardized categories for comparison and analysis. However, the increasing complexity of health information technology (HIT)–related patient safety incidents challenges the assumptions underpinning conventional classification approaches, as these incidents often emerge from dynamic, distributed, and evolving sociotechnical interactions rather than discrete, time-bounded events. In this Viewpoint, I argue that many of the challenges associated with classifying HIT-related patient safety incidents arise not simply from limitations of individual classification systems but from the inherent representational logic of classification itself. By viewing classification as a knowledge practice rather than merely a technical tool for organizing incident data, I contend that abstraction, boundary-setting, and standardization inevitably simplify complex sociotechnical processes and constrain how safety problems are represented, interpreted, and acted upon. I discuss 4 recurring representational limitations that characterize the application of patient safety classification systems to HIT-related incidents: fragmentation of sociotechnical interactions, loss of temporality and evolving processes, inadequate representation of scale and propagation across systems, and normalization of “use error” through simplified attribution of responsibility. These limitations can contribute to incomplete organizational learning, misaligned safety interventions, and challenges in interpreting and comparing classified patient safety data across health care settings. Rather than arguing against the continued use of patient safety classification systems, I propose that their strengths and limitations should be recognized simultaneously. Classification remains indispensable for surveillance, learning, and governance, but it should be interpreted as one component of a broader sociotechnical understanding of patient safety. Recognizing the representational limits of classification can support more reflexive interpretation of classification-based evidence and encourage complementary approaches that better capture the complexity of HIT-related patient safety.

    JMIR Hum Factors 2026;13:e91783

    doi:10.2196/91783

    Keywords

    sociotechnical systems; knowledge representation; incident reporting; safety science; information infrastructure; organizational learning; sensemaking; complex adaptive systems; data standardization; epistemic limits 


    Classification as Infrastructure in Patient Safety

    Classification systems are foundational to patient safety research, surveillance, and institutional learning. By transforming diverse and complex safety incidents into standardized categories, classification enables aggregation, comparison, and analysis across settings and over time. These systems underpin reporting infrastructures, research studies, quality improvement initiatives, and policy conversations, while shaping how patient safety problems are recognized and addressed [1-3].

    Diversity of Patient Safety Classification Approaches

    Patient safety classification schemes have been developed in a variety of contexts and based on diverse design logics. The International Classification for Patient Safety (ICPS), a World Health Organization (WHO)–created framework, represents a global effort to harmonize safety terminology and incident constructs [3]. Beyond ICPS, there are classification efforts tailored to specific settings or incident types (eg, primary care safety classifications developed through the LINNAEUS [Learning in an International Network About Errors and Understanding Safety] collaboration [4] and medication error taxonomies such as the NCC MERP [National Coordinating Council for Medication Error Reporting and Prevention] taxonomy [3,5,6]). Moreover, structured frameworks, such as human factors taxonomies, provide alternative logics for organizing contributing factors associated with safety events [7-9]. These varied systems share a core commitment: to render unsafe care legible through predefined categories and relationships.

    Why Health Information Technology Challenges Classification

    The expansion of health information technology (HIT) into clinical work has strained conventional classification logic. HIT-related safety incidents often involve sociotechnical interactions, distributed causality, temporal processes, and emergent behaviors that are not easily reducible to discrete event types and fixed causal categories [10,11]. Although HIT-specific classification efforts (eg, adaptations of Health Information Technology Classification System (HIT-CS) and classifications that integrate technology and device issues) have attempted to capture these phenomena, persistent difficulties in representing HIT-related incidents suggest deeper epistemic limits in how classification systems encode knowledge [1,12].

    Although HIT-specific and HIT-adapted classification schemes have been developed to address these challenges, recurring representational difficulties suggest that the problem may not lie solely in the design or implementation of particular frameworks [13]. For example, duplicate medication orders associated with computerized physician order entry (CPOE) systems have been reported as medication-related or HIT-related patient safety incidents, yet such classifications do not fully represent the interacting software, clinical workflow, human, and organizational factors that contributed to their occurrence [14,15].

    From Classification Design to Classification Limits

    Prior work on HIT-related patient safety has addressed methodological challenges in studying technology-associated incidents and articulated interpretive perspectives on HIT-related risk. Building on this foundation, the present paper shifts attention from how incidents should be classified to what classification systems, by their nature, can and cannot represent. By making these inherent limits explicit, this Viewpoint clarifies the constraints on learning from classified safety data and suggests implications for future research in HIT safety.

    Perspective Presented in This Viewpoint

    This Viewpoint examines the inherent limits of patient safety classification systems when applied to HIT-related incidents by considering classification as a representational practice rather than simply a technical tool. I argue that 4 recurring representational limitations help explain why patient safety classification systems struggle to adequately represent HIT-related incidents: (1) fragmentation of sociotechnical interactions, (2) loss of temporality and evolving processes, (3) inadequate representation of scale and propagation across systems, and (4) normalization of “use error” through simplified attribution of responsibility. Rather than proposing a new taxonomy or evaluating existing classification systems, this paper examines classification itself as an epistemic practice [16-19]. Importantly, sociotechnical perspectives on HIT safety are well established in the literature, and the present Viewpoint builds on this body of work by examining how classification systems interact with these broader sociotechnical perspectives.

    This Viewpoint is intended for patient safety researchers, HIT researchers, patient safety classification developers, health care quality and safety professionals, and policymakers involved in the design, interpretation, and use of patient safety classification systems.


    Classification as Knowledge Production

    Patient safety classification systems are often presented as neutral tools for organizing information about adverse events and near misses. In practice, however, classification operates as a representational lens that actively shapes how safety problems are understood [1,6,20]. By defining categories and boundaries, classification systems shape what is recognized as an incident, how causality is attributed, and which dimensions of harm are foregrounded or obscured [6,20,21].

    The term “system” is used in several ways throughout this paper. Specifically, it may refer to the clinical work system, the HIT system, the incident-reporting system, or the patient safety classification system. Unless otherwise specified, this perspective focuses on classification systems as representational frameworks while recognizing their interaction with broader sociotechnical and reporting systems.

