Since HIFU emerged as a treatment option for PCa, related meta-analyses published to date have focused on the oncological outcome. To our knowledge, the performance of MRI to predict recurrence after HIFU has never been the subject of systematic review or meta-analysis. With control biopsy as the reference standard, MRI showed decent results with pooled sensitivity and specificity of 0.81 and 0.91, and an area under the SROC curve of 0.81. However, significant heterogeneity was found between individual studies, suggested by Cochran’s Q- and Higgins I2 tests and large degree of scattering of individual studies in the SROC curve. This limits the generalizability of good diagnostic performance of post-HIFU MRI in predicting recurrence. Meta-regression analysis provided hints on some potential factors that could explain this heterogeneity.
Relatively larger studies, including more than 50 patients20, 25–27 were fewer in number, and have lower sensitivity and specificity than smaller studies. This also seems to be related to the publication bias revealed in the funnel plot. In other words, prior researches on this topic have been primarily small-scale, and only optimistic results may have been published, contributing to overestimating the diagnostic performance of MRI after HIFU. The possibility that MRI's performance was exaggerated is also suggested by the fact that another covariate that had a statistically significant effect on its diagnostic performance was the nadir PSA level. Studies reporting nadir PSA levels of higher than 1 ng/mL17,20,21,25,26,28,29 after HIFU had significantly poorer diagnostic performance than studies reporting lower nadir values. This difference in performance appeared to be due to lower sensitivity rather than specificity. A high nadir PSA level within the study may indicate a high proportion of relapsed patients or a significant burden of recurrent cancer. The low sensitivity of the MRI in this situation may pose the risk of missing a significant number of patients with recurrent cancer. Although slightly deviated from statistical significance, the trend toward better MRI performance in studies reporting low recurrence rate (< 30%) is also interpreted in line with this. In studies with low prevalence of recurrence14,16,17,21,23, 27–29, there was tendency of higher specificity than the studies with high recurrence rate, without overlapping confidence interval. If the cancer detection ability of MRI after HIFU is insufficient, it will mainly read negative, which can cause low sensitivity when higher tumor burden is expected, and high specificity when recurrence rate is low. Although it is understandable that the accuracy of diagnostic tests is affected by disease prevalence31, recalling that cancer detection is more important with high nadir PSA level or high likelihood of recurrence, there is a question as to whether MRI is the best surveillance tool in this situation.
Another covariate that deviated very slightly from statistical significance in explaining the inter-study heterogeneity was the presence or absence of a definition of recurrent cancer in MRI. Interestingly, studies with predefined suspicious MRI findings tended to have poorer diagnostic performance. This may suggest that the definition of suspicious findings on MRI to date may be inappropriate in predicting recurrent PCa after HIFU. Some studies used PIRADS categorization for the evaluation of post-HIFU MRI4,17, 23–25,27,28, an indicator designed for treatment-naïve patients and thus may be inappropriate in this setting. This supports the recent claim that so-called post-partial gland ablation PIRADS is required separately7.
Some covariates that did not show statistical significance for inter-study heterogeneity in meta-regression imply factors to be considered in creating this new standard. First, whether the MRI was taken with the so-called multiparametric protocol did not affect the diagnostic performance. Second, difference in magnetic field strength did not bring about a significant difference in diagnostic performance of MRI. Rather, the study using only the 3T scanner had sensitivity of lower point estimate, but had a very wide confidence interval without statistically significant difference. Third, MRI showed no significant performance difference in distinguishing whether recurrent PCa was clinically significant, with relatively higher sensitivity and lower specificity for prediction of CS PCa. This contradicts with the high sensitivity and low specificity of MRI for detection of CS PCa in patients without history of HIFU treatment32. Taken together, the currently established prostate multiparametric MRI protocols, magnetic field strength, and definition of CS PCa may be insufficient or warrant further discussion for patients following HIFU treatment.
We acknowledge that this study has some limitations. First, there have been few large-scale studies on this topic. This is probably because HIFU is not the standard treatment for PCa. This led to continuity correction for some zero-cells in the 2x2 table, which can generally lead a downward bias on the test accuracy33. It is noteworthy that the overall pooled diagnostic performance was nevertheless excellent. Second, the included individual studies have used needle biopsy as a reference standard, which cannot be flawless. However, this is a fundamental problem that cannot be resolved unless salvage prostatectomy is performed, and is difficult to reconcile as the purpose of HIFU itself is to avoid prostatectomy.