Diagnostic value of the Risk of Ovarian Malignancy Algorithm (ROMA) index in the detection of ovarian cancer in postmenopausal women: a systematic review and meta-analysis.

This meta-analysis was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA). The protocol for this systematic review and meta-analysis has been registered in PROSPERO under ID: CRD42024599579, which is available at: https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42024599579 . The study was approved by the Ethical Committee of the Urmia University of Medical Sciences (Approval No: IR.UMSU.REC.1403.385). A comprehensive search of the scientific literature was conducted on December 5, 2024, covering all years of the Web of Science, PubMed (MEDLINE), Google Scholar, Scopus, and Embase databases. To ensure comprehensive coverage, the references cited in the identified articles were also reviewed. The search strategies are available in the Additional file 1. full-text publications; 2) the study must involve human subjects; 3) the results of the ROMA index in postmenopausal women must be reported; 4) no language restrictions; 5) studies in which the values of the 2 × 2 table can be calculated; 6) studies in which the patient group included ovarian cancer cases, and the control group included borderline cases and healthy individuals. This study was conducted based on all published articles available in specified databases. Therefore, no geographical restrictions were applied to the study locations or populations. In addition, a literature search was performed without any time restrictions. Data related to postmenopausal women were extracted from studies that included only postmenopausal women, or both premenopausal and postmenopausal women. full-text publications; 2) the study must involve human subjects; 3) the results of the ROMA index in postmenopausal women must be reported; 4) no language restrictions; 5) studies in which the values of the 2 × 2 table can be calculated; 6) studies in which the patient group included ovarian cancer cases, and the control group included borderline cases and healthy individuals. This study was conducted based on all published articles available in specified databases. Therefore, no geographical restrictions were applied to the study locations or populations. In addition, a literature search was performed without any time restrictions. Data related to postmenopausal women were extracted from studies that included only postmenopausal women, or both premenopausal and postmenopausal women. animal studies; 2) Articles without full-text availability; 3) Articles lacking ROMA index data in postmenopausal women; 4) studies in which the values of the 2 × 2 table could not be extracted. animal studies; 2) Articles without full-text availability; 3) Articles lacking ROMA index data in postmenopausal women; 4) studies in which the values of the 2 × 2 table could not be extracted. Quality assessment of the articles was performed using the QUADAS-2 tool, which is specifically designed to evaluate the quality of meta-analyses involving diagnostic tests. This tool assesses articles in terms of risk of bias across four domains: Patient Selection, Index Test, Reference Standard, and Flow and Timing. Additionally, it evaluates applicability concerns in three domains, including Patient Selection, Index Test, and Reference Standard [ 24 ]. The GRADEPRO tool ( https://gradepro.org/ .) was used to evaluate and grade the quality of the evidence obtained from the meta-analysis. This tool is based on the GRADE (Grading of Recommendations Assessment, Development and Evaluation) framework, which enables researchers to assess the quality of evidence in a systematic and transparent manner. Overall, the quality of evidence was rated moderate, indicating a moderate level of confidence in the effect estimates. This suggests that the true effect is likely to be close to the estimated effect, but there is a possibility that it may be substantially different. The primary reason for downgrading the certainty of the evidence was the high heterogeneity among the studies included in the meta-analysis. This heterogeneity resulted in a downgrade owing to serious inconsistencies, as defined within the GRADE framework. Considerable variability in the results of the primary studies, differences in study populations, measurement tools, and intervention methods were identified as factors contributing to the reduced consistency of the findings. Comprehensive details of the evidence grading based on the five GRADE domains (study limitations, inconsistency, imprecision, indirectness, and publication bias) are presented in the corresponding table in Additional File 2. The extracted information included the first author, country, continent, year of publication, study design, and values for True Positive (TP), False Positive (FP), False Negative (FN), True Negative (TN), sensitivity, specificity, Positive Likelihood Ratio (LR +), Negative Likelihood Ratio (LR −), Diagnostic Odds Ratio (DOR), and inverse LR − for postmenopausal women based on the optimal cut-off. These values were extracted from all the articles. The Midas and Metandi packages [ 25 , 26 ] in Stata 18.0 (Stata Corporation, TX, USA) were used for statistical analysis of the indicators related to the study, including sensitivity, specificity, DOR, LR +, LR-, and 1/LR-. The positive likelihood ratio (LR⁺) is the true-positive rate divided by the false-positive rate, while the negative likelihood ratio (LR⁻) is the false-negative rate divided by the true-negative rate. Likelihood ratios ranged from zero to infinity, with values further from 1.0, providing stronger diagnostic evidence. An LR⁺ value greater than 10 strongly supports ruling in a diagnosis, whereas an LR value close to zero helps rule out disease. An LR⁺ of 1.0, is uninformative, as it means that the test does not distinguish between diseased and healthy individuals. The diagnostic odds ratio (DOR), defined as LR⁺ divided by LR⁻, quantified the overall test performance. It ranges from 0 to infinity, with higher values indicating better discrimination between patients and healthy individuals. A DOR of 1 implies no diagnostic value, whereas values below 1 suggest misleading test interpretation. Subgroup analysis was also performed based on continent, study method, and cutoff value. The DOR value ranges from zero to infinity, with higher values indicating a greater discriminatory power of the ROMA index in diagnosing ovarian cancer. The SROC curve was constructed using sensitivity and specificity, and the closer the area under the curve (AUC) was to one, the higher the diagnostic validity of the ROMA index. The I 2 test was used to assess heterogeneity, and Fagan’s nomogram was used to examine the relationship between prior and posterior probabilities.

