Near Miss Reporting and Organizational Learning in Health Care: Conceptual Framework Development Study

Introduction

Background

Near miss events, which reveal systemic hazards before harm occurs, are fundamental to proactive safety governance. Global health policies now emphasize their systematic analysis, a principle operationalized in frameworks such as the National Health Service England’s Patient Safety Incident Response Framework [1-3]. Consequently, leaders must treat near miss data as a crucial leading indicator of systemic safety rather than as a secondary metric [1].

Effective safety oversight requires analyzing trends rather than isolated event counts. Best practices advocate interpreting safety indicators (eg, rates of near misses, medication errors, patient falls, or infection events) alongside workforce culture and patient experience metrics to discern meaningful signals from statistical noise, thereby strengthening governance [4,5]. However, translating this into practice is fraught with methodological challenges. Voluntary reporting systems suffer from significant underascertainment due to reporting culture, workload, and fear [6,7]. These barriers disproportionately suppress near miss submissions compared to harm incidents, distorting data and undermining valid comparisons [8-10].

Analytical choices add further uncertainty. Raw event counts are confounded by patient volume and reporting behavior, while meaningful rates require domain-aligned denominators [4]. Furthermore, overdispersion is common in health care data, meaning that conventional statistical process control (SPC) charts can produce false signals if used without adjustment [11]. Although corrective methods such as the Laney p’ chart exist [12], they have not been embedded within a structured near miss analytic framework that links SPC to event classification, severity weighting, and organizational learning—a gap this study addresses.

Objectives

These challenges create a critical interpretive dilemma for leaders: does a rise in reported near misses signify deteriorating safety or an improving reporting culture? Without standardized terminology, risk-adjusted denominators, and dispersion-robust control charts, leaders risk misinterpreting these vital data, misallocating resources, and failing to address systemic vulnerabilities [1,4,13,14].

From a human factors perspective, this interpretive dilemma reflects a classic problem of decision-making under uncertainty within complex sociotechnical systems [15]. Leaders must derive meaning from incomplete, socially mediated, and variably coded safety signals. In the absence of structured analytic support, interpretation depends on informal sensemaking [16] and locally constructed mental models [17], increasing the risk of inconsistent prioritization and resource allocation.

Therefore, the objective of this study was to develop a conceptual framework that provides a structured, evidence-based methodology for near miss analysis. To achieve this, we designed a 3-level maturity model that integrates standardized terminology, SPC, severity weighting, and novel learning metrics. This structured approach aims to provide organizational leaders with reliable, decision-grade intelligence to guide and sustain safety improvement initiatives. The 3-level near miss framework (NM³) does not treat reporting frequency as a safety outcome; rather, it provides a structured analytic pathway that transforms near miss data into interpretable, learning-oriented intelligence for governance and improvement.

Methods

Design and Approach

This study reports the development of a conceptual framework for analyzing near miss safety events, following the integrative methodology by Jabareen [18]. This method proceeds through a series of iterative phases: mapping selected data sources, reading and categorizing the data, identifying and naming concepts, deconstructing and categorizing those concepts, integrating them into a coherent framework, and validating the result through an iterative process of rethinking and synthesis. The approach constructs interrelated concepts by synthesizing theory and evidence rather than through empirical data collection. The aim was to create a structured, 3-level model for near miss analytics that offers decision-grade intelligence to support organizational safety governance. The process was exclusively analytic and integrative and did not involve human participants.

Evidence and Theoretical Sources

The framework integrates principles from peer-reviewed research and authoritative institutional guidance. To align with established international standards, its foundational definitions were derived from the World Health Organization (WHO) International Classification for Patient Safety (ICPS) and Agency for Healthcare Research and Quality (AHRQ) Common Formats for Event Reporting–Hospital (version 2.0) [19,20]. The analytical components were informed by literature on the application of SPC in health care, including tutorials on chart construction and reviews of its effectiveness [20-22]. Methods for weighting event severity were adapted from the Institute for Healthcare Improvement (IHI) Severity Assessment Code (SAC) Matrix. This tool is widely used in root cause analysis [23]. To account for known limitations in reporting data, the design was informed by recent systematic reviews that identify common barriers to near miss reporting [6,8]. Conclusively, to ensure contemporary policy relevance, the framework incorporates strategic priorities for safety measurement from key governance documents issued by the WHO and National Health Service England [1,3,19]. Only peer-reviewed publications and official institutional reports were used in the synthesis.

