Risk-stratified CA125 screening integrating CA125 trajectories, trajectory-specific progression and transvaginal ultrasound for ovarian cancer.

After a median follow-up of 14.8 years for OC incidence and 23.8 years for OC mortality, a total of 250 OC cases and 218 OC deaths were documented among 28,456 participants included in this study. The tdAUC for 10-year OC incidence risk based on baseline CA125 was 0.663. The optimal cut-off value of baseline CA125 for predicting 10-year OC risk was determined to be 13.00 U/ml (Fig.  1 A). Bootstrap resampling with 2000 iterations confirmed this cut-off value (Supplementary Table S2). Participants, as well as incident OC cases and non-cases, were categorized by CA125 levels of < 13 U/ml, 13–<35 U/ml, and ≥ 35 U/ml at different measurement times (Supplementary Table S3). The proportion of women with CA125 levels ≥ 35 U/ml increased from 1.2% in the first-round CA125 test to a peak of 1.8% in the third round, and then stabilized at 1.6% through the sixth round. Furthermore, in rounds two through six, the proportion of newly diagnosed OC patients with CA125 levels above 35 U/ml, as well as those with levels between 13 and 35 U/ml, was significantly higher compared to the non-cancer group. Fig. 1 Time-dependent receiver operating characteristic curves of 10-year OC risk with CA125 ( A ) and Kaplan-Meier curves of OC incidence ( B ) and mortality ( C ) under different CA125 trajectories Note: CA125 SN , stable negative CA125; CA125 LP , loss of positive CA125; CA125 SP , stable positive CA125; CA125 GP , gain of positive CA125 Time-dependent receiver operating characteristic curves of 10-year OC risk with CA125 ( A ) and Kaplan-Meier curves of OC incidence ( B ) and mortality ( C ) under different CA125 trajectories Note: CA125 SN , stable negative CA125; CA125 LP , loss of positive CA125; CA125 SP , stable positive CA125; CA125 GP , gain of positive CA125 Based on the optimal cut-off value, participants were categorized into four groups based on their CA125 trajectories: 17,099(60.1%) women with CA125 SN , 1,740(6.1%) women with CA125 LP , 6,069(21.3%) women with CA125 SP , and 3,548(12.5%) women with CA125 GP . As detailed in Supplementary Table S1 , compared to CA125 SN , the CA125 GP group exhibited elder age at recruitment, a higher proportion of white American, lower BMI, more smokers, fewer previous users of oral contraceptive, more previous users of hormone replacement therapy, younger age at menopause, fewer prior hysterectomy, more live births, and shorter time interval between CA125 tests (all P values < 0.05). As shown in Fig.  1 B and C and detailed in Table  1 , Kaplan-Meier life table analyses revealed that women with CA125 GP had the highest crude rates of OC incidence and mortality (25.92 and 11.36 per 10,000 person-years).This was following by women with CA125 SP (16.35 and 6.63 per 10,000 person-years), CA125 LP (7.35 and 2.91 per 10,000 person-years), and CA125 SN (8.00 and 4.23 per 10,000 person-years). Both incidence and mortality rates differed significantly among groups, with all p-values for the log-rank test being < 0.001. Table 1 Associations of CA125 trajectories with OC incidence and mortality CA125 trajectories Participants, N (%) Events, N (%) Follow-up, 10,000 PYs Rates of events, per 10,000 PYs Unadjusted HR (95%CI) P value* Adjusted HR (95%CI) † P value † OC incidence 28,456(100.0) 250(100.0) 20.82 12.01  CA125 SN 17,099(60.1) 101(40.4) 12.62 8.00 Ref. < 0.001 Ref.  CA125 LP 1740(6.1) 9(3.6) 1.22 7.35 0.93(0.47–1.85) 0.92(0.46–1.81) 0.799  CA125 SP 6069(21.3) 70(28.0) 4.28 16.35 2.09(1.54–2.84) 2.00(1.47–2.73) < 0.001  CA125 GP 3548(12.5) 70(28.0) 2.70 25.92 3.17(2.34–4.30) 3.06(2.25–4.16) < 0.001 OC mortality 28,456(100.0) 218(100.0) 39.53 5.52  CA125 SN 17,099(60.1) 102(46.8) 24.13 4.23 Ref. < 0.001 Ref.  CA125 LP 1740(6.1) 7(3.2) 2.41 2.91 0.69(0.32–1.48) 0.72(0.34–1.55) 0.404  CA125 SP 6069(21.3) 54(24.8) 8.14 6.63 1.58(1.13–2.19) 1.58(1.13–2.21) 0.007  CA125 GP 3548(12.5) 55(25.2) 4.84 11.36 2.69(1.94–3.74) 2.60(1.87–3.62) < 0.001 Note: PY, person-year; IR, incidence rate; HR (95%CI), hazard ratio (95% confidential interval); *,log-rank test; †, adjusted available factors associated with CA125 trajectories as observed in Supplementary Table S1 Associations of CA125 trajectories with OC incidence and mortality Note: PY, person-year; IR, incidence rate; HR (95%CI), hazard ratio (95% confidential interval); *,log-rank test; †, adjusted available factors associated with CA125 trajectories