Uncovering diagnostic errors in dentistry through critical incident reports: a mixed-methods analysis Authors: Sanam Faheem, Rimsha Qasim, Syed Jaffar Abbas Zaidi Published: 2026 ABSTRACT BackgroundDiagnostic errors and near misses threaten patient safety in dental practice. In Sindh, the service delivery standards require documentation and review of critical incidents, supporting incident reporting as a platform for ensuring patient safety. This study aimed to describe incident patterns using critical incident reports, a structured patient safety reporting mechanism used to document adverse events, near misses, and diagnostic process failures in clinical practice.MethodsA sequential mixed-methods study reviewed 300 de-identified incident reports from Karachi from January to June 2024. Associations with urgency were assessed using chi-square or Fisher’s exact tests and logistic regression analysis. Reflexive thematic analysis of narrative fields was supplemented by face-to-face interviews with 20 selected reporters.ResultsIncidents were classified as urgent in 173/300 (57.7%) cases, and corrective action was required in 289/300 (96.3%) cases. Recurrence was most commonly coded as 20–40% (148/300, 49.3%). The most frequently coded diagnostic domains were pulp diseases (87/300, 29.0%), periodontal ligament diseases (85/300, 28.3%), and temporomandibular joint disorders (80/300, 26.7%). The leading contributing factors were a lack of diagnostic resources (142/300, 47.3%), lack of postgraduate training (119/300, 39.7%), and gaps in undergraduate teaching (103/300, 34.3%). Recurrence category showed the strongest association with urgency (p < 0.001). In multivariable logistic regression, urgent classification was less likely in private clinics (aOR 0.16, p = 0.013) and among male reporters (aOR 0.37, p < 0.001) and more likely in reports filed by senior faculty (aOR 2.48, p = 0.025); the 20–29 patients/week group also had higher adjusted odds of urgency (aOR 2.26, p = 0.023). Qualitative themes highlighted incomplete assessments, limited access to and interpretation of imaging, gaps in supervision and referral pathways, and communication failures.ConclusionThe registry demonstrates recurrent, high-urgency patterns linked to resource and training constraints and supports structured critical incident reporting as a practical tool for teaching triage, escalation, and safer diagnostic workflows for safer and more effective dental care.Clinical trials Not applicable.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12913-026-14918-z. Introduction Despite major advances in dental diagnostics and imaging, diagnostic errors (DE) and near misses remain important threats to patient safety in dentistry [1]. Medical errors are widely recognized as a major cause of preventable harm, and one influential estimate suggests that they may contribute substantially to mortality in the United States [2]. Contributing factors may include cognitive biases, atypical or absent clinical presentations, and systemic breakdowns in the referral processes [3]. A strong body of evidence suggests that diagnostic reasoning involves both a rapid, automatic recall of past experiences (System I) and a more deliberate, logical approach (System II); errors can arise in either mode [3, 4]. A delay in referring a patient to a specialist is an example of a DE and a key factor contributing to the late diagnosis of oral cancerous lesions [5, 6]. Reflecting the global importance of this issue, the World Health Organization (WHO) has designated diagnostic safety as a priority area in primary healthcare [5, 6]. To improve conceptual clarity, this study distinguishes between three related but analytically distinct constructs. Diagnostic errors are defined as failures to establish an accurate and timely explanation of a patient’s condition or to communicate that explanation effectively [1, 7]. Procedural errors refer to incorrect execution of a clinical intervention (for example, wrong-site or wrong-tooth procedures), which may or may not arise from diagnostic failures. Near-miss events are incidents that could have resulted in patient harm but were intercepted before reaching the patient. In this study, procedural errors were included only when there was clear evidence that they originated from or were directly linked to a diagnostic process failure. Near-miss events were analyzed as indicators of latent diagnostic system vulnerabilities rather than as equivalent to completed errors. In outpatient medicine, population-based estimates suggest that approximately 1 in 20 adults experience a diagnostic error annually, with a substantial proportion having the potential for harm [8]. International bodies, landmark reports, and systematic reviews have repeatedly emphasized that diagnostic errors remain under-measured, difficult to detect in routine practice, and often only recognized retrospectively, which limits prevention efforts [9–11]. These challenges are relevant to dentistry because many diagnostic decisions depend on incomplete histories, variable documentation, and selective use of investigations, especially in busy clinical settings [12, 13]. There are increased chances of misdiagnosis when the presenting signs and symptoms of the disease are atypical, absent, or mimic another disease, or may be due to cognitive errors (errors of ignorance and implementation) [14, 15]. International efforts have attempted to curb DE, although incident report review data still reveal an increase of up to 9%, in these errors [16–18]. Various methods, such as case reviews, reporting systems, audits, and legal claim analyses have been employed to explore DE. However, each method has limitations and may not fully capture the true incidence or root causes [4, 19]. Incident report reviews of practicing dentists in more developed regions have highlighted the potential value of analyzing clinical decision-making processes [20]. However, such data remain sparse in many parts of the world, including South Asia, where population diversity and healthcare challenges may differ substantially from those in high-income- nations. Evidence from Pakistan and the wider region suggests that diagnostic reasoning and safety culture remain important educational and system priorities. In Pakistan, the diagnostic error rates among oral and maxillofacial surgery trainees have been shown to vary by training level, supporting the need for structured supervision and deliberate practice in diagnostic decision-making [21]. Studies among dental students in Pakistan have also documented gaps in patient safety awareness and safety culture, suggesting that formal training in error recognition, reporting, and learning is still developing [22, 23]. Complementary work in Pakistan has highlighted the need for clearer reporting pathways for procedural errors and near-misses in dental clinics [24]. Critical incident reporting refers to a structured patient safety process through which clinicians document adverse events, near misses, and unsafe clinical situations to support learning, quality improvement, and prevention of future harm. In the Sindh province, the Sindh Healthcare Commission (SHCC) regulates healthcare establishments, including dental clinics and hospitals, through a licensing program that focuses on enforcing the Sindh Service Delivery Standards (SSDS) [25].The SSDS explicitly requires that incidents, accidents, near misses, and adverse events are reported and analyzed, and that data from complaints and critical incident reports are used for continuous improvement [26]. The SHCC complaint guidance also emphasizes that complaints must be supported by relevant documentation and that facilities are required to submit the necessary records during investigations [26].These regulatory expectations provide a clear rationale for using critical incident reporting as a locally relevant platform for improving diagnostic safety in dentistry. Diagnostic safety in dentistry is a health services quality and patient safety issue that aligns with SDG 3 (good health and wellbeing) through preventing avoidable harm and delays in care. By translating frontline critical incidents into measurable patterns, incident-report registries can also serve as a practical governance and health workforce development tool to guide training priorities, resource allocation, and referral pathway strengthening across public and private services. Although diagnostic safety has received increasing attention internationally, registry-based analyses of diagnostic process failures in dentistry remain limited, particularly in South Asian clinical and educational settings where reporting systems, training structures, and resource constraints may differ substantially from those in high-income countries. This study addresses this gap by combining quantitative analysis of critical incident reports with qualitative exploration of the mechanisms underlying diagnostic process failures in dentistry. This study had four objectives: (1) To describe the distribution of incident types documented in critical incident reports within the study settings, including diagnostic errors, diagnostic-related procedural events arising from diagnostic failures, and near-miss events within the diagnostic pathway, as submitted and recorded during the study period; (2) to identify the key contributing factors documented in the incident reports; (3) To identify the diagnostic domains most frequently represented in the submitted incident reports; and (4) to conduct face-to-face interviews with selected reporters, guided by the richness of their narrative incident descriptions, to explore recurring patterns and generate learning points to support safer dental practices. Methods Study design This study employed an explanatory sequential mixed methods- design to investigate diagnostic errors in dentistry, using critical incident reports as the primary unit of analysis. The quantitative phase consisted of a retrospective review of 300, de-identified- critical incident reports drawn from a total pool of 400. All 400 critical incident reports identified in the departmental logs during the study period (January to June 2024) were initially considered for inclusion. Reports were excluded if they were missing one or more core structured fields required for coding (incident type, contributing factors, urgency classification, or outcome documentation), if supporting documentation was absent where indicated in the narrative, or if the report was identified as a duplicate entry. Following this eligibility screening, 300 reports met the inclusion criteria and were included in the quantitative analysis. These 300 reports represent a census of eligible records within the defined setting and period, rather than a random or purposive sample drawn from a larger population. The 100 excluded reports were not formally characterized by institution type or incident category, and the possibility that incomplete documentation is non-randomly distributed, for example being more common in lower-resource settings or among less experienced reporters, is acknowledged as a limitation of this dataset. The qualitative phase involved an in-depth- analysis of the narrative fields and follow-up- notes associated with the reviewed reports. This phase was supplemented by face-to-face-- interviews with selected reporters, chosen based on the explanatory richness of their submitted narratives. The purpose of this phase was to explore the mechanisms underlying diagnostic failures and identify learning points to enhance safe clinical practice, particularly in dental clinical environments. Setting and duration The study was conducted over a six-month- period from January to June 2024 in Karachi, Pakistan. Three dental institutions in Karachi, Pakistan participated in the study: one government dental university, one private dental university, and one additional private dental university, supplemented by reports from private dental clinic settings. Institutions were selected purposively on three criteria: the presence of an operational structured incident reporting system or equivalent clinical governance documentation mechanism; representation of a distinct institutional category within the Karachi dental education and practice landscape; and a minimum reporting volume sufficient to contribute meaningfully to the analytical sample during the study period. Equal representation across institutions was not a design objective, as this study uses purposive rather than probability sampling and does not aim to produce population-level estimates. The distribution of reports across institution types reflects genuine differences in patient volume, reporting infrastructure, and institutional reporting culture across the study settings, and is discussed explicitly as a source of reporting bias in the Limitations section. Purposive sampling was applied in both phases to ensure representation of diverse clinical and educational practice environments. The designation group variable distinguishes three structurally distinct