Effect of Screening on Ovarian Cancer Mortality:  The Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Randomized Controlled Trial FREE

In the United States, ovarian cancer is among the 5 leading causes of cancer death in women.¹ The high case-fatality ratio of ovarian cancer may be attributed in part to its vague and nonspecific symptoms, which usually appear when the disease has reached an advanced stage. Ovarian cancer confined to the ovary has a 5-year survival of 92%. However, most women with ovarian cancer are diagnosed with advanced stage disease, which has a 5-year survival of only 30%.¹ The recognition that early detection of ovarian cancer may have the potential to improve prognosis prompted the development of randomized controlled trials, including the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial, to evaluate the efficacy of transvaginal ultrasound and serum cancer antigen 125 (CA-125) as screening tools to reduce ovarian cancer mortality.^2- 4

The PLCO trial's independent data and safety monitoring board, which periodically reviews the cumulative trial data, recently concluded that the predetermined end point had been reached for the ovarian cancer component of the trial and recommended that the ovarian cancer–specific mortality results be reported.

The design and methods of the PLCO trial have been reported.⁵ In brief, the PLCO trial was designed to determine the effect of specific cancer screening tests on cause-specific mortality. Enrollment began in November 1993 and concluded in July 2001. Planned follow-up was for up to 13 years from randomization. Women were randomized to either the intervention group or the usual care group. Women in the intervention group were offered annual screening with transvaginal ultrasound and CA-125, whereas those randomized to the usual care group were offered no interventions and only received their usual medical care. Women were considered eligible if they were aged 55 to 74 years and had no previous diagnosis of lung, colorectal, or ovarian cancer. The 2 initial exclusion criteria of previous oophorectomy and current tamoxifen use were dropped in 1996 and 1999, respectively.⁶ Women who had undergone previous bilateral oophorectomy were screened for lung and colorectal cancer but not for ovarian cancer. These women are not included in this analysis.

Participants were recruited at 10 screening centers across the United States. Recruitment targeted individuals from the general population residing within the catchment area of each of the screening centers. The various localities of the screening centers are reflected in the ethnic diversity of the PLCO participants. The recruitment strategies have been described.^7- 8

Eligible individuals were randomized with a scheme that used blocks of random permutations of varying lengths and was stratified by screening center, age, and sex. Random allocation was achieved using compiled software and encrypted files loaded to the screening center computers by the PLCO coordinating center. Due to the invasive nature of the screening procedures, participants were not blinded to their randomization allocation. Each institution obtained annual approval from its institutional review board to perform the study, and all participants provided written informed consent. At study entry, participants completed a self-administered baseline questionnaire, which included demographics (such as race/ethnicity classified as white, black, Asian/Pacific Islander, American Indian/Alaskan Native, or Hispanic origin), general risk factors, and screening and medical histories.

Women in the intervention group underwent screening with a CA-125 blood test and transvaginal ultrasound at baseline, an annual transvaginal ultrasound for an additional 3 years, and an annual CA-125 for an additional 5 years. The original protocol specified that women undergo a CA-125 annually for only 4 years. However, women were eligible by 1999 to have CA-125 screening for the fifth and sixth study years. Therefore, depending on when women were randomized, they may have had the opportunity to attend 4, 5, or 6 screening rounds. Bimanual examination of the ovaries was originally part of the screening procedures but was discontinued in December 1998 because no cancers were detected solely by ovarian palpation; in the usual care group, a high proportion of women underwent bimanual examination with ovarian palpation. The CA-125 screening assays were analyzed centrally at the Immunogenetics Laboratory (University of California, Los Angeles). The CA-125 assay results that were 35 U/mL or greater were classified as abnormal.⁶ In addition, blood and tissue samples were collected and stored in a central biorepository.⁹

