{"paper_id":"8b9277de-7f54-484e-b85d-37155fe819b0","body_text":"1\nIdentifying the best diagnostic test for Ovarian cancer in premenopausal women with \nnon-specific symptoms – results from the ROCkeTS prospective, multicentre, cohort \nstudy.  \n \nSudha Sundar \n1,2, Ridhi Agarwal 3, 17, Katie Scandrett 3,17, Clare Davenport3, 17, Ben V an \nCalster10,18, Susanne Johnson4, Partha Sengupta 5, Radhika Selvi-Vikram6, Fong Lien Kwong \n1,3, Sue Mallett7, Caroline Rick8, Sean Kehoe9, Dirk Timmerman10,11, Tom Bourne12, Hilary \nStobart12, Richard D Neal13, Usha Menon14, Aleksandra Gentry-Maharaj14, 15, Lauren \nSturdy16, Ryan Ottridge16, Jonathan J Deeks3,17 for ROCkeTS collaborators \n \n1PanBirmingham Gynaecological Cancer Centre, Sandwell and West Birmingham Hospitals NHS trust, \nBirmingham, UK \n2Department of Cancer and Genomic Sciences, University of Birmingham, Birmingham, UK \n3Department of Applied Health Science, University of Birmingham, Birmingham, UK \n4- Southampton University Hospitals, NHS trust \n5 Durham and Darlington NHS trust \n6 West Hertfordshire Hospitals NHS trust \n7Centre for Medical Imaging, University College London, London, UK \n8 University of Nottingham \n9 St Peter's College, University of Oxford, Oxford, UK \n10Department of Development and Regeneration, KU Leuven, Leuven, Belgium \n11Department of Obstetrics and Gynecology, University Hospitals KU Leuven, Leuven, Belgium \n12Faculty of Medicine, Department of Metabolism, Digestion and Reproduction, Imperial College London, \nLondon, UK \n12Patient Representative, Birmingham, UK \n13University of Exeter Medical School, University of Exeter, Exeter, UK \n14Department of W omen’s Cancer, Elizabeth Garrett Anderson Institute for W omen’ s Health, University College \nLondon, London, UK \n15MRC Clinical Trials Unit, Institute of Clinical Trials and Methodology, University College London, London, \nUK \n16 Birmingham Clinical Trials unit, University of Birmingham, UK \n17 NIHR Birmingham Biomedical Research Centre, University Hospitals Birmingham NHS Foundation Trust \nand University of Birmingham, Birmingham, UK  \n18 Leuven Unit for Health Technology Assessment Research (LUHTAR), KU Leuven, Leuven, Belgium \n \n \nAuthor for correspondence  \n \nProf Sudha Sundar  \nProfessor in Gynaecological Cancer and  \nHon NHS Consultant in Gynaecological Oncology  \nUniversity of Birmingham and  \nPan Birmingham gynaecological cancer centre, Sandwell and West Birmingham Hospitals \nNHS Trust,  \nUnited Kingdom \ns.s.sundar@bham.ac.uk \n \n \n \n \n \n \n \n \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \nNOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice.\n\n \n \n2\nAbstract  \nObjective \nDiagnosing ovarian cancer in premenopausal women is challenging due to the rarity of \ncancer and the ubiquity of symptoms, ovarian cysts on ultrasound, and raised serum CA125 \ntumour marker levels. We investigated the accuracy of risk prediction models and scores for \ndiagnosing ovarian cancer in premenopausal women presenting to secondary care with \nsymptoms and abnormal tests.  \n \nMethods \nA cohort of premenopausal women presenting with non-specific symptoms, and raised \nCA125 or abnormal imaging, were prospectively recruited in 23 hospitals in the UK between \nJune 2015 and March 2023, predominantly referred through the NHS urgent suspected cancer \npathway from primary to secondary care. A head-to-head comparison of the accuracy of the \nsix risk prediction models and scores was conducted using donated blood and ultrasound \nscans performed by NHS staff trained in the use of IOTA imaging terminology. Index tests (at \npre-stated thresholds) were: RMI1 (200, 250); ROMA (7.4%, 11.4%, 12.5%, 13.1%); IOTA \nADNEX (3%, 10%); IOTA SRRisk (3%, 10%); IOTA simple rules; and CA125 (87 IU/ml).  \nParticipants were classified as having primary invasive ovarian cancer versus having benign \nor normal pathology according to the reference standard determined from surgical specimens, \nbiopsies or cytology, by histology if undertaken, or else by 12-month follow-up.  After June \n2018, because of COVID restrictions and concerns about sample size, ongoing recruitment \nwas restricted to only women undergoing surgery within 3 months of presentation (a selected \ngroup in whom ovarian cancer was more likely).  \n \nResults  \nOf 1,211 premenopausal recruited women 88 were diagnosed with primary OC, 857 in the \npre-June 2018 cohort (prevalence of 5.7% (49 /857)) and 354 in the post-June 2018 cohort \n11.0% (39/354).  \n \nFor the diagnosis of primary ovarian cancer, (n=799 women after exclusion of n=58 other \ndiagnoses), RMI1 at the 250 threshold had a sensitivity of 42.6%, 95% confidence interval \n28.3 to 57.8, and specificity of 96.5%, 94.7 to 97.8. Compared to RMI1/250, CA125 and all \nother models had higher sensitivity (CA125: 55.1%, 40.2 to 69.3, p=0.06; ROMA/11.4%: \n79.2%,  65.0 to 89.5, p<0.0001; IOTA ADNEX/10%: 89.1%, 76.4 to 96.4, p<0.0001; IOTA \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n3\nSRRisk/10%: 83.0%, 69.2 to 92.4, p<0.0001; IOTA simple rules: 75.0%, 56.6 to 88.5, \np=0.01) and lower specificity (CA125: 89.0%, 86.5 to 91.2, p<0.0001; ROMA/11.4%: \n73.1%, 69.6 to 76.3, p<0.0001; IOTA ADNEX/10%: 75.1%, 71.4 to 78.6, p<0.0001; IOTA \nSRRisk/10%: 76.0%, 72.4 to 79.3, p<0.0001; IOTA simple rules 95.2%, 93.0 to 96.9, \np=0.06).  IOTA simple rules have inconclusive results in 120/799 of the participants. Analysis \nof the complete cohort (n=1,211) including the 354 premenopausal women with a higher \nlikelihood of ovarian cancer, yielded similar results.  \n \nConclusions  \nCompared to RMI 250, the current test used in NHS secondary care to triage women to \ntertiary care, most tests improve sensitivity but reduce specificity. Ultrasound triage with the \nIOTA ADNEX model at 10% in secondary care demonstrated the highest sensitivity gain with \na comparable decline in specificity to other comparator tests. Ultrasound with the IOTA \nADNEX model at 10% should be considered the new standard of care triage test for \npremenopausal women in secondary care; implementation in practice should incorporate staff \ntraining and quality assurance.   \n \n \nTrial registration – ROCkeTS is registered ISRCTN17160843 \n \nContributor and guarantor information \nSS, CD, SM, JD conceptualised and designed the study. SS, SJ, PS, RS-V recruited to the \nstudy with collaborators with CR, RO and LS coordinating the study, SM, JD, KS and RA \nanalysed results from the study,  DT and TB conducted ultrasound QA, training, BVC \nprovided insight into analysis of ultrasound models, SK, RN, UM and A G-M provided input \ninto study design and conduct. HS provided patient perspectives throughout the study from \nthe grant application, study conduct and interpretation of results. All authors reviewed the \nresults and the manuscript.  \nJD, RA and KS along with LS and RO have directly accessed and verified the underlying \ndata reported in the manuscript. \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n4\nAll authors confirm that they had full access to all the data in the study and accept \nresponsibility to submit for publication \nSS is guarantor of paper. The guarantor accepts full responsibility for the work and/or the \nconduct of the study, had access to the data, and controlled the decision to publish.  \n\"The corresponding author attests that all listed authors meet authorship criteria and that no \nothers meeting the criteria have been omitted.\" \n \nCopyright/license for publication \nThe Corresponding Author has the right to grant on behalf of all authors and does grant on \nbehalf of all authors, a worldwide licence to the Publishers and its licensees in perpetuity, in \nall forms, formats and media (whether known now or created in the future), to i) publish, \nreproduce, distribute, display and store the Contribution, ii) translate the Contribution into \nother languages, create adaptations, reprints, include within collections and create summaries, \nextracts and/or, abstracts of the Contribution, iii) create any other derivative work(s) based on \nthe Contribution, iv) to exploit all subsidiary rights in the Contribution, v) the inclusion of \nelectronic links from the Contribution to third party material where-ever it may be located; \nand, vi) licence any third party to do any or all of the above.