Diagnostic value of combined serum CEA and CA19-9 in colorectal cancer A meta-analysis Authors: Liushaoqiu Zhou, Songlin An, Yuanzhong Zhou, Yonghong Ma, Guanyu Li, Tao Zhang, Daimin Xiao Published: 2026 ABSTRACT SummaryColorectal cancer (CRC) is a leading cause of cancer-related mortality worldwide, and early detection remains critical for improving patient outcomes. Serum tumor markers carcino-embryonic antigen (CEA) and carbohydrate antigen 19-9 (CA19-9) are widely used in CRC diagnosis, yet their individual performance is limited. This meta-analysis systematically evaluated the diagnostic performance of CEA, CA19-9, and their combination across 95 studies (22,821 subjects). The combined CEA+CA19-9 testing achieved a pooled sensitivity of 0.585, specificity of 0.877, and diagnostic odds ratio of 9.99—representing improvements of 74% and 138% over CEA and CA19-9 alone, respectively. The area under the summary receiver operating characteristic (SROC) curve reached 0.783 for combined testing. Clinical utility analyses using Fagan’s nomogram and likelihood ratio scatterplots confirmed the superiority of the multi-marker strategy. These findings provide evidence-based support for implementing combined tumor marker testing in clinical CRC diagnosis.Subject areas: Oncology Introduction Colorectal cancer (CRC) ranks among the most prevalent cancers globally, with an incidence of 1–2 million new cases and approximately 700,000 fatalities annually. It follows lung, liver, and stomach cancers, thereby positioning itself as the third most common cancer and the fourth leading cause of cancer-related mortality. In terms of gender distribution, CRC is the second most prevalent cancer in women (9.2%) and the third most prevalent in men (10%). CRC encompasses a spectrum of tumors that arise from a confluence of somatic mutations, gene fusions, genetic instability, epigenetic alterations, and various other factors.1,2,3 Notably, around 70% of metastatic CRC originate from adenomatous polyps, while 25–30% are attributed to sessile serrated lesions.4 CRC is influenced by a multitude of risk factors, encompassing environmental and dietary elements, personal habits, and familial genetics.5 This complexity renders primary prevention through the elimination of causative factors particularly challenging. Research indicates that CRC typically progresses through a sequence of stages: adenoma, atypical hyperplasia, and carcinoma, with approximately 85% of CRCs originating from adenomas. The transition from adenomas to stage IV invasive cancer generally spans a period of 10–15 years, and it is noteworthy that 75% of CRCs can be averted through the endoscopic removal of adenomas.6 However, the invasive nature, intricacy, and challenges associated with bowel endoscopic surgery limit the overall benefits of such procedures for the population in terms of preclinical CRC prevention. Clinicians have observed that the 5-year survival rate for stage I CRC can reach as high as 90%, whereas it plummets to below 8% at stage IV.7 Given that the progression from epithelial hyperplasia to carcinoma occurs over a relatively extended time frame, there exists a significant window of opportunity for clinical intervention.8,9,10 Consequently, early screening, diagnosis, and treatment of CRC are crucial for enhancing the long-term prognosis of affected individuals. There are several methods available for the detection of CRC, with the most commonly employed screening techniques including fecal testing, imaging, endoscopy, and tumor markers. However, fecal testing is characterized by low specificity, imaging incurs significant costs, and endoscopy often suffers from poor patient compliance. Furthermore, the presence of clinical symptoms indicates that CRC may have progressed to a malignant stage, highlighting the urgent need for effective early detection methods for this disease.11 CRC tumor markers are highly predictive and can identify changes in the content of colon tumors prior to their manifestation, thereby providing a foundation for the early diagnosis of colon tumors. Additionally, as a novel biomarker approach, this method is economical, convenient, effective, non-invasive, and possesses other advantages that enhance its acceptability among patients.12 Currently, serum tumor markers such as carcino-embryonic antigen (CEA) and carbohydrate antigen 19-9 (CA19-9) are the most frequently utilized laboratory diagnostic tools and have been extensively applied in the early diagnosis and prognosis of CRC13,14,15 Nevertheless, in clinical practice, reliance on a single serum tumor marker does not suffice for the early detection of CRC.16,17 The pursuit of tumor markers with high specificity and sensitivity, along with their combination with other tumor markers for the early detection of CRC, represents two significant trends in the advancement of diagnostic and therapeutic technologies for this malignancy. However, the clinical diagnostic value of CEA, CA19-9, and their combination for CRC remains inconclusive; some studies18,19,20 have indicated that these markers do not enhance the sensitivity of diagnostic tests, while others have suggested that they may improve the accuracy and sensitivity of CRC diagnosis. Therefore, this study aims to perform a meta-analysis to assess the diagnostic efficacy of the combined use of CEA and CA19-9 in the early detection of CRC. Differences from previous meta-analyses include: (1) this study is the first to simultaneously include 95 studies (the largest previous sample size was 68 studies), covering populations from multiple regions including Asia, Europe, and America; (2) it is the first to combine the HSROC model with clinical utility analysis (Fagan nomogram, decision curve), rather than merely reporting pooled effect sizes. Therefore, this study aims to perform a meta-analysis to assess the diagnostic efficacy of the combined use of CEA and CA19-9 in the early detection of CRC. Results Literature search results Based on the established search strategy, a total of 7,024 documents were retrieved. Following a systematic screening process in accordance with the specified criteria, 95 documents were selected for final inclusion, all of which were case-control studies. The flowchart illustrating the literature screening process is presented in Figure 1. Literature review flowchart Literature characteristics The fundamental characteristics of the literature included in this