Transvaginal ultrasound was conducted by trained examiners using a 5- to 7.5-MHz transvaginal probe.⁶ At least 5 minutes were spent looking for each ovary before categorizing the ovaries as being nonvisualized. The following transvaginal ultrasound results were classified as abnormal: (1) ovarian volume greater than 10 cm³, (2) cyst volume greater than 10 cm³, (3) any solid area or papillary projection extending into the cavity of a cystic ovarian tumor of any size; and (4) any mixed (solid and cystic) component within a cystic ovarian tumor. Quality assurance included duplicate screening examinations on a sample of participants, independent observation of the examination, independent review of transvaginal ultrasound films, or both.¹⁰

Participants and their physicians were notified in writing about suspicious abnormalities. In accordance with standard US medical practice, it was the responsibility of the participant's primary care physician to manage the diagnostic process to assess abnormalities. The PLCO coordinating center obtained the medical records for all diagnostic and therapeutic follow-up procedures.

For the purposes of this study, cancers detected by screening were defined as those diagnosed as a result of investigations initiated after a screening test with a positive result and without a lapse in the diagnostic evaluation exceeding 9 months. A false-positive result was defined as a positive screening examination result that did not result in cancers detected by screening. Interval cancers were defined as cancers not detected by screening and diagnosed within 12 months of the woman's last expected screening examination. Other cancers not detected by screening included those diagnosed among women who never underwent a screening examination (noncompliant) and those diagnosed after the screening phase was completed.

The primary study end point was ovarian cancer–specific mortality. Secondary end points included ovarian cancer incidence, cancer stage, survival, potential harms of screening, and all-cause mortality. Ovarian, primary peritoneal, and fallopian tube cancers were considered malignant ovarian neoplasms for this report (International Classification of Diseases for Oncology, Second Revision, codes C569, C481, C482, and C570). Tumors of low malignant potential (borderline tumors) were not included in the definition of malignant ovarian neoplasms, but are included as false-positive results (n = 21 in the intervention group; n = 6 in the usual care group).

All incident cancers (both PLCO trial and other cancers) and deaths were ascertained primarily by an annual study update questionnaire mailed to participants. Population-based cancer registries also were used when possible. Additionally, to obtain more complete mortality data, annual study update follow-up was supplemented by periodic linkage to the National Death Index. All ovarian cancers and deaths from confirmed ovarian cancer known as of February 28, 2010, are included in this analysis. Medical records pertaining to diagnosed cancers were obtained by the PLCO screening centers; data on the stage, histology, and grade of PLCO cancers were abstracted by certified tumor registrars. In addition, treatment information during the first year postdiagnosis was abstracted.

The underlying cause of death was determined in a manner that was uniform and unbiased and was based on the death certificate and relevant medical records; determination of the underlying cause of death also has been described in detail elsewhere.¹¹ Briefly, deaths potentially related to a PLCO trial cancer and those of unknown or uncertain cause were reviewed by at least 1 individual from a panel with appropriate expertise (epidemiology, surgery, medicine, radiation oncology); panel members were not otherwise affiliated with the trial and reviewers were blinded to the randomization group of the deceased participant.

Early on during the death review process, each death had 2 independent reviews and discrepancies were resolved by consensus after a third reviewer had examined the records. After 2 years, the process was streamlined. As a result, a primary reviewer considered the record without having access to the death certificate. If he or she recorded an underlying cause of death different from that of the death certificate, a second reviewer independently reviewed the record together with the death certificate. Disagreement between the reviewers resulted in another independent review, which was subsequently resolved by a meeting or teleconference.

Complications directly associated with screening by transvaginal ultrasound and CA-125 (eg, bleeding, fainting, nausea, bruising) were recorded at the time of the examinations. More severe complications such as infection, cardiovascular events, and bowel injury associated with cancer diagnoses and false-positive test results were assessed from the medical records. Oophorectomy rates in each study group were assessed through the use of a supplemental questionnaire, which was sent in 2006 or 2007 to participants after all PLCO screening had been completed. Similar to the baseline questionnaire, the supplemental questionnaire inquired about general risk factors and medical history, including oophorectomy. Oophorectomy rates among those women with at least 1 ovary at baseline were then calculated in each group as the number of oophorectomies reported on the supplemental questionnaire divided by the total person-time from enrollment to completion of the supplemental questionnaire.