\" \n \nPatient consent (if applicable) – not applicable \n \nCompeting interests declaration \nSS reports a research grant from AoA diagnostics for work with samples collected in this \nstudy but not reported within this manuscript. SS reports honoraria from Astra Zeneca, \nMercke and GSK and consultancy from GSK and Immunogen, all unrelated to this work.  \nTBo reports grants, personal fees, and travel support from Samsung Medison; travel support \nfrom Roche Diagnostics; and personal fees from GE Healthcare; all outside the submitted \nwork. BVC and DT report consultancy work done by KU Leuven to help implementing and \ntesting the IOTA ADNEX model in ultrasound machines by Samsung Medison and GE \nHealthcare, outside the submitted work. Profs Timmerman and Bourne declare that they are \ncofounders of a KU Leuven spinout company, Gynaia incorporated in May 2025 (i.e post \nsubmission of ROCkeTS manuscript in Dec 2024). \n \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n5\n \nUM stock ownership awarded by University College London (UCL) until October 2021 in \nAbcodia. UM and AGM report research collaboration contracts with QIMR Berghofer \nMedical Research Institute, iLOF (intelligent Lab on Fiber), RNA Guardian, Micronoma, \nMercy Bioanalytics, Synteny Biotechnology. UM reports research support grants paid to the \ninstitution from CleoDx related to early detection of cancer especially ovarian cancer. UM \ndeclares membership of the Research Advisory Panel, Yorkshire Cancer Research (UK) \nand honorarium for membership of Tina’s Wish Scientific Advisory Board (USA). UM holds \npatent number EP10178345.4 for Breast Cancer Diagnostics.   \nSK reports honorary role as Ovacome charity trustee. DT, TBo and BVC are IOTA steering \ngroup members and developed the IOTA models. \nAll other authors declare no competing interests.  \nData availability   \nThe dataset generated including deidentified patient data and samples analysed during the \nstudy, along with additional material such as protocol, statistical analysis plan is available at \nBirmingham Clinical Trials Unit, University of Birmingham after date of publication. The \ndataset is not publicly available but maybe obtained on request to SS, review by Project \noversight group, NIHR, ethics approval and after fulfilling all data transfer requirements \nPatient and Public Involvement   \nThe study was supported by a patient co-applicant who played a crucial role throughout the \nresearch process. HS provided regular input at key stages of the study, beginning with the \napplication for funding, study design, recruitment and continuing through to the \ndissemination of findings.  Patient advocates from Target Ovarian cancer were involved as \npartners with the research team to shape the study's design, develop and review informational \nmaterials for the study, and assess the burden of questionnaires from the patient’s \nperspective. \n \nAt the conclusion of the study, HS and Target Ovarian cancer charity contributed feedback \non the findings and provided valuable input into the interpretation of results and outcome \nmeasures, ensuring that the findings were meaningful from a patient-centred perspective. \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n6\nThey also contributed to the dissemination plan with an online presentation to lay audiences, \nreview of press release of previously published manuscripts from the ROCkeTS study.  \n \nTransparency declaration  \nThe lead author (the manuscript's guarantor) affirms that the manuscript is an honest, \naccurate, and transparent account of the study being reported; that no important aspects of the \nstudy have been omitted; and that any discrepancies from the study as originally planned \n(and, if relevant, registered) have been explained. \n \nRole of funding source  \nROCKeTS was funded by National Institute of Health and Care research (NIHR) Health \nTechnology Assessment 13/13/01. The funder stipulated study design and choice of \ncomparator in a commissioned call. Funder approved protocol change after discussion in \n2018 when interim analysis showed lower than expected prevalence of Ovarian cancer. \nFunder had no role in conduct, analysis, decision to submit or interpretation of results.  \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n7\nSection 1: What is already known on this topic \n• Diagnosing ovarian cancer in premenopausal women is challenging - OC is rare, \nwhile symptoms, elevated CA125, and physiological ovarian cysts are common. \nCurrent standard of care risk prediction model used to triage women with ovarian \ncysts into low or high risk of Ovarian cancer is the Risk of Malignancy index. Several \nalternate risk prediction models and scores show promise.  \n• We searched OVID MEDLINE, OVID EMBASE, and Cochrane Library (to 14 July \n2025) using terms: ROMA, IOTA ADNEX, ORADS, IOTA simple rules, and RMI to \nidentify optimal risk prediction models for OC in premenopausal women. \n• Multiple studies were identified, but no head-to-head prospective comparisons of all \ntests exist. Studies were predominantly conducted in high-prevalence settings with \nexpert ultrasound operators, limiting generalizability to non-specialist, primary care, \nor community settings. \nSection 2: What this study adds \n• We conducted a prospective head-to-head test accuracy study of common risk \nprediction models in premenopausal women presenting to secondary care with \nsymptoms, abnormal CA125, and abnormal ultrasound.  \n• Compared to published literature, our cohort was more representative of real-world \npopulations with lower OC prevalence (5.7%). Ultrasound was performed primarily \nby NHS sonographers, enhancing applicability to practice.  \n• RMI at 250 shows poor sensitivity (42.6%) but high specificity (96.5%) in \npremenopausal women and requires replacement. Alternative tests improved \nsensitivity at the expense of reduced specificity compared to RMI 250. \n• IOTA ADNEX at 10% threshold demonstrates the highest sensitivity (89%) with \nspecificity (75%) comparable to other evaluated tests and is recommended for \npractice. \n \n \n \n \n \n \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n8\nIntroduction  \n \nOvarian cancer (OC) is a challenging disease to diagnose, with patients typically visiting \nGeneral Practitioners (GPs) or primary care multiple times before testing is initiated. 1,2 \nCurrently, most women are diagnosed at advanced stages due to non-specific symptoms and \nsuboptimal diagnostic pathways. Diagnosis is especially challenging in premenopausal \nwomen due to low OC prevalence (approximately 93,000 women diagnosed under 49 years \nglobally, (age-standardized rate of 2.2-3.6/100,000\n3), ubiquitous symptoms, non-specifically \nelevated CA125, for example, during menses and physiological ovarian cysts on ultrasound. \nIn England and Wales, 40% of women with OC are admitted as an emergency 4 weeks prior \nto diagnosis and are five times more likely to die within six months than women referred \nthrough urgent suspected cancer pathways. OC survival in the UK is significantly lower than \nin other western countries. \n \nThe National Institute for Health and Care Excellence (NICE) guidelines recommend \nsequential testing using serum CA125 and pelvic ultrasound for women presenting to their \nGP with symptoms such as persistent abdominal distension, feeling full, pelvic pain, \nincreased urinary urgency, unexplained weight loss, fatigue, or changes in bowel habit.\n4 \nWomen with elevated CA125 or abnormal ultrasound findings are referred to gynaecologists \nin secondary care hospitals through the urgent suspected cancer pathway in the UK NHS: \npatients referred to hospital receive a cancer/non-cancer diagnosis within 28 days of referral, \nand patients diagnosed with cancer receive first treatment within 62 days of referral.\n5,6  \n \nPoor performance of current diagnostic testing contributes to the challenge of timely, accurate \ndiagnosis - CA125 identifies only 50% of early-stage OC and can be elevated in other benign \nconditions, while ultrasound in primary care lacks standardization or quality control and is \nassociated with long waiting times.\n7,8  NICE guidance recommends the Risk of Malignancy \nIndex (RMI) algorithm to triage women referred with suspected OC in hospital, combining \nage and menopausal status to generate a risk score.