Article type
Year
Abstract
Background: In children presenting with acute ankle injuries, radiographs are often ordered to augment the physical exam; however, only 15% reveal a fracture. The Ottawa Ankle Rules (OAR) is a validated clinical decision rule for reliably ruling out ankle and foot injuries in adults and minimizing x-ray usage. There has been an ongoing interest in assessing the accuracy of the OAR in children. Currently there is no readily available software for clinical researchers to combine results from diagnostic studies using up-to-date methodologies; we discuss the challenges of conducting a meta-analysis of studies of diagnostic accuracy.
Objectives: To conduct a comprehensive systematic review and meta-analysis to assess the accuracy of the OAR in ruling out fractures in children with ankle and/or foot injuries.
Methods: We searched electronic databases for published and gray literature and hand searched reference lists. Two reviewers independently applied inclusion/exclusion criteria (children [2-18 years] admitted to the ED with an acute blunt ankle or midfoot injury, x-ray or proxy measure to confirm/exclude a fracture, sufficient data to create a 2x2 table). We used QUADAS to assess the quality of the included studies. Data were extracted by one reviewer using a standard form and checked by a second reviewer. Data synthesis: We tested the variation in diagnostic threshold using both Spearman's rho correlation and Lettenberg and Moses method. Sensitivities/specificities were pooled using an approximation of the inverse variance approach; 95% confidence intervals (95% CI) were calculated using the exact method. Heterogeneity was checked using Chi-squared test and/or Fisher's exact test. Subgroup analyses were conducted to explore the heterogeneity based on study design, Salter-Harris 1 fractures and injury location. Likelihood ratios (LR+/-) and diagnostic odds ratios (DOR) were combined under DerSimonian and Laird random effects model. Receiver operator characteristic (ROC) curve analysis was conducted to assess the overall performance and the accuracy of the OAR. Outliers were searched by using a Galbraith plot. Publication bias was assessed using the effective sample size funnel plot and associated regression test for asymmetry. Statistical calculations were made using SAS, Stata and RevMan. Figures were drawn using Splus, and area under a ROC curve was calculated using Maple.
Results: Eleven studies (9 prospective cohort, 1 retrospective cohort, 1 cross section; n=3,072 patients) were included. Threshold effect was not significant (p=0.902). The overall sensitivity was 97.9% (95% CI: 96.1%, 99%), and the overall specificity was 29.5% (95% CI: 27.8%, 31.3%). Heterogeneity was substantial (p<0.0001) across studies for both sensitivity and specificity; subgroup analyses did not explain the heterogeneity. Based on a prevalence of 14.03%, only 2.23% of patients with a negative test result would have a fracture (Negative LR=0.14; 95% CI: 0.07, 0.30; I2=30.5%). The odds of having a fracture were 9 times higher for patients with a positive test result by OAR compared to those with a negative test result (DOR=9.34; 95% CI: 4.68, 18.66; I2=14.2%). The area under the ROC curve is 0.82 with confidence limits of 0.74 to 0.88. Using OAR would decrease x-ray examination by 25.7%. There was evidence of publication bias (p=0.03).
Conclusions: In children, the OAR allows clinicians to accurately exclude ankle and midfoot fractures. This would reduce x-ray utilization and health care expenditures in the management of pediatric ankle and foot injuries. We recommend that systematic reviews of diagnostic accuracy studies should incorporate up-to-date statistical methods and analyses.
Objectives: To conduct a comprehensive systematic review and meta-analysis to assess the accuracy of the OAR in ruling out fractures in children with ankle and/or foot injuries.
Methods: We searched electronic databases for published and gray literature and hand searched reference lists. Two reviewers independently applied inclusion/exclusion criteria (children [2-18 years] admitted to the ED with an acute blunt ankle or midfoot injury, x-ray or proxy measure to confirm/exclude a fracture, sufficient data to create a 2x2 table). We used QUADAS to assess the quality of the included studies. Data were extracted by one reviewer using a standard form and checked by a second reviewer. Data synthesis: We tested the variation in diagnostic threshold using both Spearman's rho correlation and Lettenberg and Moses method. Sensitivities/specificities were pooled using an approximation of the inverse variance approach; 95% confidence intervals (95% CI) were calculated using the exact method. Heterogeneity was checked using Chi-squared test and/or Fisher's exact test. Subgroup analyses were conducted to explore the heterogeneity based on study design, Salter-Harris 1 fractures and injury location. Likelihood ratios (LR+/-) and diagnostic odds ratios (DOR) were combined under DerSimonian and Laird random effects model. Receiver operator characteristic (ROC) curve analysis was conducted to assess the overall performance and the accuracy of the OAR. Outliers were searched by using a Galbraith plot. Publication bias was assessed using the effective sample size funnel plot and associated regression test for asymmetry. Statistical calculations were made using SAS, Stata and RevMan. Figures were drawn using Splus, and area under a ROC curve was calculated using Maple.
Results: Eleven studies (9 prospective cohort, 1 retrospective cohort, 1 cross section; n=3,072 patients) were included. Threshold effect was not significant (p=0.902). The overall sensitivity was 97.9% (95% CI: 96.1%, 99%), and the overall specificity was 29.5% (95% CI: 27.8%, 31.3%). Heterogeneity was substantial (p<0.0001) across studies for both sensitivity and specificity; subgroup analyses did not explain the heterogeneity. Based on a prevalence of 14.03%, only 2.23% of patients with a negative test result would have a fracture (Negative LR=0.14; 95% CI: 0.07, 0.30; I2=30.5%). The odds of having a fracture were 9 times higher for patients with a positive test result by OAR compared to those with a negative test result (DOR=9.34; 95% CI: 4.68, 18.66; I2=14.2%). The area under the ROC curve is 0.82 with confidence limits of 0.74 to 0.88. Using OAR would decrease x-ray examination by 25.7%. There was evidence of publication bias (p=0.03).
Conclusions: In children, the OAR allows clinicians to accurately exclude ankle and midfoot fractures. This would reduce x-ray utilization and health care expenditures in the management of pediatric ankle and foot injuries. We recommend that systematic reviews of diagnostic accuracy studies should incorporate up-to-date statistical methods and analyses.