    Abstraction and Standardization

    At a fundamental level, all classification systems rely on abstraction and standardization. Classification makes complex phenomena easier to analyze by reducing them to predefined units. These units may include event types, contributing factors, or outcomes that can be applied consistently across cases. This abstraction enables aggregation and comparison across cases, supporting patient safety surveillance and learning. At the same time, abstraction necessarily compresses detail, strips away contextual nuance, and stabilizes fluid processes into discrete representational forms [21-23].

    Boundary-Setting and Assumptions About Causality

    Classification systems also impose boundaries. They delineate what counts as an “incident,” what falls within the scope of patient safety, and where responsibility or causation is presumed to reside. These boundaries are embedded in category definitions, hierarchies, and exclusion criteria [21-24].

    Classification, Sensemaking, and Governance

    Classification influences not only data organization but also sensemaking and governance. Classified data travel across organizational and institutional contexts, informing research syntheses, quality improvement initiatives, and policy decisions [25,26]. As categories are reused and institutionalized, they can stabilize particular narratives about safety problems while marginalizing others. Over time, this can create feedback loops in which certain types of incidents are repeatedly recognized and addressed, while others remain poorly articulated or invisible [25,27,28].

    Inherent Limits of Classificatory Work

    These characteristics of classification, including abstraction, boundary-setting, and stabilization, are not flaws but constitutive features of classificatory work. As such, they introduce limits that cannot be fully eliminated through refinement or expansion of taxonomies [23,25]. Adding categories, increasing granularity, or tailoring classifications to specific domains may improve representation of particular incident features. However, these refinements do not overcome the underlying logic of classification, which continues to stabilize complex sociotechnical processes into discrete, comparable categories [28]. Instead, they redistribute rather than eliminate representational constraints [27].

    In this paper, the term epistemic limits refers to constraints that originate from the fundamental logic of classification itself. These limits differ from design trade-offs, methodological limitations, or pragmatic constraints associated with reporting systems. Design trade-offs involve decisions about how a classification system is constructed or executed, while methodological limitations stem from analytic techniques or data availability. Pragmatic constraints reflect the institutional or operational needs of incident-reporting systems. In contrast, epistemic limits arise from the core features of classification as a knowledge practice, especially abstraction, boundary-setting, and stabilization, which inevitably convert complex, evolving sociotechnical processes into discrete, comparable categories. These limits, therefore, endure across various taxonomies and implementation settings. Figure 1 illustrates the representational trade-offs inherent in patient safety classification. The same structural properties that enable standardized reporting and organizational learning also constrain the representation of complex sociotechnical processes.

    Figure 1 illustrates how abstraction, categorization, and standardization simultaneously enable organizational learning while constraining the representation of complex sociotechnical processes.

    ‎
    Figure 1. Representational trade-offs in patient safety classification.

    Implications for a Sociotechnical Perspective on Patient Safety

    Understanding classification as a representational lens is essential for examining patient safety in complex sociotechnical settings. In such contexts, the tension between the need for standardized representation and the reality of distributed, evolving processes becomes particularly pronounced [18,29,30]. Figure 2 illustrates representational compression in patient safety classification, whereby complex sociotechnical processes are transformed into discrete, standardized incident representations that support recording, categorization, comparison, and analysis.

    Taken together, Figure 2 illustrates the core argument of this paper: the difficulties encountered in applying patient safety classification systems to HIT-related incidents arise from fundamental properties of classification itself. By converting evolving sociotechnical processes into discrete, stabilized representations, classification simultaneously enables learning at scale and imposes limits on what can be known, compared, and acted upon.

    ‎
    Figure 2. Representational compression in patient safety classification. CDS: clinical decision support; EHR: electronic health record.

    Illustrative Classification Traditions and Design Logics

    Patient safety classification systems have been developed across diverse institutional, clinical, and analytic contexts. Although they differ in scope, purpose, and intended users, they share common representational commitments, relying on categorization, abstraction, and standardization to render safety incidents legible for analysis, comparison, and organizational learning [20,30,31]. The following examples illustrate how these shared design logics shape the representation of HIT-related safety problems.

    Global and General Patient Safety Frameworks

    Global patient safety classification efforts aim to establish common conceptual foundations across health care systems. The ICPS was developed to harmonize terminology and incident constructs internationally [23,32,33].

    Such frameworks are deliberately generic, prioritizing breadth and interoperability over contextual specificity. Consequently, they typically conceptualize safety incidents as discrete events that can be decomposed into causal elements and outcomes. Although this design supports aggregation and high-level analysis, it also embeds assumptions about temporality, agency, and system boundaries that become consequential when applied to complex sociotechnical phenomena [1,23,32,33].

    Reporting-Oriented Classification Systems

    Many patient safety classifications are embedded in incident-reporting infrastructures and shaped by practical requirements for data collection, aggregation, and governance. In the United States, the Agency for Healthcare Research and Quality (AHRQ) Common Formats provide standardized categories for reporting adverse events, near misses, and unsafe conditions across care settings [25,34].

    Reporting-oriented systems often emphasize clarity, usability, and consistency, favoring predefined categories that frontline reporters can apply. HIT-related issues are typically incorporated as contributing factors or device-related elements rather than as evolving sociotechnical processes [1,25,34].

    Domain-Specific and Error-Focused Taxonomies

    Some classification systems focus on specific clinical domains or types of harm, such as medication safety. The taxonomy developed by the NCC MERP categorizes medication errors by the stages of the medication-use process and the severity of patient harm [1,5].

    Although not specifically designed for HIT, such systems are frequently used for incidents involving electronic prescribing, decision support, and medication administration technologies. In doing so, they often normalize distinctions between “use error” and system malfunction, reflecting broader classificatory tendencies to localize error within identifiable actions rather than within distributed interactions [3,5].