According to the PRISMA flowchart presented in (Fig.  1 ), a total of 99 articles were identified through the search strategy. Of these, 21 articles were duplicates, 25 articles contained unrelated data, and nine articles were not relevant to postmenopausal women; thus, they were excluded from the study. After a thorough full-text review and assessment of data consistency, 34 studies were finally included in the analysis. Fig. 1 A flowchart of the study selection process(PRISMA) A flowchart of the study selection process(PRISMA) The quality of the studies included in the analysis was assessed using the QUADAS-2 tool (Table  1 ). As shown in (Fig.  2 ), the highest scores for both the"risk of bias"and"concerns regarding applicability"domains were attributed to patient selection. Table 1 The quality of included articles according to the QUADAS-2 guidelines ID STUDY Risk of Bias Applicability Concerns Patient Selection Index Test Reference Standard Flow and Timing Patient Selection Index Test Reference Standard 1 GONG Shipeng2019 [ 27 ] LOW LOW LOW LOW LOW LOW LOW 2 Laisheng Li2016 [ 28 ] UNCLEAR LOW UNCLEAR UNCLEAR HIGH LOW HIGH 3 Srinivas Kondalsamy2013 [ 29 ] HIGH UNCLEAR HIGH UNCLEAR HIGH UNCLEAR HIGH 4 A. Stiekema2014 [ 30 ] HIGH LOW UNCLEAR HIGH HIGH LOW LOW 5 Elisabetta Bandiera2011 [ 31 ] LOW LOW LOW UNCLEAR LOW LOW LOW 6 Maria Lycke2018 [ 32 ] UNCLEAR LOW LOW UNCLEAR HIGH LOW LOW 7 Giuseppina Ruggeri2011 [ 33 ] HIGH UNCLEAR UNCLEAR UNCLEAR HIGH LOW UNCLEAR 8 Nguyen Vu Quoc Huy2018 [ 14 ] HIGH LOW LOW UNCLEAR LOW LOW LOW 9 Khawla Al Musalhi2016 [ 34 ] UNCLEAR LOW LOW UNCLEAR HIGH HIGH UNCLEAR 10 Mona Aarenstrup Karlsen2012 [ 35 ] UNCLEAR HIGH UNCLEAR UNCLEAR LOW LOW HIGH 11 Z.G. Dikmen2014 [ 2 ] LOW LOW LOW UNCLEAR LOW LOW LOW 12 Toon Van Gorp2012 [ 36 ] HIGH UNCLEAR UNCLEAR UNCLEAR HIGH HIGH HIGH 13 Shoichiro Yamanaka2022 [ 1 ] UNCLEAR UNCLEAR LOW UNCLEAR HIGH UNCLEAR HIGH 14 Lei Zhang2018 [ 6 ] HIGH UNCLEAR UNCLEAR UNCLEAR HIGH LOW UNCLEAR 15 Zvjezdana Špacir Prskalo2018 [ 37 ] LOW UNCLEAR UNCLEAR UNCLEAR HIGH UNCLEAR UNCLEAR 16 SU WEI2016 [ 38 ] LOW LOW LOW UNCLEAR LOW LOW LOW 17 Doan Tu Tran2021 [ 39 ] UNCLEAR LOW UNCLEAR UNCLEAR HIGH LOW UNCLEAR 18 Yong-ning CHEN2020 [ 40 ] LOW UNCLEAR UNCLEAR UNCLEAR LOW HIGH LOW 19 Abha Hada2020 [ 41 ] UNCLEAR UNCLEAR UNCLEAR UNCLEAR HIGH UNCLEAR UNCLEAR 20 Fariba Behnamfar2023 [ 42 ] HIGH UNCLEAR UNCLEAR UNCLEAR HIGH UNCLEAR UNCLEAR 21 Zhen Shen2021 [ 43 ] HIGH UNCLEAR UNCLEAR UNCLEAR UNCLEAR LOW LOW 22 ANITA CHUDECKA‑GŁAZ2016 [ 44 ] LOW LOW LOW UNCLEAR LOW LOW LOW 23 Elisabetta Bandiera2011 [ 31 ] LOW LOW UNCLEAR LOW UNCLEAR LOW LOW 24 Blanca Ortiz-Muñoz2014 [ 45 ] LOW LOW UNCLEAR UNCLEAR HIGH LOW UNCLEAR 25 Karen K.L. Chan2012 [ 46 ] UNCLEAR LOW LOW LOW HIGH UNCLEAR LOW 26 Farah Farzaneh2014 [ 47 ] HIGH UNCLEAR UNCLEAR UNCLEAR HIGH UNCLEAR UNCLEAR 27 Katarzyna M. Terlikowska2016 [ 48 ] UNCLEAR LOW LOW UNCLEAR UNCLEAR LOW LOW 28 Rafael Molina2011 [ 49 ] UNCLEAR LOW UNCLEAR UNCLEAR UNCLEAR LOW UNCLEAR 29 Wen-Ting Chen2014 [ 50 ] UNCLEAR LOW LOW UNCLEAR HIGH UNCLEAR LOW 30 Xinliang Chen2014 [ 51 ] HIGH UNCLEAR UNCLEAR HIGH HIGH UNCLEAR UNCLEAR 31 Ying Xu2015 [ 52 ] LOW LOW UNCLEAR LOW LOW LOW UNCLEAR 32 Ângela Melo2018 [ 53 ] UNCLEAR HIGH LOW UNCLEAR HIGH UNCLEAR LOW 33 Hyun-Jin Choi2019 [ 54 ] HIGH LOW LOW UNCLEAR HIGH LOW LOW 34 ZDENEK NOVOTNY2012 [ 23 ] UNCLEAR LOW LOW UNCLEAR HIGH LOW LOW Fig. 2 Risk of bias summary and Applicability Concerns graph(QUADAS-2) The quality of included articles according to the QUADAS-2 guidelines Risk of bias summary and Applicability Concerns graph(QUADAS-2) Table 2 presents the general information of all studies, including the country where the study was conducted, continent, study design, and validation parameters (sensitivity, specificity, LR +, LR-, DOR, and 1/LR-). Table 2 The main characteristics of the included studies in the meta-analysis Id Author Country Cancer Type Time Frame Continent Design N TP FP FN TN Sensitivity Specificity LR +  LR- DOR 1/LR- 1 GONG Shipeng 2019 [ 27 ] China EOC,BOT Sep 2014_Nov 2016 Asia Cross-sectional 719 68 10 6 48 0.918919 0.827586 5.32973 0.097973 54.4 10.2069 2 Laisheng Li 2016 [ 28 ] China OC Sep 2012_Apr 2014 Asia Cohort 1189 70 5 12 85 0.853659 0.944444 15.36585 0.15495 99.16667 6.453704 3 Srinivas Kondalsamy 2013 [ 29 ] Australia EOC NA Australia Case control 158 27 13 11 34 0.710526 0.723404 2.568826 0.400155 6.41958 2.499033 4 A. Stiekema 2014 [ 30 ] Netherlands EOC Feb 1994_Nov 2008 Europe Cohort 361 106 18 111 59 0.488479 0.766234 2.089606 0.667578 3.13013 1.497952 