review are presented in Tables 1 and 2. The studies encompassed a total of 22,821 subjects. In total, 95 articles were included, comprising 65 articles in Chinese and 30 articles in English. Table 1Characteristics of individual tumor marker assaysLiterature inclusionCEACA19-9MethodsThreshold (ng/mL)TP/FP/FN/TnMethodsThreshold (ng/mL)TP/FP/FN/TnBhawna Bagaria et al. 201321ELISA3.438/0/12/50ELISA3513/0/37/50P Fan et al. 201722ELISA1045/101/70/239ELISA2755/105/60/235L.Fermindez-Fermindez et al. 199523CLIA5.659/4/68/66CLIA5126/4/101/66Juan-Juan Gao et al. 202224ELISA556/4/54/87ELISA3917/3/93/88Yanfeng Gao et al. 201825ELISA3.4130/15/149/59ELISA3941/8/238/66Cuijuan Hao et al. 202126CLIA590/12/66/40CLIA3781/11/75/41László Herszényi et al. 200827ELISA416/4/40/31ELISA3710/3/46/32Shenghuai Hou et al. 202328ELISA3509/18/471/852ELISA20343/79/637/791Zhengyuan Huang et al. 202229CLIA579/1/148/122CLIA3736/0/191/123P Kuusela et al. 199130RIA343/6/34/24RIA3726/0/51/30Nanyang Li et al. 202331ECLI521/7/9/23ECLI3521/7/9/23Pingxia Lu et al. 202232ECLI5270/5/346/117ECLI3792/5/524/117Franco Lumachi et al. 201233ELISA4.944/10/58/89ELISA32.622/72/80/27Barbara Mroczko et al. 200734MEIA551/0/81/65MEIA3025/0/107/65Shufang Ning et al.201835ECLI6.5108/7/59/68ECLI37103/9/64/66Jianhong Peng et al. 201836ECLI5173/7/253/87ECLI3584/2/341/92F Safi et al. 198737RIA361/50/32/186RIA3749/15/44/221Cheng Shi et al. 201938ECLI1.83174/51/37/52ECLI11132/72/79/31Swarnima Singh et al. 202039ECLI6.534/6/19/47ECLI3732/6/21/47Yanfang Song et al. 201840ECLI5321/0/462/331ECLI37133/0/650/331Y F Song et al. 201741ELISA558/7/79/165ELISA3727/4/110/168Ratko Tomašević et al. 201642CLIA5139/45/42/146CLIA35.4125/51/56/140Chun-Xiao Wang et al. 202143ECLI5.260/4/76/46ECLI3719/1/117/49Jia Wang et al. 201444ECLI6.540/5/73/84ECLI3925/4/88/85Li Xie et al. 201845ECLI5.243/47/80/78ELISA3922/56/101/69Weili Yang et al. 201846CLIA520/20/30/30CLIA3727/4/23/46Dongdong Yu et al. 202147CLIA549/10/84/269CLIA3716/0/117/279Yixiang Zhan et al. 202348ECLI5164/385/281/57ECLI39418/18/27/424Shiyan Zhang et al. 201549ECLI589/12/49/99ECLI3765/11/73/100Xianwen Zhao et al. 200550ELISA2.3873/7/61/193ELISA27.3945/15/89/185Min Ai et al. 202051ECLI558/37/17/38ECLI3429/30/46/45Ji Ban et al. 202052ELISA583/20/17/81ELISA3578/19/22/82Shangdang Cai et al. 201653ECLI5.1428/4/18/33ECLI37.0117/7/29/30Lihui Cao et al. 201954CLIA531/15/19/35CLIA3734/18/16/32Yaping Cao et al. 202155ECLI5576/89/329/335ECLI35359/52/546/372Longfei Zeng et al. 201756ECLI6152/69/79/202ECLI37148/28/83/243Junfeng Chang et al. 201457CLIA3.445/7/25/63CLIA2724/8/46/62Bin Chen et al. 201058ECLI568/2/129/64ECLI3721/2/186/64Deyu Chen et al. 200859MEIA535/2/67/38MEIA3738/1/64/39Feiying Chen et al. 201760CLIA1078/23/79/144CLIA3775/26/82/141Jie Chen et al. 200561MEIA522/11/26/21MEIA3718/12/30/20Jinjin Chen et al. 202162CLIA557/25/23/140CLIA3748/16/32/149Kaijie Chen et al. 200563MEIA521/10/24/20MEIA3717/11/28/19Liwan Chen et al. 202164CLIA552/34/53/66CLIA3551/33/54/67Shuying Chen et al. 201765ECLI2.4164/1/17/29ECLI15.0551/0/30/30Yanping Chen et al. 201766CLIA546/6/54/94CLIA3940/15/60/85Yuxue Deng et al. 201767ECLI4.323/9/27/62ECLI2719/19/31/52Yujiao Ding et al. 202368ECLI624/11/6/19ECLI3721/9/9/21Jinlang Du et al.201469ECLI543/4/48/29ECLI3559/6/32/27Ping Fan et al. 201770ECLI1045/101/70/239ECLI2755/105/60/235Xiaojing Fan et al. 201971ECLI526/29/24/76ECLI3723/32/27/73Chenxi Feng et al. 202272ECLI548/12/27/29ECLI3548/18/27/23Yixi Feng et al.201773CLIA521/2/29/48CLIA3513/2/37/48Qunhao Gu et al. 201274MEIA518/4/27/26MEIA3715/3/30/27Yahui Guo et al. 202275ECLI574/38/23/66ECLI3542/13/55/91Sui He et al. 200476ECLI4.626/2/30/28ECLI2720/1/36/29Hualin He et al. 201977ECLI5119/9/60/98ECLI39114/18/65/89Ying Huang et al. 201778CLIA539/39/21/21CLIA3536/18/24/42Mengying Jiang et al. 202379ELISA524/13/18/55ELISA2717/0/25/68Kai Lei et al. 201980ECLI1040/34/33/36ECLI3730/27/43/43Xiong Lei et al. 201181ECLI6.523/6/22/48ECLI2718/4/27/41Chunguang Li et al. 202082CLIA529/2/13/38CLIA3724/3/18/37Peng Li et al. 202283ECLI1060/21/40/29ECLI3759/20/41/30Shijun Li et al. 201084MEIA560/26/98/134MEIA3743/38/115/122Zhiguo Li et al. 202185CLIA1035/8/19/38CLIA3733/7/21/39Yan Liang et al. 200686ECLI3.432/9/45/25ECLI2234/11/43/23Li Lin et al. 202087CLIA1032/4/28/96CLIA3531/7/29/93Chanrong Lu et al. 201288ECLI552/5/39/53ECLI3759/20/32/38Shaolei Luo et al. 201789Protein chip551/3/83/290Protein chip3537/5/97/288Zhongchuan Lv et al. 201390CLIA5.988/17/72/83CLIA3759/22/101/78Junkong Ma et al. 202291ECLI526/4/29/51ECLI2735/10/20/45Lijuan Pan et al. 202092ELISA522/45/23/78ELISA3721/72/24/51Fenghua Pei et al. 201093ELISA413/0/23/183ELISA376/0/30/183Zhengjie Qu et al. 202194ECLI2.80552/5/28/45ECLI3928/10/52/40Mengzhen Shi et al. 202095ECLI3.414/6/20/44ECLI2713/7/21/43Guo Wang et al. 199896RIA1519/21/23/171RIA3726/14/16/178Kun Wang et al. 201797CLIA7.246/3/55/76CLIA2741/3/60/76Yanyan Wang et al. 201998CLIA549/27/29/72CLIA3746/21/32/78Yisheng Wei et al. 200699Protein chip525/1/57/40Protein chip3528/1/54/40Yi Wu et al. 2022100ECLI525/4/40/65ECLI2733/6/32/63Mingyue Xiang.2007101MEIA537/1/42/49MEIA3739/2/40/48Qiong Xu et al. 2022102CLIA540/11/32/61CLIA3527/12/45/60Qinghui Yan et al. 2016103CLIA524/1/12/29CLIA32.425/2/11/28Liping Yang et al. 2020104CLIA3.7374/2/64/48CLIA14.4067/8/71/42Maowen Yu et al. 2018105ECLI5133/30/59/50ECLI39131/26/61/54Guoliang Zhang et al. 2023106CLIA3.1342/8/33/43ECLI12.8334/0/41/51Guo Yu et al. 2011107ECLI517/4/55/31ECLI2725/2/47/33Hua Zhang et al. 2020108MEIA5.9659/3/71/47MEIA21.163/10/67/40Wanzhong Zhang et al. 2007109RIA523/2/7/28RIA3716/0/14/30Ming Zhao et al. 2005110RIA1523/1/26/37RIA3727/2/22/36Ming Zhao et al. 2006111RIA1545/8/11/79RIA3739/9/27/78Aiqing Zhou et al. 2007112ECLI3.453/16/36/91ECLI3751/10/38/97Xujun Zhou et al. 2014113ECLI3.429/3/15/27ECLI2731/2/13/28Meiiqin Zhu et al. 2016114CLIA533/13/28/104CLIA3728/10/33/107Zili Zhu et al. 2006115RIA2032/10/35/45RIA3736/11/31/44CLIA: chemiluminescence immunoassay; ECLI: electrochemiluminescence immunoassay; ELISA: enzyme-linked immunosorbent assay; RIA: radioimmunoassay; MEIA: microbial enzyme immunoassay; protein chip; TP: true positive; FP: false positive; FN: false negative; Tn: true negative. Characteristics of individual tumor marker assays CLIA: chemiluminescence immunoassay; ECLI: electrochemiluminescence immunoassay; ELISA: enzyme-linked immunosorbent assay; RIA: radioimmunoassay; MEIA: microbial enzyme immunoassay; protein chip; TP: true positive; FP: false positive; FN: false negative; Tn: true negative. Table 2Characteristics of combined tumor marker assaysStudy inclusionVintagesCEA+CA19-9(TP/FP/FN/Tn)Bhawna Bagaria.et a21201336/0/14/50P Fan et al.22201766/95/49/245L.Fernández-Fernández et al.23199565/7/62/63Yanfeng Gao et al.252018138/16/141/58László Herszényi et al.27200818/2/38/33Shenghuai Hou et al.282023656/140/324/730Zhengyuan Huang et al.292022169/21/58/102Pingxia Lu et al.322022343/10/273/112Barbara Mroczko et al.34200761/0/71/65F Safi et al.37198770/22/23/214Yanfang Song et al.402018392/10/391/321Y F Song et al.412017121/68/16/104Ratko Tomašević et al.422016133/32/48/159Dongdong Yu et al.47202152/10/81/269Shi-Yan Zhang et al.49201598/19/40/92Xian-wen Zhao et al.50200579/17/55/183Yaping Cao et al.552021560/72/345/352Longfei Zeng et al.562017164/31/67/240Bin Chen et al.58201074/2/120/64Deyu Chen et al.59200849/2/53/38Jinjin Chen et al.62202161/25/19/140Yanping Chen et al.66201757/18/43/82Yuxue Deng et al.67201719/11/31/60Ping Fan et al.70201766/95/49/245Xiaojing Fan et al.71201941/16/9/89Sui He et al.76200430/3/26/27Hualin He et al.772019117/4/62/103Xiong Lei et al.81201132/4/13/50Yan Liang et al.86200634/11/43/23Shaolei Luo et al.89201768/8/66/285Fenghua Pei et al.93201014/0/22/183Yisheng Wei et al.99200639/2/43/39Mingyue Xiang et al.101200746/3/33/47Guo Yu et al.107201123/6/49/29Ming Zhao et al.110200538/3/11/35Aiqing Zhou et al.112200766/21/23/86TP: true positive; FP: false positive; FN: false negative; Tn: true negative. Characteristics of combined tumor marker assays TP: true positive; FP: false positive; FN: false negative; Tn: true negative. Risk of bias assessment Risk of bias assessment was performed using the QUADAS-2 tool across four domains: patient selection, index test, reference standard, and flow and timing. The results are presented in Table 3. The majority of studies were rated as low risk across all domains. High risk of bias in patient selection was noted in 13 studies, while unclear risk was noted in the gold standard domain for several earlier publications. All studies were consistently rated as low risk for the index test domain. Table 3Risk of bias assessment of included studiesStudy inclusionAssessment of biasSuitabilityCase selectionTests to be evaluatedGold StandardCase flow and progressCase selectionTests to be evaluatedGold StandardBhawna Bagaria et al. 201321low risklow risklow risklow risklow risklow risklow riskP Fan et al. 201722low risklow risklow risklow risklow risklow risklow riskL.Fermindez-Fermindez et al. 199523low risklow riskunknownlow risklow risklow riskunknownJuan-Juan Gao et al. 202224low risklow risklow risklow risklow risklow risklow riskYanfeng Gao et al. 201825unknownlow riskunknownunknownunknownlow riskunknownCuijuan Hao et al. 202126low risklow risklow riskunknownlow risklow risklow riskLászló Herszényi et al. 200827high risklow risklow riskunknownlow risklow risklow riskShenghuai Hou et al. 202328low risklow riskunknownunknownlow risklow riskunknownZhengyuan Huang et al. 202229low risklow riskunknownunknownlow risklow riskunknownP Kuusela et al. 199130high risklow riskunknownunknownhigh risklow riskunknownNanyang Li et al.202331low risklow risklow risklow risklow risklow risklow riskPingxia Lu et al. 202232low risklow risklow risklow risklow risklow risklow riskFranco Lumachi et al. 201233unknownlow riskunknownunknownlow risklow riskunknownBarbara Mroczko et al. 200734unknownlow riskunknownunknownunknownlow riskunknownShufang Ning et al. 201835low risklow riskunknownunknownlow risklow riskunknownJianhong Peng et al. 201836unknownlow riskunknownunknownunknownlow riskunknownF Safi et al. 198737unknownlow riskunknownunknownunknownlow riskunknownCheng Shi et al. 201938low risklow risklow risklow risklow risklow risklow riskSwarnima Singh et al. 202039unknownlow riskunknownunknownunknownlow riskunknownYanfang Song et al. 201840low risklow riskunknownunknownlow risklow riskunknownY F Song et al. 201741high risklow riskunknownunknownunknownlow riskunknownRatko Tomašević et al. 201642low risklow risklow risklow risklow risklow risklow riskChun-Xiao Wang et al. 202143unknownlow riskunknownunknownlow risklow riskunknownJia Wang et al. 201444low risklow risklow risklow risklow risklow risklow riskLi Xie et al. 201845low risklow riskunknownunknownlow risklow riskunknownWeili Yang et al. 201846low risklow risklow risklow risklow risklow risklow riskDongdong Yu et al. 202147low risklow risklow risklow risklow risklow risklow riskYixiang Zhan et al. 202348unknownlow risklow risklow risklow risklow risklow riskShi-Yan Zhang et al. 201549low risklow riskhigh risklow risklow risklow riskHigh riskXian-wen Zhao et al. 200550low risklow riskunknownunknownlow risklow riskunknownMin Ai et al. 202051low risklow risklow risklow risklow risklow risklow riskJi Ban et al. 202052unknownlow riskunknownunknownlow risklow riskunknownShangdang Cai et al. 201653low risklow riskunknownunknownlow risklow riskunknownLihui Cao et al. 201954low risklow riskunknownunknownlow risklow riskunknownYaping Cao et al. 202155unknownlow risklow risklow risklow risklow risklow riskLongfei Zeng et al. 201756unknownlow risklow risklow risklow risklow risklow riskJunfeng Chang et al. 201457low risklow risklow risklow risklow risklow risklow riskBin Chen et al. 201058low risklow risklow risklow risklow risklow risklow riskDeYu Chen et al. 200859unknownlow risklow risklow risklow risklow risklow riskFeiying Chen et al. 201760high risklow risklow risklow risklow risklow risklow riskJie Chen et al. 200561unknownlow risklow risklow risklow risklow risklow riskJinjin Chen et al. 202162unknownlow risklow risklow risklow risklow risklow riskKaijie Chen et al. 200563low risklow risklow risklow risklow risklow risklow riskLiwan Chen et al. 202164low risklow riskhigh riskunknownlow risklow riskhigh riskShuying Chen et al. 201765low risklow risklow risklow risklow risklow risklow riskYanping Chen et al. 201766low risklow risklow risklow risklow risklow risklow riskYuxue Deng et al. 201767low risklow risklow risklow risklow risklow risklow riskYujiao Ding et al. 202368low risklow risklow risklow risklow risklow risklow riskJinlang Du et al. 201469low risklow risklow risklow risklow risklow risklow riskPing Fan et al. 201770low risklow risklow risklow risklow risklow risklow riskXiaojing Fan et al. 201971low risklow risklow risklow risklow risklow risklow riskChenxi Feng et al. 202272low risklow risklow risklow