Diagnostic Accuracy of Clinical Biomarkers for Preoperative Prediction of Lymph Node Metastasis in Endometrial Carcinoma: A Systematic Review and Meta‐Analysis

Current preoperative risk estimation for lymph node metastasis in endometrial carcinoma is inadequate. Optimized use of clinical biomarkers might be a first step toward improved selection for tailored treatment. This review reports on the diagnostic accuracy of preoperative clinical biomarkers for the prediction of lymph node metastasis in patients with endometrial carcinoma.

Abstract

Background.

In endometrial carcinoma (EC), preoperative classification is based on histopathological criteria, with only moderate diagnostic performance for the risk of lymph node metastasis (LNM). So far, existing molecular classification systems have not been evaluated for prediction of LNM. Optimized use of clinical biomarkers as recommended by international guidelines might be a first step to improve tailored treatment, awaiting future molecular biomarkers.

Aim.

To determine the diagnostic accuracy of preoperative clinical biomarkers for the prediction of LNM in endometrial cancer.

Methods.

A systematic review was performed according to the Meta‐analysis of Observational Studies in Epidemiology (MOOSE) guidelines. Studies identified in MEDLINE and EMBASE were selected by two independent reviewers. Included biomarkers were based on recommended guidelines (cancer antigen 125 [Ca‐125], lymphadenopathy on magnetic resonance imaging, computed tomography, and ¹⁸F‐fluorodeoxyglucose positron emission tomography/computed tomography [¹⁸FDG PET‐CT]) or obtained by physical examination (body mass index, cervical cytology, blood cell counts). Pooled sensitivity, specificity, area under the curve (AUC), and likelihood ratios were calculated with bivariate random‐effects meta‐analysis. Likelihood ratios were classified into small (0.5–1.0 or 1–2.0), moderate (0.2–0.5 or 2.0–5.0) or large (0.1–0.2 or ≥ 5.0) impact.

Results.

Eighty‐three studies, comprising 18,205 patients, were included. Elevated Ca‐125 and thrombocytosis were associated with a moderate increase in risk of LNM; lymphadenopathy on imaging with a large increase. Normal Ca‐125, cytology, and no lymphadenopathy on ¹⁸FDG PET‐CT were associated with a moderate decrease. AUCs were above 0.75 for these biomarkers. Other biomarkers had an AUC <0.75 and incurred only small impact.

Conclusion.

Ca‐125, thrombocytosis, and imaging had a large and moderate impact on risk of LNM and could improve preoperative risk stratification.

Implications for Practice.

Routine lymphadenectomy in clinical early‐stage endometrial carcinoma does not improve outcome and is associated with 15%–20% surgery‐related morbidity, underlining the need for improved preoperative risk stratification. New molecular classification systems are emerging but have not yet been evaluated for the prediction of lymph node metastasis. This article provides a robust overview of diagnostic performance of all clinical biomarkers recommended by international guidelines. Based on these, at least measurement of cancer antigen 125 serum level, assessment of thrombocytosis, and imaging focused on lymphadenopathy should complement current preoperative risk stratification in order to better stratify these patients by risk.

Introduction

Endometrial carcinoma (EC) is the most common gynecological malignancy in industrialized countries [1], [2]. Most patients present with early‐stage disease and have a favorable prognosis. However, approximately 10% of clinical early‐stage patients have lymph node metastasis (LNM) [3], [4]. Primary treatment of EC consists of hysterectomy with bilateral salpingo‐oophorectomy. Pelvic and para‐aortic lymphadenectomy can be performed to enable tailored adjuvant therapy based on the presence of LNM. Yet, routine lymphadenectomy in clinical early‐stage EC has not demonstrated to improve outcome and is associated with 15%–20% surgery‐related morbidity, underlining the need for improved risk stratification [5], [6], [7], [8].