5 Elisabetta Bandiera 2011 [ 31 ] Italy EOC 2003_2010 Europe Cohort 419 81 15 6 81 0.931034 0.84375 5.958621 0.081737 72.9 12.23438 6 Maria Lycke 2018 [ 32 ] Sweden EOC,BOT Sep 2013_Feb 2016 Europe Clinical trial 638 102 2 10 8 0.910714 0.8 4.553571 0.111607 40.8 8.96 7 Giuseppina Ruggeri 2011 [ 33 ] Italy EOC 2006 Europe Cohort 259 72 7 3 22 0.96 0.758621 3.977143 0.052727 75.42857 18.96552 8 Nguyen Vu Quoc Huy 2018 [ 14 ] Vietnam EOC Jan 2016_Nov 2017 Asia Cross-sectional 277 14 1 3 29 0.823529 0.966667 24.70588 0.182556 135.3333 5.477778 9 Khawla Al Musalhi 2016 [ 34 ] Oman EOC Mar 2014_Apr 2015 Asia Cross-sectional 213 25 5 2 19 0.925926 0.791667 4.444444 0.093567 47.5 10.6875 10 Mona Aarenstrup Karlsen 2012 [ 35 ] Denmark EOC,BOT Sep 2004_Jan 2010 Europe Cohort 1218 198 70 5 209 0.975369 0.749104 3.887544 0.03288 118.2343 30.41362 11 Z.G. Dikmen 2014 [ 2 ] Turkey OC NA Europe Cohort 249 32 1 1 12 0.969697 0.923077 12.60606 0.032828 384 30.46154 12 Toon Van Gorp 2012 [ 36 ] Belgium EOC Aug 2005_Mar 2009 Europe Cohort 432 101 35 10 50 0.90991 0.588235 2.209781 0.153153 14.42857 6.529412 13 Shoichiro Yamanaka 2022 [ 1 ] Japan EOC,BOT Jan 2007_Jul 2021 Asia Cohort 171 45 0 6 5 0.882353 1 N/A 0.117647 N/A 8.5 14 Lei Zhang 2018 [ 6 ] China EOC,BOT Jul 2016_Jul 2017 Asia Cohort 373 103 5 15 55 0.872881 0.916667 10.47458 0.138675 75.53333 7.211111 15 Zvjezdana Špacir Prskalo 2018 [ 37 ] Croatia EOC,BOT May 2015_May 2017 Europe Cohort 159 32 5 4 53 0.888889 0.913793 10.31111 0.121593 84.8 8.224138 16 SU WEI 2016 [ 38 ] China OC,BOT Sep 2013_May 2015 Asia Cohort 158 34 1 3 32 0.918919 0.969697 30.32432 0.083615 362.6667 11.9596 17 Doan Tu Tran 2021 [ 39 ] Vietnam EOC Jan 2018_Jun 2020 Asia Cohort 475 28 1 9 67 0.756757 0.985294 51.45946 0.246874 208.4444 4.050654 18 Yong-ning CHEN 2020 [ 40 ] China EOC,BOT Sep 2014_Nov 2016 Asia Cohort 719 68 10 6 48 0.918919 0.827586 5.32973 0.097973 54.4 10.2069 19 Abha Hada 2020 [ 41 ] China EOC,BOT Jan 2014_Dec 2014 Asia Cohort 155 15 0 4 23 0.789474 1 N/A 0.210526 N/A 4.75 20 Fariba Behnamfar 2023 [ 42 ] Iran EOC,BOT 2020_2021 Asia Cross-sectional 203 19 0 2 42 0.904762 1 N/A 0.095238 N/A 10.5 21 Zhen Shen 2021 [ 43 ] China EOC Jul 2005_Jun 2018 Asia Cohort 509 113 8 8 42 0.933884 0.84 5.836777 0.078709 74.15625 12.705 22 ANITA CHUDECKA-GŁAZ 2016 [ 44 ] Poland OC,BOT Nov 2012_Dec 2014 Europe Cohort 619 116 15 8 71 0.935484 0.825581 5.363441 0.078146 68.63333 12.79651 23 Elisabetta Bandiera 2011 [ 31 ] Italy EOC 2003–2010 Europe Cohort 419 81 15 6 81 0.931034 0.84375 5.958621 0.081737 72.9 12.23438 24 Blanca Ortiz-Muñoz 2014 [ 45 ] Spain EOC Jan 2014_Apr2014 Europe Cohort 218 18 5 1 80 0.947368 0.941176 16.10526 0.055921 288 17.88235 25 Karen K.L. Chan 2012 [ 46 ] Asia EOC 2009–2010 Asia Cohort 414 40 6 3 47 0.930233 0.886792 8.217054 0.078674 104.4444 12.71069 26 Farah Farzaneh 2014 [ 47 ] Iran EOC Mar 2012_Mar 2013 Asia Clinical trial 99 18 0 4 9 0.818182 1 N/A 0.181818 N/A 5.5 27 Katarzyna M. Terlikowska 2016 [ 48 ] Poland EOC 2012_2016 Europe Cohort 224 56 2 7 39 0.888889 0.95122 18.22222 0.116809 156 8.560976 28 Rafael Molina 2011 [ 49 ] Spain OC NA Europe Cohort 527 80 10 4 49 0.952381 0.830508 5.619048 0.057337 98 17.44068 29 Wen-Ting Chen 2014 [ 50 ] China EOC Jun 2010_Mar 2013 Asia Cohort 297 63 3 6 12 0.913043 0.8 4.565217 0.108696 42 9.2 30 Xinliang Chen 2014 [ 51 ] China OC Mar 2012_Mar 2014 Asia Cohort 232 37 1 3 21 0.925 0.954545 20.35 0.078571 259 12.72727 31 Ying Xu 2015 [ 52 ] China EOC,BOT Jul 2013_Now 2014 Asia Cohort 1021 81 4 22 43 0.786408 0.914894 9.240291 0.233461 39.57955 4.283366 32 Ângela Melo 2018 [ 53 ] Portugal EOC,BOT Jan 2013_ Dec 2016 Europe Cohort 247 20 4 5 60 0.8 0.9375 12.8 0.213333 60 4.6875 33 Hyun-Jin Choi 2019 [ 54 ] Korea EOC,BOT Mar 2010_Apr 2014 Asia Cohort 649 215 2 37 76 0.853175 0.974359 33.27381 0.150689 220.8108 6.636175 34 ZDENEK NOVOTNY 2012 [ 23 ] Czech OC 2010_2011 Europe Cohort 277 20 31 1 225 0.952381 0.878906 7.864823 0.05418 145.1613 18.45703 N/A not applicable , NA not available in study , N Sample Size , OC Ovarian Cancer , EOC Epithelial Ovarian Cancer , BOT Borderline Ovarian Tumors The main characteristics of the included studies in the meta-analysis N/A not applicable , NA not available in study , N Sample Size , OC Ovarian Cancer , EOC Epithelial Ovarian Cancer , BOT Borderline Ovarian Tumors As shown in (Fig.  3 ), The geographic distribution of the studies is depicted on the map. No studies were found in Africa or the Americas; however, the distribution of studies across Asia and Europe includes