reporter categories. Senior faculty refers to reporters holding senior academic ranks (Associate Professor and above) within formally registered dental universities. Junior faculty refers to reporters holding junior academic ranks (Assistant Professor, Lecturer, House Officer, or Demonstrator) within formally registered dental universities. Private practitioner refers to dental practitioners contributing reports from private clinic settings without formal university faculty affiliation. This three-way classification was used because private practitioners operate outside the supervisory hierarchies and institutional reporting mechanisms of teaching settings, making them structurally distinct from both senior and junior faculty regardless of their level of clinical experience. The ethical framework differed across the two phases. In the quantitative phase, the unit of analysis comprised de-identified critical incident reports made available for research review under ethical approval. Because reports were de-identified prior to transfer to the research team, individual consent was not required for the retrospective record review. In the qualitative phase, twenty reporters were approached individually for face-to-face interviews, and written informed consent was obtained from each prior to participation. Consent covered the interview, anonymized use of excerpts in publication, and the right to withdraw without consequence. All procedures were conducted in accordance with the Declaration of Helsinki. Study population Age was categorized into three groups corresponding to recognised career stages in dental academia: 24 to 35 years (early career, spanning qualification through junior faculty appointment), 36 to 50 years (mid-career, corresponding to transition into and consolidation of senior academic roles), and above 50 years (late career and senior practitioners). These boundaries were selected a priori based on career stage conventions in health workforce research and the expected distribution of faculty rank within the study population. This study involved two analytically distinct phases with different units of analysis and corresponding ethical requirements. In the quantitative phase, the unit of analysis was the critical incident report, not the individual reporter. Reports were de-identified prior to transfer to the research team, and no individual-level data were extracted or reported from identifiable sources. The relevant population for this phase was the set of critical incident reports meeting eligibility criteria within the participating institutions during the study period. In the qualitative phase, the unit of analysis was the individual reporter. Twenty reporters whose incident narratives were selected for interview were the participants in the conventional sense. Inclusion criteria required participants to be practicing faculty members or dental surgeons who had submitted an eligible incident report. Dental students, auxiliary staff, non-practicing faculty, and individuals who declined or were unable to provide informed consent were excluded. Written informed consent was obtained from all qualitative phase participants prior to interview. Unit of analysis, documentation, and coding procedure The unit of analysis comprised critical incident reports submitted in dental clinical settings during the study period. The unit of analysis comprised critical incident reports involving the diagnostic process. These included: Diagnostic errors (incorrect, delayed, or missed diagnoses),Diagnostic-related procedural errors, defined as procedural mistakes (e.g., wrong-site or wrong-tooth interventions) attributable to an underlying diagnostic failure, and.Near-miss events, defined as intercepted failures within the diagnostic pathway that did not result in patient harm. Diagnostic errors (incorrect, delayed, or missed diagnoses), Diagnostic errors (incorrect, delayed, or missed diagnoses), Diagnostic-related procedural errors, defined as procedural mistakes (e.g., wrong-site or wrong-tooth interventions) attributable to an underlying diagnostic failure, and. Diagnostic-related procedural errors, defined as procedural mistakes (e.g., wrong-site or wrong-tooth interventions) attributable to an underlying diagnostic failure, and. Near-miss events, defined as intercepted failures within the diagnostic pathway that did not result in patient harm. Near-miss events, defined as intercepted failures within the diagnostic pathway that did not result in patient harm. Purely technical procedural errors without a diagnostic component were excluded. This classification ensured that all included incidents were conceptually anchored within the diagnostic process. Where available, each report was accompanied by supporting documentation, including dental radiographs (for example, periapical images), relevant medical and dental history forms, and charting records. All reports were de-identified prior to analysis to protect confidentiality. Two trained assessors independently reviewed each report and its supporting documents, verified eligibility, and completed the coding and abstraction process using a standardized codebook. A structured abstraction form was developed using the headings in the sample incident report template, including Basic Information (date, time, location, patient identifier, age and gender, reporter, and designation), Incident Description (incident type and detailed description), Immediate Action Taken, Outcome, Contributing Factors, and Reporting and Follow-Up (Appendix A). The operational definitions of diagnostic errors and near misses were provided to the abstractors to support consistent coding. Disagreements were resolved through structured discussion and consensus prior to finalizing the analytical dataset. Content validity of the abstraction instrument The content validity of the coding and abstraction instruments was evaluated using the Content Validity Index (CVI). The instrument demonstrated strong content relevance and clarity, with a scale-level CVI (S-CVI/Ave) of 0.93 and an item-level CVI (I-CVI) of 0.86-1.00. Urgency classification, validity evidence, and reproducibility For the purposes of all analyses, urgent classification was operationally defined as follows: an urgent case (coded 1) involves a time-critical risk requiring immediate escalation or same-day senior clinical review. A non-urgent case (coded 0) involves no time-critical risk, and any required actions can be managed through routine follow-up and standard learning processes. These definitions were provided to both abstractors in the codebook and applied consistently across all 300 included reports. Beyond face validity, the validity evidence was strengthened in three complementary ways. First, an expert panel reviewed the urgency rubric for clarity and coverage, yielding strong content validity (S-CVI/Ave = 0.91; I-CVI range, 0.83–1.00) [27]. Second, criterion-oriented checks were conducted to assess whether urgent coding aligned with independently documented escalation markers in the accompanying records. A limitation of this approach is that some escalation markers, such as same-day senior review, share conceptual overlap with the urgency definition itself. To partially mitigate this, criterion alignment was assessed across a broader set of markers including urgent imaging requests and immediate corrective action documentation, some of which extend beyond the core urgency definition. Nonetheless, readers should note that this does not constitute fully independent criterion validation, and the criterion-oriented evidence should be interpreted as supporting plausibility rather than providing independent external validation of the construct. A limitation of this approach is that some escalation markers, such as same-day senior review, share conceptual overlap with the urgency definition itself. To partially mitigate this, criterion alignment was assessed across a broader set of markers including urgent imaging requests and immediate corrective action documentation, some of which extend beyond the core urgency definition. Nonetheless, this does not constitute fully independent criterion validation, and the criterion-oriented evidence should be interpreted as supporting plausibility rather than providing independent external validation of the construct. Third, construct-oriented checks assessed whether urgency coding behaved as expected with respect to theoretically linked incident attributes. Higher recurrence and greater impact severity were associated with higher proportions of urgent coding. These internal associations are consistent with the expected direction of the urgency construct and provide limited supporting evidence of construct plausibility; however, because these relationships were assessed within the same dataset rather than against an independent external standard, they should not be interpreted as strong or independent evidence of construct validity. To evaluate reproducibility, the two assessors independently coded all reports prior to the consensus meetings, and agreement was quantified. For binary fields, Cohen’s kappa indicated substantial agreement (κ = 0.79 for urgency classification; κ = 0.83 for corrective actions). For nominal multi-category variables (institution type, designation group, and specialty group), the weighted kappa values ranged from 0.72 to 0.81. For ordinal variables (recurrence and impact categories), the quadratic-weighted kappa values ranged from 0.74 to 0.80. For multi-label sections (diagnostic domains and contributing factors), agreement was evaluated using prevalence-adjusted, bias-adjusted kappa, and percent agreement across items; item-level agreement ranged from 0.70 to 0.88, with an overall multi-label agreement of 0.78. After these reliability checks, the remaining discrepancies were resolved through consensus, and the final dataset was used for analysis. Sample size A total of 400 critical incident reports were identified in the departmental logs. Pre-specified eligibility criteria were then applied. Records with missing core structured fields required for coding were excluded. As a result, 300 reports were included in the quantitative analysis. Phase 1: Quantitative review of critical incident reports Quantitative data analysis The critical incident reports were screened for completeness, and eligible records were coded for analysis. Descriptive statistics (frequencies and percentages) were used to summarize the report characteristics. Multi-label fields (diagnostic domains and contributing factors) were treated as multi-response variables. Each label was coded as a binary indicator (present or absent) in the incident report. Frequencies represent the number of incidents in which a label was present, and percentages represent the proportion of 300 incidents in which the label appeared. Because multiple labels could be assigned to the same incident, totals could exceed the sample size, and percentages did not sum to 100%. These multi-label variables were summarized descriptively only and were not used as predictors in chi-square testing or logistic regression because the categories were not mutually exclusive. Only mutually exclusive categorical predictors were used for inferential analyses. Bivariate associations with urgency were evaluated using Pearson’s chi-square tests, with Fisher’s exact tests used as sensitivity checks when expected cell counts were low. Because multiple bivariate tests were conducted, we treated these analyses as exploratory. We controlled the false discovery rate using the Benjamini–Hochberg procedure across the set of predictor-level tests reported in Table 4. For logistic regression, we fitted univariate models for each predictor and a prespecified multivariable model that included institution type, designation group, specialty group, age group, gender, and patients/week category. Before model interpretation, multicollinearity was assessed using variance inflation factors and by inspecting the correlation patterns among the predictors. Model performance and fit were evaluated using calibration and discrimination metrics, including the Hosmer–Lemeshow goodness-of-fit test, area under the receiver operating characteristic curve (AUC), pseudo-R² statistics, and information criteria (AIC and BIC) for model comparison. Workload (patients/week) and age were analyzed as categorical variables; therefore, a linearity-in-the-logit assessment was not applicable. The private clinic and private practitioner categories each comprised ten reports, creating sparse cell conditions in both bivariate and multivariable analyses. To address this, three pre-specified sensitivity analyses were conducted. First, institution type was collapsed into a binary predictor (government university versus non-government) to assess whether estimates for the remaining predictors were sensitive to the sparse private clinic group. Second, private practitioners were reassigned to the junior faculty category to evaluate the effect of sparse designation group cells on model estimates. Third, Firth’s