Approximately 80% of patients are preoperatively diagnosed with low‐risk histology, i.e., grade 1 or 2 endometrioid endometrial carcinoma, facing 5%–9% risk of LNM [4], [9]. Around 20% are preoperatively diagnosed with high‐risk histology, i.e., grade 3 or nonendometrioid endometrial carcinoma, facing 18%–24% risk of LNM [4], [9], [10]. When classifying patients as low risk or high risk of LNM solely based on preoperative histology, a substantial number of patients with LNM will be missed (Fig. 1). Furthermore, in up to 33% of all patients, preoperative histology is discordant with postoperative histology, resulting in a subsequent incorrect risk estimation of LNM [11].

Figure 1.

Population of 100 patients with endometrial carcinoma with a population risk of approximately 10% on lymph node metastasis (LNM). Risk stratification based on preoperative tumor histology classifies 80% of patients as low‐risk, with a 5%–9% risk of LNM, and 20% of patients as high‐risk, with an 18%–24% risk of LNM. The patients within the red frame are classified as high‐risk based on preoperative tumor histology. This figure illustrates that patients are being misclassified in both risk groups.

Recently, The Cancer Genome Atlas (TCGA) identified four distinct subgroups based on genomic background, including an “ultramutated” subgroup associated with mutations in the exonuclease domain of polymerase‐ε (POLE); a microsatellite‐instable subgroup, with deficiency of one or more mismatch repair proteins; a copy number‐high subgroup with frequent p53 mutations; and a copy number‐low subgroup [12]. These molecular subgroups categorize patients with distinct prognoses, but so far TCGA classification has not been studied in relation to the risk estimation of LNM [13], [14], [15], [16]. Several guidelines, including the European Society of Medical Oncology, European Society of Gynecological Oncology, European Society for Radiotherapy and Oncology (ESMO‐ESGO‐ESTRO) consensus conference guideline, incorporate measurement of cancer antigen 125 (Ca‐125) and/or assessment of lymphadenopathy by imaging as part of preoperative workup [8], [17]. Furthermore, clinical biomarkers obtained during standardized preoperative workup, including assessment of body mass index (BMI), cervical cytology, and analysis of hemocytometric parameters, are demonstrated to carry prognostic information about the risk of LNM and may specifically be valuable because they are widely available as part of the routine workup [18], [19], [20], [21].

In summary, current preoperative risk estimation for LNM in EC is only moderate, and optimized use of clinical biomarkers might be a first step toward improved selection awaiting future molecular biomarkers. Therefore, this systematic review investigates the reported diagnostic accuracy of preoperative clinical biomarkers for the prediction of LNM in patients with EC.

Materials and Methods

Search Strategy and Selection Criteria

The systematic review was performed according to the Meta‐analysis of Observational Studies in Epidemiology (MOOSE) guidelines for meta‐analyses of observational studies in medicine [22]. A literature search was performed in MEDLINE and EMBASE from June 1997 to January 14, 2019. The following keywords and all known synonyms for these keywords were used: “endometrial cancer,” “BMI,” “cervical cytology,” “Ca‐125,” “hemoglobin,” “leukocytes,” “thrombocytes,” “MRI,” “CT,” “PET‐CT,” and “lymph node metastases.” Parameters were included when incorporated in the ESMO‐ESGO‐ESTRO or Society of Gynecologic Oncology guidelines (Ca‐125, computed tomography [CT], magnetic resonance imaging [MRI], ¹⁸F‐fluorodeoxyglucose positron emission tomography/computed tomography [¹⁸FDG PET‐CT]) or when judged as being routinely obtained by physical examination (BMI, cervical cytology, blood cell counts including hemoglobin, leukocytes, and thrombocytes). The search strategy can be found in supplemental online Appendix 1. Additional searches were performed by manual cross‐referencing in included studies and systematic reviews.

