Improved early detection of ovarian cancer using longitudinal multimarker models

Abstract

BACKGROUND: Ovarian cancer has a poor survival rate due to late diagnosis and improved methods are needed for its early detection. Our primary objective was to identify and incorporate additional biomarkers into longitudinal models to improve on the performance of CA125 as a first-line screening test for ovarian cancer. METHODS: This case-control study nested within UKCTOCS used 490 serial serum samples from 49 women later diagnosed with ovarian cancer and 31 control women who were cancer-free. Proteomics-based biomarker discovery was carried out using pooled samples and selected candidates, including those from the literature, assayed in all serial samples. Multimarker longitudinal models were derived and tested against CA125 for early detection of ovarian cancer. RESULTS: The best performing models, incorporating CA125, HE4, CHI3L1, PEBP4 and/or AGR2, provided 85.7% sensitivity at 95.4% specificity up to 1 year before diagnosis, significantly improving on CA125 alone. For Type II cases (mostly high-grade serous), models achieved 95.5% sensitivity at 95.4% specificity. Predictive values were elevated earlier than CA125, showing the potential of models to improve lead time. CONCLUSIONS: We have identified candidate biomarkers and tested longitudinal multimarker models that significantly improve on CA125 for early detection of ovarian cancer. These models now warrant independent validation.

Biomarkers and Strategies for Early Detection of Ovarian Cancer

“…In published studies to date, more than 35 different biomarkers have been reported to improve the sensitivity of CA125 for detecting early stage disease. 21,22,23,24,25 Many of these biomarkers detect only a small fraction of cases missed by CA125; most studies include a relatively small number of early stage cases; and often the utility of combinations has not been confirmed with an independent validation set.…”

“…Using preclinical serum samples in the UKCTOCS biobank from women destined to develop ovarian cancer, investigators at Manchester, UK, have identified three panels of biomarkers in combination with CA125: lecithin-cholesterol acyltransferase (LCAT) and insulin-like growth factor-binding protein 2 (IGFBP2); 23 phosphatidylcholine-sterol acyltransferase, vitamin Kdependent protein Z and C-reactive protein; 24 and HE4, CHI3L1, PEBP4 and/or AGR2. 25 Each panel detected a fraction of cases missed by CA125 and produced lead time over CA125 of 5-6 months 23 to a year 25 or more 24 . Finding an adequate number of pre-clinical serum samples to validate these panels and to identify the optimal combinations of biomarkers will be a challenge.…”

Secretion of pro-oncogenic AGR2 protein in cancer

“…In this instance, AGR2 concentrations were found to be higher in stages II and III patients and were similarly elevated in patients with both serous and non-serous tumours [ 107 ]. Recently, incorporation of AGR2 in a panel of markers that includes CA125, HE4, CHI3L1, PEBP4 has been reported to improve the early detection of ovarian cancer up to 1 year before diagnosis [ 108 ]. This panel of markers demonstrated 85.7% sensitivity and 95.4% specificity and provided higher predictive values compared to using the classical serum cancer antigen CA125 alone.…”

The Human Body as a Super Network: Digital Methods to Analyze the Propagation of Aging

“…It was demonstrated in the recent United Kingdom Collaborative Trial of Ovarian Cancer Screening (UKCTOCS; Menon et al, 2015), that it is not an individuals' CA125 measurement that indicates cancer development, rather a deviation from personal baseline. Therefore, recent approaches in ovarian cancer are directed towards constructing personalized baselines based on patients' serial measurements with analyzing further sequential measurements from the perspective of previous history (Whitwell et al, 2020). For example, three approaches have been applied to longitudinal serological data from ovarian cancer: (1) the methods of mean trends (MMT) algorithm (Blyuss et al, 2018) which evaluates the dynamics of longitudinal markers using weighted derivatives of marker changes as well as the average area under the time series, coefficient of variation and ''center of mass'' as predictors in logistic regression; (2) The Risk of Ovarian Cancer Algorithm (ROCA), that fits Bayesian hierarchical changepoint model on CA125 serial data (Skates et al, 2001); and (3) Parametric Empirical Bayes (PEB) that evaluates deviation from normality based on population characteristics such as the population mean and within-subject and between-subject variances.…”