Diagnostic Reasoning in General Internal Medicine Consultations: A Text Mining Analysis of Consultation Requests Authors: Ren Kawamura, Yukinori Harada, Taro Shimizu Published: 2026 ABSTRACT ABSTRACTBackgroundDiagnostic reasoning is often difficult to evaluate because it is largely internal, and linguistic features associated with consultation quality remain poorly understood. Consultation requests may provide a useful window into real‐time diagnostic thinking. We used text mining to examine how language in consultation requests reflects diagnostic reasoning in general internal medicine.MethodsWe conducted a single‐center retrospective observational study of diagnostic consultation requests from the Department of Diagnostic and Generalist Medicine at a Japanese university hospital in 2021. Consultations were grouped using an outcome‐based operational definition of appropriateness, centered on whether the first specialty department established the correct final diagnosis, with an additional concordance requirement when a specific diagnosis was strongly suspected. Text mining using KH Coder included co‐occurrence network analysis, correspondence analysis, and keyword‐in‐context (KWIC) concordance analysis.ResultsAmong 352 consultations, 229 were classified as appropriate and 123 as inappropriate. Appropriate consultations were characterized by terms indicating clear abnormalities and disease progression, whereas inappropriate consultations were associated with terms reflecting limited diagnostic direction, such as “unknown” and “negative”. KWIC analysis showed two uses of “negative”: revising the differential diagnosis after negative results and maintaining suspicion despite negative findings. In inappropriate consultations, these patterns were accompanied by potentially insufficient assessment/documentation and diagnostic uncertainty.ConclusionsText mining of consultation requests revealed linguistic features associated with consultation appropriateness and showed that “negative” reflected distinct diagnostic reasoning moves. Consultation request text may provide a scalable window into diagnostic reasoning and identify opportunities to improve diagnostic communication and consultation quality.Keywords: clinical communication, consultation requests, diagnostic excellence, diagnostic reasoning, general internal medicine, text mining Introduction In recent years, the pursuit of “diagnostic excellence”—defined as providing accurate and timely diagnoses that enhance patients' experiences, manage uncertainty, and use minimal resources—has been recognized as a core component of high‐quality healthcare [1, 2]. Achieving diagnostic excellence often requires collaboration across specialties, and general internists in academic hospitals frequently play a central role in coordinating this process [1, 3]. One common setting in which this collaborative diagnostic work occurs is consultation from general internal medicine (GIM) to specialty departments. Previous studies suggest that a substantial proportion of such consultations are suboptimal or could be improved [4, 5]. However, the factors that distinguish more effective from less effective diagnostic consultations remain insufficiently understood, particularly with respect to the diagnostic reasoning conveyed by the referring physicians. Diagnostic reasoning is difficult to study because much of it remains internal. Although some aspects of reasoning are documented in the medical records, such documentation is often heterogeneous and may be revised after the clinical course becomes clearer. In contrast, consultation requests usually summarize the patient's history, physical findings, test results, and the referring physician's diagnostic considerations at the time the consultation is initiated, making them a useful source for examining real‐time diagnostic thinking. Text mining offers a practical way to analyze such consultation requests at scale. By combining qualitative and quantitative analysis of large text datasets, this approach can help identify linguistic patterns associated with more and less effective consultations. It also allows both successful and problematic consultation processes to be examined, providing complementary Safety‐II and Safety‐I perspectives [6, 7]. Accordingly, this study aimed to examine how diagnostic reasoning expressed in GIM consultation requests was associated with consultation appropriateness, defined in relation to whether the first specialty department ultimately established the correct final diagnosis. Using text mining analysis, we compared the linguistic features of appropriate and inappropriate diagnostic consultations and explored how these features reflect different diagnostic reasoning patterns. Methods Study Design This single‐center, retrospective observational study was conducted at Dokkyo Medical University Hospital between January 1 and December 31, 2021. The hospital is a Japanese university medical center with 29 clinical departments. The Department of Diagnostic and Generalist Medicine (DGM) functions as the institution's general internal