PURPOSE: Screening for ovarian cancer has been notably ineffective.  We developed a model to explore the interaction of screening interval, test characteristics, and assumptions about natural history in evaluating screening strategies for reducing ovarian cancer mortality.

METHODS: We constructed a Markov cohort model of the natural history of ovarian cancer in the US general population.  Health states include Healthy, Undetected Ovarian Cancer (Stages I-IV), Diagnosed Ovarian Cancer (Stages I-IV), Benign Oophorectomy, Death from Ovarian Cancer, and Death from Other Causes.  Probabilities for most parameters were obtained from the literature and publicly accessible databases.  To reflect the uncertainty about the natural history of ovarian cancer, we constructed two separate models: one which required all cancers to progress to Stage II prior to development of Stage III, and one which allowed some cancers to progress directly from Stage I to Stage III.  Values for transition probabilities between stages of cancer, and from undetected to diagnosed cancer were imputed by fitting values until cancer incidence, stage distribution, and mortality approximated data from the National Cancer Institute's SEER database.  Screening test sensitivity varied from 80-99%, specificity from 95-99%, and screening interval from 3 to 24 months.  Main outcomes were percent reduction in ovarian cancer mortality and positive and negative predictive values of screening. 

RESULTS: Under both assumptions about natural history, reduction in ovarian cancer mortality was more dependent on screening interval than test sensitivity—at any given screening interval, the estimated mortality reduction achieved by increasing sensitivity from 80% to 99% was equivalent to the reduction achieved by decreasing the screening interval by 3 months.  To achieve a reduction in ovarian cancer mortality of ≥50%, a screening interval of 12 months or less is necessary even at 99% sensitivity.  However, at this level of screening frequency, positive predictive values were ≤3% even at 99% specificity.   

CONCLUSIONS: Screening at intervals of ≤12 months is required for substantial reduction in ovarian cancer mortality.  The combination of frequent screening and relative rarity of ovarian cancer mean that false positive rates are remarkably high, even at high levels of test specificity.  The failure to date of ovarian cancer screening may be more attributable to a natural history which is unfavorable to screening rather than the lack of an appropriate test.