    HIT-Specific and HIT-Adapted Classification Schemes

    In response to growing recognition of technology-related risk, HIT-specific classification schemes have been developed to capture incidents involving electronic health records, clinical decision support, and other digital tools. These systems typically expand contributing-factor categories to include software design, interface issues, workflow mismatches, and organizational context. They are often explicitly informed by sociotechnical perspectives [10,35-37].

    Despite this specialization, HIT-focused classifications retain core classificatory features shared with broader patient safety taxonomies. Incidents are still represented as bounded events, categorized according to predefined dimensions, and analyzed through stabilized causal structures. Consequently, these systems illuminate how far domain-specific adaptation can go without escaping the underlying logic of classification [10,18,38].

    Recently, classification systems have incorporated more detailed categories, mechanisms for linking related events, and structured narrative components to better capture temporal dynamics and sociotechnical interactions [39,40]. Although these developments improve representation, they do not overcome the representational limitations inherent in classification as a mode of organizing patient safety information.

    Local and Adapted Classification Approaches

    Beyond formal frameworks, many health care organizations and research teams develop local or adapted classification schemes to address perceived gaps in existing systems. These may involve extending categories, reinterpreting definitions, or combining elements from multiple frameworks [6,13,20,41].

    The proliferation of adapted schemes underscores both the flexibility and the limits of classification. Efforts to tailor systems to local sociotechnical realities frequently reintroduce familiar representational challenges, suggesting that these challenges are not confined to particular taxonomies but arise from shared classificatory commitments [6,22,35,36].

    Table 1 summarizes the illustrative classification design logics discussed in this section and highlights the shared representational commitments that shape the rendering of HIT-related safety problems.

    Table 1. Illustrative classification design logics and their representational commitments.
    Classification design logicPrimary unit of analysisAssumptions about causalityTreatment of temporalityTypical handling of HITa
    Global patient safety frameworksDiscrete incidentLinear or decomposableEvent-basedContributing factor
    Reporting-oriented systemsReportable eventAction- or device-linkedSnapshot at report timeEmbedded category
    Domain-specific taxonomiesProcess step or errorLocalized failureStage-basedUse error or malfunction
    HIT-specific schemesIncident with technical contextSociotechnical (bounded)Limited process captureExpanded factors

    aHIT: health information technology.


    Overview

    Patient safety classification systems vary in scope, purpose, and design. However, when applied to HIT-related incidents, they often encounter similar representational difficulties. These difficulties do not arise primarily from isolated design choices or implementation failures. Instead, they reflect recurrent representational tensions that emerge from the structural logic of classification itself [1,18].

    To illustrate the representational challenges discussed in this section, consider a simplified example involving a duplicate medication order generated through CPOE [14]. A physician enters a medication order, but due to interface design and workflow constraints, the system allows the same order to be entered twice. The duplicate order is later detected and corrected before it can cause harm. Depending on the classification system used, this incident might be coded as a medication error, a user-entry mistake, a workflow issue, or a software design problem [14,42,43]. Figure 3 illustrates where patient safety classification systems encounter recurring representational challenges across the life cycle of HIT-related incidents, from system design and clinical use to propagation and recognition of harm.

    ‎
    Figure 3. Representational challenges across the life cycle of health information technology–related incidents.

    Fragmentation of Sociotechnical Interactions

    HIT-related safety incidents typically emerge from interactions among multiple system components, including software functionality, clinical workflows, organizational routines, and professional judgment. Classification systems, however, require incidents to be decomposed into discrete elements, such as event types, contributing factors, or actor roles, to render them categorizable [18,44]. In the CPOE duplicate-order example described previously, the incident could be attributed to user error, interface design, or workflow conditions, depending on the classificatory structure, illustrating how sociotechnical interactions become fragmented across multiple categories [14,42,43].

    This decomposition breaks sociotechnical interactions into separate components. Consequently, the relational dynamics through which risk emerges may become less visible. Interdependencies between technology and practice are often redistributed across multiple categories, weakening the visibility of interaction effects. Consequently, the classified representation may suggest a set of parallel contributing factors rather than a coupled sociotechnical process [45,46].

    This fragmentation is not an error of representation but a consequence of classificatory abstraction. Classification systems are designed to isolate elements that can be consistently named and compared, even when the phenomena of interest are fundamentally relational. For HIT-related incidents, this leads to representations that understate the role of interaction and overstate the independence of contributing components [18,26,46].

    Loss of Temporality and Process

    Many HIT-related safety problems unfold over extended periods. Design decisions made during system configuration may interact with evolving patterns of use, work-arounds, and adaptation before contributing to harm that becomes apparent only much later. Classification systems, however, typically represent incidents as temporally bounded events, anchored to a moment of recognition or reporting [15,46,47]. In the duplicate-order example, the classification record may capture the moment the error was detected, while the preceding sequence of interface interactions and workflow adaptations that enabled the duplication remains largely invisible [14,42,43].

    This event-based representation collapses trajectories into snapshots. Temporal sequencing, latency, and cumulative effects are difficult to encode within categorical structures that privilege discrete occurrences. Even when temporal fields are included, they rarely capture how risks develop, propagate, and transform over time [47].

    The loss of temporality limits classification systems’ ability to represent HIT-related incidents as processes rather than outcomes. Consequently, downstream analysis may focus on proximal events while obscuring the longer-term dynamics through which sociotechnical risk accumulates [15,47].

    Several patient safety classification systems have attempted to address these issues by introducing mechanisms intended to capture temporal and relational dynamics [48,49]. Although these features improve representation, they do not eliminate the underlying constraint that classification ultimately requires incidents to be stabilized into discrete, comparable categories.

    Poor Handling of Scale and Propagation

    HIT-related incidents frequently propagate beyond the local context in which they originate. A configuration error, interface design issue, or workflow mismatch may affect multiple users, units, or care settings, with effects that scale unevenly and interact with local practices [46,47]. If similar configurations or interface conditions exist across multiple clinical units, duplicate orders may recur, yet classification systems may record them as independent incidents rather than manifestations of a shared systemic issue [12,50].

