ONE-STEP OPTIMIZATION OF SCREENING FOR OVARIAN CANCER

点击翻译按钮获取中文摘要

The long term objectives are to lessen the mortality impact on women of ovarian cancer through efficient screening techniques, and to develop methodology for optimizing screening, and more generally monitoring, of chronic diseases with quantitative markers. Ovarian cancer mortality rate in the US alone is over 12,000 per year. CA 125 is a quantitative serum measurement that has potential as a marker for ovarian cancer; women with higher levels of CA 125 have a higher probability of ovarian cancer. Since ovarian cancer has an average incidence of only 50/100,000 for postmenopausal women, no single cutoff level for CA 125 has the required sensitivity and specificity to be used as a screening test. However, CA 125 increases exponentially with time in cases of ovarian cancer, and essentially remains level for non-cases. This property will be used to develop a more sensitive and specific screening modality based on serial CA 125 levels. To derive algorithms based on this criterion, stochastic models of the longitudinal behavior of CA 125 in cases and non-cases will be developed. Serial data have been obtained from two CA 125 screening trials of approximately 5,000 and 22,000 women respectively. The methods used to test the appropriateness and estimate the parameters in these models from the data are (i) Bayesian inference for continuous time ARMA processes, and (ii) Bayesian inference for classification analysis. The specific aims are: 1. To develop one step algorithms for the conduct of screening programs for ovarian cancer which maximizes expected years of life saved, while constraining the fraction of unnecessary surgeries; 2. develop stochastic models for: (i) the natural history of the ovarian cancer, (ii) the longitudinal behavior of CA 125 levels, (iii) the length of survival given the stage at detection of ovarian cancer, (iv) the variability of radioimmunoassays, due to radioactivity & experimental variability; 3. develop algorithms for statistical inference in screening programs to: (i) calculate the probability of ovarian cancer given the woman's longitudinal CA 125 data, (ii) calculate expected years of life saved given surgical intervention and CA 125 data, (iii) derive predictive distributions for a woman's future CA 125 values, (iv) accurately quantify radioimmunoassay variability; and 4. develop computer packages based on the above design and inference algorithms, and an optimal easily remembered rule so that the full potential for applicability in clinical practice can be realized.