as observed in Supplementary Table S1 After adjusting for factors associated with CA125 trajectories, as detailed in Supplementary Table S1 , the HRs (95%CI) for OC incidence were 3.06(2.25–4.16) for CA125 GP , 2.00(1.47–2.73) for CA125 SP , and 0.92(0.46–1.81) for CA125 LP compared to CA125 SN . For OC mortality, the HRs (95%CI) were 2.60(1.87–3.62) for CA125 GP , 1.58(1.13–2.21) for CA125 SP , 0.72(0.34–1.55) for CA125 LP (Table  1 ). As depicted in Fig.  2 , among women with CA125 SP , the tdAUC for RelVel of 10-year OC incidence risk (0.712) was significantly higher than that for AbsInc, MaxAbsInc and MaxRelInc (all P-values < 0.05), though it was comparable to other trajectory-specific progression indicators. Similarly, for women with elevated CA125 in CA125 GP , the tdAUC for RelVel (0.767) was significantly greater than that for MaxAbsVel and MaxAbsInc (all P value < 0.05), while remaining comparable to other indicators. Thus, RelVel was selected as the optimal trajectory-specific progression indicators for both CA125 SP (Fig.  2 A) and CA125 GP (Fig.  2 B). Based on tdROC, the optimal cut-off values for RelVel were determined to be a 9% annual increment for CA125 SP and a 32% annual increment for CA125 GP . Bootstrap resampling confirmed these results (Supplementary Table S2). Fig. 2 Time-dependent receiver operating characteristic curves of 10-year incidence of ovarian cancer with trajectory-specific progression indicators among women with stable positive CA125 ( A ) and gain of positive CA125 ( B ) Time-dependent receiver operating characteristic curves of 10-year incidence of ovarian cancer with trajectory-specific progression indicators among women with stable positive CA125 ( A ) and gain of positive CA125 ( B ) Based on these cut-off values, participants with either CA125 SP or CA125 GP were reclassified into negative and positive progression groups. As shown in Supplementary Figure S2, cumulative OC incidence and mortality were significantly higher in the positive progression group compared to the negative progression group within both CA125 SP and CA125 GP categories. After adjusting for confounder variables, positive progression remained significantly associated with increased risk of OC incidence [3.83(1.96–7.51) for CA125 SP and 7.26(4.00-13.17) for CA125 GP ] and mortality [6.04(2.78–13.14) for CA125 SP and 8.03(4.15–15.56) CA125 GP ] (Table  2 ). Additionally, compared to women with CA125 SN , both CA125 LP and trajectory-specific negative progressions [including RelVel(-) within CA125 SP and RelVel(-) within CA125 GP ] were associated with similar risk of OC(Supplementary Figure S3, Table S4). Table 2 Associations of trajectory-specific progression indexes with OC incidence and mortality CA125 trajectories Trajectory-specific progression Participants, N (%) Event, N Follow-up, 10,000 PYs Event rate, per 10,000 PYs Adjusted HR (95%CI) † P value † OC incidence  CA125 SP RelVel (-) 2860(74.8) 17 1.98 8.56 Ref. RelVel (+) 965(25.2) 29 0.72 40.08 3.83(1.96–7.51) < 0.001  CA125 GP RelVel (-) 2891(81.5) 25 2.18 11.46 Ref. RelVel (+) 657(18.5) 45 0.52 86.54 7.26(4.00-13.17) < 0.001 OC mortality  CA125 SP RelVel (-) 2860(74.8) 11 3.85 2.86 Ref. RelVel (+) 965(25.2) 24 1.24 19.36 6.04(2.78–13.14) < 0.001  CA125 GP RelVel (-) 2891(81.5) 20 4.03 4.97 Ref. RelVel (+) 657(18.5) 35 0.81 43.00 8.03(4.15–15.56) < 0.001 Note: PY, person-year; HR (95%CI), hazard ratio (95% confidential interval); -, negative; +, positive. RelVel, relative velocity; † , adjusted available factors associated with CA125 trajectories as observed in Supplementary Table S1 Associations of trajectory-specific progression indexes with OC incidence and mortality Note: PY, person-year; HR (95%CI), hazard ratio (95% confidential interval); -, negative; +, positive. RelVel, relative velocity; † , adjusted available factors associated with CA125 trajectories as observed in Supplementary Table S1 Table  3 summarized the performance metrics of various joint screening strategies. The sensitivity, specificity and positivity for TVU alone were 55.6%, 51.4% and 48.7%, respectively. Compared to traditional strategy one, which used TVU alone, traditional strategy two, which combined TVU and CA125 testing, demonstrated significant higher sensitivity (60.8%) and positivity (50.0%), but lower specificity (50.1%). When the positive criterion for CA125 was adjusted to the optimal cut-off value, optimized strategy one achieved increased sensitivity (79.6%) and positivity (67.9%), though the