Study Selection

Eligible studies investigated the diagnostic accuracy of one of the preoperative biomarkers in patients with EC. The reference standard had to be histological lymph node assessment by lymphadenectomy. Exclusion criteria were conference abstracts, case reports, studies comprising fewer than five patients, case‐control studies, publications older than 20 years or not in English, review articles, outcome measures other than LNM, and (partial) absence of the reference standard. In publications in which diagnostic performance metrics were reported without the raw data to construct a contingency table, the authors were contacted. In case of overlapping patient cohorts, the study with most outcome data were included. Two investigators (C.R., F.S.) independently reviewed each study for eligibility based on title and abstract. The full text of presumably eligible studies was evaluated to decide if the study fulfilled the inclusion criteria. In case of discrepancies, consensus was made after discussion with a third reviewer (J.P.). The agreement on the selection of studies between the investigators was calculated by means of a weighted κ.

Assessment of Study Quality

The revised tool for Quality Assessment of Diagnostic Accuracy Studies (QUADAS‐2) was used for assessment of methodological quality of the studies by two investigators (C.R., F.S.) [23]. In case of disagreement, consensus was achieved with a third investigator (J.P.). Risk of bias was assessed as “low risk,” “high risk,” or “unclear risk” in four domains: patient selection, index test, reference standard, and flow and timing. The first three domains were also assessed in terms of applicability. Studies were classified as “low risk of bias” or “low risk regarding applicability” in a specific domain, when all subdomains were scored as “low risk.” For bias in “patient selection,” it was assessed whether consecutive inclusion was present and whether patients were inappropriately excluded based on age, histology, or International Federation of Gynecology and Obstetrics (FIGO) stage. For “index test” and “reference standard” it was assessed whether patients received the same reference standard (pelvic and para‐aortic lymphadenectomy). For “flow and timing” it was assessed whether interval was acceptable (<2 weeks).

Data Extraction

A data extraction form was designed with information about study design, patient characteristics (age, menopausal status), preoperative biomarkers, number of patients with and without LNM, and tumor characteristics (grade and histological subtype).

Statistical Analysis

Contingency tables containing the number of patients with LNM as assessed by the preoperative test and by lymphadenectomy were constructed, and the number of true positives, false negatives, false positives, and true negatives were extracted. From these, likelihood ratios (LRs), sensitivity and specificity, and their 95% confidence intervals (CIs) were calculated for each study and visualized by means of forest plots. Pooled estimates for LRs, sensitivity and specificity, and area under the curve (AUC) were calculated by means of bivariate random‐effects meta‐analysis [24]. An AUC above 0.75 was considered clinically relevant [25]. Using sensitivity, specificity, and prevalence, the pooled positive predictive values (PPVs) and negative predictive values (NPVs) were also estimated:

$$\text{Pooled}\mathrm{PPV}\hspace{0.17em}=\hspace{0.17em}\left(\text{sensitivity}\hspace{0.17em}\times \hspace{0.17em}\text{prevalence}\right)\hspace{0.17em}÷\hspace{0.17em}\left[\text{sensitivity}\hspace{0.17em}\times \hspace{0.17em}\text{prevalence}\hspace{0.17em}+\hspace{0.17em}\left(1\hspace{0.17em}-\hspace{0.17em}\text{specificity}\right)\times \left(1\hspace{0.17em}-\hspace{0.17em}\text{prevalence}\right)\right]$$

$$\text{Pooled}\mathrm{NPV}\hspace{0.17em}=\hspace{0.17em}\left[\text{specificity}\hspace{0.17em}\times \hspace{0.17em}\left(1\hspace{0.17em}-\hspace{0.17em}\text{prevalence}\right)\right]\hspace{0.17em}÷\hspace{0.17em}\left[\text{specificity}\hspace{0.17em}\times \hspace{0.17em}\left(1\hspace{0.17em}-\hspace{0.17em}\text{prevalence}\right)+\left(1\hspace{0.17em}-\hspace{0.17em}\text{sensitivity}\right)\hspace{0.17em}\times \hspace{0.17em}\text{prevalence}\right]$$

The median LNM prevalence of all included studies (13.4%) was used to estimate PPV and NPV.