    Classification systems, however, are typically organized around localized incident units. They privilege representations that tie incidents to specific locations, actors, or moments, making it difficult to capture cross-boundary propagation or system-wide effects. When scale is represented, it is often treated as an attribute of individual incidents rather than as a dynamic property of sociotechnical systems [1,51].

    This mismatch between localized incident units and distributed effects constrains the conceptual reach of classification-based data. Propagation is rendered as repetition rather than as a feature of system structure, limiting insight into how HIT-related risks amplify across contexts [46,47].

    Normalization of “Use Error”

    A recurrent feature of patient safety classification systems is the distinction between technical failure and human or “use” error. In HIT contexts, this distinction often becomes a default mode of representation, particularly when classification categories prioritize observable user actions over design-use coupling [11,52,53].

    By treating “use error” as a categorical outcome, classification systems may normalize interpretations that assign responsibility to individual behavior, even when user actions are shaped by interface design, workflow constraints, or organizational pressures. Over time, this can reinforce a conceptual framework that treats sociotechnical misalignment as user deviation rather than as a property of system design [10,11,52].

    This normalization effect does not require explicit attribution of blame. It emerges from the repeated use of categories that foreground action over interaction and outcome over process. As such, it reflects a deeper classificatory tendency to individualize causality in contexts where agency is distributed [1,38,51].

    Representational Limitations Are Structural, Not System-Specific

    Taken together, these breakdowns illustrate that the difficulties classification systems encounter with HIT-related incidents are not confined to particular taxonomies or domains. Fragmentation, loss of temporality, scale mismatch, and normalization of use error recur across classification approaches precisely because they arise from shared classificatory commitments [54,55].

    Efforts to address these issues through increased granularity or domain-specific adaptation may shift how representational limitations appear, but they do not eliminate the underlying tension between standardized representation and sociotechnical complexity [1,38,51].

    The increasing use of large language models (LLMs) and natural language processing (NLP) for analyzing unstructured patient safety reports does not fundamentally eliminate the epistemic limits discussed in this Viewpoint. Although these technologies may improve the extraction, organization, and classification of information from free-text narratives, they ultimately produce structured representations that remain subject to abstraction, boundary-setting, and representational compression [40,54,56]. Consequently, automated text classification may enhance the efficiency and consistency of incident analysis, but it does not overcome the inherent limits of classification as a knowledge practice. Instead, it may automate and scale the same structural constraints that shape how complex sociotechnical incidents are represented and interpreted.

    Table 2 summarizes the 4 recurring representational limitations discussed in this Viewpoint and their implications for the classification of HIT-related incidents.

    Table 2. Recurring representational limitations in classifying health information technology–related incidents.
    Representational limitationsWhat is lost or distorted?Why does classification produce this limit?
    Sociotechnical fragmentationInteractions and dependenciesForced attribution to components
    Loss of temporalityIncident trajectoriesEvent-based representation
    Scale and propagationSystem-wide effectsLocalized incident units
    Normalization of use errorDesign-use couplingPerson-centered categories

    Overview

    The representational constraints described in the previous section have important consequences for how patient safety data are interpreted, compared, and acted upon in both research and practice [1,57].

    Classification-based incident data rarely function as the sole basis for safety improvement. Instead, they typically serve as starting points for root cause analysis, sociotechnical assessment, and organizational learning. In HIT, these investigations are often informed by sociotechnical frameworks such as the SAFER (Safety Assurance Factors for Electronic Health Record Resilience) Guides and related evaluation methods for electronic health record safety [18,56,58].

    Misleading Learning and Partial Explanations

    Classification systems support learning by enabling aggregation and comparison across incidents. However, when HIT-related safety problems are represented as fragmented, temporally bounded events, the resulting learning may be partial or misleading. Aggregated patterns can obscure the sociotechnical interactions and trajectories through which harm emerges, encouraging interpretations that privilege proximal causes over longer-term or distributed dynamics [1,25,57]. When such interactions are flattened into discrete categories, learning may focus on symptoms rather than on the structural conditions that shape risk [1,25,57].

    Misaligned Interventions and Corrective Actions

    The way incidents are classified influences not only interpretation but also intervention. When classified representations foreground individual actions or isolated failures, they can channel responses toward training, compliance, or local fixes, even when underlying issues are rooted in system design or organizational structure [30,47].

    In HIT contexts, interventions derived from classification-based analyses may therefore focus on user behavior while leaving design-use mismatches or workflow constraints unaddressed. Over time, this can lead to repeated corrective efforts that fail to engage with the sources of sociotechnical risk, reinforcing cycles of reporting without commensurate improvement [30,46].

    Comparability Problems Across Studies and Settings

    One of the primary motivations for classification is to enable comparability across studies, organizations, and jurisdictions. However, when HIT-related incidents are forced into stabilized categories that inadequately capture sociotechnical complexity, apparent comparability may mask substantive differences in how incidents unfold across contexts [22,33].

    Differences in local workflows, system configurations, and organizational practices can lead to divergent interpretations of similar classified categories. Consequently, studies drawing on classified incident data may appear comparable at the level of codes while differing substantially in what those codes represent in practice. This complicates synthesis across studies and limits the interpretability of comparative findings in HIT-related patient safety research [22,23,45].

    Consequences as Structural, Not Correctable by Refinement

    Taken together, these consequences indicate that the challenges associated with classifying HIT-related incidents are structural rather than contingent. Although greater granularity, additional categories, and domain-specific adaptations may improve representation, they do not eliminate the fundamental trade-offs inherent in classification-based representation [18,22].

    Recognizing these consequences is not an argument against classification but a call for interpretive caution. Classification systems remain indispensable for surveillance, aggregation, and governance, yet their outputs should be understood as partial representations shaped by underlying design logics. Without this perspective, classification-based knowledge may be overextended beyond its representational limits [18,22].