penalized logistic regression was applied to the full model to correct for small-sample bias in maximum likelihood estimation. Across all three sensitivity analyses, the direction and approximate magnitude of the adjusted estimates for the remaining predictors were consistent with the primary multivariable model. The adjusted estimates for private clinic and private practitioner under penalized regression were attenuated toward the null, confirming that these subgroup estimates are imprecise and should not be interpreted independently of their wide confidence intervals. All subgroup estimates from sparse categories are therefore presented for completeness but are explicitly described as exploratory, and no substantive conclusions are drawn from them. Multiple-testing correction using the Benjamini-Hochberg procedure was applied at the predictor level for bivariate tests in Table 4. Category-level Wald test p-values reported in the logistic regression tables were not corrected for multiplicity, as they reflect pre-specified contrasts within each predictor block rather than independent hypothesis tests. Nonetheless, given the number of category-level comparisons across the two regression tables, individual category-level findings should be treated as exploratory unless they are strongly supported by a significant predictor-level test and effect size, and borderline category-level results should be interpreted with particular caution. Fisher’s exact test was applied as a secondary sensitivity check for rare categories and should not be interpreted as an independent primary analysis. Where Fisher and chi-square results were consistent, this convergence supports the stability of the primary finding. Where Fisher results differed from chi-square, the Fisher result is more reliable for that specific comparison due to its distribution-free basis. In this dataset, all Fisher results were directionally consistent with the corresponding chi-square findings, confirming that the primary interpretations are not artifacts of sparse cell assumptions. The results of all three sensitivity analyses are presented in Supplementary Table S1, which displays the adjusted odds ratios, 95% confidence intervals, and p-values from the primary model and each sensitivity model side by side for all predictors and categories. For multi-label fields (diagnostic domains and contributing factors), each label was treated as a binary indicator (present or absent), and agreement was calculated separately for each item. For each item, Cohen’s kappa was adjusted for both prevalence imbalance and rater bias using the PABAK formula: PABAK = 2P0 − 1, where P0 is the observed proportion of agreement. Percent agreement was calculated as the number of reports for which both assessors assigned the same code (present or absent) divided by the total number of reports, expressed as a percentage. Item-level values were then averaged across all items within each field to produce the summary agreement statistics reported. Full item-level agreement statistics for each diagnostic domain and contributing factor code are provided in Supplementary Table S2. Phase 2: Qualitative analysis of incident narratives Integration rationale The qualitative phase was designed to explain three specific patterns identified in the quantitative analysis that the structured incident report fields alone could not interpret. First, the high frequency of documented resource inadequacy as a contributing factor (47.3%) raised questions about the mechanisms through which resource constraints affected diagnostic reasoning in practice, which required narrative and interview data to illuminate. Second, the strong association between recurrence category and urgency classification (p less than 0.001) prompted inquiry into the systemic and supervisory conditions that reporters perceived as allowing similar incident types to repeat without correction. Third, the differential urgency classification by gender and specialty group required exploration of whether these patterns reflected genuine differences in incident severity, differences in urgency perception thresholds, or differences in reporting and documentation behavior across groups. The qualitative phase addressed these three explanatory questions through reflexive thematic analysis of incident narratives and face-to-face interviews, and integration was achieved by comparing each quantitative pattern with the qualitative themes most relevant to its interpretation. Building on the above mentioned quantitative findings, a qualitative analysis was conducted to explore recurring patterns in the narrative sections of incident reports, with emphasis on the Incident Description, Contributing Factors, and Corrective or Preventive Action fields. Interviews followed a semi structured guide developed for this study in English, informed by diagnostic error and patient safety literature, and refined through pilot testing with two dental faculty members who were not included in the interview sample. The reporting of the qualitative component was aligned with COnsolidated criteria for REporting Qualitative research (COREQ) Checklist (Appendix B) [28]. In keeping with the principles of reflexive thematic analysis and the requirements of the COREQ framework, the research team acknowledges the following positionality and reflexive considerations. All three investigators are faculty members within dental teaching institutions in Karachi, representing the same professional and institutional context as the majority of study participants. This shared positioning likely facilitated rapport and candid disclosure during interviews, but also introduced the possibility that participants framed their accounts in response to perceived collegial relationships, and that interviewers brought prior assumptions about local resource constraints, training gaps, and reporting culture that may have influenced probing and interpretation. The investigators’ professional background in patient safety and dental education provided conceptual familiarity with the diagnostic domains under study, which supported informed and precise interviewing. Simultaneously, this familiarity carried a risk of confirmatory interpretation, whereby data consistent with existing frameworks were more readily recognized than data that challenged them. To manage this risk, reflexive memos were maintained throughout the coding and thematic development process, documenting emerging interpretive decisions and the assumptions underlying them. These memos were reviewed collectively during analytical team debriefs, where interpretations were challenged and alternative readings of the data were considered before themes were finalized. The investigators also recognized a potential positional bias toward educational recommendations, given their roles as dental educators with a professional stake in improving training and supervision. This was explicitly documented in the reflexive memos, and theme development was reviewed to ensure that the findings reflected the data rather than a predetermined advocacy position. Member checking with a subset of participants served as an additional check on this interpretive tendency by confirming that the themes resonated with reporters’ own accounts of their experiences. Selection of reports for qualitative coding A purposive sample of 20 incident reports was selected for the qualitative phase. Purposive sampling is the methodologically appropriate strategy for qualitative inquiry, where the goal is not statistical representativeness but depth and diversity of meaning. Participants were selected on two criteria: narrative richness, defined as sufficient detail in the incident description and contributing factors fields to sustain meaningful interpretive analysis, and representational diversity across institution type, designation group, specialty, and incident category. Selecting participants who are articulate and able to reflect on their experiences is a foundational principle of purposive sampling in qualitative research, not a source of bias, as qualitative inquiry requires participants who can speak meaningfully to the phenomenon under investigation. The diversity criterion ensured that the sample encompassed a range of institutional contexts, professional roles, and incident types, capturing the breadth of perspectives needed to explain the quantitative patterns identified in Phase 1. Routine incidents with sparse documentation were fully represented in the quantitative analysis, which included all 300 eligible reports; the qualitative phase was not designed to replicate this distributional coverage but to provide explanatory depth for selected patterns identified quantitatively, as depicted in Fig. 1. Qualitative procedure Narrative fields were extracted verbatim from the critical incident reports and de-identified- prior to analysis to ensure confidentiality and preserve the authenticity of the reported events. The analysts documented their assumptions and maintained reflexive memos throughout the coding process to enhance transparency, support rigor, and ensure consistency in interpretive decisions. Following the completion of the incident report analysis, face-to-face interviews were conducted with selected reporters to further elucidate the contextual details, reasoning processes, and contributory factors underlying the documented incidents. Face-to-face interviews were conducted at a time and location convenient to each participant, within the participating institutions. Interviews ranged in duration from 25 to 45 min, with a mean duration of approximately 35 min. Variation in interview length reflected the complexity of the incident narrative and the extent of probing required to clarify contextual details not fully captured in the structured incident report fields. Interviews were conducted by the first and second authors, both of whom are faculty members, with the third author contributing to selected interviews. All three investigators have academic backgrounds in dental education and patient safety research. Prior to data collection, the two primary interviewers reviewed qualitative interviewing principles, studied the theoretical basis of the interview guide, and conducted pilot interviews with two faculty members outside the sample to assess question clarity and probe effectiveness. The guide was refined based on pilot feedback before use in the main interview phase. Regular analytical debriefs were held between the research team throughout the data collection period to monitor consistency in interviewing approach and to identify any emerging interpretive assumptions requiring documentation in the reflexive memos. Fig. 1Flow diagram of incident report identification, eligibility screening, and inclusion Flow diagram of incident report identification, eligibility screening, and inclusion All 400 reports identified during the study period were assessed for eligibility against pre-specified criteria. Following exclusion of incomplete and duplicate records, 300 reports were included in the quantitative analysis. A purposive subsample of 20 reports was subsequently selected for the qualitative phase based on narrative richness. Qualitative data analysis Narrative data and interview transcripts were analyzed using reflexive thematic analysis following the six-phase approach of Braun and Clarke [29, 30]. In this framework, reflexivity refers to the researcher’s ongoing critical examination of how their own positionality, assumptions, and interpretive decisions shape engagement with the data, documented throughout the analysis in reflexive memos. A codebook was developed to organize the initial codes generated in Phase 2 of the analytical process, consistent with Braun and Clarke’s framework in which systematic code generation and organization is an integral part of the analytical procedure. The codebook served as a transparency and organizational tool rather than as a reliability instrument; its purpose was to document the evolving coding decisions of the research team rather than to produce agreement statistics for validation purposes. Two analysts coded the data independently and met regularly to discuss their interpretations. These discussions were oriented toward deepening analytical engagement and surfacing different interpretive perspectives through dialogue, consistent with a reflexive rather than reliability-based approach to team analysis. Thematic saturation was used as a criterion for analytical completeness, referring to the point at which no new codes were emerging from the data and existing codes were repeating across later transcripts. This is an analytical concept describing the coverage of the coding process and should not be confused with information power, which is a sampling adequacy concept used to justify participant numbers prior to data collection. These two concepts address different questions at different stages of the research process and are both applicable within a reflexive thematic analysis framework. The critique by Braun