risklow risklow risklow riskYixi Feng et al. 201773low risklow riskunknownunknownlow risklow riskunknownQunhao Gu et al. 201274low risklow risklow risklow risklow risklow risklow riskYahui Guo et al. 202275low risklow risklow risklow risklow risklow risklow riskSui He et al. 200476low risklow risklow risklow risklow risklow risklow riskHualin He et al. 201977low risklow risklow risklow risklow risklow risklow riskYing Huang et al. 201778low risklow risklow risklow risklow risklow risklow riskMengying Jiang et al. 202379low risklow risklow risklow risklow risklow risklow riskKai Lei et al. 201980low risklow risklow risklow risklow risklow risklow riskXiong Lei et al. 201181low risklow risklow risklow risklow risklow risklow riskChunguang Li et al. 202082low risklow risklow risklow risklow risklow risklow riskPeng Li et al. 202283low risklow risklow risklow risklow risklow risklow riskShijun Li et al. 201084unknownlow risklow risklow risklow risklow risklow riskZhiguo Li et al. 202185low risklow risklow risklow risklow risklow risklow riskYan Liang et al. 200686high risklow risklow risklow risklow risklow risklow riskLi Lin et al. 202087low risklow risklow risklow risklow risklow risklow riskChanrong Lu et al. 201288low risklow risklow risklow risklow risklow risklow riskShaolei Luo et al. 201789low risklow risklow risklow risklow risklow risklow riskZhongchuan Lv et al. 201390low risklow risklow risklow risklow risklow risklow riskJunkong Ma et al. 202291low risklow riskunknownunknownlow risklow riskunknownLijuan Pan et al. 202092low risklow risklow risklow risklow risklow risklow riskFenghua Pei et al. 201093low risklow risklow risklow risklow risklow risklow riskZhengjie Qu et al. 202194high risklow risklow risklow risklow risklow risklow riskMengzhen Shi et al. 202095low risklow risklow risklow risklow risklow risklow riskGuo Wang et al. 199896high risklow risklow risklow risklow risklow risklow riskKun Wang et al. 201797low risklow risklow risklow risklow risklow risklow riskYanyan Wang et al. 201998low risklow risklow risklow risklow risklow risklow riskYisheng Wei et al. 200699low risklow risklow risklow risklow risklow risklow riskYi Wu et al. 2022100high risklow risklow risklow risklow risklow risklow riskMingyue Xiang.2007101low risklow risklow risklow risklow risklow risklow riskQiong Xu et al. 2022102low risklow risklow risklow risklow risklow risklow riskQinghui Yan et al. 2016103low risklow risklow risklow risklow risklow risklow riskLiping Yang et al. 2020104low risklow risklow risklow risklow risklow risklow riskMaowen Yu et al. 2018105low risklow risklow risklow risklow risklow risklow riskGuoliang Zhang et al. 2023106low risklow risklow risklow risklow risklow risklow riskGuo Yu et al. 2011107low risklow risklow risklow risklow risklow risklow riskHua Zhang et al. 2020108low risklow risklow risklow risklow risklow risklow riskWanzhong Zhang et al. 2007109high risklow risklow risklow risklow risklow risklow riskMing Zhao et al. 2005110high risklow riskunknownunknownlow risklow riskunknownMing Zhao et al. 2006111high risklow risklow risklow risklow risklow risklow riskAiqing Zhou et al. 2007112high risklow risklow risklow risklow risklow risklow riskXujun Zhou et al. 2014113low risklow risklow risklow risklow risklow risklow riskMeiqin Zhu et al. 2016114high risklow risklow risklow risklow risklow risklow riskZili Zhu et al. 2006115low risklow risklow risklow risklow risklow risklow risk Risk of bias assessment of included studies Main research findings Quantitative analysis indicated that the diagnostic odds ratio (DOR) for the combined test reached 9.99, reflecting significant increases of 74% and 138% compared to CEA alone (5.73) and CA19-9 alone (4.19), respectively. The results of the heterogeneity assessment consistently demonstrated high heterogeneity characteristics among the studies included in the meta-analysis. The inter-study heterogeneity indicators for all three testing strategies significantly surpassed critical thresholds: I2 values were recorded at 91.2% for CEA testing (95 studies), 91.4% for CA19-9 testing (also 95 studies), and 91.2% for the combined CEA+CA19-9 testing (36 studies). Analysis of sample characteristics revealed that individual studies on CEA and CA19-9 testing had substantial sample sizes, with case groups comprising 12,514 and 12,533 subjects, and control groups consisting of 10,307 and 10,298 subjects, respectively. In contrast, studies involving combined detection exhibited relatively smaller sample sizes, with 6,906 cases and 5,823 controls (Figure 2). Heterogeneity analysis An in-depth analysis of the performance characteristics across multiple diagnostic dimensions revealed that, in terms of sensitivity, the combination of CEA and CA19-9 exhibited optimal performance, with a pooled sensitivity of 0.585 (95% CI: 0.539–0.634). This represents a 10.0% improvement over CEA alone (0.532, 95% CI: 0.506–0.558) and a substantial 32.7% enhancement compared to CA19-9 alone (0.441, 95% CI: 0.408–0.476). The evaluation of specificity similarly demonstrated the advantages of combined testing, yielding a specificity of 0.877 (95% CI: 0.845–0.908). This indicates a 5.2% improvement over CEA alone (0.834, 95% CI: 0.801–0.862) and a 4.3% enhancement compared to CA19-9 alone (0.841, 95% CI: 0.809–0.871). Analyses of the likelihood ratio and DOR further substantiated the clinical value of the combined testing strategy. The positive likelihood ratio of the CEA and CA19-9 combination reached 4.73, significantly surpassing that of CEA alone (3.21) and CA19-9 alone (2.78), indicating that a positive result from the combined test provides stronger confirmatory power for disease diagnosis. Concurrently, the negative likelihood ratio of the combined test decreased to 0.47, lower than that of CEA (0.56) and CA19-9 (0.66), demonstrating a higher reliability of negative results in ruling out the disease. The DOR, serving as a comprehensive indicator of diagnostic performance, clearly illustrates the significant advantages of combined testing. The DOR for the CEA+CA19-9 combination reached 9.99, indicating a 74.3% improvement over CEA alone (5.73) and a 138.4% enhancement compared to CA19-9 alone (4.19). The area under the receiver operating characteristic curve (AUC) analysis indicated that the combined testing exhibited an AUC of 0.581 (95% CI: 0.500–0.581), representing an improvement relative to CEA alone (0.482, 95% CI: 0.462–0.503) and