    Overview

    Recognizing the structural limits of classification systems has important implications for how future analyses of HIT-related patient safety are conceptualized and interpreted. These implications are particularly relevant for researchers analyzing patient safety datasets, HIT safety specialists interpreting incident reports, and health care organizations using classification-based data to guide quality improvement or governance activities [1]. Rather than seeking a single, more comprehensive classification framework, future research may benefit from approaches that explicitly acknowledge the partial and situated nature of classificatory representations [25,59]. Classification systems are best viewed as one part of a larger safety analysis ecosystem rather than as complete representations of complex sociotechnical events.

    Why No Single Framework Is Sufficient

    The recurrent representational constraints discussed in this Viewpoint suggest that no single classification framework can fully capture the complexity of HIT-related safety incidents. Sociotechnical phenomena are characterized by interaction, emergence, and temporal evolution, whereas classification systems necessarily stabilize phenomena into bounded, comparable units. This tension cannot be resolved through additional categories or greater granularity alone [1,51,60]. Consequently, classification-based data should be interpreted as partial representations of complex sociotechnical processes. Although such data remain essential for identifying patterns of harm, prioritizing interventions, and comparing safety performance across settings, they should be interpreted alongside complementary approaches that preserve temporal, organizational, and sociotechnical context.

    Efforts to design more inclusive or detailed taxonomies may improve the visibility of certain dimensions of risk, but they simultaneously introduce new boundary problems and interpretive constraints. Consequently, attempts to “solve” classificatory limitations by expanding the framework risk reproducing the same representational trade-offs in different forms. These observations suggest the value of a more reflexive interpretation of classification-based safety data, particularly in sociotechnical contexts [1,51,60]. Recognizing the limits of classification can help analysts interpret incident data more cautiously, particularly when drawing conclusions about causality, system-level risk, or the effectiveness of interventions.

    The Need for Reflexive Use of Classification Data

    Future patient safety analyses that rely on classified incident data should adopt a reflexive stance toward the categories they use. Rather than treating classifications as transparent representations of underlying events, analysts should consider how category definitions, incident boundaries, and causal structures shape what becomes visible and actionable [18,51]. Without this reflexive orientation, there is a risk that classification outputs will be overinterpreted as direct reflections of sociotechnical reality [26,45,60].

    Layered and Complementary Analytic Perspectives

    Given the limits of any single representational approach, future research on HIT-related patient safety may benefit from layered perspectives that combine classification-based data with complementary approaches. These may include qualitative inquiry, sociotechnical modeling, process tracing, or narrative analysis, each of which foregrounds dimensions of interaction, temporality, context, and emergence that classification tends to suppress [1,51,61].

    Classification systems may remain central for surveillance and aggregation, while complementary approaches are used to explore interaction, temporality, and emergence. Together, these perspectives can support a more nuanced understanding of sociotechnical risk without relying on a single conceptual framework [1,51,61].

    From Representational Limits to Analytic Awareness

    The implications of this Viewpoint are not methodological prescriptions but a call for more reflexive interpretation of classification-based evidence. By making the representational limits of classification explicit, future research can better align its interpretations and inferences with what classification-based data can reasonably support. This awareness can help prevent overgeneralization, misplaced attribution, and premature conclusions about causality or the effectiveness of interventions [1,51,62].

    Treating classification as one analytic lens among many, rather than as a definitive account of safety, may enable more responsible and context-sensitive interpretations of HIT-related patient safety data [1,51,62].


    Patient safety classification systems have become indispensable for surveillance, organizational learning, research, and governance because they provide a standardized means of organizing complex safety information. However, I argue that their value should not obscure an important limitation: classification is not simply a neutral mechanism for recording events but a representational practice that shapes how safety problems are perceived, interpreted, and addressed [13,51]. This distinction is particularly important for HIT-related patient safety incidents, which emerge through dynamic interactions among technologies, clinical workflows, organizational contexts, and human decision-making rather than as discrete, isolated events [35,36].

    From this perspective, the recurring representational limitations discussed in this Viewpoint should not be understood as shortcomings of particular classification systems or reporting infrastructures. Rather, they reflect the inherent logic of classification itself. Processes such as abstraction, boundary-setting, and standardization make comparison and aggregation possible, but they also simplify the complexity of sociotechnical systems. Consequently, aspects of HIT-related incidents, including evolving interactions, temporal dynamics, distributed causation, and contextual influences, may be compressed or excluded during classification. These trade-offs are intrinsic to representational systems and cannot be eliminated solely by developing more detailed or domain-specific taxonomies.

    Recognizing these representational limits has important implications for patient safety research and practice. Classification-based datasets are often interpreted as comprehensive accounts of safety events and are subsequently used to inform organizational learning, quality improvement, policy development, and regulatory decision-making. However, when classified data are treated as complete representations of underlying events, they may support only a partial understanding of complex safety problems. This can contribute to interventions that address visible categories of harm while overlooking the sociotechnical conditions that produced them, and it may complicate comparisons across organizations and health care settings where contextual factors differ substantially.

    My argument is therefore not that patient safety classification systems should be replaced or that more comprehensive classification schemes will fully resolve these challenges. Greater granularity and HIT-specific refinement can undoubtedly improve representation, but they cannot eliminate the fundamental trade-offs inherent in any attempt to translate complex, evolving practices into stable categories. Instead, classification should be understood as one conceptual perspective among several, with its outputs interpreted alongside complementary approaches that preserve temporal, organizational, and sociotechnical context.

    Although this Viewpoint focuses on HIT-related patient safety, the underlying argument extends more broadly to contemporary health care. As digital technologies become increasingly embedded within clinical practice and health systems continue to rely on standardized data infrastructures for surveillance, benchmarking, accountability, and learning, similar representational tensions are likely to emerge wherever complex practices are reduced to standardized forms of knowledge. Appreciating these tensions is therefore essential not only for interpreting patient safety data but also for understanding the broader strengths and limitations of data-driven governance in health care.

