Given the complexity of the decision and the absence of a randomized, controlled trial to evaluate screening strategies for ovarian cancer, we turned to decision analysis to examine this issue. This allowed quantification of the effect of relevant factors and specification of test characteristics required for an effective screening strategy. The specific question we pose is, "will a one-time screen improve the average life expectancy in a cohort of 40-year-old women?"

Natural History of Ovarian Cancer

Incidence, Risk, and Prognostic Factors

The annual incidence of ovarian cancer in the United States is 13.8/100 000 with an annual incidence of death of 8.5/100 000 [2]. Its cause remains unknown, and survival rates have been relatively stable in the past 35 years [3]. The strongest risk factor for the disease is increasing age. Other risk factors are family history of ovarian cancer, residence in a Western industrialized country, and nulliparity [4]. Established protective factors are multiparity and oral contraceptive use [5].

The most important prognostic factor for ovarian cancer is stage of disease at diagnosis [6]. The present staging system for ovarian cancer was devised by the International Federation of Gynecology and Obstetrics. The system defines stage I as disease limited to the ovaries, stage II as disease limited to the true pelvis, stage III as disease that involves intraperitoneal spread, and stage IV as disease with distant metastasis. Other independently prognostic factors are the size of residual tumor remaining after a surgical debulking procedure and the histologic grade of the tumor [6].

Clinical Presentation and Disease Course

Ovarian cancer is often silent in the early stages of disease, yet aggressive and resistant to treatment late in its course when most cases become clinically apparent. Ovarian cancer first spreads locally with metastasis to the contralateral ovary, the uterus, and the fallopian tubes. Further dissemination occurs through the peritoneum to involve the omentum, bowel, and retroperitoneal lymph nodes. Hematogenous spread is rare but can result in distant metastasis. As the disease progresses, gastrointestinal and urinary tract obstruction are common. These late complications are often resistant to effective medical or surgical intervention because of the extensive amount of disease. Most deaths occur within 5 years of diagnosis [7].

Early diagnosis of ovarian cancer is difficult because symptoms are often vague and nonspecific. Presenting symptoms include abdominal swelling and pain, abnormal uterine bleeding, and gastrointestinal and urinary complaints [8]. A retrospective study of 277 patients with primary ovarian cancer reports the average patient delay in seeking medical evaluation after the onset of symptoms to be 9.1 months and the average physician delay in performing pelvic examination in symptomatic patients to be 9.6 months [9]. The clinical significance of these delays is not known.

Available Screening Tests

The optimal characteristics of a screening test include ease of performance, patient acceptability, limited expense, lack of complications, and high sensitivity and specificity. Sensitivity is defined as the proportion of patients with the target disorder who have a positive test result. Specificity is defined as the proportion of patients without the target disorder who have a negative test result^[10]. Screening tests that have been proposed for ovarian carcinoma include manual pelvic examination, cytologic detection, serologic testing with monoclonal antibodies, abdominal ultrasound, and transvaginal sonography (TVS). Of these, the monoclonal antibody to the CA 125 antigen and TVS are the most promising because of their high specificity and reproducibility and ease of performance. The test characteristics of these modalities are described below.

Pelvic Examination and Cytologic Detection

No evidence exists that annual pelvic examinations for detecting cervical cancer in asymptomatic women have increased case finding for early ovarian cancer. Compared with abdominal ultrasonography, the pelvic examination reportedly detects 34% of adnexal masses and over-reads 8% [11]. The cervical Papanicolaou smear, which is an effective test for the early detection of cervical cancer, is not a good screening test for ovarian carcinoma. In a population of 164 patients diagnosed with ovarian carcinoma, only 11% had a positive Papanicolaou smear [12]. Peritoneal and cul-de-sac lavage have also been investigated as possible screening tests for ovarian cancer but are cumbersome to perform and have poor test performance [13, 14].

Abdominal Ultrasonography

Abdominal ultrasonography has been evaluated extensively as a screening test for ovarian cancer. Uncontrolled screening trials have provided information on test characteristics but have failed to show a convincing benefit in terms of the detection of early-stage disease. The specificity of abdominal ultrasound scans in the detection of ovarian cancer was estimated to be 97.7% (95% CI, 96.4% to 99.0%) in a study of 5000 asymptomatic women aged 18 to 78 years (mean age, 52 years) who were screened with abdominal ultrasound for 3 consecutive years [15]. The sensitivity of abdominal ultrasound scan to detect a pelvic mass was found to be 90% in a study of 72 women with a diagnosis of pelvic mass who subsequently had laparotomy [16]. Abdominal ultrasonography has the disadvantages of being time consuming and expensive and requiring significant patient preparation.