a considerable number of countries, which may serve as representative samples of these two continents. Therefore, when comparing continents, Europe and Asia received more attention. The map indicates that the majority of studies conducted in Asia on the ROMA index are from China, Iran, and Vietnam, whereas in Europe, most studies are focused on Italy, the Netherlands, and Spain. Fig. 3 The geographic distribution of studies The geographic distribution of studies The SROC diagram and Forest plot presented in (Fig.  4 ) illustrate the diagnostic accuracy of the ROMA index in postmenopausal women. The results (Fig.  4 A) show that the highest sensitivity, 98%, was reported in the study by Aarenstrup (2012, Denmark), while the lowest sensitivity, 49%, was observed in the study by A. Stiekema (2014, Netherlands). The pooled sensitivity across studies was 90%, which aligns with the value reported in a study by Behnamfar (2023, Iran). Regarding specificity, the highest value of 100% was observed in studies by Farzaneh (2014, Iran), Abha Hada (2020, China), Shoichiro Yamanaka (2022, Japan), and Fariba Behnamfar (2023, Iran). The lowest specificity (59%) was found in the study by Toon Van Gorp (2012, Belgium). The pooled specificity was 89%, which is consistent with the value calculated by Karen K.L. Chan (2012, Asia). The heterogeneity values for sensitivity and specificity were I 2  = 93.06, p  = 0.00, and I 2  = 81.67, p  = 0.00, respectively. These results suggest a high degree of heterogeneity in the sensitivity and specificity across the included studies. The SROC plot (Fig.  4 B) displays the intersection points of sensitivity and specificity for the ROMA index based on the studies reviewed. This plot includes a summary point representing the combined sensitivity and specificity of all studies. The number of circles in the SROC diagram represents the number of included studies. The dashed line around this summary point denotes the confidence interval (CI) within which the true values are expected to fall with a 95% probability. Two studies, Lei Zhang (2018, China) and Zvjezdana Špacir Prskalo (2018, Croatia), were within this confidence interval. The dotted line in the plot represents the prediction region, indicating that, if the ROMA index is applied to new individuals, there is a 95% probability that the values will fall within this region. Furthermore, the SROC chart showed an AUC of 0.95, indicating high diagnostic accuracy of the ROMA index for detecting ovarian cancer. Fagan's nomogram (Fig.  4 C) illustrates that with a pre-test probability of 25% and PLR of 9, the post-test probability (PPV) increased to 74%. This finding suggests that a positive ROMA index result significantly increases the likelihood of the disease being present. Fig. 4 A  = forest plot B  = SROC models C  = Fagan’s plot for postmenopausal woman A  = forest plot B  = SROC models C  = Fagan’s plot for postmenopausal woman The diagnostic performance of the ROMA index varies depending on factors such as the continent where the study was conducted, type of study design, and cut-off value used. These factors contributed to the observed heterogeneity (Table  3 ), and understanding their impact is crucial for interpreting the overall findings of this meta-analysis. Investigating these variations will help optimize the application of the ROMA index in different clinical settings and populations. Table 3 Summary of Total and subgroup analysis of ROMA index for postmenopausal women Postmenopausal woman Total Continent Study design Cut-offs Asia Europe Cross sectional Cohort Cut-off = (10–25) Cut-off = (25.1–40) Optimal cut off Sensitivity 0.89 0.878 0.914 0.901 0.902 0.922 0.897 Specificity 0.89 0.932 0.843 0.934 0.894 0.779 0.896 DOR 68 98.972 57.085 128.658 77.331 41.621 75.557 LR +  8 12.992 5.81 13.609 8.497 4.177 8.659 LR- 0.119 0.131 0.102 0.106 0.11 0.1 0.115 1/LR- 8 7.618 9.826 9.454 9.101 9.965 8.726 Summary of Total and subgroup analysis of ROMA index for postmenopausal women Analysis of the SROC curves for postmenopausal women of Asian and European descent revealed that the summary point for Asian women was closer to the top-left corner (0,1) of the curve. This indicates the superior diagnostic performance of the ROMA index in this population compared to that in European women(18 studies vs. 15 studies) (Fig.  5 A, B). This observation was further supported by a comparison of the Diagnostic Odds Ratio (DOR), which was 99 for Asian women and 57 for European women. A higher DOR reflects better discriminatory ability between