medicine service, providing outpatient, inpatient, and emergency care. Ethical Consideration This study complied with the Declaration of Helsinki and was approved by the Research Ethics Committee of the Dokkyo Medical University Hospital (R‐93‐8J). All data were anonymous and aggregated; therefore, the requirement to obtain written informed consent from patients was waived by the Research Ethics Committee under the condition that we used an opt‐out method. Participants and Data Source Consultation requests generated by the DGM were reviewed. All recorded diagnostic consultation requests submitted to specialty departments were screened. Eligible consultations included patients aged ≥ 15 years for whom a diagnostic evaluation was explicitly requested. Patients who refused to participate were excluded. Classification of Consultations Diagnostic consultations were classified into two categories based on the level of diagnostic certainty of the DGM physicians: diagnosis with strong diagnostic suspicion (DSS) and diagnosis for others (DFO). DSS referred to consultations in which DGM physicians explicitly documented a specific suspected diagnosis in the consultation request, whereas all remaining consultations were categorized as DFO. Definition of Appropriate Diagnostic Consultations Appropriateness was assessed based on diagnostic outcomes. Consultations were considered appropriate when the correct final diagnosis was established by the first specialty department. For consultations classified as DSS, appropriateness additionally required that the correct final diagnosis was documented in the consultation request. Consultations that did not meet these criteria were classified as inappropriate. The rationale for this asymmetric operational definition was that DSS and DFO represented different diagnostic tasks at the time of consultation. In DSS consultations, DGM physicians explicitly documented a specific suspected diagnosis; therefore, appropriateness required concordance between the suspected diagnosis in the original request and the final diagnosis, in addition to diagnostic confirmation by the first specialty department. In contrast, DFO consultations were initiated without a specific diagnosis being strongly suspected; therefore, appropriateness was defined by whether the first specialty department established the correct final diagnosis. This operational definition was used to classify consultations for text‐mining comparison and was not intended to fully capture the clinical necessity, timeliness, or documentation quality of each consultation. This study used the same underlying dataset as another manuscript currently under review, which examined the prevalence of appropriate diagnostic consultations and factors associated with inappropriateness. In contrast, the present study examined linguistic features of consultation requests and their relationship to diagnostic reasoning using text mining methods. Accordingly, consultation appropriateness was used in this study primarily as an analytic grouping variable for text comparison rather than as the primary outcome. Text Mining Analysis Text mining analysis quantifies textual data through morphological analysis by calculating word frequencies and visualizing relationships and occurrence patterns among words [8]. This approach enables both quantitative evaluation and qualitative semantic interpretation of large text corpora that would be impractical to analyze through manual review alone [8, 9]. In this study, text mining analysis was conducted using KH Coder (version 3.03a), a software package widely used in medical research [8, 9, 10]. Three analytic approaches were employed to explore linguistic patterns and associations within the consultation requests: co‐occurrence network analysis, correspondence analysis, and keyword‐in‐context (KWIC) concordance analysis. Text Processing and Pre‐Analysis Procedures All diagnostic consultation request documents were processed before text mining analysis. The raw text was exported from the electronic medical record system in Microsoft Excel format. First, formulaic expressions unrelated to consultation content, including greetings and the requester's department and name, were removed from the consultation request documents. These elements were excluded because of their frequent occurrence across requests and their potential to bias frequency‐based analyses. For subsequent analyses, three Microsoft Excel files were prepared: one including all diagnostic consultation requests, one including appropriate consultations, and one including inappropriate consultations. Because the original language of the consultation requests was Japanese, morphological analysis was performed to segment the text into meaningful lexical units. Tokenization and part‐of‐speech tagging were conducted automatically using KH Coder, which incorporates the Japanese morphological analyzer ChaSen. The need for a minimal set of force pick‐up and force ignore terms was then assessed to minimize analytical bias. Based on the extracted results, the terms “kika” (your department), “chief