\n4,9 W omen w ith an RMI score >250 are \nreferred to tertiary care hospitals (gynaecological cancer centres) to be operated on by \ngynaecological cancer surgeons, while those with an RMI < 250 are managed in the referring \nsecondary care hospital with surgery or surveillance by gynaecologists. The current pathway \nis summarised in Figure 1. Rates of surgery and additional imaging (MRI/CT) in referred \nwomen are high with the current pathway.\n10 \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n9\n \nAccurate triage is important - triaging a woman as high risk for OC preoperatively enables \nappropriate surgery at first attempt in a tertiary, specialist cancer centre, which improves \nsurvival.\n4,11–13 Improving diagnostic pathways for premenopausal women is a key unmet \nneed. Alternatives (to RMI 250) risk prediction models endorsed variably by professional \nsocieties include the Risk of Malignancy Algorithm (ROMA) (combining He4 and CA125 \nbiomarkers, stratified by menopausal status), and several ultrasound based models(the \nOvarian-Adnexal Reporting and Data system (ORADS); the Assessment of Different \nNEoplasias in the adneXa (ADNEX); and the International Ovarian Tumour Analysis (IOTA) \nIOTA Simple rules).\n14–20  \n \nA 2022 Cochrane systematic review highlighted the absence of high-quality, head-to-head \ncomparative test accuracy studies applicable to a primary care referred population with \nincluded studies conducted in high-prevalence settings (16-27%).\n21 The Cochrane review also \ndemonstrated variation in performance of risk prediction models between pre- and \npostmenopausal women, reflecting differences in disease prevalence and spectrum.\n21  \nAdditionally, the trade-offs between identifying true positives (enabling early diagnosis and \nbetter survival) versus limiting false positives (reducing unnecessary referral, anxiety and \nsurgery) are different for pre compared to postmenopausal women, where fertility \npreservation and ovarian function are critical considerations.  \n \nThe ROCkeTS study aimed to improve on existing evidence by identifying the best \ndiagnostic test for women referred to secondary care hospitals with symptoms and abnormal \nCA125 and/ or ultrasound. We evaluated the accuracy of alternative risk prediction models: \nIOTA simple rules, IOTA SRRisk model, ROMA, ORADS, IOTA ADNEX, compared to the \nRMI 250 standard of care triage test in the UK (Box 1 and Supplementary Appendix C) \nagainst a reference standard of histology or follow-up for referral to tertiary care for \nsurgery/biopsy. Results for postmenopausal women have been previously reported.\n22 This \nmanuscript presents results for premenopausal women.   \n \n \n \n \n \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n10\n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n11\nMethods  \nOur report adheres to the STARD and TRIPOD checklists (Supplementary Tables 10-11).23,24 \nThe trial protocol is on https://www.birmingham.ac.uk/research/bctu/trials/pd/rockets and has \npreviously been published 25  Methods for ROCkeTS has been previously described (we \nreproduce this briefly below) and in the statistical analysis plan (Supplementary Appendix \nD). ROCkeTS received ethical approval from NHS West Midlands REC (14/WM/1241) and \nis registered on the controlled trials website (ISRCTN17160843). All participants provided \ninformed consent prior to participation in the study.  \nParticipants \nThis prospective cohort study consecutively recruited newly presenting women aged 16-90 \nyears who were referred to hospitals within the UK. Research nurses from the UK national \ncancer collaborative research network screened patients attending hospital clinics referred \nfrom primary care through urgent suspected cancer pathways, ultrasound clinics, \ngynaecology clinics or presenting through emergency admissions. To be eligible, women \nneeded to have  symptoms consistent with NICE guidance for OC (including but not \nrestricted to persistent or frequent abdominal distension, feeling full (early satiety) and/or loss \nof appetite, pelvic or abdominal pain, increased urinary urgency and/or frequency\n4) as well as \nelevated CA125 levels or abnormal ultrasound findings. At recruitment, participants self-\nreported sex and ethnicity and completed a structured questionnaire to obtain medical and \ngynaecological history. Exclusion criteria comprised pregnancy, existing non-ovarian \nmalignancy, and previous ovarian malignancy.  \nParticipants were categorised as pre- or postmenopausal status at recruitment by their age \n(<50, 51+ years) and whether they had periods in the last 12 months. Post recruitment, \nperimenopausal women were re-categorised based on a self-reported history of vaginal \nbleeding to allow analysis with existing risk prediction models that incorporate different \nthresholds or covariates based on menopausal status.  \nThe study protocol underwent significant evolution from initial publication.\n25 The ROCkeTS \npremenopausal study initially recruited consecutive eligible women who were scheduled for \nsurgical intervention within 3 months of referral to hospital or who were being managed \nthrough conservative approaches, including surveillance or discharge as clinically indicated, \nwith planned questionnaire-based assessment of wellbeing at 12 months.  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n12\nAt an interim analysis, OC prevalence was observed to be 3%, sufficient for estimating \nspecificity but inadequate to estimate comparisons of sensitivity between tests and models. \nConsequently, adaptations to the protocol were introduced in March 2018, excluding women \nwith simple ovarian cysts <5cm and normal CA125 levels due to their extremely low cancer \nrisk, and from June 2018 recruitment focused solely on newly presenting premenopausal \nwomen scheduled for surgery within 3 months. From June 2018, IOTA ultrasound scan was \ndeemed optional, acknowledging both the scheduling challenges for women undergoing \nsurgery for suspected cancer and ethical concerns about requiring additional hospital visits \nduring the COVID-19 pandemic (2020-2022). \nThe study cohort comprised two groups: the pre-protocol change cohort (labelled Cohort 1) \nrecruited up until June 2018, which includes both pre-surgical and conservatively managed \npatients, and the post-protocol change cohort, post June 2018, which comprises only pre-\nsurgical patients (labelled Cohort 2). A sensitivity analysis was undertaken to account for \npotential changes in spectrum, with the pre-protocol change cohort regarded as most likely to \nbe representative and used in the primary analysis.  \nThree patients were deemed ineligible after recruitment – one due to a previously unknown \npregnancy, one due to being diagnosed with non-ovarian malignancy prior to recruitment and \none due to planned surgery not going ahead. No screen eligible and willing patients were \nexcluded. The current NHS standard of care triage test, RMI 250, was used to manage \npatients.  \n \nIndex tests \nAll patients completed a symptom questionnaire, donated a blood sample and underwent \ntransabdominal and a transvaginal USS scan. Serum samples were collected, processed, and \nstored according to predefined standard operating procedures detailed in a laboratory manual \ndistributed to participating sites, consistent with consensus guidance from the Early Detection \nResearch Network. (Appendix D). \n26 Samples were transported and stored at -80°C until \nanalysis at NHS South Tyne and Wear Pathology Services laboratories. For consistent \nanalysis, samples were thawed in batches and tested for CA125 and HE4 using Roche Cobas \ne802 modules. HE4 and CA125 measurements employed electrochemiluminescence \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n13\nimmunoassay (ECLIA) technology, adhering to manufacturer recommendations. Roche \nElecsys assay kits were obtained from Roche Diagnostics. \n \nThe following index tests were evaluated (Box 1, \nSupplementary Appendix C) :  \n• ROMA, (a combination of CA125 and He4 tumour markers) at the manufacturer \nrecommended threshold of 11.4% for premenopausal women and at previously reported \nthresholds of: 7.4%, 12.5%, 13.1%.27  \n• The Risk of Malignancy Index 1 (RMI 1), (a combination of CA125 and limited \nultrasound features), at thresholds of 200 and 250. The 250 threshold was chosen as the \ncomparator test, representing the current standard of care diagnostic approach for triaging \npatients to gynaecological cancer centres within the UK National Health Service.