Women were expected to undergo their annual screening if they had not been diagnosed with ovarian cancer, had not died, and had not undergone oophorectomy prior to the screening date. The rate of compliance with screening was calculated as the number of women actually screened divided by the number expected for the test. The use of screening outside of the trial protocol (contamination) was measured in 2 ways. Baseline contamination (defined as testing with either CA-125 or transvaginal ultrasound at least twice during the 3 years prior to trial entry) was determined from the baseline questionnaire. After trial enrollment, annual or biennial surveys were conducted in a 1% random sample of women in the usual care group to estimate the subsequent level of contamination. Contamination during the trial was defined as having reported a screening by CA-125 or transvaginal ultrasound in the prior year.

The primary analysis was a comparison of ovarian cancer mortality rates between the 2 study groups by intention to perform screening. Secondary aims included comparison of ovarian cancer incidence, cancer stage, survival, potential harms of screening, and all-cause mortality between the 2 groups.

The trial had 88% power to detect a 35% reduction in ovarian cancer mortality at a 1-sided α level of .05.⁵ These calculations included adjustment for expected contamination in the usual care group and noncompliance in the intervention group. The projected compliance rate in the screening group was greater than 90% for CA-125 and greater than 85% for transvaginal ultrasound; the projected contamination in the usual care group was less than 10% for both screening methods.

An interim analysis plan was used to monitor the primary end point for efficacy and futility. The plan used a weighted log-rank statistic with the weights increasing in proportion to the pooled ovarian cancer mortality. The weighted statistic was chosen because of the presumed delay in effect of screening on ovarian cancer mortality. The efficacy boundary was constructed via the Lan-DeMets approach using an O’Brien-Fleming spending function, and a nonbinding futility boundary was constructed via stochastic curtailment.¹²

After taking into consideration the fact that the futility boundary for the monitoring statistic had been crossed, the PLCO trial's data and safety monitoring board concluded in November 2009 that the primary aim of the ovarian component had been achieved and recommended that the results be reported. Further details about the interim analyses are provided in the eTable. Statistical analyses were performed using SAS software version 9 (SAS Institute Inc, Cary, North Carolina) and R statistical software.¹³

Event rates were defined as the number of events (diagnoses of ovarian cancer or deaths) divided by the person-time at risk for the event. Person-time was measured from randomization to the earliest date of diagnosis, death, or censoring for incidence rates and to the earliest date of the death or censoring for mortality rates. The censoring date was the earliest of the date of last contact, the 13-year anniversary of randomization, or February 28, 2010.

Rate ratios (RRs) were derived as the unadjusted ratio of event rates. The 95% confidence interval (CI) for the ovarian mortality RR was derived based on the sequential design according to the weighted method used to monitor the trial, which allows for a varying RR.¹⁴ The 95% CIs for the other RRs and the nominal 95% CI for the ovarian mortality RR were calculated using a single test normal critical value, which assumed a Poisson distribution for the number of events and a normal distribution for the logarithm of the ratio according to asymptotic methods.¹⁵ Additional details and references are provided in the eMethods and eReferences)

The current analysis excludes women with a pretrial history of bilateral oophorectomy. Women with a history of bilateral oophorectomy were included in the trial after a protocol change in 1996 permitted enrollment of such women (12% of the participants). From the original enrollment allocation of 39 105 women in the intervention group and 39 111 women in the usual care group, 4852 and 4807, respectively, were removed, leaving a total of 34 253 in the intervention group and 34 304 in the usual care group who were included in this analysis (Figure 1). The baseline characteristics of the participants in the 2 groups were well balanced (Table 1). Because randomization was performed between 1993 and 2001, some participants did not reach 13 years of follow-up by the cutoff date (February 28, 2010). However, all participants, regardless of follow-up time, were included in the analysis. The median follow-up was 12.4 years (25th-75th percentile, 10.9-13.0 years).