complaint”, and “collagen disease” were designated as force pick‐up terms in the analysis of all diagnostic consultations. In analyses stratified by appropriateness, “kika” (your department), “touka” (our department), and “chief complaint” were designated as force pick‐up terms. No force ignore terms were applied in any analysis. Co‐Occurrence Network Analysis A co‐occurrence network is a semantic network that visualizes relationships between words that appear together within a text. In this network, words were represented as nodes, and edges represented co‐occurrence relationships between words. This approach enabled the identification of semantic structure and detection of thematic clusters [8, 11]. Node size reflected word frequency, edge thickness represented the strength of co‐occurrence, and closely connected clusters indicated related terms. The 60 strongest co‐occurrence links, ranked by Jaccard coefficient, were included in the analysis. The threshold of 60 links was chosen to balance thematic coverage, visual interpretability, and clinical interpretability of the network. In preliminary outputs, including a larger number of links increased label overlap and introduced more peripheral co‐occurrence links, which reduced the clarity of the major thematic clusters; conversely, fewer links provided insufficient coverage of the main co‐occurrence structures. Separate co‐occurrence networks were generated for consultation requests classified as appropriate or inappropriate. Themes were then identified from each network using a combined approach with KWIC concordance analysis, as described later. Each co‐occurrence network was then compared to explore whether co‐occurrence patterns differed by consultation appropriateness. Correspondence Analysis Correspondence analysis projects words into a coordinate space based on predefined external variables, enabling visual comparison of term distributions across groups. In this method, neither axis has inherent meaning; interpretation relies on the distance and direction of each term relative to the origin. Words plotted near the origin are less distinctive and show weak associations with external variables, whereas those positioned farther in the direction of a variable are considered characteristic of that variable. The absolute distance between terms and variables does not have interpretive meaning. In this study, correspondence analysis was performed using consultation appropriateness as an external variable to compare characteristic terms between appropriate and inappropriate diagnostic consultations. KWIC Concordance Analysis KWIC concordance analysis was used to examine the contextual nuance and interpretive meaning of selected terms by reviewing the surrounding text in which they appeared [12], based on words identified through the co‐occurrence network and correspondence analysis. After the co‐occurrence network and correspondence analyses identified “negative” as a term of interest, an exploratory inductive KWIC review was conducted to clarify its contextual meaning. The first author reviewed all KWIC contexts containing the term “negative” and developed contextual categories based on how the term functioned within each consultation request. Through this process, two contextual uses were identified: use of negative test results to suggest alternative diagnoses, and continued suspicion of a disease despite negative findings. The category definitions and interpretations were subsequently reviewed and discussed among all authors to ensure clinical interpretability. Any uncertainties in interpretation were resolved through author discussion. Results A total of 352 diagnostic consultations were included in the analysis, including 116 classified as DSS and 236 as DFO. For the purpose of text mining comparison, consultations were grouped as appropriate (n = 229) and inappropriate (n = 123) according to the operational definition described above. The cross‐tabulation of DSS/DFO classification and consultation appropriateness is shown in Table 1. The distribution of consulted specialty departments according to consultation appropriateness is shown in Table S1. TABLE 1Relationship between DSS/DFO classification and consultation appropriateness.Consultation certaintyAppropriate, n (%)Inappropriate, n (%)Total, n DSS83 (71.6)33 (28.4)116DFO146 (61.9)90 (38.1)236Total229 (65.1)123 (34.9)352 Note: Percentages are row percentages.Abbreviations: DFO, diagnosis for others; DSS, diagnosis with strong diagnostic suspicion. Relationship between DSS/DFO classification and consultation appropriateness. Note: Percentages are row percentages. Abbreviations: DFO, diagnosis for others; DSS, diagnosis with strong diagnostic suspicion. After preprocessing, 33,561 tokens were retained for text mining analysis in the corpus of all diagnostic consultation requests. In separate analyses, 22,062 tokens were analyzed for appropriate consultations and 11,612 tokens for inappropriate consultations. These tokens served as the basis for the co‐occurrence network, correspondence analysis, and KWIC concordance. Co‐Occurrence