\n4,17  \n• Serum CA125 measurement at a threshold of 87 IU/ml was selected based on its \nassociation with a positive predictive value of 3% in primary care as detailed by Funston \net al.28  \n• USS based tests: Three tests developed by the IOTA consortium were evaluated, two \nmodels IOTA ADNEX and the IOTA simple rules risk (SRRISK) model at a primary \nthresholds of 10% and secondary thresholds of 3% and one classifier IOTA simple \nrules.\n16,19,29–31. IOTA ADNEX is now considered a medical device and is manufactured by \nGynaia.32  \n• ORADS was evaluated in a post-hoc analysis using IOTA variables from the ROCkeTS \nultrasound case report form retrospectively mapped to the ORADS lexicon 1-3 versus 4-5 \nusing previously described methodological approaches\n19,33 for this analysis. \nWe define IOTA ultrasound scan as one where IOTA terminology has been used to describe \nthe ultrasound findings.34 IOTA terminology is a set of terms, definitions and measurements \nused to precisely describe ultrasound features seen in adnexal masses. 34 Ultrasound \nexaminations in ROCkeTS were conducted by sonographers who underwent comprehensive \ntraining in IOTA terminology, including one-day in-person and online instruction, followed \nby formal examination. A mandatory quality assessment process was implemented, with a \nsample of ultrasound images and reports centrally reviewed by the IOTA team of ultrasound \nexperts has been described previously.\n22 No prior minimum ultrasound experience \nrequirement was stipulated. Scans were primarily performed by level II (non-medical) \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n14\nsonographers which mirrors real-world clinical practice, where pelvic ultrasounds are \nperformed by sonographers with varying levels of experience.  \n \nAll diagnostic tests and subsequent surgical interventions or biopsies were required to be \ncompleted within three months of patient recruitment. All tests were conducted blinded to the \nreference standard.  \n \nReference standard \nFor Cohort 1, the reference standard was histology of surgical specimens, biopsies or \ncytology or 12-month surveillance for women who did not undergo surgery. In Cohort 2, \nstudy participation ended at surgery/biopsy/cytology, which served as the reference standard. \n \nPathology data were sourced from specialist gynecological pathology reports from the 40 \ncancer centres in the UK where women undergoing surgery for suspected OC are discussed at \nspecialist gynaecological oncology multidisciplinary team meetings.  \n \nFor participants under surveillance, 12-month well-being was ascertained through both \npatient self-completed postal questionnaires and research nurse-completed questionnaires \nutilizing hospital records from a clinic visit or by contacting patients by telephone. No \ndiagnosis of cancer or histology results were based on self-report alone. Both information \nsources were cross-referenced to identify any cancer diagnoses occurring within 12 months of \nstudy recruitment.  \n \nReference standard results were not available to sonographers or the research team prior to \ntrial entry, although clinical information was available as this is considered usual clinical \npractice.   \n \nOutcomes \nThe primary outcome focused on the diagnostic accuracy of index tests for identifying OC. \nThis was defined as a binary outcome: primary invasive malignant neoplasms diagnosed \nthrough surgical or biopsy histology, versus benign, normal, or surveillance findings. The \ndefinition of primary invasive OC included ovarian, fallopian tube, and primary peritoneal \ncancers. \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n15\n \nThe secondary outcome included a broader spectrum of malignancies encompassing primary \ninvasive cancers, secondary malignant neoplasms metastatic to the ovary, borderline \nneoplasms, and neoplasms of uncertain or unknown behaviour diagnosed through surgery, \nbiopsy, or cytology, versus benign, normal on follow-up findings. Analysis of the secondary \noutcome was also performed with borderline tumours grouped with benign/ normal follow-up \nfindings rather than with malignancies.  \n \nStatistical analysis.  \nClinical study data was entered in structured case report forms onto an electronic study \nmanagement platform hosted by Medscinet, https://rockets.medscinet.com/. Data was entered \nby research nurses at sites. Data cleaning and addressing data queries was performed by the \nBirmingham Clinical Trials unit with trial statisticians. The statistical analysis plan is \nincluded in the supplementary materials. Diagnostic accuracy was assessed using sensitivity, \nspecificity and the positive and negative predictive values (PPV and NPV). Risk prediction \ntools were dichotomised at different thresholds. The difference in sensitivity and specificity \n(and their corresponding 95% confidence intervals) comparing tests was assessed using the \nexact McNemar’s test with asymptotic confidence intervals.  Multiple testing was accounted \nfor by use of the Bonferroni correction (11 pairwise comparisons, p=0.0045). (21, 22). No \nsingle measure of accuracy was designated as primary endpoint a priori. This approach was \nchosen to fully evaluate the trade-offs inherent in performance of diagnostic tests. \n \nGlobal accuracy performance was further assessed in terms of discrimination, using a c-index \nand receiver operating characteristic (ROC) plot, and calibration, using calibration plots and  \ncalibration slope.  We used the ‘pmcalplot’ command in Stata to generate the calibration \nplots.\n35  \n \nWhere index tests produced inconclusive results (IOTA Simple rules), we opted not to \nclassify inconclusive results as positive, as this could have led to an overestimation of test \nperformance for sensitivity estimates.  \n \nWomen with missing index test data were excluded from both primary and secondary \nanalyses. Participants with missing or inconclusive reference standard results were excluded \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n16\nfrom the primary outcome definition of presence or absence of OC but included in the \nsecondary outcome to make the best use of participant data.  \n \nA sensitivity analysis was conducted for both the primary and secondary defin itions of OC \nwhere missing index test results were imputed using the multiple imputation by chained \nequations (MICE) for predictors of index test combinations by replacing missing values with \nplausible values based on the distribution of the observed data.\n36 Multiple imputation was \nperformed using the ‘mi’ package in Stata 17.37 \n \nInitial analysis was performed for primary and secondary outcomes in the pre-protocol \nchange cohort, in a combined cohort grouping Cohort 1 and Cohort 2 together.   \n \nSample size  \n \nThe original sample size was based on local audit data (unpublished) to estimate the \nperformance of RMI in premenopausal women, as previously published systematic reviews \ndid not provide separate estimates for pre- and postmenopausal women. Based on the \nperformance of RMI having a sensitivity of 72% and specificity of 46%, the study was \ndesigned to detect increases in sensitivity of 10% and in specificity of 10%, assuming a \nprevalence of OC in premenopausal women referred to secondary care of 10%.\n38 A sample \nsize of 1000 would provide 100 OC events in which to build new models combining \nsymptom and test data (adequate events to model 10 predictor variables) and will provide \n90% power to detect an increase in specificity of 8% (from 46% for RMI to 54%). With a \npredicted loss to follow-up of up to 5%, the final sample size required is 1050 women.\n39 \n \nIn an interim analysis in 2018, a much lower prevalence (3%) in premenopausal women was \nobserved than previously assumed (10%). Therefore, both sample size and inclusion criteria \nwere adapted to ensure that the study had adequate power to estimate the difference in \nsensitivities of the index tests by recruiting 105 women (accounting for 5% dropout) \nidentified as having OC. Following the change of the inclusion criteria, whilst a prevalence of \n14.9% was expected in women who undergo surgery, the revised sample size of 880 patients \nwas based on the assumption that only 11.9% (80%) actually undergo surgery.   \n \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n17\nResults  \n \nCharacteristics of study population  \n \nThe ROCkETS study recruited 2,268 eligible pre- and postmenopausal women referred to 23 \nUK hospitals between June 30, 2015 and March 23, 2023, and followed up to March 31, 2023   \n(Figure 2). Results of the 1,057 post-menopausal cohort have been reported separately \n22. \nHere, we report results for the cohort of 1,211 premenopausal participants, comprising 857 \nwomen recruited up to June 2018 under the initial protocol (Cohort 1), and 354 women \nrecruited under the adapted protocol restricted to women scheduled for surgery within 3 \nmonths (Cohort 2).  The majority of the participants in Cohort 1 were recruited from primary \nto secondary care through the urgent suspected cancer pathway (574/857;  67%).  The main \nanalysis focuses on the results of Cohort 1, as it represents ‘real-world’ practice and has low \nmissingness of IOTA USS. Findings of Cohort 2 are included in a sensitivity analysis. \n \nRecruited women in the first cohort had a median age of 44.1 (IQR 35.0-48.7) years. The \nmajority were of white ethnicity (n=732, 85.4%), had never smoked (n=476, 55.5%),  and \nwere not using contraception (n=502, 58.6%)  (Table 1).  Close to over-half of the women \n(n=444, 51.8%) had comorbidities: 137 (16.0%) with fibroids, 135 (15.8%) with \nendometriosis and 121 (14.1%) with irritable bowel syndrome. 161 (18.8%) had undergone \nprevious surgery: 44 (5.1%) hysterectomy, 82 (9.6%) cystectomy, 33 (3.9%) salpingectomy \nand 25 (2.9%) oophorectomy; and  217 (25.3%) reported a family history of cancer: 42 \n(4.9%) ovarian, 98 (11.4%) breast, 48 (5.6%) colon and 16 (1.9%) uterus (Table 1).  We did \nnot observe any systematic differences in patient characteristics between Cohort 1 and Cohort \n2.  \n \nPrevalence of ovarian cancer \n \nEighty-eight of the 1211 women (7.3%) were diagnosed with the primary outcome of ovarian \ncancer: 47 (53.4%) at FIGO Stage I, 6 (6.9%) at Stage II, 24 (27.2%) at Stage III and one \n(1.1%) at Stage IV (stage was missing for 10 ).  Five of the cases were diagnosed during the \n12-months follow-up, 83 by surgery or biopsy histology. Seventy-three cases had epithelial \nhistology types (n=73, 88.0%)  (Table 2).  \n \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n18\nIn the 857 women in Cohort 1, 49 (5.7%) of were diagnosed with OC and 58 (6.8%) were \nreported as ‘other’ and were excluded from estimation of accuracy for the primary outcome \n(13 had a missing primary outcome, 22 had borderline neoplasm, 7 had no histology, 10 had \nsecondary malignant neoplasm, 1 had a diagnostic category of ‘Other’ and 5 reported a \ndiagnosis of non-OC identified at 12-months follow up).   \n \nAccuracy of index tests \n \nOf the 799 women included in the analysis for the primary OC classification in Cohort 1, 581 \n(72.7%) had complete data for all index tests. Results were available for CA125 for 795 \n(99.5%), ROMA 750 (93.9%), RMI 1 for 672 (84.1%), IOTA SRRisk model 668 (83.6%), \nIOTA ADNEX 617 (77.2%) (Table 3, Supplementary Table 1).  For the IOTA simple rules, \nresults were only available for 553 (69.2%), as 126 had missing and 120 had inconclusive \nresults. \n \nThe pattern of values of sensitivity and specificity across the tests showed a threshold effect, \nwith tests with the lowest sensitivity having the highest specificity, and vice versa (Table 3).  \nRMI 1 at thresholds of 200 and 250 had the lowest sensitivities of 48.9% (95% CI: 34.1 to \n63.9) and 42.6% (95% CI: 28.3 to 57.8), respectively. Compared to RMI 1 at 250, CA125 and \nall other models had higher sensitivity (CA125 55.1%,, 95% CI  40.2 to 69.3, p=0.06; IOTA \nsimple rules: 75.0%, 95% CI 56.6 to 88.5, p=0.01; ROMA/11.4%: 79.2%, 95% CI 65.0 to \n89.5, p<0.0001; IOTA SRRisk/10%: 83.0%, 95% CI 69.2 to 92.4, p<0.0001; IOTA \nADNEX/10%: 89.1%, 95% CI 76.4 to 96.4, p<0.0001).   RMI 1 at thresholds of 200 and 250 \nhad the highest specificities of 95.4% (95% CI: 93.4 to 96.9) and 96.5% (95% CI: 94.7 to \n97.8), respectively. Compared to RMI 1 at 250, all other tests had lower specificity: IOTA \nsimple rules 95.2%, 93.0 to 96.9, p=0.06; CA125/87U/ml: 89.0%, 86.5 to 91.2, p<0.0001; \nIOTA SRRisk/10%: 76.0%, 72.4 to 79.3, p<0.0001; IOTA ADNEX/10%: 75.1%, 71.4 to 78.6, \np<0.0001; ROMA/11.4%: 73.1%, 69.6 to 76.3, p<0.0001).   \n \nIOTA ADNEX demonstrated the highest global accuracy with a C-index (area under the \ncurve, AUC) of 0.89 (95% CI: 0.83 to 0.96), followed by IOTA SRRisk 0.86 (0.80 to 0.93); \nRMI 1 0.85 (0.79 to 0.91); IOTA Simple rules 0.85 (0.77 to 0.93); ROMA 0.84 (0.77 to 0.92) \nand CA125 0.80 (0.72 to 0.87) (Table 3, Figure 3A). The calibration of the models for \nROMA, IOTA ADNEX and IOTA SRRisk model demonstrated under-prediction (Figure 3B).  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n19\n \nAll index tests had high negative predictive values, ranging from 95.7% (95% CI: 93.8 to \n97.2) for RMI1 at 250 to 98.9% (95% CI: 96.7 to 99.8) for IOTA ADNEX at 3%, driven by \nthe low prevalence. Positive predictive values were from 10.4% (95% CI: 7.6 to 13.7) for \nROMA at 7.4% to 49.0% (95% CI: 34.4 to 63.7) for IOTA simple rules (Table 3).  \n \nSecondary outcome – presence of any cancer  \n \nIn Cohort 1, 86/857 (10.0%) women were diagnosed with the secondary outcome definition \nof the presence of any cancer (Supplementary Table 2).  Estimates of sensitivity were lower \nthan for the primary outcome, with sensitivity values being lower by 4% for RMI 1 at 250 \nand 14% for IOTA ADNEX. RMI 1 of 250 reported the lowest sensitivity of 39.0% (95%: \n28.4 to 50.4)), whilst IOTA ADNEX reported the highest sensitivity of 75.3% (95%: 64.5 to \n84.2). Specificity values for the secondary outcome were very similar to those for the primary \noutcome. \n \nRMI 1, ROMA, IOTA ADNEX and IOTA sRisk reported very similar global accuracy C-\nindex values of 0.81, whilst values were lower for IOTA simple rules (0.76) and CA125 \n(0.73) (Supplementary Table 2, Supplementary Figure 1A).  The ROMA model demonstrated \ngood calibration, whilst the IOTA ADNEX and IOTA sRisk probabilities were \nunderpredictions (Supplementary Figure 1B). \n \nFor the secondary outcome, negative predictive values ranged from 91.8% (RMI 1 at a \nthreshold of 250) to 97.9% (ROMA at a threshold of 7.4%), lower than for the primary \noutcome.  Positive predictive values ranged from 17.0% (ROMA at a threshold of 7.4%) to \n56.9% (IOTA simple rules).  \n \nSensitivity analyses \nFor the primary outcome, the accuracy estimates were comparable in sensitivity analyses that \ncombined participants from the Combined Cohorts, with only small changes in point \nestimates but no changes in the ranking of performance (Supplementary Table 3, \nSupplementary Figures 2A and 2B), and in analyses using imputation for missing data. \n(Supplementary Table 4). \n \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n20\nSensitivity analyses of the secondary outcome showed consistent results from the Combined \nCohorts (Supplementary Table 5; Supplementary Figures 3A and 3B) and in analyses using \nimputation for missing data. (Supplementary Table 6). \nWe also analysed the diagnostic accuracy of the index tests, including borderline tumours \nwith benign tumours as normal in the Combined Cohorts analysis, findings were consistent \nwith main results (Supplementary Table 7).   \n \nResults according to quality assurance passed sonographers or high-volume recruiting centres \nwere also consistent with findings of the main primary and secondary analyses (data not \nsubmitted but available on request).  \n \nPost-hoc analysis of ORADs \n \nIn post-hoc analyses, in a Combined Cohorts analysis we compared the primary outcome \n(Supplementary Table 8) and secondary outcome (Supplementary Table 9) for ORADS at a \n10% threshold with RMI 1 at 250. ORADS demonstrated better sensitivity than RMI 1 250 \nbut with lower specificity for both outcomes. For the primary outcome, ORADS 4 had a \nsensitivity of 81.3% (95% CI: 69.5 to 89.9) and a specificity of 82.2% (95% CI: 79.5 to 84.6). \nSimilarly for the secondary outcome ORADS had a sensitivity of 68.9% (95% CI: 60.3 to \n76.7) with a specificity of 82.2% (95% CI: 79.5 to 84.6).  \n \n \n \n \n \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n21\nDiscussion  \n \nStatement of principal findings \n \nThe cancer prevalence in pre-menopausal women observed in ROCkeTS is low (5.7% in the \npre-protocol Cohort 1, 7% in Combined Cohorts). Only 1% of women referred less than 40 \nyears old for urgent suspected cancer are diagnosed with OC. \n40 Achieving a high sensitivity \n(minimising false negative (missed) diagnoses) is therefore important and challenging. \nConversely, reducing unnecessary investigations and surgery due to false positive diagnoses \n(maintaining specificity) is also important. We have previously demonstrated a high level of \nanxiety in women referred through urgent suspected cancer pathways, which persists at 12 \nmonths post-referral, despite a non-cancer diagnosis.\n40 The desire for fertility and ovarian \nfunction is also likely to be valuable to this group:  90 women (7.4%) were pursuing fertility, \nand only 345 (33.0%) were using contraception.  \n \nThe ROCkeTS study investigated diagnostic tests for OC in premenopausal women with \nsymptoms and abnormal CA125, ultrasound or both. The majority of women were referred \nfrom primary care through the urgent suspected cancer pathway to hospital clinics. Results \nshowed that the current UK standard triage test, RMI 1 at a threshold of 250, has a poor \nsensitivity of 42.6% (95% CI: 28.3 to 57.8), despite a good specificity of 96.5% (95% CI: \n94.7 to 97.8). IOTA ADNEX at thresholds of 10% and 3%, ROMA at 11.4% and 7.4%, and \nIOTA SR Risk models at 10% and 3% significantly improve on RMI 1's sensitivity but with a \nsignificant fall in specificity. Compared to RMI 1, IOTA ADNEX at 10% achieved the \nhighest sensitivity at 89.1% (95% CI: 76.4 to 96.4) with a relatively limited loss of \nspecificity, 75.1% (95% CI: 71.4 to 78.6).  \n \nThese results were consistent for the detection of primary OC, metastatic cancers, and \nborderline tumours combined (secondary outcome analysis), and across sensitivity analyses.  \n \nAmongst comparator tests, IOTA Simple Rules appeared to offer an improvement in \nsensitivity compared to RMI 1 to 75.0% (95% CI 56.6 to 88.5) whilst maintaining a high \nspecificity of 95.2% (95% CI 93.0 to 96.9). However, the test classified 120/799 women \n(15.0%) as inconclusive. Inconclusive results are seldom random; they typically represent \nhard-to-diagnose, borderline presentations with a higher underlying risk of malignancy. \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n22\nExcluding them introduces spectrum and missing-not-at-random biases, making sensitivity \nand specificity appear better than would be observed in practice.  \n \nFollowing the study, we were able to undertake a post-hoc analysis of ORADS at a 10% \nthreshold, showing a sensitivity of 81.3% (95% CI 69.5 to 89.9) and a specificity of 82.2% \n(95% CI 79.5 to 84.6%), with a lower gain in sensitivity than IOTA ADNEX but higher \nspecificity. Further investigation of the performance of ORADS in a prospective study is \nneeded.  \n \nComparison with existing literature \n \nA search on 14 th July 2025 showed no new prospective head-to-head comparative tests \naccuracy studies evaluating all relevant models for the diagnosis of OC. Our 2022 Cochrane \nsystematic review of risk prediction models for the diagnosis of OC demonstrated a similar \npattern of results in premenopausal women of a higher sensitivity but lower specificity of \nROMA (27 studies, 4463 participants) and IOTA ADNEX (4 studies, 1696 participants) \ncompared to RMI 1 (17 studies, 5233 participants).\n21 Differences in accuracy estimates  can \nbe explained by the highly selected participants in included studies (mean prevalence OC 16-\n27%) compared to a prevalence of 5.7% for the ROCkeTS main analysis. Results of \nROCkeTS in premenopausal women are also consistent with our previous report from \npostmenopausal women where we identified IOTA ADNEX ultrasound as the triage test \nachieving the highest gain in sensitivity over RMI triage with a reduction of specificity \ncomparable to other comparator tests.\n22 \n \nImplications for practice for clinicians and policymakers  \nResults need to be interpreted in the context of current practice. ROCkeTS recruited \npredominantly from urgent suspected cancer pathway referrals (67%), and 60% or primary \nOC diagnoses were stage I/II (pelvis confined) when the cancer is likely curable with \nstandard of care treatment.\n41 A risk prediction model with high sensitivity would improve \nsurvival, ensuring women with OC are triaged appropriately to receive surgery from trained \nspecialist gynaecological cancer surgeons whilst enabling women identified as ‘low risk’ to \nbe managed without surgery and with reassurance alone.\n13  \nData from ROCkeTS demonstrates high surgery rates in women triaged using RMI 1 (64.7%, \n551/857 patients) despite its relatively high specificity, at least in part because current \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n23\nguidance advocates surgery even in pre-menopausal women triaged as ‘low risk’ due to the \nknown low sensitivity of RMI in this group. Balancing gains in sensitivity to detect OC at an \nearly stage against reductions in specificity, and given the limitations of the IOTA simple \nrules, we recommend IOTA ADNEX at 10% to replace RMI1 at 250 as the UK standard of \ncare OC triage test in secondary care. The sensitivity and specificity of the IOTA ADNEX \nultrasound model was achieved using non-specialist, appropriately trained, certified and \nquality assured sonographers and IOTA ultrasound training resources have been established \nthrough ROCkeTS and are available for all NHS staff.\n42 \n \nIn ROCkeTS, to robustly evaluate algorithm performance, we evaluated IOTA ADNEX \nmodel performance in all patients.  In practice, a two-step strategy, initially triaging out \nwomen with a benign appearances on scan (< 1% risk of cancer over 2 years) and then using \nIOTA ADNEX to calculate risk of OC in the remaining women has been shown to improve \nspecificity. \n43–45 For patients with IOTA ADNEX scores of 10-50%, additional MRI imaging \ncan help prevent unnecessary surgery. We have previously discussed strategies needed to \nsuccessfully implement IOTA ADNEX ultrasound and a flowchart for practice \n(Supplementary Figure 4) \n46 Our results also suggest that earlier, accurate OC diagnosis is \npossible using IOTA ADNEX ultrasound as a first test in primary care, potentially concurrent \nwith CA125 testing and this approach needs evaluating in further research (Supplementary \nFigure 5)  \n \nStrengths  \n \nTo our knowledge, ROCkeTS is the first multi-site, blinded, prospective head-to-head \ncomparison study of all commonly used candidate risk prediction models for the diagnosis of \nOC in newly presenting premenopausal women. Selection bias was minimized by recruiting \nthrough the UK National cancer collaborative research network infrastructure.  Quality \nassured index tests were compared against a common reference standard and a predefined \nstatistical analysis plan included appropriate handling of missing data. Differentiation \nbetween primary (OC) and secondary (all cancer types including borderline) outcomes \nallowed investigation of test performance without inflation by inclusion of borderline \ntumours, which although common in younger women, have little impact on survival.  The \nstudy mirrored real life and practice where some patients who are referred undergo surgery \nand others are kept under surveillance or discharged. Inclusion of 12-month follow-up allows \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n24\nus to investigate false negatives of risk prediction models that would be used in clinical care \nto make triage decisions in women with suspected OC. \n \nIn contrast to previous studies the low prevalence of OC in ROCkeTS is applicable to a \nprimary care referred population 21 and most scans were performed by level II, non-medical \nNHS sonographers rather than  clinically qualified experts (gynaecologists or radiologists)  \nparticipating from centers of excellence.47  \n \nLimitations  \n \nAlthough ROCkets recruitment is consistent with UK census patterns, (mainly white \nethnicity), results may be less applicable to more diverse ethnicities. Although we are unable \nto identify differences in patient characteristics across pre- and post-protocol change cohorts, \nour decision to restrict recruitment to pre-surgical patients following observation of low OC \nprevalence in an interim analysis is likely to have led to systematic differences between the \ntwo cohorts. Further, not scheduling an additional hospital visit for a transvaginal IOTA \nultrasound post-protocol change to reduce risk during COVID and to improve recruitment led \nto high missingness, particularly of ultrasound variable data, in the cohort of women recruited \npost-protocol change.  \nAlthough we encouraged sites to obtain the IOTA ultrasound scan at the same time as routine \nscans for clinical care; we did not collect information on how this was delivered across sites. \nIt is possible that the sonographers who completed the scan to calculate RMI were different \nin some sites from the sonographers who collected the IOTA ultrasound scan.  \nDespite these issues, the consistency of results across pre-protocol, combined cohort, and \nimputed analyses across primary and secondary analyses demonstrates the robustness of the \nstudy's findings. The challenges of conducting ROCkeTS and estimating sample size are \ncommon to evaluations of test accuracy in low prevalence populations – we believe that our \nstudy has learnings for others designing a diagnostic test accuracy study to diagnose rare \nsignificant events amongst commonly occurring backgrounds.   \nImplications for research \nThe performance of risk prediction models in ethnically diverse populations needs further \nresearch.  Alternative methods of evaluating the impact of diagnostic test use on clinical \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n25\nutility or net benefit may be highly relevant and will be conducted as next steps in the \nROCkeTS study.48  Exploring the effect of threshold on test performance using the area under \nthe curve may play a role in determining optimal trade-offs in sensitivity and specificity. \nHowever, limitations of this approach includes not providing the threshold needed to realise \nthe optimal trade-off. This is essential for clinicians and health systems to make decisions \nabout further management.\n The use of decision curve analysis to compare decisions at the \nsame thresholds across risk prediction models was not possible because RMI 1 does not \nproduce probabilities.  \n \nLong waiting times for ultrasound, absence of standardisation and quality assurance are key \nchallenges to achieving IOTA USS implementation at scale in the NHS. 7 Artificial \nIntelligence (AI) enabled solutions for ultrasound along with quality assurance and training \nfor sonographers may deliver timely ultrasound availability in practice for women with non-\nspecific symptoms and significantly improve outcomes by early diagnosis. The impact of AI \nenabled USS with IOTA ADNEX in primary care practice on expediting diagnostic intervals \nand improving early detection of OC needs investigation. Understanding facilitators and \nbarriers to implementing change in diagnostic pathways, including a one-stop model for \nIOTA USS, is currently being investigated. (SONATA study, (NCT06129968).  \n \nCompared to RMI 1 at 250, in post-hoc analysis, ORADS at 10% achieved a smaller gain in \nsensitivity but a lower drop in specificity than IOTA ADNEX at 10%.  However, the latest \nversion of ORADS v2 with additional variables was not evaluated. 30 Prospective multicentre \nresearch studies investigating the diagnostic accuracy of ORADS for OC is necessary.  \n \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n26\nConclusion  \n \nThe current triage test for OC, RMI 1, demonstrates poor sensitivity in premenopausal \nwomen and should be replaced. IOTA ADNEX at 10% delivered by trained and quality-\nassured NHS sonographers achieves significantly higher sensitivity with limited specificity \nreduction in a real-world cohort and should be considered the new standard of care for \nsecondary care triage. Primary care implementation could potentially improve survival \nthrough earlier detection in symptomatic premenopausal women (Supplementary Figure 5); \nthis requires further research alongside investment in sonographer training and quality \nassurance.  \n \n \nDissemination plans \n \nThese results have been presented at the International gynaecological cancer society meeting \nin Dublin, October 2024 and at multiple national meetings. Results will be summarised in lay \nlanguage and placed on the ROCkeTS trial website. We do not plan to contact participants \nindividually to communicate the results of the study \n \n \n \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n27\nReferences \n \n1.  Lyratzopoulos G, Neal RD, Barbiere JM, Rubin GP, Abel GA. Variation in number of \ngeneral practitioner consultations before hospital referral for cancer: findings from the \n2010 National Cancer Patient Experience Survey in England. Lancet Oncol. 2012 Apr \n1;13(4):353–65.  \n2.  Lyratzopoulos G, Wardle J, Rubin G. Rethinking diagnostic delay in cancer: how difficult \nis the diagnosis? BMJ. 2014 Dec 10;349:g7400.  \n3.  The International Agency for Research on Cancer (IARC). Global Cancer Observatory \n[Internet]. 2024 [cited 2024 Dec 18]. Available from: https://gco.iarc.fr/ \n4.  UK National Institute for Health and Care Excellence. Ovarian cancer: recognition and \ninitial management | Clinical Guideline [CG122] [Internet]. NICE; 2011 [cited 2024 Dec \n18]. Available from: https://www.nice.org.uk/guidance/cg122 \n5.  NHS England. NHS England Faster diagnosis [Internet]. [cited 2025 Oct 8]. Available \nfrom: https://www.england.nhs.uk/cancer/faster-diagnosis/ \n6.  NHS England. NHS England Faster diagnostic pathways: Implementing a timed \ngynaecology cancer diagnostic pathway. Guidance for local health and care systems.  \n7.  NHS England. NHS Diagnostic Waiting Times and Activity Data [Internet]. 2023 [cited \n2024 Dec 18]. Available from: https://www.england.nhs.uk/statistics/wp-\ncontent/uploads/sites/2/2023/05/DWTA-Report-March-2023_OLEX2.pdf \n8.  Charkhchi P, Cybulski C, Gronwald J, Wong FO, Narod SA, Akbari MR. CA125 and \nOvarian Cancer: A Comprehensive Review. Cancers. 2020 Dec 11;12(12):3730.  \n9.  Jacobs I, Oram D, Fairbanks J, Turner J, Frost C, Grudzinskas JG. A risk of malignancy \nindex incorporating CA 125, ultrasound and menopausal status for the accurate \npreoperative diagnosis of ovarian cancer. Br J Obstet Gynaecol. 1990 Oct;97(10):922–9.  \n10.  Samuel J Perry, Griffin D, Williams E, Roberts T, Kwong A, Williams S, et al. A cost \nconsequence analysis of six diagnostic strategies for ovarian cancer: A model-based \neconomic evaluation. Under review British Journal of Obseterics and Gynaecology, \nManuscript ID BJOG-25-0376R1.  \n11.  Cummins C, Kumar S, Long J, Balega J, Broadhead T, Duncan T, et al. Investigating the \nImpact of Ultra-Radical Surgery on Survival in Advanced Ovarian Cancer Using \nPopulation-Based Data in a Multicentre UK Study. Cancers. 2022 Sep 7;14(18):4362.  \n12.  Engelen MJA, Kos HE, Willemse PHB, Aalders JG, de Vries EGE, Schaapveld M, et al. \nSurgery by consultant gynecologic oncologists improves survival in patients with ovarian \ncarcinoma. Cancer. 2006 Feb 1;106(3):589–98.  \n13.  Vernooij F, Heintz P, Witteveen E, van der Graaf Y . The outcomes of ovarian cancer \ntreatment are better when provided by gynecologic oncologists and in specialized \nhospitals: a systematic review. Gynecol Oncol. 2007 Jun;105(3):801–12.  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n28\n14.  Moore RG, McMeekin DS, Brown AK, DiSilvestro P, Miller MC, Allard WJ, et al. A \nnovel multiple marker bioassay utilizing HE4 and CA125 for the prediction of ovarian \ncancer in patients with a pelvic mass. Gynecol Oncol. 2009 Jan;112(1):40–6.  \n15.  Van Calster B, Van Hoorde K, Valentin L, Testa AC, Fischerova D, V an Holsbeke C, et al. \nEvaluating the risk of ovarian cancer before surgery using the ADNEX model to \ndifferentiate between benign, borderline, early and advanced stage invasive, and \nsecondary metastatic tumours: prospective multicentre diagnostic study. BMJ. 2014 Oct \n15;349:g5920.  \n16.  Timmerman D, Ameye L, Fischerova D, Epstein E, Melis GB, Guerriero S, et al. Simple \nultrasound rules to distinguish between benign and malignant adnexal masses before \nsurgery: prospective validation by IOTA group. BMJ. 2010 Dec 14;341:c6839.  \n17.  Royal College of Obstetricians and Gynaecologists. Ovarian Masses in Premenopausal \nWomen, Management of Suspected (Green-top Guideline No. 62) [Internet]. RCOG. \n[cited 2024 Dec 18]. Available from: https://www.rcog.org.uk/guidance/browse-all-\nguidance/green-top-guidelines/ovarian-masses-in-premenopausal-women-management-\nof-suspected-green-top-guideline-no-62/ \n18.  American College of Obstetricians and Gynecologists’ Committee on Practice \nBulletins—Gynecology. Practice Bulletin No. 174: Evaluation and Management of \nAdnexal Masses. Obstet Gynecol. 2016 Nov;128(5):e210–26.  \n19.  Andreotti RF, Timmerman D, Strachowski LM, Froyman W, Benacerraf BR, Bennett GL, \net al. O-RADS US Risk Stratification and Management System: A Consensus Guideline \nfrom the ACR Ovarian-Adnexal Reporting and Data System Committee. Radiology. 2020 \nJan;294(1):168–85.  \n20.  Timmerman D, Planchamp F, Bourne T, Landolfo C, du Bois A, Chiva L, et al. \nESGO/ISUOG/IOTA/ESGE Consensus Statement on preoperative diagnosis of ovarian \ntumors. Ultrasound Obstet Gynecol Off J Int Soc Ultrasound Obstet Gynecol. 2021 \nJul;58(1):148–68.  \n21.  Davenport C, Rai N, Sharma P, Deeks JJ, Berhane S, Mallett S, et al. Menopausal status, \nultrasound and biomarker tests in combination for the diagnosis of ovarian cancer in \nsymptomatic women. Cochrane Database Syst Rev. 2022 Jul 26;7(7):CD011964.  \n22.  Sundar S, Agarwal R, Davenport C, Scandrett K, Johnson S, Sengupta P , et al. Risk-\nprediction models in postmenopausal patients with symptoms of suspected ovarian cancer \nin the UK (ROCkeTS): a multicentre, prospective diagnostic accuracy study. Lancet \nOncol. 2024 Oct;25(10):1371–86.  \n23.  Bossuyt PM, Reitsma JB, Bruns DE, Gatsonis CA, Glasziou PP, Irwig L, et al. STARD \n2015: an updated list of essential items for reporting diagnostic accuracy studies. BMJ. \n2015 Oct 28;351:h5527.  \n24.  Collins GS, Reitsma JB, Altman DG, Moons KGM. Transparent Reporting of a \nmultivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD): The \nTRIPOD Statement. Ann Intern Med. 2015 Jan 6;162(1):55–63.  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n29\n25.  Sundar S, Rick C, Dowling F, Au P, Snell K, Rai N, et al. Refining Ovarian Cancer Test \naccuracy Scores (ROCkeTS): protocol for a prospective longitudinal test accuracy study \nto validate new risk scores in women with symptoms of suspected ovarian cancer. BMJ \nOpen. 2016 Aug 9;6(8):e010333.  \n26.  Tuck MK, Chan DW, Chia D, Godwin AK, Grizzle WE, Krueger KE, et al. Standard \noperating procedures for serum and plasma collection: early detection research network \nconsensus statement standard operating procedure integration working group. J Proteome \nRes. 2009 Jan;8(1):113–7.  \n27.  Moore RG, Jabre-Raughley M, Brown AK, Robison KM, Miller MC, Allard WJ, et al. \nComparison of a novel multiple marker assay vs the Risk of Malignancy Index for the \nprediction of epithelial ovarian cancer in patients with a pelvic mass. Am J Obstet \nGynecol. 2010 Sep;203(3):228.e1-6.  \n28.  Funston G, Hamilton W, Abel G, Crosbie EJ, Rous B, Walter FM. The diagnostic \nperformance of CA125 for the detection of ovarian and non-ovarian cancer in primary \ncare: A population-based cohort study. PLOS Med. 2020 Oct 28;17(10):e1003295.  \n29.  Moore RG, McMeekin DS, Brown AK, DiSilvestro P, Miller MC, Allard WJ, et al. A \nnovel multiple marker bioassay utilizing HE4 and CA125 for the prediction of ovarian \ncancer in patients with a pelvic mass. Gynecol Oncol. 2009 Jan;112(1):40–6.  \n30.  Strachowski LM, Jha P, Phillips CH, Blanchette Porter MM, Froyman W, Glanc P , et al. \nO-RADS US v2022: An Update from the American College of Radiology’s Ovarian-\nAdnexal Reporting and Data System US Committee. Radiology. 2023 \nSep;308(3):e230685.  \n31.  Timmerman D, V an Calster B, Testa A, Savelli L, Fischerova D, Froyman W, et al. \nPredicting the risk of malignancy in adnexal masses based on the Simple Rules from the \nInternational Ovarian Tumor Analysis group. Am J Obstet Gynecol. 2016 \nApr;214(4):424–37.  \n32.  Gynaia [Internet]. [cited 2025 Oct 8]. Available from: https://gynaia.com/ \n33.  Timmerman S, V alentin L, Ceusters J, Testa AC, Landolfo C, Sladkevicius P , et al. \nExternal Validation of the Ovarian-Adnexal Reporting and Data System (O-RADS) \nLexicon and the International Ovarian Tumor Analysis 2-Step Strategy to Stratify \nOvarian Tumors Into O-RADS Risk Groups. JAMA Oncol. 2023 Feb 1;9(2):225–33.  \n34.  Timmerman D, V alentin L, Bourne TH, Collins WP, Verrelst H, Vergote I. Terms, \ndefinitions and measurements to describe the sonographic features of adnexal tumors: a \nconsensus opinion from the International Ovarian Tumor Analysis (IOTA) group. \nUltrasound Obstet Gynecol. 2000 Oct;16(5):500–5.  \n35.  Collins GS, Dhiman P, Ma J, Schlussel MM, Archer L, V an Calster B, et al. Evaluation of \nclinical prediction models (part 1): from development to external validation. BMJ. 2024 \nJan 8;384:e074819.  \n36.  van Buuren S, Boshuizen HC, Knook DL. Multiple imputation of missing blood pressure \ncovariates in survival analysis. Stat Med. 1999 Mar 30;18(6):681–94.  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint \n\n \n \n30\n37.  White IR, Royston P, Wood AM. Multiple imputation using chained equations: Issues and \nguidance for practice. Stat Med. 2011 Feb 20;30(4):377–99.  \n38.  Rai N, Nevin J, Downey G, Abedin P, Balogun M, Kehoe S, et al. Outcomes following \nimplementation of symptom triggered diagnostic testing for ovarian cancer. Eur J Obstet \nGynecol Reprod Biol. 2015 Apr;187:64–9.  \n39.  Collins GS, Ogundimu EO, Altman DG. Sample size considerations for the external \nvalidation of a multivariable prognostic model: a resampling study. Stat Med. 2016 Jan \n30;35(2):214–26.  \n40.  Kwong FL, Kristunas C, Davenport C, Aggarwal R, Deeks J, Mallett S, et al. \nInvestigating harms of testing for ovarian cancer - psychological outcomes and cancer \nconversion rates in women with symptoms of ovarian cancer: A cohort study embedded \nin the multicentre ROCkeTS prospective diagnostic study. BJOG Int J Obstet Gynaecol. \n2024 Sep;131(10):1400–10.  \n41.  Kwong FLA, Kristunas C, Davenport C, Deeks J, Mallett S, Agarwal R, et al. Symptom-\ntriggered testing detects early stage and low volume resectable advanced stage ovarian \ncancer. Int J Gynecol Cancer Off J Int Gynecol Cancer Soc. 2024 Sep 25;ijgc–2024–\n005371.  \n42.  IOTA Plus. IOTA Plus NHS Training [Internet]. 2024. Available from: \nhttps://iotaplus.org/en/iota-plus-nhs-training \n43.  Landolfo C, Bourne T, Froyman W, V an Calster B, Ceusters J, Testa AC, et al. Benign \ndescriptors and ADNEX in two-step strategy to estimate risk of malignancy in ovarian \ntumors: retrospective validation in IOTA5 multicenter cohort. Ultrasound Obstet Gynecol \nOff J Int Soc Ultrasound Obstet Gynecol. 2023 Feb;61(2):231–42.  \n44.  Borges AL, Brito M, Ambrósio P, Condeço R, Pinto P, Ambrósio B, et al. Prospective \nexternal validation of IOTA methods for classifying adnexal masses and retrospective \nassessment of two-step strategy using benign descriptors and ADNEX model: Portuguese \nmulticenter study. Ultrasound Obstet Gynecol Off J Int Soc Ultrasound Obstet Gynecol. \n2024 Oct;64(4):538–49.  \n45.  Froyman W, Landolfo C, De Cock B, Wynants L, Sladkevicius P , Testa AC, et al. Risk of \ncomplications in patients with conservatively managed ovarian tumours (IOTA5): a 2-\nyear interim analysis of a multicentre, prospective, cohort study. Lancet Oncol. 2019 \nMar;20(3):448–58.  \n46.  Sundar S. Addressing the challenges of implementation of IOTA ADNEX in clinical \npractice. Lancet Oncol. 2024;accepted, DOI awaited.  \n47.  V an Calster B, Valentin L, Froyman W, Landolfo C, Ceusters J, Testa AC, et al. \nValidation of models to diagnose ovarian cancer in patients managed surgically or \nconservatively: multicentre cohort study. BMJ. 2020 Jul 30;370:m2614.  \n48.  V an Calster B, Vickers AJ. Calibration of risk prediction models: impact on decision-\nanalytic performance. Med Decis Mak Int J Soc Med Decis Mak. 2015 Feb;35(2):162–9.  \n \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted October 18, 2025. ; https://doi.org/10.1101/2025.10.17.25338220doi: medRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}