To quantify the test impact on the posttest probability, LRs were categorized: “small increase,” “moderate increase,” and “large increase” for positive LRs between 1.0 and 2.0, 2.0 and 5.0, and 5.0 and 10.0, respectively; “small decrease,” “moderate decrease,” and “large decrease” for negative LRs between 0.5 and 1.0, 0.2 and 0.5, and 0.1 and 0.2, respectively [26], [27]. Posttest probabilities were calculated using Fagan's nomogram [28].

The I² statistic was used for all variables to estimate the amount of heterogeneity between the studies [29]. Heterogeneity was graduated: null for I² = 0%, minimal for 0% < I² ≤ 25%, low for 25% < I² ≤ 50%, moderate for 50% < I² ≤ 75%, and high for I² > 75%. In case of high heterogeneity, we investigated whether the heterogeneity could be explained by the use of various definitions of positive test results by performing subgroup analyses based on these definitions (Table 1).

Table 1. Preoperative biomarkers, including the definitions used in the included studies for negative and positive test results.

^aFor lymphadenopathy on imaging, a short axis lymph node diameter of 1.0 cm was set as cutoff value.

Abbreviations: ¹⁸FDG PET‐CT, ¹⁸F‐fluorodeoxyglucose positron emission tomography/computed tomography; BMI, body mass index; Ca‐125, cancer antigen 125; CT, computed tomography; MRI, magnetic resonance imaging.

Furthermore, separate analyses were performed for studies including low‐risk and high‐risk populations. A study population was considered low risk when the study prevalence of LNM was below the median (13.4%) and high‐risk when the prevalence of LNM was above the median (13.4%). To estimate the corresponding PPVs and NPVs, the median prevalence of all low‐risk and high‐risk population studies, respectively, was used (10.5% and 19.0%). The statistical software R was used for the statistical analysis (version 3.3.2) with the MADA package (1.9‐9) [30].

Results

Study Selection

A total of 7,072 studies were retrieved, of which 6,453 remained after removal of duplicates. Based on title and abstract, 525 studies were relevant. After full‐text screening, 83 studies, comprising 18,205 patients, were included in the systematic review (supplemental online Appendix 2) [18], [19], [20], [21], [31], [32], [33], [34], [35], [36], [37], [38], [39], [40], [41], [42], [43], [44], [45], [46], [47], [48], [49], [50], [51], [52], [53], [54], [55], [56], [57], [58], [59], [60], [61], [62], [63], [64], [65], [66], [67], [68], [69], [70], [71], [72], [73], [74], [75], [76], [77], [78], [79], [80], [81], [82], [83], [84], [85], [86], [87], [88], [89], [90], [91], [92], [93], [94], [95], [96], [97], [98], [99], [100], [101], [102], [103], [104], [105], [106], [107], [108], [109]. Three studies lacked raw data, and because the authors did not respond after being contacted, these publications were excluded [68], [110], [111]. Two studies had overlapping study cohorts, and one of these was excluded [67], [112]. One additional article was identified by extensive cross‐checking of the reference lists [103]. Ten studies were analyzed multiple times because they included two or three biomarkers [53], [59], [68], [75], [76], [77], [78], [80], [87], [104]. The agreement of the two reviewers (C.R. and F.S.) regarding eligibility was 94% (κ = 0.89; 95% CI, 0.85–0.93). Characteristics of the included studies are shown in supplemental online Appendix 3.

Assessment of Study Quality

The risk of bias and the applicability was evaluated by means of the QUADAS‐2 tool (supplemental online Appendices 4 and 5). Ten studies were assessed for two or three biomarkers, so in total 95 evaluations were performed. The risk of bias was low in 4 (4.2%) evaluations, unclear in 81 (85.2%) evaluations, and high in 10 (10.5%) evaluations. Forty‐three (45.3%) evaluations had an unclear risk of bias in patient selection, attributable to the absence of selection criteria or no description of consecutive inclusion. Furthermore, in a majority of studies, the mean number of lymph nodes resected by lymphadenectomy was varying. Applicability concerns were mainly attributable to patient selection: the histological diagnoses were not described properly or only patients with either low‐risk or high‐risk histology were included.