Transvaginal Sonography

The use of TVS has been proposed as an alternative to abdominal ultrasound in ovarian cancer screening. Transvaginal sonography was developed to improve the resolution of abdominal ultrasound and provides detailed imaging of the ovary and masses confined to the true pelvis. It is easy to perform, well accepted by patients, and shows strong interobserver interpreter agreement [17, 18]. In premenopausal women, a normal ovary is defined as having a volume of 18 cm³ or less and being uniformly hypogenic or entirely cystic. If the ovary is enlarged but the structure is normal, the scan is repeated 1 week after menses. If the scan remains abnormal, the test is considered positive. In postmenopausal women, a normal ovary is defined as having a volume of 8 cm³ or less and a uniformly hypogenic internal structure. Any ovary that exceeds this volume or shows complex or solid areas on sonography is defined as abnormal. The specificity of TVS for ovarian carcinoma was found to be 98.1% (CI, 97.4% to 98.8%) in a study of 1300 asymptomatic postmenopausal women who underwent screening with TVS [17]. The sensitivity of TVS in the detection of an ovarian mass was found to be 90% in a study of postmenopausal women undergoing elective surgery for reasons unrelated to adnexal disease [19].

Transvaginal sonography with color flow imaging is under investigation as a screening modality [20]. This test can detect intraovarian vascular changes and measure impedance to blood flow in ovarian masses as potential indices of early malignancy. This technique may increase the specificity of conventional ultrasound examinations.

Monoclonal Antibodies

The development of monoclonal antibodies reactive to tumor-specific antigens has been applied to the diagnosis and management of ovarian cancer. The most promising monoclonal antibody with respect to ovarian cancer screening is 0C 125, which reacts to the CA 125 antigen. It was originally reported that serum CA 125 was greater than 35 U/mL in 83% of patients with epithelial ovarian cancer and 1% of presumably healthy blood donors [21]. CA 125 was also found to be elevated in nongynecologic carcinomas and benign abdominal disorders such as pancreatic pseudocyst, uterine fibroids, and endometriosis [21]. The sensitivity of CA 125 for ovarian cancer increases with the clinical stage at diagnosis, 50% for stage I and II neoplasms and 90% for stage III and IV neoplasms [22, 23]. The specificity of CA 125 for ovarian cancer was determined to be 97.6% in a prospective study of 5550 women ages 40 years or older who were screened with CA 125. In the study, an elevated level of CA 125 was followed with physical examination, abdominal ultrasound, and serial CA 125 levels, with further management based on clinical and sonographic findings [24].

Methods

Top Methods Results Discussion Author & Article Info References

The Decision Analysis Model

We designed a decision tree to test the strategy of a one-time screen versus no screen for a population of healthy 40-year-old women (Figure 1). We modeled the use of CA 125 and TVS as individual and combined screening tests. When used in combination, an abnormal test was required of both modalities to define a positive screen. A decision tree is a diagram that describes a clinical decision and possible outcomes of that decision. The tree begins with a decision node that represents an active choice that is to be made by the patient or physician. Each point of uncertainty is represented by a chance node. Branches of chance nodes indicate possible health states. Each health state is associated with a probability. Each branch ends in a terminal node, or final outcome, representing the average life expectancy of that health state. The estimated value of a strategy is determined by weighting the value of the final outcomes by the probability of its occurrence. To examine uncertainty in the model, each probability is examined independently for a range of plausible values.

View larger version (27K): [in this window] [in a new window]   Figure 1. Ovarian cancer decision tree. The decision tree is designed to test the strategy of no-screen compared with a one-time screen for a population of healthy 40-year-old women. There are two options at the decision node: no-screen or screen. A series of chance nodes represent the following points of uncertainty: the likelihood of disease, the percentage of prevalent disease in the early stage at the time of the screen, the clinical detection rate of early disease, the detection rate of early disease with the screening strategy, the specificity of the screening test, and the mortality rate associated with diagnostic laparotomy. Assigned to each terminal node is the life expectancy for an individual whose experience corresponds to that path in the decision tree. These include the life expectancy of a 40-year-old woman with no disease, with early-stage disease, and with late-stage disease. For health states that require a laparotomy, 1 week is subtracted from life expectancy. For health states in which disease had progressed from early- to late-stage before diagnosis, 1 year is added to life expectancy.