Network Analysis Separate co‐occurrence networks were generated for appropriate and inappropriate consultations, and nine shared themes (a–i) were identified across both networks (Figure 1). In the inappropriate network, the words “antibody” and “negative” frequently co‐occurred. Co‐occurrence networks of diagnostic consultation requests. (A) Appropriate consultations. (B) Inappropriate consultations. Node size reflects word frequency, and edge thickness reflects the strength of co‐occurrence. Nine themes were identified: (a) patient background, (b) medical visit history, (c) symptoms, (d) symptom progression, (e) test results, (f) physician's reasoning, (g) consultation process, (h) patient preferences, and (i) template expressions. Correspondence Analysis Correspondence analysis was performed using consultation appropriateness as the external variable (Figure 2). Because consultation appropriateness consisted of two categories, the correspondence analysis yielded a single dimension; Dimension 1 explained 100% of the total inertia (inertia = 0.0482). In appropriate consultations, characteristic terms included those describing clear abnormalities (e.g., swelling, rash, aneurysm) and indicators of disease progression (e.g., progression, worsening). In contrast, inappropriate consultations were characterized by terms reflecting a lack of diagnostic direction (e.g., unknown, negative). Frequencies, relative frequencies, and relative frequency ratios of these selected terms are shown in Table S2. Correspondence analysis of terms associated with consultation appropriateness. Because consultation appropriateness consisted of two categories, the analysis yielded a single dimension; Dimension 1 explained 100% of the total inertia (inertia = 0.0482). Terms positioned farther from the origin in the direction of each category are interpreted as more characteristic of that category. The axis does not have inherent semantic meaning. KWIC Concordance Findings In the KWIC concordance analysis, the term negative was used in two distinct ways: (1) to suggest alternative diagnoses based on negative test results, and (2) to continue suspecting a disease despite negative findings (Table 2). In appropriate consultations, Type (1) typically reflected an explicit shift in the working diagnosis or differential diagnosis following negative results, whereas Type (2) was characterized by the presence of sufficient supporting findings that justified continued suspicion despite partial test negativity (Table S3). TABLE 2Contextual uses of negative in consultation requests and interpretive patterns identified by KWIC review.Contextual use of “negative”Appropriate consultations: typical interpretationInappropriate consultations: potentially insufficient assessmentInappropriate consultations: diagnostic uncertaintyType (1): “Negative” used to suggest alternative diagnosesNegative results were used to revise the working diagnosis/differential and propose an alternative explanation consistent with available findingsRequests sometimes suggested limited clinical assessment or test interpretation, such that the “alternative” diagnosis appeared insufficiently supported (e.g., incomplete physical examination or incomplete interpretation of prior workup)Requests reflected uncertainty in which negative results prompted consideration of alternatives, but definitive direction remained unclear because of evolving/discordant findings or inability to reach a final diagnosisType (2): “Negative” used to continue suspecting a disease despite negative findingsContinued suspicion was typically accompanied by additional supportive findings (clinical course, imaging, physical findings) that were consistent with persistence despite partial negativityPersistence despite negativity sometimes appeared to occur with insufficiently documented supporting assessments, including incomplete physical examination or incomplete interpretation of prior workupPersistence despite negativity also occurred under legitimate uncertainty, such as discordant results, evolving disease course, or incomplete diagnostic confirmation despite ongoing evaluation Contextual uses of negative in consultation requests and interpretive patterns identified by KWIC review. In inappropriate consultations, both uses of negative were observed; however, KWIC‐based review identified two overarching features regardless of whether the usage corresponded to Type (1) or Type (2) (Table S4). First, several consultations suggested potentially insufficient assessment (e.g., incomplete physical examination, limited information gathering, or inadequate interpretation of test results). Second, other consultations reflected diagnostic uncertainty, such as insufficient diagnostic evidence at the time of referral, the need for longitudinal reassessment by the specialty service because the diagnosis could evolve over time, or inability to complete the evaluation owing to death during the diagnostic workup. Discussion In this single‐center retrospective study, we applied text mining to general internal medicine consultation requests