Revisiting the Diagnostic Accuracy of Ottawa Ankle Rules for Ankle Fractures: A Systematic Review and Meta-Analysis Authors: Danial Sharifi Razavi, Mohammadali Seiri, Mehrdad Farrokhi Published: 2026 ABSTRACT Introduction:The Ottawa Ankle Rules (OAR) have demonstrated conflicting results across studies, with generally high sensitivity but relatively low and variable specificity. This updated meta-analysis aims to evaluate the overall diagnostic characteristics of the OAR for early identification of ankle fractures in adult patients.Methods: In this systematic review and meta-analysis, a comprehensive literature search was conducted from database inception through May 2026 across three electronic databases, including Medline, Web of Science, and Scopus. The MIDAS package in STATA and Meta-Disc software were used to pool the findings of the diagnostic accuracy studies. Clinical application of the OAR was assessed using Fagan’s nomogram and scattergram.Results:The pooled sensitivity and specificity of OAR were 0.92 (95% confidence interval (CI): 0.91–0.93) and 0.35 (95% CI: 0.34–0.36), respectively. The pooled positive likelihood ratio (PLR) and negative likelihood ratio (NLR) were 1.76 (95% CI: 1.46–2.13) and 0.13 (95% CI: 0.09–0.19), respectively. Furthermore, the pooled diagnostic odds ratio (DOR) was 16.21 (95% CI: 10.15–25.89).Conclusion: In conclusion, the OAR demonstrate excellent sensitivity and a very low negative likelihood ratio, confirming their value as a reliable screening tool for ruling out ankle fractures in clinical practice. With pooled sensitivities exceeding 90% and a post-test probability reduced to approximately 1% following a negative result, the OAR can safely reduce unnecessary radiographic imaging. 1. Introduction: Ankle injuries, which demonstrate a high prevalence, represent the second most common form of lower-extremity trauma presenting to emergency departments and are among the most frequently encountered conditions managed by emergency clinicians (1). Moreover, in the United States, approximately 14.6% of the 117 million emergency department visits were reported to involve injuries primarily affecting the lower extremities (2). Following physical examination, imaging, most commonly plain ankle radiography, constitutes the primary diagnostic modality for the early detection of ankle injuries, including fractures and sprains (3). Nevertheless, the routine use of radiographic imaging for all patients presenting with ankle injuries contributes to unnecessary radiation exposure, increased burden on emergency and radiology departments, and elevated healthcare expenditures (4, 5). These concerns underscore the importance of reducing unnecessary imaging, particularly among patients with a low probability of ankle fracture (5). The Ottawa Ankle Rules (OAR), a well-established clinical decision-making tool originally introduced by Stiell et al. (6), is widely utilized by healthcare professionals to reduce the number of unnecessary ankle radiographs through the early exclusion of ankle fractures (7). These clinical criteria have been disseminated across various countries as a decision-support tool for clinicians, with the aim of minimizing exposure to ionizing radiation and reducing patient burden in emergency departments through the decreased use of unnecessary radiographic imaging (5, 6). However, the OAR have demonstrated conflicting results across studies, with generally high sensitivity but relatively low and variable specificity (5, 8, 9). This variability underscores the importance of conducting secondary research to synthesize and pool the findings of existing studies. Therefore, this updated meta-analysis aims to evaluate the overall diagnostic characteristics of the OAR for the early identification of ankle fractures in adult patients. 2. Methods 2.1 Search strategy The protocol for this meta-analysis was registered in PROSPERO (CRD420261404225). A comprehensive literature search was conducted from database inception through May 2026 across three electronic databases, including Medline, Web of Science, and Scopus. Keywords related to the OAR, ankle injuries, and diagnostic performance, along with Medical Subject Headings (MeSH) terms for PubMed, were combined using Boolean operators (Appendix 1). To minimize the likelihood of missing eligible studies, supplementary search strategies were employed, including gray literature searches and manual screening of the reference lists and citation records of the included studies. 