Study Results

I² for pooled positive LR and negative LR estimates was <50% for all markers, indicating absence of moderate heterogeneity (Table 2). The next section will focus on pooled LR estimates and AUCs; however, information about pooled sensitivity, specificity, and PPVs can be found in Table 2. Individual study data with forest plots for sensitivity, specificity, and predictive values can be found in supplemental online Appendix 6.

Table 2. Pooled diagnostic test characteristics of preoperative biomarkers for the prediction of LNM.

^aReferences can be found in supplemental online Appendix 3.

^bThis value is extracted from the study with the median prevalence of LNM. Range includes the minimum and maximum prevalence.

^cPooled PPVs and NPVs were estimated using a median prevalence of lymph node metastasis (13.4%).

^dFor these variables, included studies used different definitions of a positive test result. These can be found in Table 1.

^eFor these biomarkers, no subgroup analysis was performed because of small study numbers.

Abbreviations: ¹⁸FDG PET‐CT, ¹⁸F‐fluorodeoxyglucose positron emission tomography/computed tomography; AUC, area under the curve; BMI, body mass index; Ca‐125, cancer antigen 125; CI, confidence interval; CT, computed tomography; LNM, lymph node metastasis; LR, likelihood ratio; MRI, magnetic resonance imaging; NPV, negative predictive value; PPV, positive predictive value.

BMI

In patients with a low BMI, the pooled positive LR point estimate was 1.26, indicating a small increase in risk of LNM (Table 3). In patients with high BMI, the negative LR estimate was 0.97, indicating a small decrease in risk of LNM. Similar estimates were found in the low‐risk population studies. LR estimates in the high‐risk population studies were not significantly different from 1 (Table 2, Fig. 2). The AUC was found to be 0.543.

Table 3. Effect on posttest probability of lymph node metastasis categorized according to the likelihood ratio value.

^aEffect on posttest probability was categorized according to the likelihood ratio (LR) values: “small increase,” “moderate increase,” and “large increase” for LRs between 1.0 and 2.0, 2.0 and 5.0, and 5.0 and 10.0, respectively; “small decrease,” “moderate decrease,” and “large decrease” for LRs between 0.5 and 1.0, 0.2 and 0.5, and 0.1 and 0.2, respectively.

Abbreviations: ¹⁸FDG PET‐CT, ¹⁸F‐fluorodeoxyglucose positron emission tomography/computed tomography; BMI, body mass index; Ca‐125, cancer antigen 125; CT, computed tomography; LR, likelihood ratio; MRI, magnetic resonance imaging.

Figure 2.

Forest plots displaying pooled likelihood ratios with corresponding 95% confidence intervals.

Abbreviations: ¹⁸FDG PET‐CT, ¹⁸F‐fluorodeoxyglucose positron emission tomography/computed tomography; BMI, body mass index; Ca‐125, cancer antigen 125; CT, computed tomography; LR, likelihood ratio; MRI, magnetic resonance imaging.

Cervical Cytology

In patients with abnormal cervical cytology, the pooled positive LR estimate was 1.73, and in patients with normal cervical cytology, the pooled negative LR was 0.44. In analysis of low‐risk and high‐risk population studies, similar results were found (Table 2). LRs varied depending on the definition of abnormal cervical cytology. When the presence of endometrial cells was regarded as abnormal, regardless of whether these were atypical, pooled positive LR and negative LR estimates were 1.30 and 0.21, respectively. When the presence of atypical endometrial cells was regarded as abnormal, pooled positive LR and negative LR estimates were 2.06 and 0.66, respectively (supplemental online Appendix 7). The AUC was found to be 0.701.