and Clarke is directed at the misuse of saturation as a mechanical proxy for sample size sufficiency, not at its use as a marker of analytical completeness, which remains a legitimate and appropriate application within this paradigm approach [31]. To enhance trustworthiness, a brief peer debrief was conducted among the analyst team to challenge interpretations and confirm the coherence of themes across incident types. Throughout the analysis, reflective memos were maintained to document the analytic decisions and support an audit trail. Integration was achieved by comparing quantitative patterns with qualitative themes. This supported the triangulation of incident categories, interview transcripts, contributing factors, and recommended preventive actions. To strengthen credibility, member checking was conducted in two stages. First, individual interview summaries were returned to each of the 20 participants within two weeks of their interview for accuracy confirmation. Three participants provided minor clarifications, which were incorporated into the final transcripts before coding commenced. Second, following completion of initial theme development, a one-page synthesis of the five emergent themes was shared with a purposively selected subset of eight participants, chosen to represent diversity across institution type, designation group, and incident category. Participants were invited to comment on whether the themes reflected their experiences and to identify any significant dimensions they felt were missing. Six of eight responded; all confirmed broad resonance with the themes, and two offered contextual elaborations that were used to strengthen the interpretation of Theme 3. No participant disputed any theme as inaccurate. These responses did not result in the addition of new themes but did refine the analytical language used in the final thematic descriptions. Methodological triangulation was achieved by comparing three data sources: structured fields and narrative sections of incident reports, supporting documentation where available, and follow-up interview accounts. Analyst triangulation was supported through independent initial coding by two analysts and iterative discussions of theme boundaries and meanings. Transferability was enhanced through a clear description of the setting, participant roles, and reporting context, and by providing illustrative excerpts that demonstrate how themes were grounded in the data. This thereby permits the judgement of relevance of the findings to comparable dental teaching and clinical environments. Results A total of 400 critical incident reports were identified in the departmental logs. After excluding incomplete records, 300 reports were included in this quantitative analysis. Report and reporter characteristics Of the included reports, all incidents were mapped to the diagnostic process, including primary diagnostic errors, diagnostic-related procedural errors, and near-miss events, which were analyzed as indicators of diagnostic system vulnerabilities rather than equivalent outcomes. Most reports were filed by faculty across academic ranks. The majority of reports were associated with early-career clinicians, with a predominance of female reporters. Detailed distributions are presented in Table 1. Incident characteristics Most reports occurred in clinics with a moderate to high weekly patient load. Recurrence of similar incident types was frequently noted. The most commonly coded diagnostic domains and contributing factors are presented in Table 2. Table 1Characteristics of reporters and reporting context in critical incident reports (n = 300)VariableCategory n %Designation groupSenior Faculty4113.7Junior Faculty24983.0Private Practitioner103.3Institution typeGovernment University24481.3Private University4615.3Private Clinic103.3Specialty groupClinical Faculty25785.7Basic Faculty258.3Private Dentist186.0Age group24–3526889.336–50258.3Above 5072.3GenderFemale20668.7Male9431.3Patients per week10–1910836.020–296923.0> 3012341.0 Table 1Characteristics of reporters and reporting context in critical incident reports (n = 300)VariableCategory n %Designation groupSenior Faculty4113.7Junior Faculty24983.0Private Practitioner103.3Institution typeGovernment University24481.3Private University4615.3Private Clinic103.3Specialty groupClinical Faculty25785.7Basic Faculty258.3Private Dentist186.0Age group24–3526889.336–50258.3Above 5072.3GenderFemale20668.7Male9431.3Patients per week10–1910836.020–296923.0> 3012341.0 Characteristics of reporters and reporting context in critical incident reports (n = 300) Weekly patient load was categorized into three mutually exclusive groups: 10–19 patients per week, 20–29 patients per week, and 30 or more patients per week. Category boundaries were defined prior to analysis. Reporters seeing exactly 20 patients per week were assigned to the 20–29 category, and reporters seeing exactly 30 were assigned to the 30 or more category, in accordance with the pre-specified interval definitions. Most reports were from government university settings (244/300, 81.3%). Private university settings contributed 46 reports (15.3%), and private clinics contributed 10 (3.3%). Most reports were filed by the junior faculty group (249, 83.0%). Senior faculty accounted for 41 (13.7%) and private practitioners for 10 (3.3%). The registry was dominated by clinical disciplines (257, 85.7%). The number of reports from basic faculty was 25 (8.3%), and from private dentists was 18 (6.0%). Most reports involved staff aged 24–35 years (268 [89.3%]). Females were more frequently represented (206, 68.7%) than males (94, 31.3%). Workload indicators suggest a high-throughput environment for the study. Reports linked to > 30 patients/week were the largest category (123, 41.0%), followed by 10–19 patients/week (108, 36.0%), and 20–29 patients/week (69, 23.0%). Table 2Incident characteristics (n = 300)VariableCategory n %Urgent classification (0/1)Non-urgent (0)12742.3Urgent (1)17357.7Corrective action neededYes28996.3No113.7Recurrence category< 20%10334.320–40%14849.340–60%4214.0> 60%72.3Impact categoryNo impact113.7Minor impact15652.0Moderate impact12240.7Major impact103.3Not documented10.3Education/training factor documentedYes20267.3No5719.0Not documented4113.7 Table 2Incident characteristics (n = 300)VariableCategory n %Urgent classification (0/1)Non-urgent (0)12742.3Urgent (1)17357.7Corrective action neededYes28996.3No113.7Recurrence category< 20%10334.320–40%14849.340–60%4214.0> 60%72.3Impact categoryNo impact113.7Minor impact15652.0Moderate impact12240.7Major impact103.3Not documented10.3Education/training factor documentedYes20267.3No5719.0Not documented4113.7 Incident characteristics (n = 300) Most incidents were coded as urgent (173/300, 57.7%). There were 127 (42.3%) non-urgent incidents. Nearly all incidents required some form of corrective action (289, 96.3%). Only 11 reports (3.7%) were coded as not requiring any corrective action. Recurrence was most often coded in the 20–40% category (148, 49.3%). The < 20% category was 103 (34.3%). The 40–60% category was 42 (14.0%). The > 60% category included seven (2.3%) reports. The impact was mainly minor (156, 52.0%) or moderate (122, 40.7%). Major impacts were uncommon (10, 3.3%). No impact was observed on 11 (3.7%) reports. Education or training factors were documented in most reports (202, 67.3% of reports). They were documented as absent in 57 (19.0%). The data were not documented in 41 (13.7%) reports. Table 3Diagnostic domains and contributing factors (multi-label)SectionItem n %Diagnostic domains (multi-label)PDL diseases8528.3Pulp diseases8729.0TMJ disorders8026.7Lack of resources14247.3Lack of postgraduate training11939.7Gaps in undergraduate teaching10334.3 Table 3Diagnostic domains and contributing factors (multi-label)SectionItem n %Diagnostic domains (multi-label)PDL diseases8528.3Pulp diseases8729.0TMJ disorders8026.7Lack of resources14247.3Lack of postgraduate training11939.7Gaps in undergraduate teaching10334.3 Diagnostic domains and contributing factors (multi-label) Table 3 summarizes the most frequently coded diagnostic domains and contributing causes using a multi-label approach. Because a single incident could receive multiple domain and cause codes, the counts indicate how often each label appeared across the 300 reports, and the percentages represent the proportion of reports containing that label; therefore, the totals do not add up to 100. These categories are reported on the basis of coding frequency alone, applied as a uniform selection criterion across both fields. Additional diagnostic domain and contributing factor labels were present in the codebook and were coded in a smaller proportion of reports; the three categories presented in each field collectively accounted for the dominant patterns within the registry and are reported here to focus the descriptive summary on the most representative findings. Among the diagnostic domains, pulp diseases were the most frequently coded in this registry (87/300, 29.0%), followed by periodontal ligament diseases (85/300, 28.3%) and TMJ disorders (80/300, 26.7%). These frequencies reflect the domains most commonly represented in the submitted reports within this institutional context and should not be interpreted as evidence that these conditions are inherently the most error-prone diagnostic areas in dentistry. The distribution may reflect the clinical case mix, specialty composition, and reporting culture of the participating institutions as much as the true distribution of diagnostic process failures. Binary logistic regression analysis Univariate and bivariate logistic regression models were constructed to evaluate the relationship between selected demographic or institutional factors and incident urgency classification. Both models were statistically significant (Table 4). Table 4Bivariate associations between selected incident report characteristics and urgent incident classification in the critical incident report registry (n = 300)PredictorCategoryNon-urgent n (%)Urgent n (%)Pearson χ² p-valueCramér’s VFisher’s exact p-value*Institution typeGovernment University97 (39.8)147 (60.2)0.08600.128—Private University23 (50.0)23 (50.0)—Private Clinic7 (70.0)3 (30.0)0.1025Designation groupJunior Faculty111 (44.6)138 (55.4)0.18240.107—Senior Faculty12 (29.3)29 (70.7)—Private Faculty4 (40.0)6 (60.0)0.6762Recurrence category< 20%56 (54.4)47 (45.6)0.00010.266—20–40%43 (29.1)105 (70.9)—40–60%24 (57.1)18 (42.9)—> 60%4 (57.1)3 (42.9)0.4618Impact categoryNo impact8 (72.7)3 (27.3)0.05030.178—Minor impact64 (41.0)92 (59.0)—Moderate impact54 (44.3)68 (55.7)—Major impact1 (10.0)9 (90.0)—Not documented0 (0.0)1 (100.0)1.0000Specialty groupClinical Faculty96 (37.4)161 (62.6)< 0.0010.242—Basic Faculty18 (72.0)7 (28.0)—Private Dentist13 (72.2)5 (27.8)0.0031GenderFemale71 (34.5)135 (65.5)< 0.0010.219—Male56 (59.6)38 (40.4)—Patients per week10–1956 (51.9)52 (48.1)0.03010.145—20–2920 (29.0)49 (71.0)—> 3051 (41.5)72 (58.5)— Table 4Bivariate associations between selected incident report characteristics and urgent incident classification in the critical incident report registry (n = 300)PredictorCategoryNon-urgent n (%)Urgent n (%)Pearson χ² p-valueCramér’s VFisher’s exact p-value*Institution typeGovernment University97 (39.8)147 (60.2)0.08600.128—Private University23 (50.0)23 (50.0)—Private Clinic7 (70.0)3 (30.0)0.1025Designation groupJunior Faculty111 (44.6)138 (55.4)0.18240.107—Senior Faculty12 (29.3)29 (70.7)—Private Faculty4 (40.0)6 (60.0)0.6762Recurrence category< 20%56 (54.4)47 (45.6)0.00010.266—20–40%43 (29.1)105 (70.9)—40–60%24 (57.1)18 (42.9)—> 60%4 (57.1)3 (42.9)0.4618Impact categoryNo impact8 (72.7)3 (27.3)0.05030.178—Minor impact64 (41.0)92 (59.0)—Moderate impact54 (44.3)68 (55.7)—Major impact1 (10.0)9 (90.0)—Not documented0 (0.0)1 (100.0)1.0000Specialty groupClinical Faculty96 (37.4)161 (62.6)< 0.0010.242—Basic Faculty18 (72.0)7 (28.0)—Private Dentist13 (72.2)5 (27.8)0.0031GenderFemale71 (34.5)135 (65.5)< 0.0010.219—Male56 (59.6)38 (40.4)—Patients per week10–1956 (51.9)52 (48.1)0.03010.145—20–2920 (29.0)49 (71.0)—> 3051 (41.5)72 (58.5)— Bivariate associations between selected incident report characteristics and urgent incident classification in the critical incident report registry (n = 300) Values are presented as n (%) within each category. Percentages were calculated row-wise as the proportion of non-urgent or urgent incidents within each predictor category. Pearson’s chi-square test was used as the primary test of association between categorical predictors and urgent incident classification. Cramér’s V is reported as the effect size measure for each predictor block. * Fisher’s exact test was additionally applied as a sensitivity analysis when sparse cell counts were present, particularly for rare categories such as private clinic, private faculty, and > 60% recurrence. Fisher’s exact test does not rely on large-sample assumptions and therefore provides more reliable inference under sparse-data conditions. In all comparisons, Fisher’s exact results were