2.2 Eligibility criteria We included studies that reported sufficient data to construct a 2 × 2 contingency table for evaluating the diagnostic accuracy of the OAR in detecting ankle fractures among adult patients. The analysis focused specifically on ankle fractures; however, studies reporting combined diagnostic data for both ankle and foot fractures were also included when separate ankle-specific results were unavailable. Studies conducted exclusively in pediatric populations or involving participants younger than 18 years of age were excluded. In addition, case reports, case series, conference abstracts, preprints, and other non–peer-reviewed publications were excluded from the analysis. 2.3 Data extraction and quality appraisal Two authors independently extracted both diagnostic and baseline study data, and any discrepancies were resolved through consultation with a third author. The extracted variables included the first author’s name, percentage of male participants, mean age, country of study, reference standard, sensitivity, specificity, and data required for the construction of 2 × 2 contingency tables. The methodological quality and risk of bias of the included studies were assessed using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool. 2.4 Statistical analyses The MIDAS package in STATA and Meta-Disc software were used to pool the findings of the diagnostic accuracy studies. Publication bias was assessed using Deeks’ funnel plot asymmetry test. Outlier detection and influence analyses were conducted to identify potentially influential or outlying studies. In addition, Fagan’s nomogram and scattergram analyses were employed to evaluate the clinical applicability of the OAR for the diagnosis of ankle fractures. 3. Results: 3.1 Characteristics of the include studies A systematic search of the three databases yielded a total of 98 studies, of which 21 were removed as duplicates using EndNote. Title and abstract screening of the remaining 77 articles resulted in 53 studies being considered eligible for full-text review. Following full-text evaluation, 43 studies met the eligibility criteria and were ultimately included in the meta-analysis, as illustrated in the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) flow diagram (Figure 1). A total of 43 original studies involving 16,503 patients were included in the meta-analysis. Two studies were entered twice in the analysis, with one study providing separate data for geriatric and non-geriatric populations and the other reporting independent prospective and retrospective evaluations. The included studies were published between 1993 and 2025. The majority of the studies used radiography as the reference standard, while one study additionally utilized three-dimensional computed tomography (3D CT) alongside radiography as the gold standard. Further diagnostic and baseline characteristics of the included studies are summarized in Table 1. 3.2 Quality appraisal and publication bias The main methodological concerns among the included studies were related to the risk of bias in the patient selection domain, followed by concerns regarding flow and timing, as well as applicability concerns associated with patient selection (Table 2). Furthermore, because many studies did not clearly report whether the assessors of the index test were blinded to the results of the reference standard, and vice versa, these studies were classified as having an unclear risk of bias. Asymmetry test of Deeks’ funnel plot revealed no publication bias (p = 0.79) (Figure 2). 