Serum Biomarkers

Overall, Ca‐125 was elevated in 28.9% of all patients. In patients with elevated Ca‐125, the pooled positive LR estimate was 3.17, and in patients with normal Ca‐125, the pooled negative LR estimate was 0.44, both indicating a moderate effect on the risk of LNM. These results were comparable for low‐risk and high‐risk population studies (Table 2), and predictive values did not vary depending on the chosen threshold (supplemental online Appendix 7). The AUC was found to be 0.771.

In patients with elevated Ca‐125, the pooled positive LR estimate was 3.17 and in patients with normal Ca‐125 the pooled negative LR estimate was 0.44, both indicating a moderate effect on the risk of LNM. These results were comparable for low‐risk and high‐risk population studies and predictive values did not vary depending on the chosen threshold.

In patients with preoperative anemia, leukocytosis, or thrombocytosis, the pooled positive LR estimates were 1.96, 1.79 (both small increase), and 2.66 (moderate increase), respectively. In all three biomarkers the pooled negative LR estimates were above 0.50, indicating a small decrease in risk of LNM. The AUC was found to be above 0.75 for thrombocytosis (0.785) but not for anemia and leukocytosis. No subgroup analysis could be performed because of insufficient numbers of studies.

Imaging

In most studies, a 1.5T MRI scanner was used, except in three studies that used both 1.5T and 3.0T scanners. Two studies did not provide specifications on scanner type (supplemental online Appendix 3). The pooled positive LR point estimates for CT, MRI, and ¹⁸FDG PET‐CT were 6.30, 7.29, and 7.47, respectively, indicating a large increase. The pooled negative LR point estimates were 0.67 for CT, 0.55 for MRI, and 0.39 for ¹⁸FDG PET‐CT. Subgroup analyses resulted in comparable estimates in low‐ and high‐risk population studies. Overall, the presence of LNM increased the odds of positive findings on imaging approximately seven times, whereas the absence of LNM decreased the odds approximately two times. The AUC was 0.800 for MRI, 0.773 for ¹⁸FDG PET‐CT, and 0.687 for CT.

Translation to Clinical Practice

As preoperative tumor grade currently is one of the most important preoperative predictors for LNM, the contributive value of all evaluated clinical biomarkers to preoperative tumor grade is summarized in Figure 3. Results are shown for two clinical settings based on patients presenting with low‐grade EC and an estimated 7% a priori probability on LNM, and patients presenting with high‐grade EC with an estimated 20% a priori probability. Results are classified according their increase or decrease on the odds of LNM, i.e., small, moderate, and large.

Figure 3.

Summarizing results of pre‐ and posttest probabilities for evaluated clinical biomarkers in low‐ and high risk‐risk preoperative setting. Estimates based on pooled likelihood ratios (Table 2). Thickness of bars represents the number of studies performed on each marker; range, 2–20.

†, Based on prevalence of lymph node metastasis in patients preoperatively diagnosed with low‐grade (7%, left panel) and high‐grade endometrial cancer (20%, right panel), respectively.

Abbreviations: ¹⁸FDG PET‐CT, ¹⁸F‐fluorodeoxyglucose positron emission tomography/computed tomography; BMI, body mass index; Ca‐125, cancer antigen 125; CT, computed tomography; EC, endometrial carcinoma; MRI, magnetic resonance imaging.

Discussion

In this review we have demonstrated that several clinical biomarkers are significantly associated with LNM and thus may improve current preoperative risk stratification of patients with EC. As expected, the presence of lymphadenopathy on imaging had a large impact on the risk of LNM. Elevated Ca‐125 and thrombocytosis were associated with a moderate increased risk of LNM, whereas normal cervical cytology, Ca‐125, and absence of lymphadenopathy on ¹⁸FDG PET‐CT were associated with a moderate decreased risk of LNM. All markers having a moderate or large impact were found to have clinically relevant AUCs (0.75), whereas markers with small impact did not.