specificity decreased to 32.2%. To address potential false positive, optimized strategy two excluding CA125 LP as a positive screen, resulting in a decrease in both sensitivity (76.0%) and positivity (61.8%), while specificity increased to 38.3%. Further refining the strategy to exclude trajectory-specific negative progression, optimized strategy three led to a reduction in sensitivity (64.4%) and positivity (51.3%), but an increase in specificity to 48.8%. Compared to traditional strategy one, optimized strategy three showed significantly higher sensitivity and positivity, although with lower specificity (both P values < 0.001), and improved Youden, PPV and NPV. Table 3 Comparisons of screening performances between traditional and optimized joint screening strategies for ovarian cancer Methods Cases Non-cases Total Sensitivity* Specificity* Positivity † Youden PPV NPV %(95%CI) P %(95%CI) P %(95%CI) P % %(95%CI) %(95%CI) Traditional strategy 1 (positive screen defined as only any positive TVU)  Positive 139 13,720 13,859 55.6 (49.2–61.8) Ref. 51.4 (50.8–51.9) Ref. 48.7 (48.1–49.3) Ref. 7.0 1.0 (0.85–1.19) 99.2 (99.1–99.4)  Negative 111 14,486 14,597  Total 250 28,206 28,456 Traditional strategy 2 (positive screen defined as any positive TVU and CA125 above diagnostic criteria)  Positive 152 14,074 14,226 60.8 (54.4–66.8) < 0.001 50.1 (49.5–50.7) < 0.001 50.0 (49.4–50.6) < 0.001 10.9 1.1 (0.91–1.25) 99.3 (99.2–99.4)  Negative 98 14,132 14,230  Total 250 28,206 28,456 Optimized strategy 1 (positive screen defined as any positive TVU and CA125 above optimal cut-off value)  Positive 199 19,133 19,332 79.6 (74.0-84.3) < 0.001 32.2 (31.6–32.7) < 0.001 67.9 (67.4–68.5) < 0.001 11.8 1.0 (0.89–1.18) 99.4 (99.3–99.6)  Negative 51 9073 9124  Total 250 28,206 28,456 Optimized strategy 2 (based on strategy 1 and excluding CA125 LP as positive screen)  Positive 190 17,402 17,592 76.0 (70.1–81.1) < 0.001 38.3 (37.7–38.9) < 0.001 61.8 (61.3–62.4) < 0.001 14.3 1.1 (0.93–1.25) 99.4 (99.3–99.6)  Negative 60 10,804 10,864  Total 250 28,206 28,456 Optimized strategy 3 (based on strategy 2 and excluding negative RelVel in either CA125 SP or CA125 GP as positive screen)  Positive 161 14,447 14,608 64.4 (58.1–70.3) < 0.001 48.8 (48.2–49.4) < 0.001 51.3 (50.8–51.9) < 0.001 13.2 1.1 (0.94–1.29) 99.4 (99.2–99.5)  Negative 89 13,759 13,848  Total 250 28,206 28,456 Note: CA125 LP , loss of positive CA125; CA125 SP , stable positive CA125; CA125 GP , gain of positive CA125; RelVel, relative velocity; PPV and NPV, positive and negative prediction value. *, compared with McNemer tests; † , compared with Pearson chi-square tests Comparisons of screening performances between traditional and optimized joint screening strategies for ovarian cancer 55.6 (49.2–61.8) 51.4 (50.8–51.9) 48.7 (48.1–49.3) 1.0 (0.85–1.19) 99.2 (99.1–99.4) 60.8 (54.4–66.8) 50.1 (49.5–50.7) 50.0 (49.4–50.6) 1.1 (0.91–1.25) 99.3 (99.2–99.4) 79.6 (74.0-84.3) 32.2 (31.6–32.7) 67.9 (67.4–68.5) 1.0 (0.89–1.18) 99.4 (99.3–99.6) 76.0 (70.1–81.1) 38.3 (37.7–38.9) 61.8 (61.3–62.4) 1.1 (0.93–1.25) 99.4 (99.3–99.6) 64.4 (58.1–70.3) 48.8 (48.2–49.4) 51.3 (50.8–51.9) 1.1 (0.94–1.29) 99.4 (99.2–99.5) Note: CA125 LP , loss of positive CA125; CA125 SP , stable positive CA125; CA125 GP , gain of positive CA125; RelVel, relative velocity; PPV and NPV, positive and negative prediction value. *, compared with McNemer tests; † , compared with Pearson chi-square tests Based on the above results, optimized joint screening strategy three is recommended as the most effective approach for integrating CA125 trajectories, CA125 progression indicators, and TVU for asymptomatic women. As illustrated in Fig.  3 , the proposed screening protocol begins an initial round of TVU and CA125 testing. Woman with any positive TVU was advised to undergo further examinations. Those with negative TVU results proceed to a second round of TVU and CA125 testing. In this second round, women with any positive TVU were again referred for further evaluation, while those with positive progression indicators within either CA125 SP or CA125 GP [namely RelVel(+)] are recommended for further examination. Women with RelVel(-) within both CA125 SP and CA125 GP , as well as those with CA125 LP , are advised to undergo re-evaluation of TVU and CA125. Fig. 3 The recommended flowchart of joint screening with TVU and CA125 for ovarian cancer Note: -, negative; +, positive. RelVel, relative velocity The recommended flowchart of joint screening with TVU and CA125 for ovarian cancer Note: -, negative; +, positive. RelVel, relative velocity

