Ovarian Cancer Detection by Uterine Lavage DNA and Serum Proteins: A Phase 2 Biomarker Study

Abbreviated Name
Uterine Lavage
Lead Investigator
Skates, StevenMassachusetts General Hospital
Coordinating Investigator
Feng, Ziding Fred Hutchinson Cancer Center
Involved Investigators


This project investigates the complementarity of tumor DNA (tDNA) in uterine lavage to serum proteins for detection of ovarian cancer and builds on the results from project 1. Since there is no biobank of uterine lavage a prospective study is required. Two hundred patients scheduled for surgery for suspected ovarian cancer will be enrolled at four recruitment sites (50 pts at each site) and blood drawn before anesthesia and uterine lavage obtained before surgery. tDNA will be analyzed at two sequencing sites and serum proteins measured at two immunoassay sites.


Patients: Two hundred and fifty patients scheduled for surgery for suspected ovarian cancer will be enrolled at five sites (FHCRC, JHU, UAB, KP-NC, MGH) and blood and uterine lavage collected prior to surgery (a second blood draw is optional), including prior to tissue or cytology based diagnostic procedure (e.g. diagnostic laparoscopy, percutaneous biopsy, paracentesis). Collecting biospecimens prior to a confirmed diagnosis of ovarian cancer ensures the study is PRoBE compliant. Patients known to have ovarian cancer, or any cancer, at time of enrollment confirmed by a tissue or cell-based diagnostic procedure, whether performed at the recruitment site or at a referring institution, are NOT eligible. The blood and uterine lavage will then be processed and frozen. Each site will accrue 40 patients scheduled for surgery for suspected ovarian cancer in the first cohort (investigator-estimated probability of ovarian cancer exceeding 25%), and 10 patients with a BRCA1 or BRCA2 mutation undergoing RRSO in the second cohort. We expect approximately one-quarter to one-third of patients in the first cohort will have ovarian cancer pathologically confirmed, yielding a minimum of 50 cases among the 200 women suspected of having ovarian cancer, and in up to 5 cases of ovarian cancer with incidentally found in situ or low volume early stage disease in the 50 RRSO patients forming the second cohort.

Analytic Method

As indicated in Section 7.1 we will estimate biomarker sensitivity at 95% and 98% specificity using data for the 200 women undergoing surgery for suspected ovarian cancer. For sensitivity at 95% specificity, the standard empirical ROC curve will be calculated along with 95% confidence intervals (13. Section 8.2.2). For estimating sensitivity at 98% specificity, we will model the distribution in controls using a t-distribution after assessing its fit to the data from ~150 control subjects. We note that a higher target specificity (namely 98%) will be required for population screening for ovarian cancer in order to obtain an acceptable positive predictive value. However, given the limited number of control subjects (n=150) in this phase 2 biomarker study, we cannot non-parametrically estimate the biomarker threshold corresponding to 98% specificity with sufficient precision. Therefore, we will use a parametric estimate which leverages past experience with biomarker distributions in control subjects as mostly having bell shaped distributions on the log-concentration scale. The parametric form will be verifiable for most of the distribution with the study’s data, although the estimates at 98% specificity will depend on assumptions about the tails which are not directly verifiable and instead rely on past experience with the tail behavior of biomarker distributions in controls. However, the estimates will allow us to hypothesize the performance we might observe in the subsequent phase 3 biomarker study. One check on the parametric model will be concordance within standard error limits of the parametric with the non-parametric estimates at 95% specificity. If results are promising at 95% and 98% specificity in this study and a larger phase 3 biomarker study goes forward, we will have sufficient sample sizes in the phase 3 early detection biomarker study to evaluate sensitivity at 98% specificity. Secondary data analysis plans include linear regression analyses to examine effects of covariates on biomarker values on the log-concentration scale in controls e.g. age, clinical site. In aim 4, we will develop parametric combinations of biomarkers using logistic regression analyses and estimate performance using 10-fold cross validation of the empirical ROC curve as an approach to lessen optimism bias. Other methods for combining biomarkers such as random forests will also be explored. Descriptive exploratory analyses will be used to investigate correlation and complementarity of biomarkers, particularly from UL versus blood specimens. We will compare the empirical ROC curve for blood biomarkers with the empirical ROC curve for blood biomarkers and UL biomarkers combined. The focus will be on the ROC points corresponding to 95% and 98% specificity. Descriptive data summaries will be provided for the 50 patients undergoing risk reducing Salpingo-Oophorectomy. Since we expect only 5 cancers in this cohort, more sophisticated statistical inference will not


Data Collections

  • No data available at this time for this protocol
Protocol ID
Protocol Type
Field of Research
Collaborative Group
Breast and Gynecologic Cancers Research Group
Cancer Types
  • Malignant neoplasm of ovary
Phased Status

Associated Forms