Screening outside the trial protocol (contamination) was minimal. In each study group, baseline contamination rates (ie, use of screening in the 3 years prior to trial entry) were 2.8% for CA-125 and 9.4% for transvaginal ultrasound. During the screening phase of the trial, contamination in the usual care group ranged from 2.3% to 3.2% per year for CA-125 and from 2.7% to 4.6% for transvaginal ultrasound.

Compliance with screening in the intervention group was 85% for CA-125 and 84% for transvaginal ultrasound at baseline, with compliance declining modestly to 79% and 78%, respectively, by the fourth screening. At the fifth and sixth screenings (at which transvaginal ultrasound was not administered), the proportions screened by CA-125 were 75% and 73%, respectively. Positive screening rates ranged from 1.4% to 1.8% for CA-125 during the 6 screening rounds and from 2.9% to 4.6% for transvaginal ultrasound during the 4 screening rounds.

Through the follow-up period, 212 ovarian cancer cases (5.7 per 10 000 person-years) were diagnosed in the intervention group and 176 cases (4.7 per 10 000 person-years) in the usual care group (RR, 1.21; 95% CI, 0.99-1.48). The accumulation of ovarian cancer cases over time since randomization appears in Figure 2. The excess of cases in the intervention group compared with the usual care group increased during the first 2 years and then remained approximately constant after year 3.

Examination of the ovarian cancer deaths in the final interim analysis demonstrated that the boundary for futility had been reached. There were 118 deaths caused by ovarian cancer (3.1 per 10 000 person-years) in the intervention group and 100 deaths (2.6 per 10 000 person-years) in the usual care group (mortality RR, 1.18; 95% CI, 0.91-1.54 [unadjusted] and mortality RR, 1.18; 95% CI, 0.82-1.71 [sequentially adjusted]). The cumulative numbers of ovarian cancer deaths over time appears in Figure 2.

Table 2 presents characteristics of the ovarian cancers by study group. Primary peritoneal and fallopian tube cancers, which were considered ovarian cases for this analysis, accounted for 20% of all ovarian cancers diagnosed in the intervention group (n = 43) and 14% in the usual care group (n = 25). Eliminating these cases did not alter the statistical significance of any of the findings (mortality RR, 1.04; 95% CI, 0.79-1.38). The histological subtypes were similar across groups and 55% were serous cystadenocarcinomas in the intervention group and 59% in the usual care group. The majority of cancers in each study group were high grade.

Table 2 also shows the cases in the intervention group by method and time of diagnosis. Of the 212 intervention group cases, 126 were diagnosed during the screening phase of the trial (59%), of which 58% were screen-detected cases, 29% were interval cases, and 13% were detected in those without prior screening (noncompliant).

Overall, the stage distributions were similar by study group with stage III and IV cancers comprising the majority of cases in both the intervention group (77%) and usual care group (78%) (Table 3). The absolute number of stage IV cancers was slightly higher in the usual care group (n = 54) than in the intervention group (n = 43), although this difference was not statistically significant. There was no significant association between study period and stage distribution; 76% of intervention group cases diagnosed during the screening phase of the study (years 0-5) were stage III or IV compared with 79% of those diagnosed during the postscreening phase. Among actual screen-detected cases (n = 73), 69% were stage III or IV (data not shown).

Treatments for ovarian cancer were similar across study groups overall and within each stage (Table 4). In the intervention group, 81% underwent surgery and received systemic therapy compared with 80% in the usual care group. Among women with stage III cancer, 93% in the intervention group and 87% in the usual care group underwent surgery and received systemic therapy.