diseased and non-diseased individuals. The difference in diagnostic performance may be attributed to genetic variations, biological differences, or regional differences in diagnostic procedures and healthcare practices. Fig. 5 A : SROC for postmenopausal Asian women B : SROC for postmenopausal European women A : SROC for postmenopausal Asian women B : SROC for postmenopausal European women When comparing SROC curves based on study design, cross-sectional studies showed a smaller distance from the top-left corner (0,1) than cohort studies. This suggests that cross-sectional studies provide higher estimates of the diagnostic accuracy of the ROMA index (4 studies vs. 27 studies) (Fig.  6 A, B). This finding was further validated by the DOR comparison, where cross-sectional studies reported a DOR of 129 compared to 77 in cohort studies. This variation may be due to methodological differences between study designs. Cross-sectional studies assess subjects at a single point in time, which can lead to inflated estimates of sensitivity and specificity. In contrast, cohort studies follow participants over time, which may introduce variability and lower the diagnostic estimates. These results highlight the importance of considering study design when interpreting the diagnostic accuracy of the ROMA index. Fig. 6 A : SROC for cross-sectional studies B : SROC for cohort studies A : SROC for cross-sectional studies B : SROC for cohort studies A comparison of SROC curves based on different ROMA index cut off values showed that the summary point for the 25.1–40 range was closer to the top left corner (0,1) than that for the 10–25 range. This indicates superior diagnostic performance within the 25.1–40 range (4 studies vs. 27 studies) (Fig.  7 A, B). The DOR analysis supported this finding, with a DOR of 75 for the 25.1–40 range compared to 41 for the 10–25 range. The choice of an optimal cut off value is crucial in clinical applications of the ROMA index, as inappropriate thresholds may lead to increased false-positive or false-negative results, affecting patient management and treatment decisions. Fig. 7 A : SROC based on cut-off (10–25) B : SROC based on cut-off (25.1- 40) A : SROC based on cut-off (10–25) B : SROC based on cut-off (25.1- 40) In this study, the diagnostic value of the ROMA index was evaluated based on a likelihood ratio scattergram. According to this chart, for the application of the ROMA index in postmenopausal Asian women, the summary point was located in the second quartile (Fig.  8 A) because LR + > 10 and LR − < 0.1. Therefore, based on this chart, the ROMA index is useful only for confirming the diagnosis of ovarian cancer in postmenopausal Asian women and not for excluding this disease. For postmenopausal European women, the summary point was in the fourth quartile, indicating that the ROMA index is not useful for confirming or excluding the disease. Based on the study design (cross-sectional and cohort studies), the summary point in cross-sectional studies also fell into the second quartile (Fig.  8 B), suggesting that the ROMA index is useful only for confirming the diagnosis of ovarian cancer in postmenopausal women. In cohort studies, the summary point was located in the fourth quartile, indicating that the ROMA index was not useful for confirming or excluding the disease. Fig. 8 A : likelihood ratio scattergram for postmenopausal Asian women B : likelihood ratio scattergram for cross-sectional studies A : likelihood ratio scattergram for postmenopausal Asian women B : likelihood ratio scattergram for cross-sectional studies In the Deeks'Funnel Plot Asymmetry Test, the p -value was 0.30, indicating the absence of publication bias in this study (Additional file 3).

Ovarian cancer is a malignant and aggressive cancer often referred to as the"silent killer"of women worldwide [ 11 ]. The risk of ovarian malignancy algorithm (ROMA) index is an innovative approach for diagnosing ovarian cancer that combines CA125 and HE4 levels with menopausal status [ 8 , 9 ]. Numerous studies have been conducted to evaluate the diagnostic value of the ROMA index; however, these studies often yield debatable results owing to variations in study designs and heterogeneous findings. Therefore, we conducted this meta-analysis to estimate the diagnostic performance of the ROMA index for ovarian cancer. In this meta-analysis, the ROMA index demonstrated a sensitivity of 0.89, specificity of 0.89, diagnostic odds ratio (DOR) of 68, positive likelihood ratio (LR +) of 8, and negative likelihood ratio (LR-) of 