    The contribution of this Viewpoint is not to advocate for another patient safety classification framework. Rather, it is to clarify the epistemic boundaries within which classification-based patient safety data should be interpreted. By making the limits of classification explicit, I encourage a more reflexive use of classification systems that acknowledges both their indispensable role in surveillance, organizational learning, and governance and their inability to fully capture the complexity of HIT-related patient safety. Such an approach supports a more balanced interpretation of classification-based evidence and reinforces the importance of combining standardized classification with complementary sociotechnical perspectives when investigating, learning from, and responding to digital health safety incidents.

    Conclusions

    This Viewpoint argues that patient safety classification systems are indispensable for organizing, interpreting, and learning from adverse events and near misses. However, when applied to HIT-related incidents, they encounter recurring representational limitations that cannot be fully overcome through refinement or domain-specific adaptation. These limitations arise from the inherent properties of classification itself (ie, abstraction, boundary-setting, and stabilization), which transform dynamic sociotechnical processes into stable, comparable categories.

    Viewing classification as a representational lens, rather than as a neutral container of safety information, helps explain why fragmentation of sociotechnical interactions, loss of temporality, inadequate representation of scale and propagation, and normalization of use error recur across different classificatory traditions. These limitations are not shortcomings of individual classification systems but predictable consequences of representing complex, evolving sociotechnical phenomena through standardized categories.

    The contribution of this Viewpoint is not to advocate for another classification framework but to clarify the epistemic boundaries within which classification-based patient safety data should be interpreted. Recognizing both the strengths and the representational limits of classification supports a more reflexive use of classification-based evidence while reinforcing the indispensable role of classification systems in surveillance, organizational learning, research, and governance.

    Acknowledgments

    The author would like to thank colleagues and health care professionals whose prior discussions and insights into health information technology–related patient safety informed the development of this Viewpoint. During the preparation of this manuscript, the author used ChatGPT (GPT-5.5; OpenAI) to assist with language refinement, editing, and improving the clarity and readability of the manuscript. The author takes full responsibility for the content of the manuscript and has reviewed and verified all outputs generated by the AI tool.

    Data Availability

    No new data were generated or analyzed in this study. This Viewpoint paper is informed by previously published studies and the author’s prior research. As such, data sharing is not applicable.

    Funding

    This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

    Authors' Contributions

    The author was solely responsible for the conception, conceptual development, literature review and synthesis, and writing of the manuscript.

    Conflicts of Interest

    None declared.