Framework Construction

Overview

The framework was constructed as a staged maturity model with 3 progressive levels, each representing greater analytic depth for interpreting near miss data. Construction followed a sequence of definitional alignment, operational specification, and integration into maturity levels. International taxonomies provided the definitional base, operational rules were derived from methodological literature on SPC and risk assessment, and maturity staging was informed by organizational safety models that emphasize progressive capacity building [18,19,24]. The 3 levels are summarized in Table 1, which presents their scope, core components, and associated formulas. These definitions and operational rules are further detailed and operationalized in Table 2 to ensure direct alignment between theoretical sources and the applied framework design. Together, these tables codify the design specifications of the framework and ensure reproducibility in application.

Level 1: Essential Minimum

The first level establishes a standardized foundation for organization-wide surveillance of near miss events. At this stage, events are classified using the WHO ICPS and the AHRQ’s Common Formats, which define the categories of near miss, no-harm incident, and harmful incident [20,25]. Each event is recorded with an indication of whether it reached the patient and with specification of the detection mode, classified as human intervention, automated alarm, patient self-report, or chance discovery. These data elements are defined in Table 2, which provides the operational specifications required at all 3 levels.

For rate construction, the denominator is hospital patient-days, consistent with conventions used for monitoring hospital-acquired conditions such as falls and infections [26]. The calculations specified for this level, summarized in Table 1, are the near miss rate, defined as the number of near misses per 1000 patient-days, and the reported near miss–to-harm ratio, defined as the number of reported near misses divided by the number of reported harmful incidents. Temporal analysis is performed using u-charts, which accommodate variable denominators, and the Laney U’ modification is applied when overdispersion is identified, ensuring stable control limits and valid interpretation of signals [21,26,31].

Level 2: Targeted Stratification

The second level adds analytic depth by applying stratification to high-risk or high-volume domains in which the opportunity for near misses is greatest. The framework specifies denominators that align with domain-specific exposures (Table 1): for intensive care, near misses per 1000 intensive care unit patient-days; in pharmacy services, near misses per 10,000 medication doses; in laboratory settings, near misses per 1000 specimens; in transfusion services, near misses per 1000 blood units; and in procedural areas, near misses per 1000 surgeries or diagnostic studies. These denominators are drawn from existing quality indicator conventions and are further defined in Table 2 to ensure consistent application across settings. Organizations must ensure these definitions are applied with high fidelity, as variations in how they are operationalized can compromise consistency and comparability. For example, a “procedure” or a “dispensed dose” can introduce measurement error.

Each domain generates unit-level u-charts that are interpreted longitudinally within the unit rather than across units, reflecting best practice in the application of SPC for quality improvement [21,26]. Where denominators fall below stability thresholds, data are aggregated to quarterly intervals to reduce small-number instability, which is a recognized limitation in attribute charting for health care data [26]. Aggregation applies primarily to low-volume domains such as transfusion or pharmacy services; higher-volume domains (eg, intensive care units or surgical units) maintain monthly analysis to preserve sensitivity to special-cause variation.

Level 3: Advanced Decision-Grade Analytics

Level 3 introduces prioritization and learning metrics. Severity weighting was adapted from the IHI SAC Matrix [23]. Each near miss is scored on a 4-point ordinal scale representing the worst reasonable outcome if not intercepted: minor, moderate, major, or catastrophic. The Near Miss Index (NMI) is calculated by summing severity scores across events in a reporting period. To capture the organizational response, specific learning metrics were defined. However, systematic reviews indicate that only a minority of reported incidents lead to implemented systemic changes, highlighting a critical gap between reporting and learning [27].