0.119, with the inverse of LR- (1/LR-) equal to 8. These values indicate that the ROMA index is highly effective in diagnosing ovarian cancer. Owing to significant heterogeneity across the included studies, we applied a bivariate random effects model to minimize potential bias arising from this variability. To further investigate the sources of heterogeneity, subgroup analyses were conducted based on continent (Asia and Europe), study design (cohort and cross-sectional), and cutoff value ranges (cutoff: 10–25 vs. cutoff: 25.1–40). The results indicated that the ROMA index exhibited higher diagnostic accuracy in postmenopausal Asian women than in postmenopausal European women (DOR = 13 vs. DOR = 6). Additionally, cross-sectional studies provided higher estimates of the diagnostic value compared than cohort studies (DOR = 129 vs. DOR = 77). Regarding cutoff values, the 25.1–40 range showed better diagnostic performance than the 10–25 range (DOR = 76 vs. DOR = 42). Our analysis was based on the results of studies conducted in China, Vietnam, Oman, Iran, Korea, Singapore, Taiwan, Malaysia, Japan, Thailand, Turkey, the Netherlands, Italy, Sweden, Denmark, Belgium, Poland, Croatia, Spain, Portugal, the Czech Republic, and Australia. However, the majority of studies from Asia have focused on China and Iran, which could potentially influence the results of our analysis. However, these two countries are geographically large and represent both distant and nearby Asia. Despite these limitations, this meta-analysis included all the available evidence. Moreover, the number of studies included in this meta-analysis was higher than that in previous meta-analyses. According to the findings, which suggest that the ROMA index is more effective in postmenopausal Asian women than in postmenopausal European women, the best explanation is that the effectiveness and positive predictive value (PPV) of screening tests are enhanced when the incidence of the disease increases [ 55 , 56 ]. The overall incidence of cancer in Asian countries, including China [ 57 ] and Iran [ 58 – 60 ], is increasing. Given the issue of underreporting, the actual prevalence is likely higher than that reflected in the official statistics [ 61 ]. Consequently, the performance and utility of diagnostic tests are expected to be higher in such populations. As the incidence of ovarian cancer is higher in Asian countries than in European countries [ 62 ], this may explain why the ROMA index shows better performance in postmenopausal Asian women. The 5-year survival rate in countries like China, which is the focus of most studies in this analysis, ranges from 40 to 49% [ 62 ]. This increase in survival rates compared with other regions has led to an increase in the prevalence of ovarian cancer, ultimately increasing the positive predictive value of the test [ 63 , 64 ]. Additionally, other factors, such as improved diagnostic capabilities within the healthcare systems of these countries, may contribute to increased detection rates of ovarian cancer, ultimately leading to an increase in the reported incidence. Social Determinants of Health encompass non-medical factors that play a crucial role in shaping health outcomes and directly impact health equity [ 65 ]. Among these, socioeconomic status is considered a key indicator of SDH, encompassing factors such as financial resources, education level, and access to healthcare services. Disparities in SES can contribute to inequalities in cancer diagnosis and prognosis, influencing both the accessibility of early detection and the effectiveness of treatment interventions [ 66 ]. Postmenopausal Asian women with a lower socioeconomic status may have limited access to routine screening programs, leading to the diagnosis of ovarian cancer at more advanced stages. At these stages, the Risk of Ovarian Malignancy Algorithm (ROMA) index may have a higher diagnostic value owing to the larger tumor burden and increased detectability. In contrast, postmenopausal European women with a higher socioeconomic status are more likely to have access to regular screening programs, resulting in earlier detection of malignancies. In such cases, the sensitivity of the ROMA index may be reduced because of the detection of tumors at earlier stages with lower biomarker expression [ 67 ]. Genetic factors, particularly mutations in the BRCA1 and BRCA2 genes, are associated with ovarian cancer susceptibility [ 68 ]. These mutations may vary across populations in different continents, potentially contributing to differences in the