    1. Jabin MSR. Identifying and Characterising Problems Arising From Interactions Between Medical Imaging and Health Information Technology as a Basis for Improvements in Practice [Phd Thesis]. Adelaide, Australia. University of South Australia; 2019.
    2. Wyatt KD, Benning TJ, Morgenthaler TI, Arteaga GM. Development of a taxonomy for medication-related patient safety events related to health information technology in pediatrics. Appl Clin Inform. Oct 2020;11(5):714-724. [FREE Full text] [CrossRef] [Medline]
    3. Brixey J, Johnson TR, Zhang J. Evaluating a medical error taxonomy. Proc AMIA Symp. 2002:71-75. [FREE Full text] [Medline]
    4. Klemp K, Dovey S, Valderas JM, Rohe J, Godycki-Cwirko M, Elliott P, et al. Developing a patient safety incident classification system for primary care. A literature review and Delphi-survey by the LINNEAUS collaboration on patient safety in primary care. Eur J Gen Pract. Sep 2015;21 Suppl(sup1):35-38. [FREE Full text] [CrossRef] [Medline]
    5. Zhu M, Huang Y, Gong Y, Brixey JJ, Zhang J, Turley JP. Mapping concepts in medical error taxonomies. AMIA Annu Symp Proc. 2006;2006:1159. [FREE Full text] [Medline]
    6. Taheri Moghadam S, Sheikhtaheri A, Hooman N. Patient safety classifications, taxonomies and ontologies, part 2: a systematic review on content coverage. J Biomed Inform. Dec 2023;148:104549. [FREE Full text] [CrossRef] [Medline]
    7. Roggow BJ, Zarei E. Human factors analysis and classification system (HFACS): development, tiers, nanocodes, application, and adaptations. In: Safety Causation Analysis in Sociotechnical Systems: Advanced Models and Techniques. Cham, Switzerland. Springer; Jul 2024:151-180.
    8. Holden RJ, Carayon P, Gurses AP, Hoonakker P, Hundt AS, Ozok AA, et al. SEIPS 2.0: a human factors framework for studying and improving the work of healthcare professionals and patients. Ergonomics. 2013;56(11):1669-1686. [FREE Full text] [CrossRef] [Medline]
    9. Shappell SA, Wiegmann DA. The human factors analysis and classification system–HFACS. U.S. Department of Transportation, Federal Aviation Administration. 2000. URL: https://commons.erau.edu/publication/737/ [accessed 2026-08-18]
    10. Magrabi F, Liaw ST, Arachi D, Runciman W, Coiera E, Kidd MR. Identifying patient safety problems associated with information technology in general practice: an analysis of incident reports. BMJ Qual Saf. Nov 2016;25(11):870-880. [CrossRef] [Medline]
    11. Magrabi F, Baker M, Sinha I, Ong MS, Harrison S, Kidd MR, et al. Clinical safety of England's national programme for IT: a retrospective analysis of all reported safety events 2005 to 2011. Int J Med Inform. Mar 2015;84(3):198-206. [CrossRef] [Medline]
    12. Pan D, Nilsson E, Rahman Jabin MS. A review of incidents related to health information technology in Swedish healthcare to characterise system issues as a basis for improvement in clinical practice. Health Informatics J. 2024;30(3):14604582241270742. [FREE Full text] [CrossRef] [Medline]
    13. Jabin MS. A qualitative, multi-framework methodology for analysing health information technology-related patient safety incidents. Front Digit Health. May 29, 2026;8:1786358. [FREE Full text] [CrossRef] [Medline]
    14. Horsky J, Kuperman GJ, Patel VL. Comprehensive analysis of a medication dosing error related to CPOE. J Am Med Inform Assoc. 2005;12(4):377-382. [FREE Full text] [CrossRef] [Medline]
    15. Koppel R, Metlay JP, Cohen A, Abaluck B, Localio AR, Kimmel SE, et al. Role of computerized physician order entry systems in facilitating medication errors. JAMA. Mar 09, 2005;293(10):1197-1203. [CrossRef] [Medline]
    16. Carayon P, Schoofs Hundt A, Karsh BT, Gurses AP, Alvarado CJ, Smith M, et al. Work system design for patient safety: the SEIPS model. Qual Saf Health Care. Dec 2006;15 Suppl 1(Suppl 1):i50-i58. [FREE Full text] [CrossRef] [Medline]
    17. Carayon P, Wooldridge A, Hoonakker P, Hundt AS, Kelly MM. SEIPS 3.0: human-centered design of the patient journey for patient safety. Appl Ergon. Apr 2020;84:103033. [FREE Full text] [CrossRef] [Medline]
    18. Sittig DF, Singh H. A new sociotechnical model for studying health information technology in complex adaptive healthcare systems. Qual Saf Health Care. Oct 2010;19 Suppl 3(Suppl 3):i68-i74. [FREE Full text] [CrossRef] [Medline]
    19. Bowker GC, Star SL. Sorting Things Out: Classification and Its Consequences. Cambridge, MA. MIT Press; 2000.
    20. Taheri Moghadam S, Hooman N, Sheikhtaheri A. Patient safety classifications for Health Information Technology (HIT) and medical devices: a review on available systems. Stud Health Technol Inform. May 16, 2022;293:153-160. [CrossRef] [Medline]
    21. Conceptual framework for the international classification for patient safety: final technical report. World Health Organization. Jan 15, 2009. URL: https://www.who.int/publications/i/item/WHO-IER-PSP-2010.2 [accessed 2026-08-17]
    22. Lorbeer ER. Sorting things out: classification and its consequences. Issues Sci Technol Librariansh. 2000;(25). [FREE Full text] [CrossRef]
    23. Runciman W, Hibbert P, Thomson R, Van Der Schaaf T, Sherman H, Lewalle P. Towards an International Classification for Patient Safety: key concepts and terms. Int J Qual Health Care. Feb 2009;21(1):18-26. [FREE Full text] [CrossRef] [Medline]
    24. Reason J. Human error: models and management. BMJ. Mar 18, 2000;320(7237):768-770. [FREE Full text] [CrossRef] [Medline]
    25. Vincent C. Incident reporting and patient safety. BMJ. Jan 13, 2007;334(7584):51. [CrossRef] [Medline]
    26. Braithwaite J, Runciman WB, Merry AF. Towards safer, better healthcare: harnessing the natural properties of complex sociotechnical systems. Qual Saf Health Care. Feb 2009;18(1):37-41. [FREE Full text] [CrossRef] [Medline]
    27. Waring JJ. Beyond blame: cultural barriers to medical incident reporting. Soc Sci Med. May 2005;60(9):1927-1935. [CrossRef] [Medline]
    28. Weick KE. Sensemaking in Organizations. Thousand Oaks, CA. SAGE Publications; 1995.
    29. Perrow C. Normal Accidents: Living with High Risk Technologies. New York, NY. Basic Books; 1984.
    30. Ash JS, Berg M, Coiera E. Some unintended consequences of information technology in health care: the nature of patient care information system-related errors. J Am Med Inform Assoc. 2004;11(2):104-112. [FREE Full text] [CrossRef] [Medline]
    31. Runciman B, Merry A, Walton M. Safety and Ethics in Healthcare: A Guide to Getting it Right. London, UK. CRC Press; 2007.
    32. Thomson R, Lewalle P, Sherman H, Hibbert P, Runciman W, Castro G. Towards an International Classification for Patient Safety: a Delphi survey. Int J Qual Health Care. Feb 2009;21(1):9-17. [FREE Full text] [CrossRef] [Medline]
    33. World Alliance For Patient Safety Drafting Group, Sherman H, Castro G, Fletcher M, World Alliance for Patient Safety, Hatlie M, et al. Towards an International Classification for Patient Safety: the conceptual framework. Int J Qual Health Care. Feb 2009;21(1):2-8. [FREE Full text] [CrossRef] [Medline]
    34. Device or medical/surgical supply dashboard. Agency for Healthcare Research and Quality. 2025. URL: https://www.ahrq.gov/npsd/data/dashboard/devices.html [accessed 2026-08-17]
    35. Jabin MS. Why health information technology safety problems remain invisible. Front Digit Health. May 15, 2026;8:1785141. [FREE Full text] [CrossRef] [Medline]