to examine how language reflecting diagnostic reasoning relates to consultation appropriateness as defined by our outcome‐based operational criteria. Across analytic approaches, appropriate consultations were characterized by terms describing clear abnormalities and clinical progression, whereas consultations classified as inappropriate under our operational definition were associated with terms suggesting limited diagnostic direction, such as “unknown” and “negative”. Notably, KWIC concordance analysis demonstrated that the term “negative” functioned as a marker of distinct reasoning moves rather than a unitary concept: it was used either to redirect the differential diagnosis following negative test results or to maintain suspicion for a disease despite negative findings. Furthermore, among consultations classified as inappropriate, KWIC‐based review suggested two overarching features that were evident regardless of these contextual uses: potentially insufficient assessment/documentation and diagnostic uncertainty. Interpretation of “Negative” as Distinct Reasoning Moves Our KWIC findings highlight that negative does not simply represent the absence of abnormalities; instead, it appears to index how clinicians respond to disconfirming information during diagnostic work. In appropriate consultations, Type (1) use of negative typically reflected an explicit shift in the working diagnosis or differential diagnosis after negative results. This pattern is consistent with iterative diagnostic reasoning, in which accumulating information leads clinicians to revise the working diagnosis and re‐prioritize the differential diagnosis [13]. In this context, negative results can serve as constructive constraints on the diagnostic space, encouraging clinicians to align subsequent hypotheses with the evolving evidence. In contrast, Type (2) use of negative—continuing to suspect a disease despite negative findings—was not necessarily inappropriate. In appropriate consultations, persistence despite partial negativity was characterized by the presence of sufficient supporting findings described in the request. This suggests that clinicians were not simply ignoring negative data but instead contextualizing it within a broader evidentiary picture, in which other features of the clinical presentation plausibly outweighed the disconfirming test result. Such persistence may reflect recognition of test limitations, pretest probability, timing effects, or disease heterogeneity [14, 15], and is therefore compatible with high‐quality diagnostic reasoning when the clinical rationale is clearly described in the consultation requests. However, Type (2) use of negative can also reflect vulnerability in diagnostic reasoning when the rationale for overriding negative results is not articulated. Because consultation requests function both as clinical communication and as an externalization of the referrer's diagnostic thinking, the extent to which the meaning of negative is explained may influence how effectively specialty services can understand the clinical question, evaluate alternative hypotheses, and contribute to diagnostic closure [16]. Two Overarching Features in Inappropriate Consultations A key contribution of this study is the observation that consultations classified as inappropriate under our operational definition were not homogeneous. Regardless of whether negative was used to shift the differential (Type [1]) or to persist with a suspected diagnosis (Type [2]), KWIC‐based review suggested two overarching features that could underlie the “inappropriate” classification: potentially insufficient assessment/documentation and diagnostic uncertainty. First, several consultations suggested potentially insufficient assessment. In these cases, the consultation requests appeared to contain limited Supporting Information, including incomplete physical examination findings, constrained clinical history, or insufficient interpretation of prior test results. Importantly, this pattern should be interpreted cautiously. Because we analyzed consultation requests rather than the full medical record, limited documentation may represent either genuinely incomplete assessment or incomplete communication of information that was assessed but not recorded in the request. Nonetheless, from a systems perspective, inadequate articulation of key diagnostic reasoning steps in consultation requests is itself consequential, as it may impair shared understanding of the clinical problem and reduce the efficiency and effectiveness of specialty input [16, 17, 18]. Thus, the pattern of potentially insufficient assessment/documentation can be viewed as a modifiable target for quality improvement—through training, templates, or feedback—without attributing intent or competence to individual clinicians [19, 20]. Second, other consultations reflected diagnostic uncertainty. These included situations in which diagnostic evidence was limited at the time of consultation, cases requiring longitudinal reassessment because the working diagnosis could evolve, and cases in which further evaluation