The diagnostic accuracy of human epididymis protein 4 (HE4) for discriminating between benign and malignant pelvic masses: a systematic review and meta-analysis

Introduction Many women with benign pelvic masses, suspected of ovarian cancer, are unnecessarily referred for treatment at specialized centers. There is an unmet clinical need to improve diagnostic assessment in these patients. Our objective was to obtain summary estimates of the accuracy of human epididymis protein (HE4) for diagnosing ovarian cancer and to compare the performance of HE4 with that of cancer antigen 125 (CA125). Material and methods We searched PubMed, Ovid and Scopus using search terms for "pelvic masses" and "HE4", to identify studies that evaluated HE4 for diagnosing malignant ovarian masses, in adult women presenting with a pelvic mass, suspected of ovarian cancer, and with diagnosis confirmed by histopathology. Screening, data extraction and Risk of Bias assessment with the QUADAS-2 tool were done independently by two authors. We performed a meta-analysis of the accuracy of HE4 and CA125 using a random-effects bivariate logit-normal model. A study protocol was registered at PROSPERO (CRD42020158073). Results In the 17 eligible studies, which included 3404 patients, ovarian cancer prevalence ranged from 15% to 71%. Overall, the studies were heterogeneous. All studies seemed to have recruited patients in specialized settings. A meta-analysis of seven HE4 studies resulted in a mean sensitivity of 79.4% (95% confidence interval [CI] 74.1%-83.8%) and a mean specificity of 84.1% (95% CI 79.6%-87.8%), for cut-off values of 67-72 pmol/L. Based on eight studies, the mean sensitivity of CA125 was 81.4% (95% CI 74.6%-86.2%) and the mean specificity was 56.8% (95% CI 47.9%-65.4%), at a cut-off of 35 U/ml. Given a 40% ovarian cancer prevalence, the positive predictive value (PPV) for HE4 would be 76.9% (71.9%-81.2%) versus 55.6% (50.2%-60.9%) for CA125. The negative predictive value (NPV) would be 85.9 (82.8%-88.6%) and 81.9% (76.2%-86.4%), respectively. At a 15% prevalence, the NPV would be 95.8% (95% CI 94.4%-96.7%) for HE4 and 94.4% (95% CI 92.3%-96.0%) for CA125. The PPV would be 46.9% (40.4%-53.4%) and 24.9% (21.1%-29.2%), respectively. Conclusions HE4 had higher specificity and similar sensitivity compared with CA125. At high prevalence, PPV was also higher for HE4, but at low prevalence, it had a similar NPV to CA125. The field would benefit from studies conducted in general settings.