diagnostic performance of the ROMA index. Further extensive studies are required to explore this variability and its clinical implications. The strengths of this meta-analysis include the inclusion of 34 studies, which provided greater diversity than previous studies. Additionally, subgroup analyses based on continent, study design, and cutoff values, which were not performed in any prior meta-analysis of the ROMA index, provide valuable insights. Moreover, this study utilized the"midas"and"metandi"commands, tools commonly used to assess diagnostic test accuracy, whereas earlier meta-analyses conducted in 2012 [ 69 ], 2014 [ 70 ], 2016 [ 71 ], 2019 [ 72 ] and 2021 [ 73 ] employed the"metaDisc"command. The commands used in earlier studies were less robust and yielded more limited outputs, whereas those applied in the current study allowed for the analysis of multiple dimensions of the problem, providing more accurate results. In previous studies, the highest reported estimate of the ROMA index was obtained in a 2021 study (DOR = 44), whereas in the present study, a higher estimate of the ROMA index was obtained, particularly in cross-sectional studies (DOR = 129).

However, this study had several limitations that must be considered when interpreting the results. A primary challenge was limited access to full-text articles, which restricted comprehensive subgroup analyses for more precise evaluation of ROMA's diagnostic performance across key variables such as different study methodologies, disease stages (early vs. advanced), and histological tumor types (e.g., serous, endometrioid, mucinous, and clear cell carcinomas). Additionally, limitations in the available data within the included studies restricted further subgroup analyses in areas such as lifestyle factors and precise age classification. In this meta-analysis, the researchers utilized all available scientific evidence from reputable databases to assess ROMA's diagnostic accuracy with maximal precision. Nevertheless, a concerning finding was the paucity of adequate research on this index's diagnostic value in many countries worldwide. This data gap may prevent a comprehensive understanding of ROMA's actual performance across diverse populations. For instance, in the subgroup analysis by continent (Asia vs. Europe), more studies originated from Asia (18 vs. 15). Similarly, in the subgroup analysis based on study design, the majority were cohort studies (27 vs. 4 cross-sectional studies). In terms of cutoff values, more studies used a cutoff range of 25.1–40 than 10–25 (27 vs. 4). This imbalance in the number of studies within the subgroups may affect the accuracy of the results because a smaller number of studies leads to reduced sample sizes and decreased precision of estimates, which in turn may limit the generalizability of the findings to global populations. To address these limitations, future multicenter prospective studies involving medical centers in various countries with racial, ethnic, and geographic diversity are essential. Such studies would not only provide more comprehensive data on ROMA's diagnostic accuracy across populations but also enable evaluation of potential confounding factors, including genetic variations, lifestyle factors, and healthcare access disparities. Expanding this research could advance clinical knowledge in several ways. First, it significantly enhanced the external validity of the findings. Second, it would facilitate a more accurate identification of patient subgroups that would benefit most from this diagnostic index. Third, the resulting data can provide a robust scientific foundation for refining and optimizing current diagnostic algorithms, including the ROMA index. Ultimately, these advancements could lead to the development of more precise and personalized clinical guidelines that improve diagnostic and therapeutic decision-making globally, thereby significantly enhancing the quality of care for patients with ovarian cancer. This achievement would be particularly valuable for developing countries facing greater diagnostic challenges. Based on analyses conducted using studies on the ROMA index and employing subgroup analysis and likelihood ratio scattergrams, it was concluded that the ROMA index demonstrates higher diagnostic performance in postmenopausal Asian women than in postmenopausal European women. Furthermore, the ROMA index with a cut-off (25.1–40) showed better performance than the cut-off (10–25). In addition, cross-sectional studies have provided higher estimates than cohort studies.