Learning yield (90 days) measures the proportion of near misses that generate an implemented corrective action within 90 days of reporting. This time frame is critical for ensuring timeliness; health care literature shows that prolonged delays between incident reporting and visible intervention can erode frontline staff engagement and undermine trust in the safety system [32]. Learning yield (180 days) assesses the proportion of those initial actions that are sustained at the 180-day mark. This follow-up period is a standard in implementation science used to differentiate a temporary, reactive solution from a change that has been successfully embedded into routine clinical practice. Evidence of sustainment should be determined through objective methods such as process audits or analysis of postimplementation performance data [28]. The 90- and 180-day intervals are operational thresholds specified by this framework to operationalize responsiveness and sustainment. Therefore, these metrics align with quality improvement literature by providing quantitative indicators of a functional learning system, measuring both its immediate responsiveness (action closure) and its ability to create lasting change (sustainment) [33]. Importantly, learning yield is a process metric that quantifies the timeliness and persistence of organizational response; it does not measure whether the implemented actions were effective at reducing harm. Demonstrating intervention effectiveness would require outcome-level studies with appropriate controls, which lies beyond the scope of a reporting-based analytic framework. Therefore, learning yield should be interpreted as evidence of a functioning learning loop rather than proof of harm reduction.

Finally, triangulation was embedded to contextualize near miss data. Framework design requires that near miss signals be interpreted alongside harmful incident rates, exposure denominators, reporting volumes, and scores from validated safety culture instruments such as the Safety Attitudes Questionnaire [29] and the Hospital Survey on Patient Safety Culture [30]. Triangulation addresses the limitation of voluntary reporting systems, in which fluctuations may reflect cultural variation more than underlying risk [8].

Ethical Considerations

This study did not involve human participants or patient-level data. All information was derived from peer-reviewed literature and publicly available institutional reports. Therefore, institutional review board approval was not required. Organizations implementing the framework with operational data should obtain local quality improvement or institutional review board determinations as appropriate.

Results

The outcome of this study is a 3-level framework for decision-grade analysis of near miss data. The framework integrates taxonomy, operational definitions, SPC, and severity weighting.

Framework Artifact

The framework begins with the capture of events through 3 complementary input channels: voluntary incident reporting systems, daily safety huddles, and targeted audits. In daily huddles, frontline staff verbally flag safety concerns and near miss events; a designated recorder enters qualifying events into the incident reporting system using the same WHO and AHRQ classification fields, ensuring a single integrated dataset. Audit-identified events, such as medication discrepancies found during pharmacy reconciliation, are likewise entered through the reporting system. Therefore, all inputs are coded using the WHO and AHRQ classifications to ensure consistency of terms. Data elements include event type, a patient-reached flag, detection mode, and unit or domain. The information is processed through successive levels. The first level produces organization-wide rates and ratios with SPC monitoring. The second level produces domain-specific rates using aligned denominators and unit-level charts. The third level weights events by severity, computes an NMI, and links reporting to corrective actions and culture data. The flow from inputs to outputs, with feedback to learning systems, is depicted in Figure 1.

‎

Interpretation Matrix

The framework also serves as a structured aid for leadership interpretation. Patterns in near miss reporting require context. In some instances, near misses increase while harm decreases. This suggests stronger interception and more open reporting. In other cases, near misses increase while total reports remain flat. This indicates better detection in selected domains. A third pattern is when both near misses and overall reports increase. This is more complex and may reflect cultural shifts and emerging system strain. A fourth scenario arises when both near misses and harm decrease simultaneously, which may signal declining reporting engagement rather than genuine safety improvement. A fifth scenario, in which near miss and harm rates remain stable with no SPC signals, may indicate system equilibrium but still warrants periodic review to guard against complacency. Each of these patterns is linked to diagnostic checks such as denominators, SPC variation, harm rates, and culture scores. The matrix is presented in Table 3. It structures trends into observed patterns, diagnostic considerations, and leadership implications.