CA19-9 alone (0.444, 95% CI: 0.391–0.444), further validating the efficacy of the multi-marker combination strategy in enhancing diagnostic performance (Table 4). Table 4Pooled diagnostic performance of the three testing strategiesTesting StrategyCEACA19-9CEA+CA19-9Number of Studies959536Total Cases12514125336906Total Controls10307102985823Pooled Sensitivity0.5320.4410.585Sensitivity 95% CI[0.506,0.558][0.408,0.476][0.539,0.634]Pooled Specificity0.8340.8410.877Specificity 95% CI[0.801,0.862][0.809,0.871][0.845,0.908]Positive Likelihood Ratio3.212.784.73Negative Likelihood Ratio0.560.660.47Diagnostic Odds Ratio5.734.199.99AUC 95%CI[0.462,0.503][0.391,0.444][0.500,0.581] Pooled diagnostic performance of the three testing strategies Forest plot analysis The forest plot analysis of CA19-9 testing (Figure 3) elucidates the intricate performance characteristics of this biomarker in the diagnosis of CRC. In terms of sensitivity, the 95 included studies exhibit significant inter-study variability, with individual study sensitivity estimates ranging from approximately 0.15 to 0.85. The pooled sensitivity is calculated at 0.565 (95% CI); however, the degree of overlap in the confidence intervals suggests statistical inconsistency among certain study results. Notably, approximately 15–20% of the studies report CA19-9 sensitivity below 0.3, which may be attributable to variations in patient tumor staging, histological types, or detection threshold settings across different investigations. In regard to specificity, CA19-9 testing demonstrates relatively better consistency, with the majority of studies’ specificity estimates concentrated within the 0.7–0.95 range, yielding a pooled specificity of 0.877 (95% CI). Forest plots for CEA and CA19-9 The forest plot for CEA testing (Figure 3) reveals distribution characteristics that are similar to yet distinct from those of CA19-9. In the sensitivity analysis, the individual estimates from the 95 studies also exhibit considerable variation, ranging from approximately 0.2 to 0.8; however, the overall distribution appears more concentrated compared to CA19-9, with a pooled sensitivity of 0.585 (95% CI). The proportion of studies reporting low CEA sensitivity (<0.3) is relatively smaller, accounting for approximately 10–12% of the total, which may suggest that CEA, as a traditional biomarker for CRC, maintains relatively better stability across diverse research settings. In terms of specificity, CEA testing demonstrates good inter-study consistency, with the majority of studies’ specificity estimates distributed within the 0.75–0.95 range, resulting in a pooled specificity of 0.834 (95% CI). The forest plot analysis of the combined testing strategy (Figure 4) provides compelling visual evidence for the synergistic effects of multiple biomarkers. An analysis based on 36 studies demonstrates that combined testing exhibits a significant improvement in both sensitivity and specificity compared to individual tests. The sensitivity forest plot reveals that the individual study estimates for combined testing are predominantly distributed within the 0.45–0.75 range, yielding a pooled sensitivity of 0.585 (95% CI). Notably, the proportion of studies with low sensitivity (<0.4) decreases significantly to below 5%. In the evaluation of specificity, combined testing demonstrates optimal performance, with individual study specificity estimates primarily concentrated in the 0.8–0.95 range, and a pooled specificity increasing to 0.877 (95% CI). Forest plot for the combined detection of CEA and CA19-9 Summary receiver operating characteristic curve analysis The summary receiver operating characteristic (SROC) curve for CEA testing indicates a moderate-to-low level of diagnostic accuracy, with an AUC of 0.662 (95% CI: 0.618–0.763). In contrast, CA19-9 testing exhibits relatively superior SROC performance, with the AUC rising to 0.725 (95% CI: 0.670–0.780). Furthermore, the SROC curve for the combined testing of CEA and CA19-9 demonstrates optimal diagnostic performance characteristics, achieving an AUC of 0.783 (95% CI: 0.760–0.807), which represents a substantial improvement over individual marker testing (Figure 5). ROC curves Clinical utility assessment Fagan’s nomogram analysis elucidates the clinical conversion efficiency of three testing strategies across various pre-test probability scenarios. The nomogram for CA19-9 testing indicates that in typical CRC screening settings (with an approximate pre-test probability of 10%), a positive result can elevate the post-test probability to approximately 25–30%, whereas a negative result can diminish the probability to around 5–7%. CEA testing exhibits clinical utility characteristics that are similar to, yet slightly enhanced over, those of CA19-9. Under identical pre-test probability conditions, a positive CEA result can increase the post-test probability to about 30–35%, while a negative result can reduce the probability to approximately 6–8%. The Fagan’s nomogram for the combined CEA+CA19-9 testing illustrates a significantly enhanced clinical utility. Under the same pre-test probability conditions, a positive combined test result can elevate the post-test probability to approximately 40–45%, thereby reaching the clinical threshold for high suspicion, while a negative result can reduce the probability to about 3–5%, thereby approaching the clinical standard for effectively ruling out the condition (Figure 6). Fagan’s nomogram The likelihood ratio scatterplot (Figure 7) serves as an intuitive visualization tool for evaluating the clinical discriminative value of the three testing strategies. The analysis indicates that the majority of study data points are situated within the moderate diagnostic value zone (PLR 2–10, NLR 0.1–0.5). Data points pertaining to CEA testing are predominantly clustered within the PLR 2–6 and NLR 0.3–0.7 range, reflecting a moderate-to-low clinical discriminative value. CA19-9 testing exhibits a relatively favorable distribution of likelihood ratio characteristics, with the majority of data points situated within the ranges of PLR 3–8 and NLR 0.2–0.5. The combined testing of CEA and CA19-9 reveals optimal distribution characteristics, with data points predominantly concentrated in the high-value region of PLR 4–10 and NLR 0.1–0.4. Notably, approximately 30–40% of studies utilizing combined testing fulfilled the criteria of PLR>6 and NLR<0.3, in contrast to only 