In summary, OC risk varies across different CA125 trajectories and trajectory-specific progression. Future monitoring or intervention strategies should be tailored to these factors. Integrating CA125 trajectories, trajectory-specific progression indicators, and TVU could refine the identification of high-risk population and enhance the performances of OC screening. Future research with more sophisticated design and larger sample is needed to validate the current findings.

This study is the first to determine the optimal screening cut-off value for CA125 in OC and to investigate the association between CA125 trajectory, trajectory-specific progression, and OC risk. Our findings reveal that a gain of positive CA125 is associated with a higher OC risk compared to other CA125 trajectories. Additionally, within the same trajectory, positive progression is associate with a significantly higher risk of OC than negative progression. This study also proposes and evaluates a novel screening strategy integrating CA125 trajectories, optimal screening cut-off value, trajectory-specific progression indicators, and TVU. The optimized screening strategy not only significantly reduced the number of missed OC by 3.6%, but also improved over accuracy. In summary, the strategy can effectively refine the identification of high-risk population and is recommended as a fundamental strategy to enhance OC screening performance and reduce missed diagnoses. Several previous studies have suggested that serial CA125 measurements and longitudinal algorithms, such as ROCA, could improve OC screening effectiveness under single-threshold rules [ 14 , 17 , 21 – 23 ]. However, some study have indicated that the adoption of ROCA in PLCO did not result in a significant reduction in OC mortality [ 23 ]. One potential reason for the limited benefit of ROCA-based two-stage multimodal screening (MMS) in both the PLCO and UKCTOCS might be the underestimated role of ultrasound in OC screening. ROCA is designed to support decisions following a positive CA125 test [ 24 ], but in the MMS approach, a second-line TVU examination is only conducted if their first-line CA125 test indicates increased risk [ 6 ]. In the UKCTOCS, only 41% of OC cases had a positive screen in the MMS arm, while TVU alone had a sensitivity of 32% [ 6 ]. Thus, the performance of MMS based on ROCA is limited by the effectiveness of longitudinal CA125 tests alone. Integrating CA125 and TVU screening in parallel would offer better performance than CA125 alone or CA125 and TVU in series. Secondly, the ROCA may struggle to differentiate OC risks associated with identical changes in CA125 levels but different time intervals, as it relies solely on current and past CA125 values, along with age, to estimate risk using a Bayesian change-point model. This approach does not incorporate the time intervals between longitudinal CA125 tests, unlike COX models or other discriminant models [ 10 , 14 ]. Consequently, ROCA might not detect time-dependent changes in CA125 that reflect early progression of OC [ 24 ]. In contrast, the four selected screening strategies in this study demonstrated significantly higher sensitivities compared to the 41% observed in the UKCTOCS [ 6 ]. These suggest that combining CA125 and TVU in parallel is likely to yield a lower missed diagnostic rate compared to using CA125 and TVU in series, which is comparable to the ROCA-based two-stage MMS in UKCTOCS. Furthermore, to address potential false positive associated with CA125 cut-off value below the diagnostic criteria of OC [ 2 ], the optimized screening strategy excluded women with definitive CA125 regression and those with negative progression within either CA125 SP or CA125 GP , even if they had elevated CA125 from baseline. Therefore, the redefined approach is expected to enhance detection of progressive OCs compared to traditional methods and improve screening performance over extended periods. Another key finding is that the OC risk associated with CA125 GP was significantly higher than that with CA125 SP . Although elevated CA125 levels can occur in other malignancies(e.g., breast and pancreatic cancer) [ 25 – 28 ] and various non-cancerous conditions(e.g., endometriosis, liver cirrhosis, pregnancy) [ 29 – 33 ], CA125 is typically elevated in approximately 50% of early-stage OC, 65% of mucinous OC, and 80–85% of advanced epithelial OC [ 34 – 36 ]. In OC screening, a stable positive CA125 may suggest