The lead-time bias associated with early detection of ovarian cancer is illustrated in the eFigure. The eFigure presents ovarian cancer survival from the date of diagnosis and the date of randomization, which eliminates most of the lead-time effect. A small difference in survival was seen in the date from diagnosis (P = .18) but not in the date from randomization (P = .67) (eFigure). The difference in survival between the intervention and usual care groups was not statistically significant.

Minor complications such as fainting and bruising occurred at a rate of 58.3 per 10 000 women screened with CA-125 and 3.3 per 10 000 women screened with transvaginal ultrasound. Of women diagnosed with ovarian cancer, 95 in the intervention group (45%) and 91 in the usual care group (52%) experienced at least 1 major complication associated with their diagnostic procedures (ie, infection, blood loss, bowel injury, cardiovascular events).

Of 3285 women with false-positive results, 1080 underwent surgery (32.9% for oophorectomy) as part of the diagnostic workup. Of these 1080 women, 163 (15%) experienced a total of 222 distinct major complications, which yielded a rate of 20.6 complications per 100 surgical procedures (Table 5).

Oophorectomy rates were determined for the 22 955 women in the intervention group and the 22 542 in the usual care group who completed the supplemental questionnaire. A total of 1771 women in the intervention group (7.7%) and 1304 in the usual care group (5.8%) reported oophorectomy, which yielded rates of 85.7 and 64.2 per 10 000 person-years, respectively (RR, 1.33; 95% CI, 1.24-1.43).

All-cause mortality (excluding deaths from ovarian, colorectal, and lung cancer) was similar in the 2 study groups; there were 2924 deaths (76.6 per 10 000 person-years) in the intervention group and 2914 deaths (76.2 per 10 000 person years) in the usual care group (RR, 1.01; 95% CI, 0.96-1.06). Mortality rates for the major causes of death were generally similar between the 2 study groups (Table 6).

In this randomized controlled trial, we found no statistically significant reduction in mortality from ovarian cancer in a cohort of women derived from the general population who were screened for ovarian cancer with 6 annual CA-125 tests and 4 annual transvaginal ultrasound examinations. The numbers of deaths from ovarian cancer were similar in the 2 trial groups over the entire period of follow-up, with a modestly (although not statistically significant) greater cause-specific mortality rate in the intervention group (RR, 1.18; 95% CI, 0.82-1.71).

This finding cannot be explained by differences in participant characteristics between the 2 study groups because they were virtually identical at baseline and follow-up was almost complete in both groups. Stage-specific treatments also were similar across the 2 study groups. Compliance with screening was high in the intervention group and contamination was low in the usual care group. We conclude that the screening intervention, as implemented in this trial, was not effective in reducing mortality caused by ovarian cancer.

In this trial, there was a lack of an observed stage shift. A stage shift (ie, a decrease in the absolute number of late stage [III or IV] cases in the intervention group compared with the usual care group) is thought to be necessary but not sufficient for a mortality benefit to be realized. The lack of a stage shift was apparent in the finding that the total number of advanced stage cancers was greater in the intervention group (n = 163) than in the usual care group (n = 137). This was true overall as well as when limited to the screening phase of the trial (years 0-5), even though the majority of intervention group cases diagnosed during this period were detected by screening. Among cases detected by screening, 69% of the cancers were in the late stage, which is a percentage only slightly lower than the usual care group (78%).