Application and Governance Outputs

A synthetic example was constructed to demonstrate the framework’s application, using parameters detailed in Multimedia Appendix 1 [19,23,27,28,31]. In a hypothetical quarter, 50 reported harmful incidents and 10 reported near misses were modeled across 50,000 patient-days. This yields a reported harm rate of 1.0 per 1000 patient-days and a near miss rate of 0.2 per 1000 patient-days. The resulting reported near miss-to-harm ratio is 0.2:1, reflecting the common underreporting of near misses. Severity was then assigned using a weighted scale (minor=1, moderate=2, major=3, and catastrophic=5). The 10 near misses were distributed as 5 minor, 3 moderate, 1 major, and 1 catastrophic, which computes a severity-weighted NMI of 19. Organizational response was modeled over time: at 90 days, 3 corrective actions were implemented for a learning yield of 30% (3/10); at 180 days, 2 of those actions were sustained for a sustainment rate of approximately 67% (2/3). This example illustrates how the NM³ framework converts raw and even sparse reported data into standardized rates, a risk-prioritized index, and actionable governance metrics. The detailed rationale for each parameter is provided in Multimedia Appendix 1.

The framework produces outputs that can be integrated into governance processes. The first level produces a monthly organizational dashboard with standardized rates and SPC signals. The second level provides unit-level reports with stratified denominators and longitudinal trends. The third level delivers a quarterly report with weighted indices, learning yields, and triangulated indicators. Together, these outputs move near miss reporting from descriptive tallies to decision-grade intelligence that supports prioritization and oversight.

Discussion

Principal Findings

This study developed a 3-level framework to strengthen the analysis of near miss data. The framework integrates taxonomy, rate calculation, SPC, severity weighting, and learning metrics. It addresses the current challenge that near miss reporting often produces data but not reliable interpretation. By combining definitional clarity with analytic structure, it converts descriptive data into decision-grade intelligence.

The contribution of this work lies in its ability to unify 3 methodological strands. The first is a standardized classification of events, which ensures that data reflect common definitions rather than local interpretation. The second is the application of SPC, which provides stable detection of variation over time and guards against overdispersion. The third is the use of severity weighting and learning metrics, which elevate high-consequence signals and link reporting to organizational action. Together, these strands close the gap between reporting and governance.

The NM³ framework can be understood as a decision support architecture within a health care sociotechnical system [15,34,35]. By standardizing how safety events are classified, stabilized over time, weighted by potential consequence, and interpreted in relation to exposure and culture metrics, the framework reduces cognitive ambiguity in governance decision-making. Rather than leaving interpretation to informal heuristics, NM³ provides a shared analytic structure that supports consistent situational awareness across organizational levels [36]. In this way, near miss analytics shift from a descriptive reporting exercise to a structured component of human-centered safety governance [37].

Comparison With Prior Work

Existing approaches often remain descriptive. Most incident reporting systems provide tallies and categorical breakdowns but do not resolve the dilemma of what rising or falling near miss counts mean. The WHO ICPS established a definitional foundation [25], but it does not specify analytic processes. The structure-process-outcome model by Donabedian [38] underpins quality measurement broadly, but it does not address the specific problem of near miss interpretation. More recent frameworks, such as the Value Transformation Framework in primary care [39], focus on organizational change at the system level rather than the metrics of event reporting. Therefore, this study adds a new lens that sits between operational reporting and strategic governance.

The framework also supports practical adoption. Levels are designed as a maturity pathway. Organizations at baseline can implement level 1 by adopting a taxonomy and calculating standardized rates with simple u-charts. More advanced organizations can progress to level 2, where stratification by domain exposes hot spots without generating cross-unit competition. The most mature can reach level 3, where severity weighting, NMIs, and learning yields allow prioritization of risk and assessment of system responsiveness. In this way, the framework adapts to capacity rather than demanding wholesale transformation at once. In terms of scale, level 1 analytics can be applied by any hospital with a functioning incident reporting system. For level 2 stratification, sufficient denominator volume is needed to produce stable u-charts at the unit level; where monthly event counts are very low, such as in small community hospitals, a quarterly aggregation, as specified in the framework, preserves analytic validity. Level 3 severity weighting and learning yields are most informative when the organization reports at least several dozen near misses per quarter, providing enough events for meaningful index calculation and trend detection.