non-cancerous conditions or relatively stable cancer, potentially reducing the need for aggressive intervention due to lower lethality [ 13 , 14 ]. Conversely, a significant increase or gain of positive CA125 from a baseline negative level, after excluding noncancerous conditions, likely indicates cancer incidence or progression [ 13 , 14 ]. Kobayashi et al. found that non-serous cancers showed mildly elevated CA125 levels before diagnosis, while serous ovarian cancers often develop suddenly from normal CA125 levels, with a mean interval of only 1.4 years [ 37 ]. This aspect is often overlooked in single-test practice, potentially missing high-risk population suitable for OC screening. Additionally, CA125 GP was associated with smoking, previous use of hormone replacement therapy, and older age at menopause—all factors linked to increased risk of OC [ 38 ]. These findings warrant further investigation to understand why CA125 GP associates with lower BMI and less previous use of oral contraceptive. Additionally, while distinct optimal trajectory-specific progression indicators were anticipated for CA125 SP and CA125 GP , the same indicators were observed for both trajectories. Furthermore, there was no significant difference between most indicators within the same trajectory. This may be attributed to the long-time dilution effect of including OCs diagnosed clinically after the end of screening period, as different optimal indicators emerged when analyses were censored at an earlier time (results not shown). Nevertheless, trajectory-specific progression indicators are believed to reflect different pathways for OC incidence and/or different OC subtypes associated with varying CA125 changes [ 35 , 39 ]. Importantly, these findings suggest that different monitoring and intervention strategies may be needed for women with varying CA125 trajectories in the future. For instance, for women with stable positive CA125, intervention should focus on identifying or treating potential non-cancerous condition and controlling CA125 levels. Conversely, for women with a rapid gain of positive CA125, the primary interventional should target reducing exposure to risk factors. For BRCA mutation carriers, bilateral salpingo-oophorectomy remains effective strategy to lower OC risk [ 40 , 41 ]. Future research with improved study designs and larger sample is needed to validate these results and assumptions. Despite the lower specificity of the third revised strategy, which is the recommended optimized strategy, it may still be clinically appropriate for the following three points: First, higher sensitivity facilitates early detection of cases, which is crucial for improving patient outcomes. Identifying more true positives may outweigh the concern of increased false positives, particularly in serious conditions where early intervention can significantly alter prognosis. Second, to mitigate the impact of false positives, including the need for confirmatory testing and follow-up procedures, our optimized screening strategy initially recommended an additional round of CA125 only after a previous positive result. Following comparisons between different rounds of CA125 suggest no additional procedures are needed for women with a definitive regression of CA125 or negative progression, even if their CA125 levels were elevated from the baseline. Third, we did not recommend the third revised strategy based solely on higher sensitivity or specificity. Instead, we considered the higher Youden index, which integrates both sensitivity and specificity. Since the third revised strategy demonstrated a higher Youden index compared to the traditional strategy, we recommend it as the optimized screening strategy. To address the trade-offs between sensitivity and specificity more critically, a more comprehensive assessment of the proposed screening strategy is needed, taking into account its potential impact on patients’ well-being and overall cost-effectiveness. Additionally, this is not a study where a predefined sample was collected to test a specific hypothesis but rather a post-hoc analysis based on the PLCO study. Due to the relatively small number of ovarian cancer cases, this limited case number is likely to constrain the statistical power of the current analyses. In addition to the notable findings and the advantages of long-term follow-up, large sample size, and sophisticated analyses, several limitations must be addressed. First, the absence of an independent external validation population