This lack of a stage shift suggests the possibility that the 2 screening modalities used (CA-125 and transvaginal ultrasound) with the cutoffs used in the PLCO trial for screening positivity were not effective in detecting ovarian cancers early enough when the cancers were still in a nonadvanced stage. Some evidence from modeling suggests that ovarian tumors need to be found when they are relatively small, considerably smaller than the current threshold used for transvaginal ultrasound (10 cm³ for cysts) to be in an early stage at detection.¹⁶

Similarly, had a CA-125 threshold lower than 35 U/mL been used, it may have been possible to detect cancers at an earlier stage; however, this would be at the expense of more false-positive results (and perhaps overdiagnosis of clinically indolent tumors). Other approaches to ovarian cancer screening with these same modalities, including the risk of ovarian cancer algorithm, which incorporates longitudinal CA-125 and considers changes over time rather than a single cut point, may be able to detect ovarian cancers earlier and at a reasonable cost in terms of increased false-positive results; however, the benefit of this approach has not been demonstrated.^17- 18

The risk of ovarian cancer algorithm is currently being evaluated in the randomized UK Collaborative Trial of Ovarian Cancer Screening,¹⁹ which is a 3-group trial comparing 2 strategies of screening (specifically, annual CA-125 testing interpreted using the risk of ovarian cancer algorithm with transvaginal ultrasound as a second-line test [multimodal group] vs annual transvaginal ultrasound only vs no intervention). To date, only the results of the prevalence screening from the UK Collaborative Trial of Ovarian Cancer Screening have been published. The multimodal group had 34 primary invasive epithelial cancers and 16 were stage I or II (47%); in the transvaginal ultrasound group, 50% of 24 such cancers were stage I or II. Whether these modalities will produce a stage shift and demonstrate a mortality benefit compared with usual care await further findings.

An additional randomized controlled trial of ovarian cancer screening is the Shizuoka Cohort Study of Ovarian Cancer Screening in Japan.⁴ This trial randomized women to undergo either 5 annual screenings with concurrent transvaginal ultrasound and CA-125 or to a usual care control group. Although the number of women enrolled was similar to that of the PLCO trial, only 35 cancers in the intervention group and 32 in the control group were diagnosed after a mean follow-up of 9.2 years. Mortality findings from this trial have not yet been reported.

It is possible that even an optimized program of annual screening may be insufficient to detect cancers early enough to reduce mortality. Evidence from modeling suggests that aggressive cancers progress rapidly through the early stages, limiting the ability to detect these cancers with yearly screening.²⁰ In contrast, more ovarian cancers were diagnosed in the screened group than in the usual care group (212 vs 176), suggesting that some of the additional cancers detected by screenings were not clinically important and, if left undetected, may never have caused any symptoms or affected the women during their lifetimes (ie, overdiagnosis).

The false-positive results rate in the PLCO trial was approximately 5% of those screened at each round, with about 60% of these resulting from transvaginal ultrasound (a modality that has trouble distinguishing benign adnexal masses from malignant entities).²¹ Although this rate is comparable with or slightly lower than the false-positive results rate of mammography screening, the nature of the diagnostic follow-up, which often included invasive procedures, was a serious concern.^21- 22 As evidenced by the differing rate of oophorectomies between groups in this study (33% higher rate in intervention group), false-positive results resulted in more women undergoing major surgery in the intervention group than in the usual care group. Apart from health care costs, some women who were not ultimately diagnosed with ovarian cancer experienced serious medical complications associated with their diagnostic follow-up of a false-positive screening result.

The PLCO trial had certain limitations. The trial was powered for a 35% mortality reduction based on a predicted number of mortality events (n = 226) that was essentially met. However, from a public health point of view, smaller effect sizes are still potentially worthwhile to detect. The sequentially adjusted lower 95% CI for the mortality RR was 0.82, indicating at most an 18% relative benefit within the limits of reasonable probability. Additionally, the data collected on treatment were somewhat limited. The PLCO trial neither abstracted the type of systemic therapy used, nor the type of surgeon who performed the oophorectomy (eg, gynecologic oncologist or not); both factors have been shown to be related to ovarian cancer survival. However, we have no reason to suppose that these factors differed by study group.

We conclude that annual screening for ovarian cancer as performed in the PLCO trial with simultaneous CA-125 and transvaginal ultrasound does not reduce disease-specific mortality in women at average risk for ovarian cancer but does increase invasive medical procedures and associated harms.