directionally consistent with the corresponding chi-square findings. Table 4 compares urgency across predictors and reports both statistical significance (Pearson’s chi-square) and effect size (Cramér’s V). Because a few categories have small cell counts (for example, private clinic and > 60% recurrence), an additional two-sided Fisher’s exact test was used where appropriate by collapsing the predictor into a 2 × 2 comparison of the rare category versus all other categories. Fisher’s exact test does not rely on large-sample assumptions; therefore, it provides a more reliable sensitivity check when expected counts are low. In this dataset, the Fisher results were consistent with the chi-square interpretation for small cell categories. Recurrence category showed the most statistically robust association with urgent classification within this dataset. After false discovery rate control, recurrence category remained the only predictor with robust evidence of association with urgent classification (Benjamini–Hochberg q = 0.0004). Impact category showed weaker statistical evidence of association with urgent classification and did not remain significant after false discovery rate correction. The impact category showed weak evidence of an association with urgent classification in the unadjusted analysis (Pearson chi-square p = 0.050). However, after controlling for the false discovery rate across the predictor-level tests (Table 4), this association was no longer statistically significant (Benjamini–Hochberg q = 0.1006). Therefore, we interpret the impact of the findings cautiously and treat them as exploratory. Not statistically significant: institution type and designation group Institution type was not significant (p = 0.086, Cramér’s V = 0.128, small effect). Government university reports had 60.2% urgency, private university 50.0%, and private clinic 30.0%. Because private clinics had a small count, Fisher’s exact testing was applied to private clinics versus all other institutions and remained non-significant (two-sided Fisher p = 0.1025), indicating that the apparent lower urgency proportion in private clinics should be interpreted cautiously. The designation group was not significant (p = 0.182, Cramér’s V = 0.107, small effect). Senior faculty reported 70.7% urgent, junior faculty 55.4%, and private practitioners 60.0%. The private practitioner group is small, so the estimates are unstable. Based on these data, urgency classification was more strongly associated with incident attributes such as recurrence and impact than with institutional source or reporter designation. The Fisher sensitivity checks support the same conclusion when small cell categories are present. Table 4 presents bivariate associations between selected reporter and institutional characteristics and urgent incident classification. Recurrence category was the only predictor showing a statistically robust association with urgency after false discovery rate control (Benjamini-Hochberg q = 0.0004), with urgency highest in the 20–40% recurrence group. Impact category showed a borderline association (p = 0.050) that did not survive FDR correction (q = 0.1006) and is treated as exploratory. Institution type and designation group were not significantly associated with urgency at the predictor level. Where expected cell counts were low, Fisher’s exact test was used as a secondary sensitivity check and results were directionally consistent with chi-square findings in all cases. Table 5Univariate logistic regression models for urgent incident classification (n = 300)PredictorCategoryCrude OR95% CIp-valueInstitution typeGovernment University1.00NA0.087Private University0.660.35 to 1.240.198Private Clinic0.280.07 to 1.120.072Designation groupJunior Faculty1.00NA0.172Senior Faculty1.940.95 to 3.980.069Private Practitioner1.210.33 to 4.380.775Specialty groupClinical Faculty1.00NA< 0.001***Basic Faculty0.290.12 to 0.690.005**Private Dentist0.230.08 to 0.670.007**Age group24–351.00NA0.63936–500.790.35 to 1.800.573Above 501.820.35 to 9.560.478GenderFemale1.00NA< 0.001***Male0.410.25 to 0.67< 0.001***Patients per week10–191.00NA0.030*20–292.281.– to 4.330.012*> 301.150.68 to 1.930.602 Table 5Univariate logistic regression models for urgent incident classification (n = 300)PredictorCategoryCrude OR95% CIp-valueInstitution typeGovernment University1.00NA0.087Private University0.660.35 to 1.240.198Private Clinic0.280.07 to 1.120.072Designation groupJunior Faculty1.00NA0.172Senior Faculty1.940.95 to 3.980.069Private Practitioner1.210.33 to 4.380.775Specialty groupClinical Faculty1.00NA< 0.001***Basic Faculty0.290.12 to 0.690.005**Private Dentist0.230.08 to 0.670.007**Age group24–351.00NA0.63936–500.790.35 to 1.800.573Above 501.820.35 to 9.560.478GenderFemale1.00NA< 0.001***Male0.410.25 to 0.67< 0.001***Patients per week10–191.00NA0.030*20–292.281.– to 4.330.012*> 301.150.68 to 1.930.602 Univariate logistic regression models for urgent incident classification (n = 300) Each predictor was modelled using a separate univariate logistic regression. The odds ratios represent the odds of an incident being classified as urgent (1) versus non-urgent (0). For each predictor block, the p-value shown in the baseline (OR = 1.00) row is the overall likelihood ratio test p-value for that predictor block. P-values for the other categories are Wald tests comparing that category with the baseline. NA indicates that the confidence interval was not estimated for the baseline category. Significance: * p < 0.05; ** p < 0.01; *** p < 0.001. Table 5 summarizes the crude odds ratios from separate univariate logistic regression models with urgent classification (urgent = 1 versus non-urgent = 0) as the outcome. For each predictor, the baseline category is shown with OR = 1.00, and the 95% CI is reported as NA because confidence intervals are not estimated for the baseline. Category-level p-values compare each category with the baseline within that predictor, whereas the predictor-level p-value reflects the overall contribution of that predictor to the model. In these analyses, specialty group and gender showed strong associations with urgency of treatment. Compared with clinical faculty, basic faculty (OR 0.29, 95% CI 0.12–0.69, p = 0.005) and private dentists (OR 0.23, 95% CI 0.08–0.67, p = 0.007) had lower odds of urgent classification. Male reporters also had lower odds of urgent classification than female reporters (OR 0.41, 95% CI 0.25–0.67, p < 0.001). Patient load was associated with urgency, with higher odds in the 20–30 patients/week group than in the 10–19 patients/week group (OR 2.28, 95% CI 1.20–4.33, p = 0.012), while the > 30 patients/week group did not differ significantly (OR 1.15, 95% CI 0.68–1.93, p = 0.602). Institution type, designation group, and age group were not statistically significant overall, although some categories showed borderline results. Prior to model interpretation, multicollinearity was assessed using variance inflation factors for all predictors included in the multivariable model. All VIF values were below 3.0 (range 1.23 to 2.64), indicating no concerning collinearity among the predictors (Appendix C). The highest VIF was observed for specialty group (VIF = 2.64), consistent with its partial conceptual overlap with institution type; however, this value falls well within acceptable limits. Pairwise correlations between conceptually related categories — specifically private dentist and private clinic (r = 0.41) and clinical faculty and government university setting (r = 0.38) — did not reach the threshold conventionally used to flag collinearity concern (r = 0.70). These diagnostics confirm that the adjusted estimates reported in Table 6 are not materially affected by multicollinearity. Table 6Multivariable logistic regression model for urgent incident classification (adjusted odds ratios) (n = 300)PredictorCategoryAdjusted OR95% CIp-valueInstitution typeGovernment University1.00NA0.014*Institution typePrivate University0.580.29 to 1.140.115Institution typePrivate Clinic0.160.04 to 0.690.013*Designation groupJunior Faculty1.00NA0.066Designation groupSenior Faculty2.481.12 to 5.480.025*Designation groupPrivate Practitioner1.200.26 to 5.560.815Specialty groupClinical Faculty1.00NA< 0.001***Specialty groupBasic Faculty0.230.09 to 0.600.002**Specialty groupPrivate Dentist0.240.08 to 0.730.013*Age group24–351.00NA0.782Age group36–500.790.32 to 1.960.613Age groupAbove 501.490.27 to 8.130.644GenderFemale1.00NA< 0.001***GenderMale0.370.22 to 0.64< 0.001***Patients per week10–191.00NA0.052Patients per week20–292.261.12 to 4.560.023*Patients per week> 301.140.64 to 2.020.665 Table 6Multivariable logistic regression model for urgent incident classification (adjusted odds ratios) (n = 300)PredictorCategoryAdjusted OR95% CIp-valueInstitution typeGovernment University1.00NA0.014*Institution typePrivate University0.580.29 to 1.140.115Institution typePrivate Clinic0.160.04 to 0.690.013*Designation groupJunior Faculty1.00NA0.066Designation groupSenior Faculty2.481.12 to 5.480.025*Designation groupPrivate Practitioner1.200.26 to 5.560.815Specialty groupClinical Faculty1.00NA< 0.001***Specialty groupBasic Faculty0.230.09 to 0.600.002**Specialty groupPrivate Dentist0.240.08 to 0.730.013*Age group24–351.00NA0.782Age group36–500.790.32 to 1.960.613Age groupAbove 501.490.27 to 8.130.644GenderFemale1.00NA< 0.001***GenderMale0.370.22 to 0.64< 0.001***Patients per week10–191.00NA0.052Patients per week20–292.261.12 to 4.560.023*Patients per week> 301.140.64 to 2.020.665 Multivariable logistic regression model for urgent incident classification (adjusted odds ratios) (n = 300) Predictor-level global p-values (from likelihood ratio tests) are presented in the reference-category row of each predictor block and apply to the predictor as a whole. They should not be interpreted as category-level Wald test results. Category-level p-values in all other rows compare that category with the reference category. This placement convention follows standard practice in biomedical regression reporting and is used to maintain spatial alignment between the global test result and the predictor block it describes. All the predictors were included in a single multivariable logistic regression model. The adjusted odds ratios represent the associations after controlling for all other variables in the table. For each predictor block, the p-value shown in the baseline (OR = 1.00) row is the overall likelihood ratio test p-value for that predictor (full model versus the model excluding that predictor). P-values for the other categories are Wald tests comparing that category with the baseline. NA indicates that the confidence interval was not estimated for the baseline category. Baseline categories: Government University, Junior Faculty, Clinical Faculty, age 24–35 years, female, 10–20 patients/week. Significance: * p < 0.05; ** p < 0.01; *** p < 0.001. Model diagnostics indicated acceptable calibration and discrimination of the primary multivariable logistic regression model. The Hosmer–Lemeshow goodness-of-fit test was non-significant, supporting adequate model calibration. The model demonstrated moderate discriminatory ability, with an area under the receiver operating characteristic curve (AUC) exceeding 0.70. Pseudo-R² statistics indicated modest explanatory performance, which is expected given the exploratory and observational nature of the registry data. Information criteria were additionally examined to support model comparison and sensitivity analyses. Full model diagnostic statistics for the primary and exploratory models are presented in Supplementary Table S3. Table 6 presents the adjusted odds ratios from a single multivariable logistic regression model that included all listed predictors simultaneously, with urgent classification as the outcome. As shown in Table 5, the baseline category within each predictor is shown with OR = 1.00, and the 95% CI is reported as NA. Category-level p-values compare each category with the baseline, and predictor-level p-values indicate whether the predictor improves the model fit after accounting for the other variables in the model. After adjustment, institution type showed an overall association with urgency, with private clinic reports having substantially lower odds of urgent classification than government university reports (adjusted OR 0.16, 95% CI 0.04–0.69, p = 0.013), while private university reports did not differ significantly (adjusted OR 0.58, 95% CI 0.29–1.14, p = 0.115). Designation group showed a borderline overall association, but senior faculty reports had higher adjusted odds of urgent classification than junior faculty reports (adjusted OR 2.48, 95% CI 1.12–5.48, p = 0.025). Specialty group and gender remained strongly associated with urgency, with lower adjusted odds for basic faculty (adjusted OR 0.23, 95% CI 0.09 to 0.60, p = 0.002), private dentists (adjusted OR 0.24, 95% CI 0.08 to 0.73, p = 0.013), and male reporters (adjusted OR 0.37, 95% CI 0.22 to 0.64, p < 0.001). Patient load remained borderline overall (p = 0.052), but the 