Interestingly, we found that high BMI (>25/30 kg/m²) was associated with reduced risk of LNM. This could be explained by the higher prevalence of low‐grade endometrioid ECs in obese patients who overall have a low risk of LNM [34]. Yet, we could not rule out whether lymphadenectomy was less frequently performed in obese patients with EC because of comorbidity and/or surgical difficulties, because this could underestimate the risk of LNM in obese patients [34]. However, once lymphadenectomy is performed, numbers of lymph nodes are reported to be equal in obese and nonobese patients [34].

Abnormal cervical cytology was a predictor of LNM. The presence of atypical endometrial cells in cervical cytology could be a consequence of EC cells detaching and exfoliating, as in patients with serous histology, who present with abnormal cervical cytology in 66%–88% [113], [114]. Because it is often obtained by general practitioner prior to referral for postmenopausal bleeding and can easily be added in the workup by gynecologists, further validation is worthwhile [115].

The association between Ca‐125 level and LNM has already been demonstrated in a large number of studies, and Ca‐125 has also been shown to predict the high‐risk features deep myometrial invasion (MI) and lymphovascular space invasion [48], [51], [60], [116]. Nevertheless, Ca‐125 has not yet been implemented in standardized preoperative workup, possibly because of controversies regarding the appropriate cutoff value. In this meta‐analysis we found that the diagnostic accuracy was similar using varying cutoff values.

We found that preoperative anemia, leukocytosis, and thrombocytosis were associated with a small (anemia, leukocytosis) or moderate (thrombocytosis) increased risk of LNM. Interestingly, abnormal blood cell counts are reportedly associated with poor prognosis in several malignancies [117], [118]. Whereas anemia and leukocytosis only have a small impact on risk of LNM, thrombocytosis has a moderate impact and has also been proposed as a marker to be included into a preoperative scoring system for advanced disease in EC [119]. Because of limited study numbers, further validation is needed.

Whereas anemia and leukocytosis only have a small impact on risk of LNM, thrombocytosis has a moderate impact and has also been proposed as a marker to be included into a pre‐operative scoring system for advanced disease in EC.

Interestingly, for imaging, the presence of LNM increased the odds of positive findings on imaging approximately seven times, whereas the absence of LNM decreased the odds only two times. The different imaging techniques rely on pelvic or para‐aortic lymph node enlargement (CT and MRI), or increased glucose metabolism (¹⁸FDG PET‐CT), which is only detectible when there are sufficient tumor cells present to discriminate them from physiological glucose metabolism from surrounding nonmalignant cells. Micrometastases leading to minimally increased avidity being barely detectible would typically be missed. The relatively low negative LR estimates suggest that omitting lymphadenectomy in patients with negative findings on imaging may thus lead to potential surgical undertreatment in some patients, underlining the importance of incorporating multiple predictors into risk‐stratification models. On the other hand, the very high positive LR and PPV estimates of CT, MRI, and ¹⁸FDG PET‐CT justify lymphadenectomy in patients with imaging findings indicating lymph node metastases. The ability of MRI to assess MI and cervical involvement as well supports the use of MRI in addition to ¹⁸FDG PET‐CT or CT for preoperative risk stratification [120]. Diagnostic superiority was also demonstrated in the AUC values, which were shown to be higher than 0.75 for MRI and ¹⁸FDG PET‐CT but lower than 0.75 for CT.

To our knowledge, this is the first systematic review comprehensively investigating the diagnostic accuracy of preoperative clinical biomarkers in EC for the prediction of LNM. We have performed an in‐depth analysis including subgroup analyses to improve clinical applicability. However, some limitations need to be addressed. Most studies documented a study design with only moderate quality, being retrospective and lacking a consecutive research design because of selective performance of lymphadenectomy. Although I² was <50% for all biomarkers, some clinical heterogeneity could exist between the studies. With study population and the employed thresholds being important sources of heterogeneity, separate analyses were performed based on these two.

Although we have shown the impact of several biomarkers on risk of LNM, multivariable analysis is impossible in this setting, and therefore the combined value of markers cannot be concluded from this review and requires further analysis of individual patient data.