may limit the generalizability of these results to other populations. However, bootstrap resampling with 2000 iterations produced similar results, supporting the stability of the current findings. Second, different time intervals within the same CA125 trajectory could introduce bias into the results. Nonetheless, multivariable analyses that adjusted for these time intervals persistently showed significant associations, reinforcing the stability of the findings. Third, while the optimized strategy—after excluding definite regressive trajectory and negative trajectory-specific progression—was anticipated to detect significantly more early-stage OCs compared to traditional strategy, this was affected by the low incidence of OC and limited numbers of OC cases with clear stage information (data not shown). Nevertheless, the selected optimized screening strategy, which included several improved screening indexes, is expected to offer better long-term benefits compared to traditional methods.

Ovarian cancer (OC) is a prevalent malignancy and a leading cause of cancer-related mortality among women worldwide. According to GLOBOCAN 2020, approximately 314,000 new cases and nearly 207,300 deaths from OC in 2020 [ 1 ]. Over 60% of OC cases are diagnosed at an advanced stage, resulting in 5-year survival rate of less than 30%, compared to 90% for localized disease [ 2 – 4 ]. Current OC screening methods, including serum cancer antigen 125 (CA125) testing, transvaginal ultrasound (TVU), or a combination of both, aim to detect the disease at an earlier stage and reduce OC mortality. However, significant reductions in OC mortality have not been observed in current randomized controlled trials (RCTs) [ 5 – 8 ], despite a notable shift toward earlier-stage disease in the UK Collaborative Trial of Ovarian Cancer Screening (UKCTOCS) [ 6 , 9 ]. Several factors may conjointly contributed to the lack of significant reduction in OC mortality, such as insufficient increases in early-stage cases and decreases in late-stage cases, low incidence rate resulting in low positive predictive value of screening methods, and the dilution effect of including diagnosed OC cases after screening ends [ 5 , 6 , 10 ]. One major factor could be the ineffective cut-off value of CA125 used in screening. Since most participants are healthy women, only a minority have CA125 level above the clinical diagnostic threshold of 35 U/ml. Consequently, relying on this cut-off in population-based screening may lead to missed diagnosis of preclinical OC. For low-incidence cancers, significant increases in early-stage cases and reductions in late-stage cases are necessary to demonstrate meaningful mortality reduction; otherwise, the impact of screening may remain marginal [ 10 ]. Additionally, previous studies indicated that approximately 20% of patients with OC have CA125 levels below 35 U/ml [ 11 , 12 ], with even lower levels expected in asymptomatic women. This underscores the need to redefine the optimal CA125 cut-off value for effective screening. Furthermore, studies have shown that CA125 levels increase rapidly over time in patients with preclinical OC, while levels remain relatively stable in women without OC, even if initial CA125 levels are elevated [ 10 , 13 – 15 ]. Various indices and algorithms, such as the CA125 velocity [ 16 ], the empirical Bayesian longitudinal algorithm [ 17 ], and the widely used Risk of Ovarian Cancer Algorithm (ROCA) from the UKCTOCS trial [ 10 , 13 , 14 ], have been developed to assess CA125 progression and its association with OC risk. However, few studies have investigated the associations between CA125 trajectories and OC risk, and even fewer have examined optimal CA125 progression indicators within the same trajectory. Therefore, it is crucial to explore the associations and identify the most effective progression indicators. Additionally, while TVU is also recommended for OC screening [ 18 ], it remains unclear whether integrating CA125 trajectories, trajectory-specific progression indicators, and TVU into a combined screening approach would provide better outcomes compared to traditional strategy. Therefore, based on data of multiple rounds of CA125 and TVU screening from the Prostate, Lung, Colorectal and Ovarian Cancer (PLCO) Screening Trial, this study aims to determine the optimal CA125 cut-off value for OC screening and examine the associations between different CA125 trajectories and OC risk. Then this study aimed to identify the optimal trajectory-specific progression indicators for women with elevated CA125 and explore their associations with OC risk. Finally, this study aimed to develop and compare optimized joint screening strategies that integrate CA125 trajectories, trajectory-specific progression indicators, and TVU, with traditional screening methods to assess potential improvements in screening performance.