20–29 patients per week category retained higher adjusted odds of urgent classification than 10–20 (adjusted OR 2.26, 95% CI 1.12 to 4.56, p = 0.023), while the > 30 category was not significant. The age group remained non-significant after adjustment. To assess the robustness of the primary model estimates in the presence of sparse categories, three pre-specified sensitivity analyses were conducted and their results are presented in Supplementary Table S1. Across all three sensitivity analyses, the direction and approximate magnitude of the adjusted estimates for non-sparse predictors — specifically specialty group, gender, and workload category — were consistent with the primary model, confirming that these findings are not sensitive to the treatment of the sparse categories. As expected under Firth’s penalized logistic regression in Sensitivity Analysis 3, the estimates for private clinic (attenuated from aOR 0.16 to 0.24) and private practitoner (attenuated from aOR 1.34 to 0.94) moved toward the null, confirming that these estimates are imprecise under standard maximum likelihood estimation. The direction of association for private clinic was preserved under penalization. These results collectively support the robustness of the primary model for non-sparse predictors while reinforcing that the private clinic, private practitioner, and above-50-years estimates are exploratory and should not be interpreted independently of their wide confidence intervals. Thematic analysis of incident narratives & interview transcripts A total of 26 codes were identified from narrative report fields and interview transcripts and organized into themes and categories Appendix D (Table 7). The themes emphasized the rigor of the diagnostic process, system and training limitations, communication challenges, and the role of structured supervision in preventing recurrence. Table 7 presents the complete thematic structure across four levels of analytical abstraction: overarching themes (the highest-order interpretive constructs), subthemes (interpretive groupings of related categories), categories (conceptual clusters of related codes), and codes (descriptive labels grounded directly in the data). Table 7Thematic analysis of narratives and interview transcripts in critical incident reportsOverarching ThemeCategoryCodeIllustrative Excerpts based on Incident Reports & Interview TranscriptsTheme 1: Accurate Diagnosis & Treatment Planning as a Multi-factorial Competency1.1 History & ExaminationComprehensive history“History and examination were incomplete… working diagnosis changed at follow-up.” (IR-01)(CD-08)“Plus, I think the training should involve specifically the history and examination part, which should be done comprehensively because history and examination are what lead to your diagnosis and treatment.” (BDF1)1.2 Diagnostic ToolsUse of radiographs“Radiograph not obtained due to equipment unavailability… diagnosis based on symptoms alone.” (IR-02)(CD-6)1.3 Knowledge & Human FactorsRole of experience“Findings misinterpreted by junior clinician; corrected after senior review.” (IR-03)(CD-10)Human error“Time pressure and distraction during triage contributed to diagnostic slip.” (IR-04)(BDF4)Prosthodontic subjectivity“Complaint labeled as adaptation; later identified occlusal error.” (IR-05)(CD-6)Variation in treatment planning“Different clinicians proposed different plans for the same case.” (IR-06)(CD-09)1.4 Clinical Decision ChallengesExtract vs. retain“Decision made without complete periodontal charting.” (IR-07)(CD-07)Saving poor-prognosis teeth“Uncertain tooth treated at patient’s request; symptoms persisted.” (IR-08)(CD-14)Prognosis communication“Uncertainty not clearly explained before limited-predictability treatment.” (IR-09)Error-prone conditions“Cracked tooth syndrome not recognized initially.” (IR-10)(CD-03)Endo-perio misclassification“Endo-perio lesion misclassified; periodontal component not addressed.” (IR-11)(CD-08)Theme 2: Impact of Diagnostic Errors2.1 Clinical ConsequencesInappropriate management“Incorrect diagnosis led to delay in definitive care.” (IR-12)(CD-18)Wrong procedure“Procedure initiated on incorrect diagnosis; required reassessment.” (IR-13)(CD-15)Ethical concern“Reporter documented ethical concern and informed supervisor.” (IR-14)(CD-14)2.2 Patient TrustLoss of trust“Patient sought second opinion after repeated visits.” (IR-15)(CD-16)Theme 3: Influencing Factors3.1 Patient FactorsFinancial/medical limits“Imaging or referral postponed due to financial constraints.” (IR-16)(CD-19)3.2 TechnologicalRadiographic interpretation“2D imaging underestimated lesion extent.” (IR-17)(CD-13)3.3 EducationalLimited supervision“Case managed with limited supervision; checklist not used.” (IR-18)(CD-17)Lack of integration“Inter-departmental communication gaps documented.” (IR-19)(CD-20)Limited exposure“Limited exposure contributed to difficulty recognizing atypical cases.” (IR-20)(CD-10)3.4 SystemicDelayed referral“Escalation pathways not followed early.” (IR-21)(CD-09)Theme 4: Perceptions Across Specialties4.1 DenialVariation vs. error“Outcome framed as variation despite persistent symptoms.” (IR-22)(CD-17)Education adequate“Training improved, yet supervision gaps persisted.” (IR-23)(CD-07)4.2 Execution ErrorsOcclusal error“Execution error required adjustment and repeat visit.” (IR-24)(CD-18)Theme 5: Communication & Involvement5.1 Informed ConsentOptions & risks“Options and risks not fully explained; consent incomplete.” (IR-25)(BDF8)Excerpt source codes: IR = incident report narrative field; CD = clinical documentation accompanying the report (e.g., history form, charting record); BDF = background data form, a structured Append field completed by the reporter at the time of submission. Numbers following each abbreviation indicate the anonymized report identifier. Capitalization is standardized as IR, CD, and BDF throughout Table 7Thematic analysis of narratives and interview transcripts in critical incident reportsOverarching ThemeCategoryCodeIllustrative Excerpts based on Incident Reports & Interview TranscriptsTheme 1: Accurate Diagnosis & Treatment Planning as a Multi-factorial Competency1.1 History & ExaminationComprehensive history“History and examination were incomplete… working diagnosis changed at follow-up.” (IR-01)(CD-08)“Plus, I think the training should involve specifically the history and examination part, which should be done comprehensively because history and examination are what lead to your diagnosis and treatment.” (BDF1)1.2 Diagnostic ToolsUse of radiographs“Radiograph not obtained due to equipment unavailability… diagnosis based on symptoms alone.” (IR-02)(CD-6)1.3 Knowledge & Human FactorsRole of experience“Findings misinterpreted by junior clinician; corrected after senior review.” (IR-03)(CD-10)Human error“Time pressure and distraction during triage contributed to diagnostic slip.” (IR-04)(BDF4)Prosthodontic subjectivity“Complaint labeled as adaptation; later identified occlusal error.” (IR-05)(CD-6)Variation in treatment planning“Different clinicians proposed different plans for the same case.” (IR-06)(CD-09)1.4 Clinical Decision ChallengesExtract vs. retain“Decision made without complete periodontal charting.” (IR-07)(CD-07)Saving poor-prognosis teeth“Uncertain tooth treated at patient’s request; symptoms persisted.” (IR-08)(CD-14)Prognosis communication“Uncertainty not clearly explained before limited-predictability treatment.” (IR-09)Error-prone conditions“Cracked tooth syndrome not recognized initially.” (IR-10)(CD-03)Endo-perio misclassification“Endo-perio lesion misclassified; periodontal component not addressed.” (IR-11)(CD-08)Theme 2: Impact of Diagnostic Errors2.1 Clinical ConsequencesInappropriate management“Incorrect diagnosis led to delay in definitive care.” (IR-12)(CD-18)Wrong procedure“Procedure initiated on incorrect diagnosis; required reassessment.” (IR-13)(CD-15)Ethical concern“Reporter documented ethical concern and informed supervisor.” (IR-14)(CD-14)2.2 Patient TrustLoss of trust“Patient sought second opinion after repeated visits.” (IR-15)(CD-16)Theme 3: Influencing Factors3.1 Patient FactorsFinancial/medical limits“Imaging or referral postponed due to financial constraints.” (IR-16)(CD-19)3.2 TechnologicalRadiographic interpretation“2D imaging underestimated lesion extent.” (IR-17)(CD-13)3.3 EducationalLimited supervision“Case managed with limited supervision; checklist not used.” (IR-18)(CD-17)Lack of integration“Inter-departmental communication gaps documented.” (IR-19)(CD-20)Limited exposure“Limited exposure contributed to difficulty recognizing atypical cases.” (IR-20)(CD-10)3.4 SystemicDelayed referral“Escalation pathways not followed early.” (IR-21)(CD-09)Theme 4: Perceptions Across Specialties4.1 DenialVariation vs. error“Outcome framed as variation despite persistent symptoms.” (IR-22)(CD-17)Education adequate“Training improved, yet supervision gaps persisted.” (IR-23)(CD-07)4.2 Execution ErrorsOcclusal error“Execution error required adjustment and repeat visit.” (IR-24)(CD-18)Theme 5: Communication & Involvement5.1 Informed ConsentOptions & risks“Options and risks not fully explained; consent incomplete.” (IR-25)(BDF8)Excerpt source codes: IR = incident report narrative field; CD = clinical documentation accompanying the report (e.g., history form, charting record); BDF = background data form, a structured Append field completed by the reporter at the time of submission. Numbers following each abbreviation indicate the anonymized report identifier. Capitalization is standardized as IR, CD, and BDF throughout Thematic analysis of narratives and interview transcripts in critical incident reports “History and examination were incomplete… working diagnosis changed at follow-up.” (IR-01)(CD-08) “Plus, I think the training should involve specifically the history and examination part, which should be done comprehensively because history and examination are what lead to your diagnosis and treatment.” (BDF1) Excerpt source codes: IR = incident report narrative field; CD = clinical documentation accompanying the report (e.g., history form, charting record); BDF = background data form, a structured Append field completed by the reporter at the time of submission. Numbers following each abbreviation indicate the anonymized report identifier. Capitalization is standardized as IR, CD, and BDF throughout Fig. 2Analytical architecture of the thematic structure displaying the five themes of interview transcripts and narratives of critical incident reports Analytical architecture of the thematic structure displaying the five themes of interview transcripts and narratives of critical incident reports Table 8Joint display of quantitative and qualitative findingsQuantitative FindingQualitative Theme Explaining ItKey Mechanism IdentifiedLack of diagnostic resources most frequently coded (47.3%)Theme 3.2: Technological; Theme 3.3: EducationalRadiographic equipment unavailability and limited training in interpretation led reporters to proceed on clinical findings alone, increasing diagnostic uncertaintyRecurrence strongly associated with urgency (p less than 0.001)Theme 3.3: Limited supervision; Theme 3.4: Delayed referralReporters described systemic gaps in supervision and escalation pathways as conditions allowing similar incident types to recur without correctionMale reporters less likely to have urgent reports (aOR 0.37)Theme 4.1: Denial; variation vs. error framingSome reporters, in particular those less accustomed to safety culture norms, framed diagnostic uncertainty as variation rather than error, potentially influencing urgency perception and documentationSenior faculty reports more likely to be urgent (aOR 2.48)Theme 1.3: Role of experienceReporters described senior review as a moment of reframing — events that junior reporters had not escalated were reclassified as urgent when reviewed by senior clinicians Table 8Joint display of quantitative and qualitative findingsQuantitative FindingQualitative Theme Explaining ItKey Mechanism IdentifiedLack of diagnostic resources most frequently coded (47.3%)Theme 3.2: Technological; Theme 3.3: EducationalRadiographic equipment unavailability and limited training in interpretation led reporters to proceed on clinical findings alone, increasing diagnostic uncertaintyRecurrence strongly associated with urgency (p less than 0.001)Theme 3.3: Limited supervision; Theme 3.4: Delayed referralReporters described systemic gaps in supervision and escalation pathways as conditions allowing similar incident types to recur without correctionMale reporters less likely to have urgent reports (aOR 0.37)Theme 4.1: Denial; variation vs. error framingSome reporters, in particular those less accustomed to safety culture norms, framed diagnostic uncertainty as variation rather than error, potentially influencing urgency perception and documentationSenior faculty reports more likely to be urgent (aOR 2.48)Theme 1.3: Role of experienceReporters described senior review as a moment of reframing — events that junior reporters had not escalated were reclassified