Among the strongest prognosticators for LNM are lymphovascular space invasion, tumor grade and histology, and deep MI [121], [122], [123], [124], [125]. However, preoperative assessment of these markers is accompanied with some challenges. Lymphovascular space invasion is based on postoperative histological examination of the surgical specimen and cannot be reliably assessed preoperatively. MI can be assessed preoperatively by MRI, transvaginal ultrasound, or intraoperatively by frozen section with varying diagnostic accuracies. For transvaginal ultrasound, sensitivity and specificity for deep MI are reported to be 71%–85% and 72%–90%, respectively [126]. For MRI, sensitivity and specificity are 63%–100% and 56%–100% [120], [127].

Several models have been developed and validated for prediction of LNM in order to improve existing risk stratification [58], [128], [129]. Among these, one model has an AUC of 0.75, indicating good diagnostic accuracy [130]. This model has included serum Ca‐125, together with MRI parameters (MI, lymphadenopathy, and extension beyond uterine corpus), adequately identifying 43% of patients as low risk for LNM (<4%), with a false‐negative rate of only 1.3% [128].

For future implementation, it is important to acknowledge of the increasing impact of sentinel node (SN) procedures. These procedures create the possibility to reduce treatment‐related morbidity and might lead to more adequate staging [131]. Still, it is questionable whether SN procedures should be performed in all patients with EC, with a population‐based risk of 10% for LNM. Sentinel node procedures require sufficient expertise and are generally available in large oncology centers, and thus are demanding for public health care facilities and costs. It could therefore be imagined that preoperative risk stratification will remain a crucial part in EC care in order to expose only those patients at risk to more extensive surgical procedures. In what perspective the investigated biomarkers could complement the recent TCGA classification remains to be elucidated. Even though this classification identifies four subgroups with distinct prognoses, it is unknown whether these subgroups are associated with the risk of LNM. Recent TCGA classification has raised debate about how to relate prognostic molecular classification systems to traditional histopathological criteria, and an integrated genomic‐pathologic classification was proposed as a superior classification [13]. This discussion is also relevant for implementing these clinical biomarkers complementary to molecular classification.

Conclusion

We identified clinical biomarkers that could contribute to an improved preoperative risk stratification and more individualized treatment strategy in patients with EC. Lymphadenopathy identified at preoperative imaging incurred a large risk of LNM, and as such should be incorporated into future preoperative risk‐stratification models. Furthermore, clinical biomarkers with moderate impact on risk of LNM, i.e., Ca‐125 serum levels and thrombocytosis are candidates for future preoperative risk‐stratification models in addition to the established markers. In what way they could complement molecular classification should be studied.

See http://www.TheOncologist.com for supplemental material available online.

Author Contributions

Conception/design: Casper Reijnen, Joanna IntHout, Leon F.A.G. Massuger, Fleur Strobbe, Heidi V.N. Küsters‐Vandevelde, Ingfrid S. Haldorsen, Marc P.L.M. Snijders, Johanna M.A. Pijnenborg

Collection and/or assembly of data: Casper Reijnen, Joanna IntHout, Leon F.A.G. Massuger, Fleur Strobbe, Heidi V.N. Küsters‐Vandevelde, Ingfrid S. Haldorsen, Marc P.L.M. Snijders, Johanna M.A. Pijnenborg

Data analysis and interpretation: Casper Reijnen, Fleur Strobbe, Johanna M.A. Pijnenborg

Manuscript writing: Casper Reijnen, Joanna IntHout, Leon F.A.G. Massuger, Fleur Strobbe, Heidi V.N. Küsters‐Vandevelde, Ingfrid S. Haldorsen, Marc P.L.M. Snijders, Johanna M.A. Pijnenborg

Final approval of manuscript: Casper Reijnen, Joanna IntHout, Leon F.A.G. Massuger, Fleur Strobbe, Heidi V.N. Küsters‐Vandevelde, Ingfrid S. Haldorsen, Marc P.L.M. Snijders, Johanna M.A. Pijnenborg

Disclosures

The authors indicated no financial relationships.