10–15% observed in single-marker testing. Likelihood ratio scatterplot Quality assessment CEA testing revealed the broadest distribution span along the horizontal axis (ranging from −2 to +7), indicating a significant degree of inter-study variability. The dispersed distribution of data points suggests that CEA test results are influenced by multiple study design factors, thereby necessitating careful consideration of specific testing conditions and patient characteristics in clinical applications. The distribution of CA19-9 testing is relatively concentrated in the lower right quadrant of the central area; however, it exhibits a broad vertical distribution range (from −2 to +3), indicating considerable inter-study variability in the specificity dimension. The distribution of the combined CEA+CA19-9 testing lies between those of the individual biomarkers, demonstrating a relatively more concentrated distribution pattern (Figure 8). Bivariate boxplot Deeks’ funnel plot analysis (Figure 9) provided statistical evidence for evaluating the risk of publication bias across the three testing strategies. The funnel plot for CA19-9 testing exhibited a generally symmetrical distribution of data points around the regression line, with the slope test P-value suggesting a relatively low risk of publication bias. The funnel plot for CEA testing demonstrates comparable distribution characteristics, with the majority of studies aligning along the regression line. However, a slight asymmetry is noted in the lower region, which may be linked to the selective publication of studies with smaller effect sizes. Deeks’ funnel plot Sensitivity analysis The leave-one-out sensitivity analysis (Figure 10) conducted in this study demonstrated that the pooled sensitivity and specificity estimates for all three testing strategies exhibited considerable robustness, with the exclusion of individual studies showing no substantial impact on the overall conclusions. Notably, the combination of CEA and CA19-9 testing displayed the highest stability in the sensitivity analysis, exhibiting the narrowest range of variation in pooled estimates, thereby providing further support for the reliability of conclusions regarding the combined testing strategy. The sensitivity analysis of CA19-9 and CEA The sensitivity analysis of CA19-9 testing demonstrated robust stability in results. Following the sequential removal of any individual study from the 95 included studies, the variation range of the pooled sensitivity estimates remained relatively narrow (approximately 0.01–0.02). The sensitivity analysis for CEA testing further confirmed the reliability of its diagnostic performance estimates. In the sensitivity dimension, the leave-one-out analysis demonstrated that the variation range of the pooled estimates was maintained within 0.008–0.018. In terms of specificity analysis, CEA testing exhibited superior stability characteristics, with the variation range of the pooled specificity estimates being only 0.003–0.012. The sensitivity analysis results for the combined CEA+CA19-9 testing (Figure 11) are particularly encouraging, demonstrating the most optimal robustness characteristics among the three testing strategies. Although the number of included studies is relatively limited (36 studies), the leave-one-out analysis revealed excellent stability in both pooled sensitivity and specificity estimates. The variation range for sensitivity estimates was only 0.005 to 0.015, while the variation range for specificity estimates was even more constrained within the narrow interval of 0.002–0.008. The sensitivity analysis results for the combined CEA+CA19-9 testing Discussion This study systematically evaluates the performance of individual CEA, CA19-9, and their combined testing in the diagnosis of CRC through a meta-analysis. The results clearly demonstrate that the combined testing of CEA and CA19-9 represents the optimal diagnostic strategy among the three approaches, exhibiting significant advantages over individual biomarker tests across key metrics, including sensitivity, specificity, DOR, and AUC. These findings provide robust evidence-based support for clinical decision-making in the diagnosis of CRC. Quantitative results indicate that the diagnostic advantage of combined testing possesses clear statistical and clinical significance. In terms of sensitivity, the pooled sensitivity of the combined test reached 0.585 (95% CI: 0.539–0.634), representing a 10.0% improvement over CEA alone (0.532) and a 32.7% increase over CA19-9 alone (0.441). This enhancement directly correlates with a reduction in the risk of missed diagnoses in clinical practice, suggesting that the combined test can identify a greater number of early-stage or occult CRC cases. Particularly for patients with earlier tumor stages or low expression of individual markers, the synergistic effect of multiple biomarkers aids in mitigating diagnostic delays caused by false negatives. Regarding specificity, although the improvement of the combined test (0.877, 95% CI: 0.845–0.908) over individual tests is relatively modest (a 5.2% increase over CEA and 4.3% over CA19-9), its significance is substantial in large-scale population screening scenarios. For instance, in screening high-risk populations for CRC, each 1-percentage-point increase in specificity can prevent thousands of false-positive results, thereby avoiding unnecessary invasive procedures such as colonoscopy, reducing both the healthcare burden on patients and psychological anxiety, while conserving medical resources. In terms of comprehensive performance metrics, the combined test achieved a DOR of 9.99, reflecting a 74.3% improvement over CEA (DOR 5.73) and a 138.4% increase over CA19-9 (DOR 4.19). This indicates a significantly enhanced capability to differentiate CRC patients from healthy controls. The SROC curve analysis revealed an AUC of 0.783 (95% CI: 0.760–0.807) for the combined test, which is substantially higher than that of CEA (0.662) and CA19-9 (0.725). Furthermore, the curve approaches the ideal diagnostic zone, maintaining a sensitivity greater than 0.6 even in high-specificity regions (>0.8), effectively mitigating the risk of over-diagnosis while ensuring comprehensive diagnostic coverage. All three testing strategies employed in this study exhibited high heterogeneity, with I2 values of 91.2%, 91.4%, and 91.2% for CEA, CA19-9, and combined testing, respectively. This observation