Ovarian cancer is the eighth most common cancer in women, the fifth leading cause of cancer-related death among women, and the second most prevalent malignancy in Europe and the United States [ 1 – 3 ]. Notably, it is recognized as the deadliest cancer of the female reproductive system [ 4 , 5 ]. The early stage symptoms are non-specific [ 6 , 7 ], and many patients experience gastrointestinal symptoms [ 8 ]. As a result, more than 60% of cases are diagnosed when the disease has advanced significantly [ 8 – 10 ]. Diagnosing ovarian cancer at an advanced stage leads to high mortality [ 6 , 11 ], and is often referred to as the silent killer of women worldwide [ 12 ]. Owing to its poor prognosis, the 5-year survival rate is less than 30% [ 13 , 14 ], and it is estimated that this disease will lead to the death of eight million patients with ovarian cancer between 2022 and 2050 [ 15 ]. This cancer primarily affects women during menopause, with the majority of cases occurring between the ages of 55 and 65 years [ 2 ]. Early diagnosis is crucial to improve patient survival rates [ 7 , 14 , 16 ] and leads to improved survival rates [ 17 ]. Various methods are available for diagnosing ovarian cancer, including clinical examination, pelvic imaging techniques, and serum biomarkers such as CA125 and HE4 [ 2 , 18 ]. The most commonly used biomarker for diagnosing ovarian cancer is CA125 [ 6 , 19 ]. This serum biomarker has several limitations, including elevated levels in various benign conditions in women, such as endometriosis, myoma, ovarian cysts, and pelvic inflammatory disease, as well as in non-female malignancies, including colon, pancreatic, breast, bladder, and lung cancers, and physiological conditions such as pregnancy and menstruation [ 19 – 21 ]. Additionally, the sensitivity and specificity of this biomarker are relatively low in the early stages of the disease, and its performance is better in postmenopausal than in premenopausal women [ 19 , 22 ]. Another biomarker, HE4, has been clinically utilized since 2003 and has demonstrated higher sensitivity (90% vs. 83%) and specificity (95% vs. 85%) than CA125 in identifying ovarian cancer, particularly in the early stages of the disease in premenopausal women [ 8 , 19 ]. This biomarker was approved by the US FDA for the diagnosis of ovarian cancer [ 10 , 23 ]. However, HE4 also has several limitations, including the fact that its levels gradually increase with age, which complicates the establishment of a reliable reference range, as well as an increase in cases of lung cancer and kidney failure [ 8 , 22 ]. In 2009, Moro et al. introduced a new index, ROMA, designed to enhance the diagnostic process by combining existing diagnostic methods [ 9 , 13 , 22 , 23 ]. The ROMA index is a composite of CA125 and HE4 levels along with menopausal status [ 8 , 9 ]. The ROMA index has a significant ability to differentiate ovarian cancer from benign ovarian tumors [ 5 ]. Women with a higher ROMA index are at an increased risk of ovarian cancer and should be referred to an oncology specialist for effective treatment [ 7 , 8 , 12 ]. Given the importance of early diagnosis of ovarian cancer, the heterogeneity in the results of primary studies, and the methodological limitations of some published meta-analyses on this topic, the present study utilized the most accurate statistical tools and controlled for confounders through subgroup analysis to assess the diagnostic value of the ROMA index in predicting ovarian cancer in postmenopausal women.

Additional file 1. Search strategy. Additional file 2. The Grading of Recommendations, Assessment, Development and Evaluationapproach was applied based on the characteristics of the studies included in meta-analysis. Additional file 3: Fig. 1. Deeks’Funnel Plot Asymmetry Test to Assess Publication Bias. Additional file 1. Search strategy. Additional file 2. The Grading of Recommendations, Assessment, Development and Evaluationapproach was applied based on the characteristics of the studies included in meta-analysis. Additional file 3: Fig. 1. Deeks’Funnel Plot Asymmetry Test to Assess Publication Bias.