3.3 Meta-analyses The pooled sensitivity and specificity were 0.92 (95% confidence interval (CI): 0.91–0.93) and 0.35 (95% CI: 0.34–0.36), respectively (Figure 3). The pooled positive likelihood ratio (PLR) and negative likelihood ratio (NLR) were 1.76 (95% CI: 1.46–2.13) and 0.13 (95% CI: 0.09–0.19), respectively (Figure 4). Furthermore, the pooled diagnostic odds ratio (DOR) of the OAR was 16.21 (95% CI: 10.15–25.89) (Figure 5). The area under the curve (AUC) derived from the summary receiver operating characteristic (SROC) curve was 0.87 (Figure 5). The pooled diagnostic accuracy estimated using the metaprop package was 0.52 (95% CI: 0.45–0.59) (Figure 6). The forest plot shows that sensitivity estimates were consistently high across most studies, indicating strong ability of the test to identify ankle fractures, whereas specificity demonstrated substantial variability between studies. Compared with the observed estimates, the empirical Bayes estimates were more stabilized and shrunk toward the pooled mean, reflecting improved precision through hierarchical modeling (Figure 6). The goodness-of-fit plot reveals a sigmoidal deviation from the diagonal reference line, with deviance residuals forming a flat plateau in the lower quantiles before rising sharply, suggesting suboptimal model fit and potential heterogeneity among studies (Figure 7). Similarly, the bivariate normality plot shows reasonable adherence to the reference line in the lower quantiles, but notable upper-tail divergence indicates that a subset of studies violates the bivariate normality assumption. The outlier detection plot identifies several studies flagging them as statistical outliers. Similarly, the boxplot analysis identified several studies as potential outliers among the included studies (Figure 8). Fagan’s nomogram demonstrated that a positive result on the OAR increases the probability of ankle fracture from 20% to 29%, whereas a negative result reduces this probability to approximately 1% (Figure 8). The likelihood ratio scattergram placed the OAR in the left lower quadrant, suggesting a primary role in ruling out ankle fractures rather than confirming them (Figure 8). Similarly, the probability-modifying plot indicated that a negative result from the OAR is more informative, as the curve approaches the point (1, 0) and substantially reduces the pre-test probability of ankle fractures (Figure 9). The sensitivity and specificity of the OAR for the detection of ankle fractures, based on studies that exclusively reported ankle fracture data or provided sufficient data for separate extraction of ankle fracture outcomes, were 0.91 (95% CI: 0.90–0.92) and 0.34 (95% CI: 0.33–0.35), respectively (Figure 9). The Grading of Recommendations, Assessment, Development and Evaluation (GRADE) assessment indicated moderate certainty of evidence for both sensitivity and specificity sections (Table 3). 3.4 Meta-regression Meta-regression was performed using the following covariates: age group (≥35 years vs. <35 years), population type (geriatric vs. non-geriatric), clinical setting (emergency department vs. other settings), geographic region (developing vs. developed countries), study design (prospective vs. retrospective), assessor type (physician vs. other or mixed assessors), and fracture type (ankle and foot fractures vs. ankle fractures only) (Figure 9). The covariates "clinical setting" and "geographic region" were excluded from the final meta-regression analysis because the software failed to achieve model convergence, resulting in a maximum iteration error during model estimation. Meta-regression analysis demonstrated that only fracture type was significantly associated with diagnostic performance. 4. Discussion: Given inconsistencies across previous studies and the absence of an updated meta-analysis incorporating recent evidence, the present study aimed to evaluate the diagnostic performance of the OAR for the detection of ankle fractures. The pooled sensitivity and specificity of OAR were 0.92 and 0.35, respectively. The positive likelihood ratio (PLR) and negative likelihood ratio (NLR) were estimated at 1.76 and 0.13, respectively. The area under the summary receiver operating characteristic (SROC) curve was 0.87, and the overall diagnostic accuracy was 0.52. These findings suggest that, due to its high sensitivity and low NLR, OAR is more clinically useful for ruling out ankle fractures rather than confirming their presence. However, a substantial proportion of the included studies demonstrated a high risk of bias, particularly in the domains of patient selection and flow and timing, while several studies were also rated as having unclear risk of