The PLCO Cancer Screening Trial was a multicenter randomized controlled trial (RCT) designed to investigate whether screening could reduce mortality from prostate, lung, colorectal and ovarian cancers. Details information about the trial have been provided elsewhere [ 19 ]. Briefly, from November 1993 to July 2001, 78,209 women aged 55 to 74 were recruited across ten PLCO screening centers. Eligible participants were randomly assigned to either the intervention or control groups in a 1:1 ratio. After providing informed consent, all participants completed a baseline questionnaire (BQ) that collected information on demographics, disease history, and lifestyle factors. For OC screening, the intervention group received annual CA125 tests for 6 years and annual TVU for 3 years, while the control group received usual care [ 5 , 19 , 20 ]. This study initially included 39,103 female participants from the screening arm. We excluded 1,095 women who did not have an eligible baseline questionnaire after informed consent, 4,833 women who had undergone bilateral oophorectomies, and 2,831 women who did not receive any CA125 test. This left 30,344 women who were initially eligible for this study. After further excluding 1,842 participants who had only one CA125 test before the end of screening and 46 women who had only one CA125 test prior to the diagnosis of OC, 28,456 women with at least two CA125 tests were finally included. A flowchart of participant selection was shown in Supplementary Figure S1 . The definition of a positive OC screening result has been described in previous studies [ 5 ]. Screening results were communicated to participants and their healthcare providers by mail, typically within three weeks. Participants with positive screening results are encouraged to undergo diagnostic assessments. Cancer cases and deaths were identified through the Annual Study Update (ASU) questionnaire and supplemented by an annual search of National Death Index. When the ASU indicated a probable death, the PLCO Center obtained and reviewed the death certificate, and the Data Compilation Center coded the cause of death using ICD-9. The death review committee then determined whether ovarian cancer was the cause of death [ 5 , 19 ]. Active follow-up for cancer diagnoses continued until December 2009, and extended follow-up data for deaths were updated through 2018. Therefore, the primary outcomes were censored at the date of the OC diagnosis (for OC incidence), death, loss of follow-up, or the end of the follow-up, whichever occurred first. To determine the optimal screening cut-off value of CA125 in OC, time-dependent receiver operating characteristic curve (tdROC) and the area under the curve (tdAUC) were calculated using a COX regression model. Bootstrap resampling with 2000 iterations was employed to internally validate the stability of these results, with the median cut-off value of CA125 and its 95% confidence intervals (CIs). Based on the optimal screening CA125 cut-off value, four CA125 trajectories were defined: stable negative CA125 (CA125 SN ) with CA125 below the cut-off in both first-round (FR) and last-round (LR) tests; loss of positive CA125 (CA125 LP ) with FR positive and LR negative CA125; stable positive CA125 (CA125 SP ) with positive CA125 in both FR and LR tests; and gain of positive CA125 (CA125 GP ) with FR negative and LR positive CA125. Chi-square tests were used to assess significant differences in baseline characteristics across CA125 trajectories. Among women with elevated CA125 within CA125 SP and CA125 GP categories, eight progression indicators were calculated based on baseline CA125 (i.e., FR CA125), CA125 increment, CA125 maximum, and time interval between CA125 tests. These indicators include: absolute increment (AbsInc) defined as the difference between FR and LR CA125 tests; maximum absolute increment (MaxAbsInc) as the difference between FR and the maximum CA125 tests; relative increment (RelInc) as AbsInc divided by FR CA125; maximum relative increment (MaxRelInc) as MaxAbsInc divided by FR CA125; absolute velocity (AbsVel) as AbsInc divided by the time between FR and LR