In this model, there are two options at the decision node: no-screen or screen. A series of chance nodes represent the following points of uncertainty: the likelihood of disease, the percentage of disease that is in the early stage at the time of the screening test, the clinical detection of early disease, the detection of early disease with a screening strategy, the specificity of the screening test, and the mortality rate associated with diagnostic laparotomy. Assigned to each terminal node is the life expectancy for an individual whose experience corresponds to that path in the decision tree. These include the life expectancy of a 40-year-old woman with no disease, with early-stage disease, and with late-stage disease.

Assumptions

Several assumptions are made in the model to simplify the analysis yet retain the basic clinical issues. First, we assume that survival time for early disease detected clinically is the same as survival time for early disease detected by screening, effectively correcting for potential lead-time bias [25]. Second, we assume that the morbidity and mortality rates associated with a diagnostic laparotomy are the same for patients with and without disease. Third, we disregard the potential benefit of identifying benign disease. Finally, we model only a one-time screen.

Probabilities

The baseline values and range of plausible values used in the decision tree are summarized in Table 1. The prevalence of undiagnosed disease can be calculated by the Equation m= P/I, where m is the average duration of the preclinical phase of disease and I is the incidence of disease [26]. Based on an incident rate of 14.3/100 000 and assuming a preclinical disease phase of 2 years, we estimate the prevalence of ovarian cancer in women aged 40 years to be 28.6/100 000 [2]. The range of values considered varied from 20 to 200/100 000.

There are no direct data on the distribution of early-compared with late-stage disease in the population. It was estimated that 50% of prevalent cases at any given time are in the early stage of disease. The range of values considered varied from 20% to 80%.

The percent of prevalent early-stage cases detected by current clinical practice is not known. Approximately 24% of incident ovarian cancers are diagnosed at an early stage [1]. Thus, it was estimated that without a screening intervention, 25% of disease in the early stage will be diagnosed clinically before progressing to late-stage disease. The range of values considered varied from 20% to 80%.

The sensitivity and specificity of the combined approach can be calculated from the test characteristics of the individual tests assuming conditional independence of the test modalities. This assumption is reasonable because the tests are based on different aspects of the tumor, one measuring the antigen released by tumor cells and one measuring the anatomic features of the tumor. The estimate of sensitivity for the combined strategy is 45% for early disease and 81% for late disease. We considered a plausible range of values from 20% to 80% for early disease and 50% to 100% for late disease. The estimate of specificity for the combined strategy is 99.95%; it varied from 96% to 100%.

Mortality rate for a diagnostic exploratory laparotomy, taken from the National Halothane Study, was estimated to be 0.23% [27]. The estimate is consistent with postoperative mortality rates in studies of elective staging procedures for Hodgkin disease [28]. For sensitivity analysis, this estimate varied from 0.0% to 10.0%.

Measure of Benefit

Benefit is quantified as improvement in life expectancy. Life expectancy is estimated using the Declining Exponential Approximation of Life Expectancy method [29]. The average life expectancy of a 40-year-old woman in the United States is 40.2 years [30]. The life expectancies of those with early-stage and late-stage ovarian cancer are estimated to be 38.46 years and 2.95 years, respectively. These calculations are based on studies reporting a 5-year survival rate of 81% for early-stage disease (median age of study population, 58 years) and a mean survival of 1.98 years for late-stage disease (median age of study population, 54 years) [31, 32]. For patients whose early disease progressed to late disease before diagnosis, 1 year is added to life expectancy to account for the time taken to progress from early- to late-stage disease. To account for the discomfort and inconvenience of laparotomy, 1 week is subtracted from the life expectancy of those patients undergoing surgery in the baseline case.