was precluded by death during the diagnostic workup. This finding is clinically intuitive: diagnostic processes often unfold over time, and early specialty consultation may be appropriate even when definitive diagnostic features have not yet emerged [21, 22]. In addition, some trajectories inherently resist immediate diagnostic closure [23, 24, 25]. Importantly, this uncertainty‐related feature intersects with our operational definition of appropriateness, which requires the correct final diagnosis to be established by the first specialty department. Under this definition, consultations that were clinically reasonable but could not yield immediate diagnostic confirmation—because of evolving disease, discordant results, or interrupted evaluation—may be classified as inappropriate. Accordingly, the observed uncertainty‐related feature underscores that “inappropriate consultation” may sometimes reflect limitations of outcome‐based classification rather than remediable deficiencies in diagnostic reasoning or consultation behavior. Implications for Diagnostic Excellence and Consultation Quality Our findings have practical implications for diagnostic excellence and consultation quality in academic general internal medicine settings. First, the language patterns observed in appropriate consultations suggest that high‐quality consultation requests may explicitly convey (i) salient positive evidence, (ii) the interpretation of negative findings, and (iii) the intended reasoning move—either revising the differential (Type [1]) or explaining persistence (Type [2]). Conversely, in inappropriate consultations, negative and related language often coincides with either limited articulation of reasoning or genuine uncertainty. These observations support the potential value of structured consultation request formats that prompt clinicians to make their reasoning explicit [19, 20, 26, 27, 28]. For example, when documenting negative findings, a template could encourage clinicians to specify whether the negative result is being used to redirect the differential or whether suspicion is being maintained despite negativity, and to provide brief justification (e.g., key supportive features, test limitations, timing, or competing hypotheses). Such structured documentation may improve the efficiency of specialty evaluation, promote shared situational awareness, and create opportunities for feedback and learning [20, 27, 29]. From a Safety‐I perspective, identifying text features associated with potentially insufficient assessment/documentation could support targeted interventions to reduce avoidable inefficiencies. From a Safety‐II perspective, the language patterns in appropriate consultations could be used to define exemplars of effective reasoning and communication, informing training programs for residents and junior clinicians. Relationship With Prior Work and Contribution of Text Mining Prior studies have reported that a considerable proportion of consultations may be inappropriate or improvable [4, 5], yet the underlying diagnostic thought processes and communication features contributing to such consultations have been difficult to capture. Diagnostic reasoning is largely internal, and standard medical record documentation may be heterogeneous and subject to revision. By focusing on consultation requests—which often contain patient history, salient findings, and the referrer's diagnostic considerations and are less likely to be modified after specialty input—we were able to evaluate linguistic correlates of diagnostic reasoning at scale. The combined use of co‐occurrence networks, correspondence analysis, and KWIC concordance enabled both quantitative pattern detection and qualitative interpretation of key terms within their textual context [8, 9, 10, 11, 12]. In particular, the KWIC‐based typology of negative usage provides a concrete example of how text mining can reveal distinct reasoning moves embedded in routine clinical documentation. Limitations This study has several limitations. First, the study was conducted at a single Japanese academic hospital, and consultation culture, documentation norms, and specialty access may differ across institutions and healthcare systems, limiting generalizability. Relatedly, some linguistic features, such as terms related to antibody testing, skin findings, or vascular abnormalities, may reflect differences in case mix or consulted specialty departments rather than consultation appropriateness per se. Therefore, the observed associations should be interpreted as descriptive patterns rather than specialty‐independent markers of appropriateness. Second, our appropriateness definition was outcome‐based and required the correct final diagnosis to be established by the first specialty department; for DSS consultations, appropriateness also required that the correct final diagnosis be documented in the consultation request. While these criteria provide an objective framework, they may misclassify clinically reasonable referrals as inappropriate in contexts where diagnostic confirmation is delayed, evolves over