bias. Therefore, these limitations warrant cautious interpretation of the pooled estimates. In a similar meta-analysis conducted by Gomes et al. (8), the literature search was performed up to 2020 and yielded 15 studies with a total sample size approximately half that of the present meta-analysis. They reported a pooled sensitivity of 0.91, which was comparable to the estimate observed in our analysis, whereas the pooled specificity of 0.25 was considerably lower than our estimated specificity. An important methodological difference between the two meta-analyses relates to the eligibility criteria of the included studies. In the study by Gomes et al., only studies reporting diagnostic performance specifically for ankle injuries were included, and in studies presenting combined diagnostic parameters for ankle and midfoot injuries, ankle-specific data were extracted separately. In contrast, the present meta-analysis also included studies reporting combined diagnostic parameters for ankle and midfoot injuries, as the majority of patients in those studies presented primarily with ankle trauma. Another notable difference concerns the methodological quality assessment of the included studies. Gomes et al. did not classify any study as having a high risk of bias in the patient selection domain, whereas in our assessment, 21 studies were judged to have a high risk of bias for patient selection. This issue may limit the generalizability of the findings to the broader emergency department population, as participants in several studies were selected based on restrictive exclusion criteria and may not fully reflect routine clinical practice. Interestingly, in our meta-analysis, when the sensitivity and specificity of the OAR for the detection of ankle fractures were estimated using only studies that exclusively reported ankle fracture data or studies from which ankle-specific data could be extracted separately from combined ankle and foot fracture analyses, similar pooled estimates were observed, with a sensitivity of 0.91 and a specificity of 0.34. In this regard, a meta-analysis by Bachmann et al. (7) evaluated the diagnostic performance of the OAR for the exclusion of ankle and midfoot fractures. They performed subgroup analyses separately for studies focusing on ankle fractures and those reporting combined foot and ankle fractures. In contrast to our findings, their results showed lower sensitivity and specificity in studies including combined foot and ankle fractures compared with studies restricted to ankle fractures. This discrepancy may be partly explained by differences in inclusion criteria and the number of included studies, as their analysis also incorporated pediatric populations and was based on a smaller number of eligible publications. It has been shown in another meta-analysis that the Ottawa Ankle and Foot Rules (OAFR) and the OAR demonstrate similar diagnostic performance, with sensitivity values of 0.93 versus 0.94 and specificity values of 0.45 versus 0.42, respectively (10). Although their reported sensitivity was comparable to that observed in our study, the specificity was higher than that obtained in our meta-analysis. Notably, that meta-analysis included both adult and pediatric populations, whereas the present study was limited to adult patients only. In a meta-analysis by Dowling et al. (11), the diagnostic accuracy of the OAR for detecting ankle and midfoot fractures in pediatric populations was assessed. The review included 12 studies comprising 3,130 patients identified through searches of Medline, CINAHL, Embase, and the Cochrane Library. They reported a pooled sensitivity of 98.5%, which was considerably higher than the sensitivity of 0.92 observed in the present meta-analysis. Accordingly, the higher sensitivity reported in previous meta-analyses that included both adult and pediatric populations, compared with our findings, may be partly explained by the greater diagnostic sensitivity of the OAR in pediatric patients. Table 1Characteristics of the included studies Study Country Age* (year) Gender M% Gold Standard Sensitivity Specificity Ahmad et al. (12); 2025Pakistan36.73±6.767.13Radiography0.560.95Ak et al. (13); 2022Turkey33.1±16.353.60Radiography0.970.48Anyanwu et al. (14); 