Analysis

The average life expectancy of the no-screen option compared with the screen option was calculated. A test strategy combining CA 125 and TVS was used as the baseline case. Sensitivity analysis was performed by independently changing the estimates of the following variables: disease prevalence, distribution of early- and late-stage disease, detection rate of early disease in usual clinical practice (the no-screen strategy), mortality rate for diagnostic exploratory laparotomy, transition time from early- to late-stage disease, and the sensitivity and specificity of the screening test. When the preferred strategy changed over the plausible range on any variable, a threshold value was calculated; above the threshold value one strategy was preferred; below the threshold value, the other strategy was preferred.

Results

Top Methods Results Discussion Author & Article Info References

Baseline and Sensitivity Analysis

When CA 125 and TVS were used in combination, the expected value of the screen option was greater than the expected value of the no-screen option (40.192 years compared with 40.191 years). The screen strategy increased the average life expectancy by approximately one third of a day of life. Sensitivity analysis was done to determine threshold values for the baseline screening strategy. The no-screen strategy was favored if the postoperative mortality rate exceeded 7.32% or the specificity of the test was less than 98.53% (Table 2).

Screening in the Elderly

The prevalence of ovarian cancer increases with age [33]. To assess the effectiveness of screening in the elderly, we applied the model to a population of 65-year-old women. The life expectancy of a healthy 65-year-old is 18.60 years [2]. Estimates of life expectancy for 65-year-old women with early- and late-stage ovarian cancer are 18.32 years and 2.72 years, respectively [30, 31]. The baseline variable estimates and range of plausible values for prevalence and postoperative death are given in Table 3 [2, 27]. The model favored the screen strategy in the elderly cohort as with the younger cohort, but only by a small amount. Average life expectancy in the screened population was increased by approximately three quarters of a day of life (18.589 years compared with 18.587 years). Sensitivity analysis found a threshold value for test specificity of 99.25%, below which the no-screen strategy would be favored.

Discussion

Top Methods Results Discussion Author & Article Info References

Although effective treatment for ovarian cancer is available, the disease is usually detected too late for women to gain the full benefit of treatment. Early detection remains the best hope for reducing the case-mortality rate. Some gains in early detection may be possible from prompt attention of patients and physicians to early signs and symptoms of the disease. But undoubtedly many cases will be asymptomatic until the disease has spread beyond the local stage. Thus, the role of screening is an essential question in the clinical approach to ovarian cancer.

The best tests available for ovarian cancer screening are CA 125 and TVS. However, their use is considered experimental because a decrease in the case-mortality rate from screening has not been demonstrated [34]. We used a decision analysis model to evaluate the effect of ovarian cancer screening on life expectancy. We chose to model a test strategy combining CA 125 and TVS to optimize test specificity. A high rate of test specificity is essential in designing a screening strategy for a disease with low prevalence. However, despite an estimated specificity of 99.95%, the benefit in terms of life expectancy was very small.

The morbidity and mortality rates associated with ovarian cancer screening might be decreased if an alternative to laparotomy was available to serve as a criterion standard for diagnosis. Laparoscopy, the insertion of a fiber-optic instrument in the peritoneum, might have a role in situations when the clinical work-up is suggestive of a benign mass. However, it is not always possible to distinguish a benign cyst from malignant disease before the tissue is examined histologically [35]. Because of concern that manipulation and spillage of an unexpectantly malignant tumor might occur, laparoscopy has no defined role in the diagnosis of ovarian cancer at this time [36].

Ultimately, the problem with screening for ovarian cancer is the low prevalence of disease. It is difficult to design effective screening strategies for rare conditions [37]. The prevalence of ovarian cancer is lower than that of other malignant diseases we screen for such as breast, colon, and prostate cancer. Future screening or case-finding strategies might focus on identifying a population with a higher prevalence of disease. This population would include those with multiple risk factors such as family history and nulliparity, not only age alone. Although prevalence increases with age, the increased mortality rate associated with laparotomy and the decreased life expectancy in the elderly limits the effectiveness of a screening strategy in this age group.

Mass screening for ovarian cancer will not improve average life expectancy in the population by a meaningful amount of time and cannot be recommended as an effective health policy. Primary care physicians must continue to be clinically astute in evaluating patients who have signs and symptoms suggesting ovarian cancer, such as abnormal vaginal bleeding or undiagnosed abdominal discomfort. Researchers have an ongoing challenge to identify characteristics that predict a population with a sufficiently high prevalence to make available tests useful for detection of preclinical disease.