The utility of quantitative ADC values for differentiating high-risk from low-risk prostate cancer: a systematic review and meta-analysis

Purpose: The purpose of the study is to perform a meta-analysis of studies investigating the diagnostic performance of apparent diffusion coefficient (ADC) values in separating high-risk from low-risk prostate cancer (PCa). Methods: MEDLINE and EMBASE databases were searched in December 2015 for studies reporting diagnostic performance of ADC values for discriminating high-risk from low-risk PCa and providing sufficient data to construct 2 x 2 contingency tables. Diagnostic performance was quantitatively pooled using a bivariate random-effects model including subgroup analysis and assessment of study heterogeneity and methodological quality. Results: 13 studies were included, providing 1107 tumor foci in 705 patients. Heterogeneity among studies was moderate (tau(2) = 0.222). Overall sensitivity was 76.9% (95% CI 68.6-83.6%); overall specificity was 77.0% (95% CI 69.9-82.8%); and summary AUC was 0.67. Inverse correlation between sensitivity and specificity (rho = -0.58) indicated interstudy heterogeneity was partly due to variation in threshold for test positivity. Primary biases were readers' knowledge of Gleason score during ADC measurement, lack of prespecified ADC thresholds, and lack of prostatectomy as reference in some studies. Higher sensitivity was seen in studies published within the past 2 years and studies not using b value of at least 2000; higher specificity was associated with involvement of one, rather than two, readers measuring ADC. Field strength, coil selection, and advanced diffusion metrics did not significantly impact diagnostic performance. Conclusion: ADC values show moderate accuracy in separating high-risk from low-risk PCa, although important biases may overestimate performance and unexplained sources of heterogeneity likely exist. Further studies using a standardized methodology and addressing identified weaknesses may help guide the use of ADC values for clinical decision-making.