Background
Poor diagnosis and inadequate treatment of osteochondral lesions of the talus (OLTs) may result in further pain and progression to osteoarthritis. Magnetic resonance imaging (MRI) has an important role in diagnosing and staging OLTs, but its accuracy has not been systematically established. The purpose of this systematic review with meta-analysis was to evaluate the accuracy performance of MRI (as compared to arthroscopy) in diagnosis OLTs.
Methods
We included studies published up to October 24, 2020 that reported the accuracy of MRI in diagnosing OLTs and using arthroscopy as reference standard. The risk of bias was assessed through the QUADAS-2 tool. Quantitative syntheses with 95% confidence intervals (CI) were performed to calculate the pooled sensitivity, specificity, positive likelihood ratios (LR+) and negative likelihood ratios (LR-), diagnostic odds ratio (DOR) and summary receiver operating characteristic (SROC) curves.
Results
Fourteen studies were included for qualitative analyses and twelve studies were included for quantitative synthesis. A total of 711 participants and 436 OLTs with a weighted mean age of 37.9±5.6 years (mostly males, 77%) were included. The MRI analyses were performed by one or two musculoskeletal radiologists using most commonly a 1.5T MRI. Staging of OLTs was determined using heterogenous grading scales and thus not considered for quantitative analyses. Pooled sensitivity (80%, 95% CI 67–89%), specificity (96%, 95% CI 82–98%), LR+ (19.7, 95% CI 4.16–93.40), LR- (0.20, 95% CI 0.12–0.36) and DOR (96.20, 95% CI 19.31–479.28) were obtained. The SROC curve showed an excellent area under the curve (AUC) for diagnosing OLTs (0.94, 95% CI 91–96). The hierarchical SROC was comparable to results from the bivariate model and was symmetrical (ß=0.588, 95% CI 0.342-1.520, p=0.215). The value of ? was 4.237 (95% CI, 2.98 5.49), indicating a high diagnostic accuracy. The I-squared was 79.8%, 91.3%, 83.4%, 79.4% and 94.7% for sensitivity, specificity, LR+, LR-, and DOR, respectively, indicating heterogeneity. The Cochrane Q statistic was 32.20 (P?<?0.05) for AUC estimation, indicating that heterogeneity was likely due to non-threshold effects. After inferences by the bivariate model, the proportion of heterogeneity likely due to threshold effect was 27%. The pre-test probability of 25%, 50% and 75% showed that a positive result improved the post-test probability up to 87%, 95% and 98%, while a negative result decreased the post-test probability to 6%, 17% and 38%. Sensitivity analysis (goodness-of-fit, bivariate normality, influence and outlier detection analyses) demonstrated that the bivariate model was moderately robust and did not identify outliers. Five studies presented high risk of selection bias and when excluding these 5 studies, the diagnostic features were similar; the DOR was considerably lower (67 versus 96) but the CIs were very large. Meta-regression (age, time from MRI to arthroscopy, magnetic field intensity) showed no statistical correlation. Publication bias was negligible and not statistically significant (p=0.56), showing a symmetric the funnel plot.
Conclusion
MRI examination is accurate to detect OTLs (as compared to arthroscopy) providing good diagnostic performance for DOR and AUC, and showing high sensitivity and specificity.