2025Nigeria37.4±1154.40Radiography10.37Auley et al. (15); 1998France 34 [18–90]8%Radiography0.980.45Avinca et al. (16); 2023Turkey30.93 ± 13.15436Radiography0.850.82Beceren et al. (17); 2013Turkey30.3 ± 13.260.40Radiography0.740.68Broomhead et al. (18); 2003Australia34.7[18.1–84.8]51.70Radiography10.15Can et al. (19); 2008SwitzerlandFracture present: 51±21; Fracture absent: 38±17Fracture present: 39; Fracture absent: 56Radiography10.21Daş et al. (20); 2016TurkeyMean:37.46 [18–85]61.20Radiography0.980.44Derksen et al. (a) (21); 2005NetherlandsAbove 18NRRadiography0.930.39Derksen et al. (b) (21); 2015NetherlandsMean: 37 [ 18–73]46.80Radiography0.970.29Dwivedi et al. (22); 2014Nepal35.78±11.3954.90Radiography10.036Fiesseler et al. (23); 2004USA37±1633Radiography0.920.47Glas et al. (24); 2002Netherlands35±1450Radiography8926Gomes et al. (9); 2020Australia38±18.350.40Radiography0.590.37Jia et al. (25); 2025ChinaAbove 18NRRadiography0.980.26Kelly et al. (26); 1994New ZealandNRNRRadiography0.930.11Keogh et al. (27); 1998UKNRNRRadiography10.48Leisey et al. (28); 2004Saudi ArabiaMean: 27 [19–46]82Radiography10.4Lucchesi et al. (29); 1995USAMean: 38 [18–91]44.60Radiography0.940.15Makiev et al. (30); 2025Greece39.6±15.843Radiography0.940.15Marinelli et al. (31); 2007Italy31.8±15.964.90Radiography10.43McBride et al. (32); 1997CanadaNo Fracture: 30.90±12.27; Fracture: 41.21±17.87No Fracture: 54.6; Fracture: 38.2Radiography0.970.3Morais et al. (33); 2021Portugal46.1±18.734Radiography10.26Murphy et al. (Geriatric) (34); 2020USA76.2±8.120.20Radiography0.970.33Murphy et al. (non-geriatric) (34); 2020USA39.0±13.346.30Radiography0.980.6Özbay et al. (35); 2021Turkey57.2±20.951.60Radiography0.80.63Papacostas et al. (36); 2001Greece29±9.62NRRadiography10.3Perry et al. (37); 1999UKAbove 18NRRadiography0.930.46Pigman et al. (38); 1994USAFracture: 38; non-fracture: 34Fracture: 43; non-fracture: 50Radiography10.19Pijnenburg et al. (39); 2002Netherlands35±1450Radiography0.970.38Pires et al. (40); 2014BrazilMean: 37.4 [18–88]40.90Radiography0.970.07Rosin et al. (41); 1999South KoreaMean: 24.9 [19-41]80.50Radiography0.70.73Salt et al. (42); 1997UKAbove 18NRRadiography10.31Santelli et al. (43); 2008Italy31.8 ± 15.9NRRadiography10.46Seyhan et al. (44); 2024Turkey73.6±6.926.20Radiography0.980.86Singh et al. (45); 2015IndiaMean: 34.76 [18–68]61.30Radiography0.950.97Springer et al. (46); 2000USAMean: 20.25 [18–25]80.39Radiography10.46Stiell et al. (a) (47); 1994Canada36±1647Radiography10.5Stiell et al. (b) (48); 1993Canada35±1552Radiography10.39Thakur et al. (49); 2022NepalAbove 18NRRadiography10.47Verma et al. (Prospective) (50); 1997USAAbove 18NRRadiography0.980.2Verma et al. (Retrospective) (50); 1997USAAbove 18NRRadiography10.02Wang et al. (51); 2013ChinaMean: 36.6 [18–70]44.26Radiography/3D-CT0.960.45Wynn-Thomas et al. (52); 2002New ZealandAbove 1847Radiography10.47*: Data are presented as mean ± standard deviation, median (interquartile range), or mean (range). M: Male; NR: Not reported; 3D-CT: 3-dimensional computed tomography. Characteristics of the included studies *: Data are presented as mean ± standard deviation, median (interquartile range), or mean (range). M: Male; NR: Not reported; 3D-CT: 3-dimensional computed tomography. Table 2Quality appraisal of the included studies Quality appraisal of the included studies Table 3GRADE summary of certaintyOutcomeNDesignFactors that may decrease certainty of evidenceEffect per 100 patients testedTest accuracy Risk of biasIndirectnessInconsistencyImprecisionPublication biasPTP 10%PTP 15%PTP 20% True positives(patients with Ankle Fractures)45 studies3066 patientsCross-sectional (cohort type accuracy study)SeriousNot seriousSeriousNot seriousAll plausible residual confounding would reduce the demonstrated effect9 (9 to 9)14 (14 to 14)18 (18 to 19)⨁⨁⨁◯ Moderate False negatives(patients incorrectly classified as not having Ankle Fractures)1 (1 to 1)1 (1 to 1)2 (1 to 2) True negatives(patients without Ankle Fractures)45 studies13437 patientsCross-sectional (cohort type accuracy study)SeriousNot seriousSeriousNot seriousAll plausible residual confounding would reduce the demonstrated effect32 (31 to 32)30 (29 to 31)28 (27 to 29)⨁⨁⨁◯ Moderate False positives(patients incorrectly classified as having Ankle Fractures)58 (58 to 59)55 (54 to 56)52 (51 to 53)N: number; PTP: pre-test probability; GRADE: Grading of Recommendations, Assessment, Development and Evaluation. GRADE summary of certainty N: number; PTP: pre-test probability; GRADE: Grading of Recommendations, Assessment, Development and Evaluation. Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) flowchart of the included studies. Deeks’ funnel plot and test. Pooled sensitivity and specificity of Ottawa Ankle Rules for detection of ankle fractures. CI: confidence interval. Pooled positive likelihood ratio (PLR) and negative likelihood ratio (NLR) of Ottawa Ankle Rules for detection of ankle fractures. CI: confidence interval; LR: likelihood ratio. Pooled diagnostic odds ratio (DOR) and the area under the hierarchical summary receiver operating characteristic (HSROC) curve (AUC) of Ottawa Ankle Rules for detection of ankle fractures. CI: confidence interval; OR: odds ratio. Pooled accuracy of Ottawa Ankle Rules for detection of ankle fractures (left); Forest plot of observed and empirical Bayes estimates of sensitivity and specificity across included studies (right). CI: confidence interval; Goodness-of-fit and bivariate normality plots for model evaluation, along with influence analysis and outlier detection plots for the identification of influential and outlier studies. Boxplot for evaluation of heterogeneity and identifying outliers (left); Fagan’s nomogram illustrating the effect of the Ottawa Ankle Rules on the post-test probability of ankle fractures (middle); Likelihood ratio (LR) scattergram illustrating the role of the Ottawa Ankle Rules in ruling out and confirming ankle fractures. Probability modifying plot for Ottawa ankle rule (OAR); Sensitivity and specificity of the Ottawa Ankle Rules for the detection of ankle fractures based on studies that exclusively reported ankle fracture data or provided sufficient data for separate extraction of ankle fracture outcomes; Meta-regression based on the clinical and demographic covariates. Age Group Yes: Age≥35, Age Group No: Age<35; Population Geriatric Yes: Geriatric Population; Study Design Yes: Prospective Design, Study Design No: Retrospective Design; Assessor Yes: Physician, Assessor No: Other Assessors or mixed of different assessors; Type Fracture Yes: Ankle and Foot Fractures, Tye Fracture No: Only Ankle Fractures; CI: confidence interval; LR likelihood ratio. 5. Limitations Our study has several limitations. First, the literature search was restricted to studies published in the English language, which may have introduced language bias. Second, we did not include or compare the diagnostic performance of the OAR between pediatric and adult populations. Third, studies exclusively evaluating midfoot fractures were not included; therefore, direct comparison of the diagnostic performance of the OAR for midfoot versus ankle fractures was not possible. Finally, most of the included studies used plain radiography as the reference standard, despite the fact that this imaging modality may not accurately detect all ankle fractures. 6. Conclusion: In conclusion, the OAR demonstrated excellent sensitivity and a very low negative likelihood ratio, confirming its value as a reliable screening tool for ruling out ankle fractures in clinical practice. With pooled sensitivities exceeding 90% and a post-test probability reduced to approximately 1% following a negative result, the OAR can safely reduce unnecessary radiographic imaging. However, the consistently low specificity and modest positive likelihood ratio indicate limited ability to confirm fractures, meaning that positive OAR findings should be interpreted cautiously and followed by radiographic evaluation rather than used as definitive evidence of fracture. Furthermore, the observed heterogeneity, evidence of outlier studies, and deviations from model assumptions suggest variability in study populations, clinical settings, and application of the OAR across studies. Despite these limitations, the findings consistently support the OAR as an effective exclusion tool, particularly in emergency and primary care settings where the primary clinical objective is to safely identify patients who do not require imaging.