    36. Jabin MS. Why digital health fails silently: a sociotechnical theory of health information technology-related risk. Front Digit Health. May 15, 2026;8:1785086. [FREE Full text] [CrossRef] [Medline]
    37. Jabin MS. Health information technology-related loss of central surveillance data in a heart intensive care unit: multi-framework case report. JMIR Hum Factors. Apr 09, 2026;13:e92560. [FREE Full text] [CrossRef] [Medline]
    38. Coiera E. Information system safety. In: Guide to Health Informatics. London, UK. CRC Press; 2015.
    39. Gqaleni TM, Mkhize SW, Chironda G. Patient safety incident reporting and learning guidelines implemented by health care professionals in specialized care units: scoping review. J Med Internet Res. Oct 04, 2024;26:e48580. [FREE Full text] [CrossRef] [Medline]
    40. Leistikow I, Mulder S, Vesseur J, Robben P. Learning from incidents in healthcare: the journey, not the arrival, matters. BMJ Qual Saf. Mar 2017;26(3):252-256. [FREE Full text] [CrossRef] [Medline]
    41. Runciman WB, Williamson JA, Deakin A, Benveniste KA, Bannon K, Hibbert PD. An integrated framework for safety, quality and risk management: an information and incident management system based on a universal patient safety classification. Qual Saf Health Care. Dec 2006;15 Suppl 1(Suppl 1):i82-i90. [FREE Full text] [CrossRef] [Medline]
    42. Collin S, Reeves BC, Hendy J, Fulop N, Hutchings A, Priedane E. Implementation of computerised physician order entry (CPOE) and picture archiving and communication systems (PACS) in the NHS: quantitative before and after study. BMJ. Aug 14, 2008;337:a939. [FREE Full text] [CrossRef] [Medline]
    43. Thomas S, Fylan B, Silcock J, Jabin MS. Medication management using technology-based interventions in older people at home care settings: protocol for a scoping review. JMIR Res Protoc. Oct 03, 2025;14:e75200. [FREE Full text] [CrossRef] [Medline]
    44. Castro GM, Buczkowski L, Hafner JM. The contribution of sociotechnical factors to health information technology-related sentinel events. Jt Comm J Qual Patient Saf. Feb 2016;42(2):70-76. [CrossRef] [Medline]
    45. Aarts J. Towards safe electronic health records: a socio-technical perspective and the need for incident reporting. Health Policy Technol. Mar 2012;1(1):8-15. [CrossRef]
    46. Harrison MI, Koppel R, Bar-Lev S. Unintended consequences of information technologies in health care--an interactive sociotechnical analysis. J Am Med Inform Assoc. 2007;14(5):542-549. [FREE Full text] [CrossRef] [Medline]
    47. Koppel R, Wetterneck T, Telles JL, Karsh BT. Workarounds to barcode medication administration systems: their occurrences, causes, and threats to patient safety. J Am Med Inform Assoc. 2008;15(4):408-423. [FREE Full text] [CrossRef] [Medline]
    48. Hibbert PD, Molloy CJ, Hooper TD, Wiles LK, Runciman WB, Lachman P, et al. The application of the Global Trigger Tool: a systematic review. Int J Qual Health Care. Dec 01, 2016;28(6):640-649. [CrossRef] [Medline]
    49. Pham JC, Girard T, Pronovost PJ. What to do with healthcare incident reporting systems. J Public Health Res. Dec 01, 2013;2(3):e27. [FREE Full text] [CrossRef] [Medline]
    50. Jabin MS, Wepa D, Hassoun A. A case report of system configuration issue in medical imaging due to system upgrade- changes in hardware and software. Front Digit Health. Sep 2, 2024;6:1371761. [FREE Full text] [CrossRef] [Medline]
    51. Jabin MS. The need for a refined classification system and national incident reporting system for health information technology-related incidents. Front Digit Health. Jul 26, 2024;6:1422396. [CrossRef] [Medline]
    52. Magrabi F, Ong MS, Runciman W, Coiera E. An analysis of computer-related patient safety incidents to inform the development of a classification. J Am Med Inform Assoc. 2010;17(6):663-670. [FREE Full text] [CrossRef] [Medline]
    53. Sittig DF, Wright A, Coiera E, Magrabi F, Ratwani R, Bates DW, et al. Current challenges in health information technology-related patient safety. Health Informatics J. Mar 2020;26(1):181-189. [FREE Full text] [CrossRef] [Medline]
    54. Coiera E, Magrabi F, Talmon J. Engineering technology resilience through informatics safety science. J Am Med Inform Assoc. Mar 01, 2017;24(2):244-245. [FREE Full text] [CrossRef] [Medline]
    55. Magrabi F, Ong MS, Runciman W, Coiera E. Using FDA reports to inform a classification for health information technology safety problems. J Am Med Inform Assoc. 2012;19(1):45-53. [FREE Full text] [CrossRef] [Medline]
    56. Singh H, Sittig DF. Measuring and improving patient safety through health information technology: the Health IT Safety framework. BMJ Qual Saf. Apr 2016;25(4):226-232. [FREE Full text] [CrossRef] [Medline]
    57. Jones SS, Rudin RS, Perry T, Shekelle PG. Health information technology: an updated systematic review with a focus on meaningful use. Ann Intern Med. Jan 07, 2014;160(1):48-54. [FREE Full text] [CrossRef] [Medline]
    58. SAFER guides. Office of the National Coordinator for Health Information Technology. URL: https://healthit.gov/clinical-quality-and-safety/safer-guides/ [accessed 2026-03-11]
    59. Vincent C, Burnett S, Carthey J. Safety measurement and monitoring in healthcare: a framework to guide clinical teams and healthcare organisations in maintaining safety. BMJ Qual Saf. Aug 2014;23(8):670-677. [FREE Full text] [CrossRef] [Medline]
    60. Capra F, Luisi PL. The Systems View of Life: A Unifying Vision. Cambridge, MA. Cambridge University Press; 2014.
    61. Alsaqqa HH. Healthcare organizations management: analyzing characteristics, features and factors, to identify gaps "scoping review". Health Serv Insights. May 03, 2023;16:11786329231168130. [FREE Full text] [CrossRef] [Medline]
    62. Girard MJ, Lapides J, Roe CM. The fifth discipline: a systems learning model for building high-performing learning organizations. In: Handbook of Human Performance Technology: Principles, Practices, and Potential. Hoboken, NJ. John Wiley & Sons; 2006.

    ‎
    AHRQ: Agency for Healthcare Research and Quality
    CPOE: computerized physician order entry
    HIT: health information technology
    HIT-CS: Health Information Technology Classification System
    ICPS: International Classification for Patient Safety
    LINNAEUS: Learning in an International Network About Errors and Understanding Safety
    LLM: large language model
    NCC MERP: National Coordinating Council for Medication Error Reporting and Prevention
    NLP: natural language processing
    SAFER: Safety Assurance Factors for Electronic Health Record Resilience
    WHO: World Health Organization

    Edited by S Law; submitted 20.Jan.2026; peer-reviewed by L Faria, K Fuji; comments to author 10.Mar.2026; revised version received 28.Jul.2026; accepted 13.Aug.2026; published 28.Aug.2026.

    Copyright

    ©Md Shafiqur Rahman Jabin. Originally published in JMIR Human Factors (https://humanfactors.jmir.org), 28.Aug.2026.

    This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Human Factors, is properly cited. The complete bibliographic information, a link to the original publication on https://humanfactors.jmir.org, as well as this copyright and license information must be included.