For governance, the framework strengthens the link between frontline reporting and board oversight. Dashboards at level 1 provide stable high-level indicators. Stratified reports at level 2 support managers who need to identify domain-specific risks. The advanced analytics of level 3 help leaders focus on high-consequence clusters and track whether actions are sustained. This aligns with international calls for safety metrics that are not only valid but actionable.

Future work must test and refine the framework. Empirical studies should evaluate its feasibility in practice, assess the reliability of severity scoring, and compare outputs across institutions. Research should also explore how digital systems can automate rate calculation, control chart production, and NMI computation. Benchmarking across hospitals and international contexts will determine generalizability and adaptation needs. Such validation will establish whether the framework produces consistent intelligence and supports decisions that improve safety outcomes. Furthermore, research should explore the integration of qualitative data from near miss narratives with the quantitative outputs of the NM³ framework to provide richer contextual understanding.

Limitations

This framework has important limitations. It depends on voluntary reporting systems that are known to undercount events substantially; estimates suggest that voluntary systems capture as few as 5% to 10% of patient safety incidents, with near misses particularly underrepresented [7,9]. Reporting behavior is shaped by culture, workload, and fear, and these influences may distort numerators. Denominator data also present challenges. Accurate counts of patient-days, doses, or procedures are not always available or consistent across systems. Errors in denominators can lead to misleading rates.

Event classification introduces uncertainty at level 1: even with standardized WHO and AHRQ definitions, assigning the correct event type depends on individual reviewer judgment, and variability in interpretation may introduce classification error. At level 3, severity scoring presents a similar challenge; the IHI SAC score provides structure, but interrater reliability was not tested here. Training, calibration exercises, and periodic agreement checks are recommended for both event classification and severity scoring. Learning yield metrics depend on robust action tracking, which is not consistently available in many organizations.

Finally, this study is conceptual. The framework was derived from literature and policy rather than empirical testing. Its generalizability across different health care contexts remains uncertain until validation is undertaken. Pilot applications will be necessary to confirm feasibility, reliability, and value in governance. Implementation also presupposes dedicated human resources and analytic expertise. This may not be available, particularly for level 2 stratification and level 3 severity weighting in resource-constrained settings, and organizations should assess workforce readiness before progression beyond level 1.

Implications

This framework has direct implications for health care organizations, quality leaders, and researchers. For organizations, the staged design lowers the barrier to adoption by allowing hospitals to begin with standardized baseline measurement and progress to more advanced analytics as institutional capacity grows, rather than requiring wholesale transformation at the outset.

For leadership, NM³ converts descriptive reporting into actionable intelligence that supports evidence-based governance. The tiered outputs provide a common language for safety oversight and enable boards to distinguish changes in reporting culture from genuine shifts in underlying risk, aligning with international calls for actionable patient safety metrics.

For researchers, NM³ provides a reproducible conceptual model that can be operationalized and validated across diverse health care contexts. Adoption in varied settings will help establish its role in patient safety improvement and determine the conditions under which the framework yields the greatest analytic value.

Conclusions

This study developed a 3-level framework for decision-grade interpretation of near miss data within acute hospital systems. The framework integrates taxonomy, SPC, severity weighting, and organizational learning metrics. It addresses the ambiguity that has long limited the use of near miss reports in governance.

By structuring analysis into progressive levels, the framework functions as a cognitive and organizational scaffold for interpreting safety signals in complex sociotechnical environments. It enables leaders to distinguish culture shifts from genuine changes in risk and to prioritize high-consequence signals. Adoption has the potential to strengthen human-centered safety governance and support proactive risk mitigation.