CA125 tests; maximum absolute velocity (MaxAbsVel) as MaxAbsInc divided by the time between FR and maximum CA125 tests; relative velocity (RelVel) as AbsVel divided by FR CA125, and maximum relative velocity (MaxRelVel) as MaxAbsVel divided by FR CA125. To identify the optimal trajectory-specific progression indicators, tdAUCs for these eight progression indicators were calculated using COX regression models and compared pairwise with Delong’s test. The indicators with the highest tdAUC were selected as the optimal trajectory-specific progression indicators for CA125 SP and CA125 GP . Time-dependent receiver operating characteristic curves (tdROCs) were used to determine the optimal cut-off values for these indicators, and bootstrap resampling was performed to internally validate the stability of these cut-off values. Based on these cut-off values, trajectory-specific progressions were further reclassified into positive and negative progressions. The risks of OC incidence and mortality across different CA125 trajectories and trajectory-specific progressions were analyzed using Kaplan-Meier survival curves and compared with the log-rank test. Univariate COX regression models evaluated crude associations between CA125 trajectories, trajectory-specific progression indicators, and OC risk. To investigate the independent associations of CA125 trajectories, trajectory-specific progression indicators, and OC risk, Multivariable COX regression models, adjusting for factors associated with CA125 trajectories as detailed in Supplementary Table S1 , were used to investigate independent associations. These factors included age at recruitment ( 30 kg/m 2 ), smoking status (never, current, previous), previous oral contraceptives(none, ≤ 5 years, > 5 years), previous hormone replacement therapy (none, ≤ 5 years, > 5 years), age at menopause (< 55 years, ≥ 55 years), previous hysterectomy(none, yes), live births (0, 1–2, ≥ 3 times), and time intervals between CA125 tests(1 year, 2 years, 3 years, 4 years, 5 years). Missing data were categorized as an independent group. Associations were measured with hazard ratios and 95% confidence intervals [HR (95% CIs)]. Based on the CA125 trajectories, trajectory-specific progression indicators, and TVU, three optimized joint screening strategies we proposed alongside two traditional strategies. The traditional strategy with TVU screening alone served as the reference. The other strategies explored whether adding CA125 test to TVU screening improved the OC detection accuracy. For the traditional strategy two, a positive screen was defined as any positive TVU and CA125 above diagnostic criteria. In optimized strategy one, a positive screen was defined as any positive TVU and CA125 above the optimal screening cut-off value. To reduce potential false positives associated with CA125 levels below diagnostic criteria, optimized strategy two excluded definite regression of CA125 (namely CA125 LP ). Further reducing potential false positive and focusing on detecting potentially progressive or lethal cancer, optimized strategy three further excluded negative progression in either CA125 SP or CA125 GP . Screening performances metrics including sensitivity, specificity, positivity, Youden index, positive prediction value (PPV), and negative prediction value (NPV) as well as their corresponding confidence intervals were calculated for each screening strategy. When comparing sensitivity (or specificity) between two screening methods, only individuals with cases (or non-cases) were considered. McNemar’s test was employed in this context to evaluate whether the number of discordant pairs (where one test is positive and the other is negative) for detecting cases (or non-cases) is significantly differs from what would be expected by chance. In contrast, when comparing positive rates (defined as the proportion of individuals who receive a positive results, including true positives and negative positives, out of the total number of tests conducted) between two screening methods, all individuals (both cases and non-cases) are included. The Pearson chi-square test was used to determine whether the proportion of individuals receiving a positive result from one screening method significantly differs from that of another screening method. All data analyses were conducted using R (version R 4.2.2) and SPSS software (version R26.0.0.0). Statistical significance was defined as a P value of less than 0.05.

Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1