is consistent with the prevalent scenario in diagnostic biomarker research pertaining to CRC. The primary sources of heterogeneity can be categorized into three key aspects. First, variations in study design and patient selection: among the 95 individual testing studies and 36 combined testing studies included in this analysis, inconsistencies were noted in the inclusion criteria concerning patient tumor staging (early versus advanced), histological types (adenocarcinoma versus other subtypes), and comorbidities (such as inflammatory bowel disease and hepatobiliary diseases). Second, lack of standardization in detection methods and thresholds: variations were observed across studies regarding the detection platforms (e.g., chemiluminescence immunoassay versus enzyme-linked immunosorbent assay), reagent brands, and diagnostic thresholds (e.g., CEA cut-off values of 2.5 ng/mL versus 5 ng/mL). Third, heterogeneity in control group definitions: certain studies employed healthy populations as controls, while others included patients with benign bowel diseases (e.g., intestinal polyps or enteritis). Despite the considerable heterogeneity, the large sample sizes (12,514 cases and 10,307 controls for CEA; 6,906 cases and 5,823 controls for combined testing) provided adequate statistical power for the random-effects model. Sensitivity analysis demonstrated that after sequentially excluding any individual study, the variation ranges of pooled sensitivity and specificity remained minimal (sensitivity variation range of 0.005–0.015 and specificity variation range of 0.002–0.008 for combined testing), indicating that the core conclusions are robust and not significantly influenced by individual outlier studies. Likelihood ratio and Fagan’s nomogram analyses further validate the clinical utility of combined testing. The PLR of 4.73 for the combined test was significantly higher than that of CEA (3.21) and CA19-9 alone, indicating that its positive results possess greater confirmatory value for CRC. Conversely, its NLR of 0.47 was lower than that of the individual tests, demonstrating a higher reliability of negative results in ruling out the disease. In a moderate-risk population for CRC (pre-test probability approximately 10%), a positive combined test increased the post-test probability to 40–45%, thereby reaching the high suspicion threshold. Conversely, a negative result reduced the probability to 3–5%, approaching the effectively ruled out criterion. In conclusion, this study confirms that the combined CEA+CA19-9 testing demonstrates significant advantages in the diagnosis of CRC and can be recommended as a preferred clinical strategy. However, the observed high heterogeneity and limitations of the studies indicate the necessity for further refinement through standardized, large-scale studies to better facilitate its application in precision diagnosis. Limitations of the study Several limitations should be acknowledged when interpreting these findings. First, high heterogeneity was observed across all three testing strategies (I2 > 91%), which may be attributed to variations in patient populations, tumor staging, detection platforms, and diagnostic thresholds among included studies. Second, the number of studies available for combined CEA+CA19-9 analysis (n = 36) was substantially smaller than for individual markers (n = 95), which may limit the statistical power of subgroup analyses. Third, the majority of included studies were conducted in Asian populations, particularly Chinese cohorts, which may restrict the generalizability of findings to other ethnic groups. Fourth, differences in control group definitions (healthy individuals versus patients with benign bowel disease) across studies may have introduced spectrum bias. Fifth, although Deeks’ test suggested low publication bias, selective reporting of positive results cannot be entirely excluded. Future prospective, multicenter studies with standardized detection protocols are needed to validate these findings. Resource availability Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Tao Zhang (zhangt760@zmu.edu.cn). Materials availability This study did not generate new unique reagents. Data and code availability •All data supporting the findings of this study are available within the paper and its supplemental information. The individual study-level 2 × 2 contingency table data extracted from included studies are provided in Tables 1 and 2. All data supporting the findings of this study are available within the paper and its supplemental information. The individual study-level 2 × 2 contingency table data extracted from included studies are provided in Tables 1 and 2. •This study did not generate original code. Statistical analyses were performed using STATA 12 (StataCorp, College Station, TX, USA). No custom code was developed. This study did not generate original code. Statistical analyses were performed using STATA 12 (StataCorp, College Station, TX, USA). No custom code was developed. •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Acknowledgments This work was supported by the Guizhou Provincial Basic Research Program (Natural Science) (Qiankehe JichuZK2024yiban295), Zunyi City Science and Technology Innovation Team Construction Project (KCTD(2025)61), and the Program for High-Level Innovative “Thousand Level” Talents in Guizhou Province (Thousand Level). This meta-analysis demonstrates an association between colorectal cancer and the co-testing of CEA and CA19-9. Although this meta-analysis is classified as a secondary study and the validity of its findings is contingent upon the quality of the original studies, it possesses certain limitations. Nevertheless, it offers a foundational reference for predicting precursors to colorectal cancer. Furthermore, it serves as a guide for research into the pathogenesis of CRC and provides evidence-based medical insights for the diagnosis and treatment of CRC. Author contributions Conceptualization, L.Z., S.A., T.Z., and D.X.; methodology, L.Z.; formal analysis, L.Z.; writing – original draft, L.Z.; writing – review and editing, S.A., T.Z., and D.X.; visualization, Y.M. and G.L.; supervision, Y.Z., Y.M., and G.L.; project administration, S.A.; validation, Y.Z. Declaration of interests The authors declare that they have no competing interests. Declaration of generative AI and AI-assisted technologies in the writing process The authors confirm that the use of generative AI and AI-assisted technologies was explicitly excludedfrom all stages of this research and manuscript preparation. The ideas, analyses, and conclusions presented are entirely human-generated. The authors assume full responsibility for the originality and integrity of the work. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERDeposited dataPubMedhttps://pubmed.ncbi.nim.nih.gov/N/AEMBASEhttps://www.embase.com/N/AWeb of Sciencehttps://www.webofscience.com/wos/author/searchN/ACNKIhttps://www.cnki.net/N/AWanfanghttps://www.wanfangdata.com.cn/N/ASoftware and algorithmsStata software Version 12.0Downloaded STATA softwarehttps://www.stata.com/products/Review Manager 5.4The Cochrane Collaborationhttps://revman.cochrane.org/infoEndNoteClarivate Analyticshttp://www.myendnoteweb.com/ Experimental model and study participant details This study is a systematic review and meta-analysis of previously published studies. No new patient data were collected. Inclusion criteria required studies to enroll patients suspected of having colorectal cancer, with histopathological confirmation as the gold standard. Details of sex, age, and other participant characteristics were not uniformly reported across included studies; where available, these are summarized in Tables 1 and 2. The influence of sex and age on diagnostic performance could not be assessed due to insufficient reporting in primary studies, which represents a limitation acknowledged in the limitations of the study section. Method details Search strategy and data sources Computer searches were conducted to retrieve relevant literature published from the establishment of the China Knowledge Network (CNKI), Wanfang, China Medical Library (CMCC), SinoMed, Web of Science, and PubMed databases up to January 20, 2024. Search terms were identified using the Subject Search feature within SinoMed for Chinese MeSH equivalents and the MeSH search functionality in PubMed for English synonyms. Boolean operators (AND, OR) were used to connect keywords. Retrieved literature records were exported to EndNote software for duplicate removal. Inclusion and exclusion criteria Literature inclusion criteria: (1) study subjects comprised patients suspected of having colorectal cancer; (2) sufficient data were available to construct a 2×2 contingency table (TP, FP, TN, FN); (3) the study design was clearly described, including inclusion/exclusion criteria and baseline characteristics; (4) colonoscopic biopsy histopathology was explicitly defined as the gold standard for CRC diagnosis; (5) threshold values for relevant tumor markers were established within the study. Literature exclusion criteria: (1) studies with follow-up periods exceeding one year; (2) studies with sample sizes smaller than 30; (3) studies that did not clearly specify detection methods; (4) studies that did not establish threshold values for tumor markers; (5) studies lacking a clear histopathological diagnosis as the reference standard; (6) studies reporting only sensitivity/specificity without raw 2×2 data. Data extraction A standardized data extraction form was used to collect the following information: (1) basic research information: authors, year of publication, study design; (2) basic characteristics of tumor markers: method of detection, threshold value; (3) results: TP, FP, TN, and FN values, derived directly from the original literature or calculated from reported sensitivity and specificity. Two independent reviewers screened the literature and extracted data. Discrepancies were resolved by consultation with a third reviewer. Quality assessment The methodological quality of included studies was assessed using the QUADAS-2 tool, which evaluates risk of bias and applicability concerns across four domains: patient selection, index test, reference standard, and flow and timing. Two reviewers independently assessed each study, and disagreements were resolved through consensus. Quantification and statistical analysis Pooled sensitivity and specificity for the three testing strategies were calculated using a bivariate random-effects model implemented in STATA 12 (Stata Corp, College Station, TX, USA). This model accounts for potential correlations between sensitivity and specificity, avoiding biases inherent in single-indicator analyses. Heterogeneity was quantified using the I2 statistic (I2 > 75% indicating high heterogeneity) and visualized using bivariate box plots. Summary receiver operating characteristic (SROC) curves were constructed using the HSROC model, with pooled AUC and 95% CI calculated and reported for each testing strategy. Clinical utility was assessed using Fagan’s nomogram and likelihood ratio scatter plots. Publication bias was assessed using Deeks’ test in conjunction with funnel plots (P < 0.05 indicating significant publication bias). Sensitivity analysis was conducted using the leave-one-out method. All data are presented as pooled estimates with 95% confidence intervals. Additional resources PROSPERO registration number: CRD420251101973. Footnotes Supplemental information can be found online at https://doi.org/10.1016/j.isci.2026.115639. Contributor Information Tao Zhang, Email: zhangt760@zmu.edu.cn. Daimin Xiao, Email: 372422022@qq.com. Supplemental information Document S1. Data S1 and S2 Data S3. PRISMA 2020 checklist: where to report each item in a systematic review and relevant descriptions References Associated Data Supplementary Materials Document S1. Data S1 and S2 Data S3. PRISMA 2020 checklist: where to report each item in a systematic review and relevant descriptions Data Availability Statement •All data supporting the findings of this study are available within the paper and its supplemental information. The individual study-level 2 × 2 contingency table data extracted from included studies are provided in Tables 1 and 2. All data supporting the findings of this study are available within the paper and its supplemental information. The individual study-level 2 × 2 contingency table data extracted from included studies are provided in Tables 1 and 2. •This study did not generate original code. Statistical analyses were performed using STATA 12 (StataCorp, College Station, TX, USA). No custom code was developed. This study did not generate original code. Statistical analyses were performed using STATA 12 (StataCorp, College Station, TX, USA). No custom code was developed. •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.