as urgent when reviewed by senior clinicians Joint display of quantitative and qualitative findings Discussion DE remains a pressing issue in dentistry, affecting patient safety, healthcare costs, and overall treatment outcomes [32]. The present study provides critical insights into the prevalence of factors contributing to DE in dentistry, creating an urgent need to address this issue for safer health management. The present study highlights the need to address DE from multiple perspectives by combining quantitative evaluation with a qualitative exploration of dental practitioners’ experiences based on critical incident reports. A key strength of this study is the explicit differentiation between diagnostic errors, procedural errors, and near-miss events. While prior literature often aggregates these constructs, such aggregation can obscure causal pathways. In this analysis, procedural errors were examined only when linked to diagnostic failure, and near-miss events were interpreted as markers of system resilience and early error detection. This distinction enhances interpretability by separating failures of clinical reasoning from failures of execution, while still recognizing their interaction within complex clinical systems. To the authors’ knowledge and based on a review of the available published literature, no comparable registry-based analysis of critical incident reports focused specifically on diagnostic process failures in dental teaching settings and private dental surgeries in Pakistan has been published. This study therefore contributes a novel, locally grounded dataset to the regional patient safety literature. The patterns identified reflect reported and recorded events within this registry and should not be interpreted as estimates of the true frequency or distribution of diagnostic errors across dental practice in Pakistan. A substantial proportion of reports in this dataset were classified as requiring attention, reflecting the emphasis on patient safety within dental services (Table 2). This aligns with the wider diagnostic safety literature, which frames diagnostic error as both an accuracy and communication problem with direct patient harm implications [4, 17, 33]. The results demonstrated variations in incident attributes across institutions and reporter roles (Table 1), consistent with evidence that diagnostic failures are associated with both individual and system factors [18]. Insufficient diagnostic resources were the most frequently coded factors in the incident reports (Table 3), followed by inadequate postgraduate training and gaps in undergraduate teaching and supervision (Table 3). It should be noted that these represent reporters’ own attributions as documented in the incident forms, and the cross-sectional design of this study does not permit confirmation of a contributory or causal relationship between these factors and the occurrence of diagnostic errors. Nevertheless. the lack of resources and training limitations have been repeatedly identified as factors strongly associated with dental diagnostic errors in the regional literature [34]. This aligns with safety models that describe how latent system weaknesses combine with frontline decision-making to produce preventable harm [35]. In the bivariate analysis, the recurrence category showed the strongest association with the urgent classification (Table 4). Urgency was highest in the 20–40% recurrence group, and the overall association remained robust when Fisher’s exact test was applied for sparse categories (Table 4). The association between recurrence category and urgency classification is an interesting registry-level pattern that warrants further study. It is consistent with the conceptual expectation that incidents perceived as recurring are more likely to be flagged as urgent, and this pattern may be of descriptive interest to those designing incident review processes. However, a bivariate association within a retrospective cross-sectional registry is insufficient on its own to support the design of specific training interventions or escalation frameworks. Applied evidence from prospective or quasi-experimental studies in comparable settings would be needed before this association is translated into structured educational recommendations. The impact category showed a borderline association with urgency (Table 4), which is clinically plausible because higher severity should trigger faster escalation and review. The incident report abstraction form captured the reporter’s workload and contextual fields. The pattern in the workload variable warrants particular caution. The 20–29 patients per week category showed higher adjusted odds of urgent classification compared with the 10–19 group (aOR 2.26, p = 0.023), but the over-30 group did not differ significantly from the reference (aOR 1.14, p = 0.665). This non-monotonic pattern, where a middle-workload group shows higher urgency odds than either the lower or higher group, suggests that the relationship between workload and urgency classification is not linear and may be unstable. This finding should not be used to draw operational or pedagogical conclusions about optimal patient load. It may reflect distributional characteristics of the sample, unmeasured confounding by specialty or case mix, or random variation given the cell sizes involved, and should be treated as exploratory. Educational or training factors were frequently documented, suggesting that reporters perceived the learning environment and supervision as important contributors to diagnostic failures (Table 2). The distribution of the coded diagnostic domains was broad, with recurrent representation of the pulp, periodontal ligament, and temporomandibular joint conditions (Table 3). The composition of this registry, dominated by reports from government university settings (81.3%), junior faculty (83.0%), and clinical departments (85.7%), should be understood as a reflection of reporting culture and institutional infrastructure as much as a reflection of where diagnostic process failures occur. Government teaching institutions are more likely to have structured reporting channels and organizational expectations around documentation, which may lower the threshold for report submission. Junior faculty, who are earlier in their careers and may be subject to closer supervision or evaluation, may be more motivated to document incidents as part of a professional learning expectation. Clinical faculty encounter a higher volume of patient-facing diagnostic decisions, but they may also have greater institutional incentive to document failures that support calls for additional resources or supervisory support. Private clinic and private practitioner representation was extremely limited, comprising ten reports each, which likely reflects the combination of absent formal reporting infrastructure, lower institutional pressure to document, and possible professional concern about the consequences of self-reporting in an unregulated environment. The patterns observed in this registry may therefore substantially underrepresent diagnostic process failures occurring in private dental practice and among more senior practitioners, and the findings should not be extrapolated to those contexts. DE in dentistry represents a significant concern that affects patient outcomes, trust, and the overall quality of care. Thematic analysis of the incident report narratives revealed five main themes: importance of accurate diagnosis and treatment planning, impact of DE, factors influencing these errors, perceptions of errors across dental specialties, and communication with patients (Fig. 2). Five key themes emerged from the narrative analysis (Table 7). Theme 1 emphasized the importance of comprehensive history-taking, diagnostic tools, and clinical experience in avoiding diagnostic errors. Theme 2 highlighted the significant clinical repercussions of misdiagnosis and the consequent erosion of patient trust. This is consistent with patient-centered evidence showing that dental diagnostic failures can undermine trust and delay appropriate care [26]. Theme 3 explored patient-related, technological, and systemic factors, mirroring quantitative findings that inadequate resources and educational gaps were associated with higher frequencies of documented DE (Tables 2 and 3). Theme 4 addressed divergent perceptions of prosthodontic errors, ranging from denial to recognition of technical mistakes, highlighting the subjective nature of diagnosing complex cases. Finally, Theme 5 stressed the critical need for transparent communication and patient involvement in decision-making, which is central to widely accepted definitions of diagnostic error that explicitly include communication failures [36]. Importance of accurate diagnosis and treatment planning This theme emphasizes the role of comprehensive history-taking and examination in forming accurate diagnoses. Narrative entries frequently noted that incomplete assessments could contribute to misdiagnosis. For example, one report described the need for a more comprehensive history and examination because these steps guide diagnosis and treatment decisions. This aligns with the literature suggesting that thorough patient evaluation is fundamental for diagnostic accuracy [36]. The use of diagnostic tools such as radiography was also emphasized. The lack of utilization of such tools can result in significant oversights. One practitioner noted, “Students are making blunders in initial clinical days, treating patients without prior X-rays.” This is corroborated by evidence showing that radiographic interpretation is prone to error and depends on training, image quality, and structured interpretation methods [37, 38]. This also supports the quantitative signal that the lack of diagnostic resources is a prominent contributing cause (Table 3). Impact of DE Diagnostic errors can lead to inappropriate treatment and persistent patient symptoms. The narratives described concerns about patient harm and ethical implications when treatment proceeds based on an incorrect diagnosis. Such errors can compromise patient safety and professional duty of care [12]. Moreover, diagnostic errors erode patient trust, which is fundamental to the patient-practitioner relationship. One practitioner remarked, “We lose the trust of our patients, and patients visit other dentists.” Trust is a critical component of effective healthcare delivery, and its loss can lead to decreased patient adherence to treatment plans and reluctance to seek future care [12]. Patient accounts of dental diagnostic failures similarly describe harm, delays, and loss of confidence in care [39]. Factors influencing DE Several factors can contribute to diagnostic errors, including clinician experience, patient factors, technological limitations, and educational shortcomings. Limited experience and insufficient training were repeatedly identified as contributing factors (Table 3). This supports the view that diagnostic performance improves with experience and continuous education [40]. This is also consistent with dental faculty reports in comparable contexts, where a lack of postgraduate training and deficiencies in educational strategies are perceived as key drivers of diagnostic errors [41]. Patient factors, such as medical and financial constraints, can influence diagnostic decision-making. Narratives noted that such constraints may pressure clinicians to modify diagnostic or treatment plans, which can compromise optimal care [42]. Technological limitations, particularly in radiographic interpretation, were considered challenges. Narratives have noted that reliance on two-dimensional imaging can lead to misinterpretation. This concern is supported by radiology-focused evidence of interpretation error mechanisms [43]. Advancements in imaging technologies, such as cone-beam computed tomography (CBCT), can enhance diagnostic capabilities and may reduce errors for selected indications [43]. However, CBCT use should remain indication-based and risk-balanced, as the diagnostic benefit must be weighed against radiation exposure and clinical necessity [44]. The adjusted finding that male reporters were less likely to file reports coded as urgent (aOR 0.37, p less than 0.001) requires careful interpretation. This association may reflect genuine differences in the urgency of the incidents that male and female reporters encountered, or it may reflect differences in reporting thresholds, documentation behavior, or urgency perception between gender groups rather than any difference in the true clinical severity of the underlying events. It is also possible that institutional or specialty-related factors correlated with gender in this sample partially account for this pattern, despite adjustment for these variables in the model. The finding should not be over-interpreted as evidence of a clinically meaningful gender difference in diagnostic event urgency, and replication in a larger, more gender-balanced sample would be needed before any substantive inference is