time, or cannot be completed (e.g., death during workup). Conversely, consultations classified as appropriate may still have included incomplete reasoning or documentation in the original request, because appropriateness was determined by diagnostic outcome rather than by an independent assessment of request quality. Third, text mining relies on morphological analysis and tokenization; although we minimized analytic bias by using a limited set of force pick‐up terms and no force ignore terms, results may still be influenced by dictionary performance, synonymous expressions, and context‐dependent meaning in Japanese clinical writing. Fourth, consultation requests represent only a portion of diagnostic reasoning and may not fully capture oral communication, team discussions, or data available elsewhere in the electronic record. Finally, the Type 1/Type 2 categorization and interpretation of KWIC contexts were based on qualitative review by the first author and subsequent discussion among all authors; independent blinded coding and formal inter‐rater agreement assessment were not performed. Therefore, the classification may have been influenced by reviewer interpretation, and future studies could strengthen reproducibility by pre‐specifying coding rules and assessing inter‐rater agreement. Future Directions Future work should validate these findings across multiple centers and explore whether the identified linguistic patterns are robust across specialties, disease categories, and levels of referrer experience. Further, the KWIC‐derived typology of negative could be operationalized into measurable features—for example, quantifying the extent of supportive evidence documented when suspicion persists despite negative results—and tested for association with downstream outcomes such as diagnostic delay, repeat consultation, testing intensity, or length of stay. Finally, intervention studies evaluating structured consultation templates or feedback systems could assess whether improving the explicitness of diagnostic reasoning in consultation requests enhances consultation efficiency and contributes to diagnostic excellence. Conclusions Text mining of general internal medicine consultation requests revealed distinct linguistic features associated with outcome‐based consultation appropriateness and demonstrated that the term negative reflects different diagnostic reasoning moves. Among consultations classified as inappropriate under this operational definition, KWIC‐based review suggested two overarching features—potentially insufficient assessment/documentation and diagnostic uncertainty—present across contextual uses of negative. These findings support the use of consultation request text as a scalable window into diagnostic reasoning, while highlighting the need to distinguish modifiable communication issues from diagnostic uncertainty and limitations of outcome‐based classification. Author Contributions Ren Kawamura: conceptualization, methodology, investigation, funding acquisition, visualization, project administration, writing – original draft, formal analysis, data curation. Yukinori Harada: writing – review and editing, supervision, validation, investigation. Taro Shimizu: writing – review and editing, supervision, validation. Funding This work was supported by JSPS KAKENHI Grant Number JP24K20178 and JP25K02855. Ethics Statement This study complied with the Declaration of Helsinki and was approved by the Research Ethics Committee of the Dokkyo Medical University Hospital (R‐93‐8J). Consent All data were anonymous and aggregated; therefore, the requirement to obtain written informed consent from patients was waived by the Research Ethics Committee, provided that we used an opt‐out method. Conflicts of Interest The authors declare no conflicts of interest. Supporting information Table S1: Distribution of consulted specialty departments according to consultation appropriateness. Table S2: Frequencies and relative frequencies of selected terms by consultation appropriateness. Table S3: Examples of contextual use of “negative” in consultation requests (appropriate consultations). Table S4: Examples of contextual use of “negative” in consultation requests (inappropriate consultations). Acknowledgments The authors acknowledge the use of ChatGPT (GPT‐5.5; OpenAI, accessed July 2026) to improve the English language, clarity, and flow of the manuscript. All AI‐assisted text was reviewed and revised by the authors to ensure accuracy, appropriateness, and clarity of meaning. Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request. References Associated Data Supplementary Materials Table S1: Distribution of consulted specialty departments according to consultation appropriateness. Table S2: Frequencies and relative frequencies of selected terms by consultation appropriateness. Table S3: Examples of contextual use of “negative” in consultation requests (appropriate consultations). Table S4: Examples of contextual use of “negative” in consultation requests (inappropriate consultations). Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request.