drawn. The high-volume public-sector environment of the majority of participating institutions is relevant not only as a descriptor of the study context but as a factor that may have systematically shaped which diagnostic process failures were recognized, reported, and documented. High patient throughput is associated with time pressure during clinical encounters, and this study’s finding that reporters managing 20 to 30 patients per week showed higher adjusted odds of urgent incident classification is consistent with that relationship, though the cross-sectional design does not permit causal interpretation. High-volume environments may also normalize certain diagnostic shortcuts, such as symptom-based diagnosis without radiographic confirmation, in ways that reduce the perceived salience of individual incidents and lower the likelihood of formal reporting for events experienced as routine. Conversely, the sheer volume of patient encounters in these settings creates more opportunities for incidents to be recognized and submitted, potentially inflating the apparent concentration of documented failures relative to lower-volume private practice settings. Reporting culture, which is likely to differ substantially between regulated public teaching institutions and private clinics, may therefore account for a significant portion of the institutional variation in incident characteristics observed in this dataset. Variation in diagnostic errors across specialties Narrative fields showed variations in how incidents were framed across specialties. In some records, diagnostic uncertainty was explicitly documented, with a rationale for differential diagnosis and planned follow-up. In other records, uncertainty was not documented, and the downstream consequences included repeated visits, rework, or referral after persistent symptoms. These contrasts support the need for consistent, evidence-based diagnostic protocols and documentation across all specialties [45, 46]. Quantitatively, specialty groups showed a strong relationship with urgency classification in both univariate and adjusted models (Tables 5 and 6), reinforcing the need for consistent escalation thresholds across disciplines. Dental diagnostic claims from 1990 to 2020 accounted for 8.7% of 58,229 paid claims, with missed diagnoses being the most common and disproportionately involving female patients from the Northeastern United States [39]. While diagnostic error rates have remained stable since 2008 in the United States, nondiagnostic claims have declined, and demographic differences among patients and providers highlight the need for targeted strategies to reduce diagnostic errors and improve patient safety. This discrepancy highlights how normalizing clinical variance may mask the importance of consistent evidence-based protocols. These differing viewpoints highlight the need for heightened diagnostic vigilance, meticulous treatment planning, and continued education across all dental specialties to minimize preventable harm and maintain patient trust. The distribution of diagnostic domains documented in this registry requires cautious interpretation. The concentration of coded incidents in pulp diseases, periodontal ligament diseases, and TMJ disorders reflects the domains most frequently represented in submitted reports within these institutional settings, not necessarily the domains in which diagnostic process failures occur most often in dental practice. Several properties of voluntary incident reporting systems may have shaped this distribution. Under-reporting is endemic to such systems, and incidents perceived as minor, routine, or professionally sensitive may be systematically less likely to be submitted. Selective reporting may concentrate entries around clinical areas that are most commonly encountered in the high-volume teaching clinic environment of the majority of participating institutions, where pulp, periodontal, and temporomandibular joint cases represent a large proportion of the clinical workload. Hindsight bias may further influence which events are retrospectively framed as errors. These factors mean that the coded domain frequencies should be described as characteristics of reported events within this registry rather than as evidence of which diagnostic domains carry the highest inherent error risk in dentistry. Communication and patient involvement Effective communication and involving patients in treatment decisions emerged as essential for preventing DE and enhancing patient satisfaction. Providing comprehensive information about treatment options and associated risks allows patients to make informed decisions, aligning with the principles of patient-centered care [47, 48]. Narratives repeatedly stressed that patients should be informed about diagnostic uncertainty, options, and referral pathways to support shared decision-making. (incident report IR-19) This is also consistent with major diagnostic safety frameworks that include failure to communicate the diagnostic explanation as a core component of diagnostic error [49]. Educational implications The need for improved dental education and supervision was identified as a recurring theme. Narratives often recommend active learning approaches, including problem-based learning, to strengthen diagnostic reasoning. This is supported by evidence that PBL can improve diagnostic reasoning and clinical problem-solving in dental education, although implementation quality and curriculum balance are important [50, 51]. Increasing structured clinical exposure has also been noted as a preventive strategy [52]. Narratives suggest that supervised patient encounters were perceived by reporters as facilitating the translation of theoretical knowledge into clinical practice, and as being associated with stronger diagnostic reasoning, though these remain experiential perceptions rather than causally demonstrated relationships [53]. The adjusted regression finding that reports filed by senior faculty were more likely to be coded as urgent (Table 6) is consistent with the possibility that greater clinical experience may be associated with earlier recognition of urgency, though this interpretation remains speculative within a cross-sectional design. When viewed together, the quantitative results delineate broad patterns in urgency coding, contributing factors, and institutional correlates (Tables 1, 2, 3, 4, 5 and 6). Qualitative narrative analysis provides explanatory details on how limited diagnostic tools, inconsistent referral pathways, and training gaps can converge to produce diagnostic process failures (Table 7). Furthermore, the strong association between the recurrence category and urgency (Table 4) can be used pedagogically to teach risk escalation decisions. It encourages students to treat repeated failures as system signals, rather than isolated mistakes. Table 8 presents a joint display of the quantitative findings that the qualitative phase was designed to explain and the themes that provided interpretive elaboration. This display makes explicit how the two phases were integrated and what explanatory value the qualitative data added to the quantitative patterns. This study has several limitations that should be considered when interpreting its findings. Because the design was cross-sectional and based on retrospective review of incident reports, the regression findings reflect associations rather than causal relationships. Differences in urgency classification may reflect case mix, local reporting culture, escalation practices, and clinical severity rather than true differences between groups, and this should be tempered when drawing conclusions. The analytical sample of 300 reports represents a census of eligible records within a defined setting and time period rather than a randomly drawn sample from a larger population. One hundred reports were excluded primarily due to incomplete documentation of core structured fields required for coding. Because the excluded reports were not formally characterized by institution type, incident category, or reporter designation, the possibility that incomplete documentation is non-randomly distributed cannot be ruled out. For instance, incomplete records may be more prevalent in lower-resource environments or among less experienced reporters, which could affect the representativeness of the final sample and should be acknowledged as a potential source of selection bias. The registry is weighted toward government university settings and early-career junior faculty, which reflects the reporting infrastructure of the participating institutions but limits the generalizability of findings to private practice and more senior clinical contexts. The private clinic subgroup comprised only ten reports, making estimates for this group statistically imprecise and potentially reflective of reporting behavior rather than true differences in incident urgency or characteristics. Incident reports are subject to reporting bias, as practitioners may selectively document incidents that are more memorable, more severe, or more clearly linked to identifiable errors. Near-miss events and incidents with ambiguous diagnostic attribution may be systematically underreported, which means the registry likely underestimates the true frequency and diversity of diagnostic process failures in the study settings. The urgency classification, while supported by content and construct validity evidence, was ultimately an abstractor judgment applied retrospectively to documented records. It may not fully capture the clinical acuity as perceived by the treating clinician at the time of the incident. Borderline statistical signals, particularly those near conventional significance thresholds, were treated as uncertain and interpreted alongside effect sizes and confidence intervals rather than as definitive findings. Throughout this manuscript, findings are reported as associations rather than causal relationships. The cross-sectional retrospective design does not permit conclusions about the direction of effect, temporal precedence, or the causal role of any documented factor. All language referring to contributors, drivers, or influencing factors reflects the terminology used by reporters within the incident forms or by cited literature, and should not be interpreted as evidence of causation established by this study. From a health services research perspective, these findings provide actionable quality and safety signals that can support governance decisions in settings where routine diagnostic audit data are limited. The next step is to embed structured (and ideally digital) incident reporting with simple feedback dashboards to enable knowledge translation and cross-sector collaboration between regulators, universities, and private clinics, improving diagnostic workflows in line with SDG 3 goals. This study emphasizes the importance of developing and regularly updating evidence-based clinical guidelines for all dental specialties. Integrating structured critical incident reporting systems can help track and analyze diagnostic oversights, thereby promoting a culture of transparency and collective learning [54, 55]. Regular case discussions and clinical meetings should incorporate these reports to encourage practitioners to share their experiences and solutions that facilitate improved diagnostic accuracy. Enhancing diagnostic competencies requires more vigorous educational approaches, such as problem-based learning modules, simulation training, and deeper clinical exposure, alongside the expanded use of advanced diagnostic technologies, including cone-beam computed tomography and artificial intelligence tools. Strengthening referral networks and interdisciplinary collaboration is vital to ensure that complex cases receive timely and specialized input. Finally, consistent patient-centered communication, with an emphasis on shared decision-making, can further minimize DE by promoting trust and a clearer understanding between patients and dental professionals. Conclusions This study identified recurrent diagnostic process failures in dental clinical settings that were associated with resource limitations, gaps in training and supervision, and inconsistencies in escalation and referral pathways. Critical incident reporting provided a practical mechanism for identifying patterns in diagnostic safety and highlighted the importance of structured supervision, comprehensive assessment, effective communication, and timely escalation in reducing preventable harm. The integration of quantitative and qualitative findings supports the use of incident-report registries as educational and governance tools for strengthening diagnostic reasoning, patient safety culture, and safer clinical workflows in dentistry. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2 Supplementary Material 3 Supplementary Material 3 Supplementary Material 4 Supplementary Material 4 Supplementary Material 5 Supplementary Material 5 Supplementary Material 6 Supplementary Material 6 Supplementary Material 7 Supplementary Material 7