Molecular Detection of Tumor Cells in Regional Lymph Nodes Is Associated With Disease Recurrence and Poor Survival in Node-Negative Colorectal Cancer: A Systematic Review and Meta-Analysis

The following databases were searched in August 2011: MEDLINE, BIOSIS, Science Citation Index, EMBASE, CCMed, publisher databases, American Society of Clinical Oncology Abstracts, and the clinical trial registries WHO International Clinical Trials Registry Platform and ClinicalTrials.gov. We also checked reference lists of relevant articles and review articles. No language restrictions and time limits were applied to the initial search. Search strategies, databases, and date ranges are provided in supplemental material (Data Supplement).

Two reviewers (N.N.R. and U.B.) independently screened the identified abstracts for eligibility and retrieved full articles for detailed assessment. We included studies that evaluated the association of molecular tumor-cell detection (ie, ITCs and/or MMs) with overall survival (OS; date of surgery to date of death as a result of any cause), disease-specific survival (DSS; date of surgery to date of death as a result of CRC), disease-free survival (DFS; date of surgery to date of first recurrence or death; ie, events were death or recurrence, whichever came first), or recurrence-free survival (date of surgery to date of first recurrence; ie, events were locoregional and/or systemic recurrence) of patients with node-negative CRC. Molecular detection techniques may have included any form of RT-PCR or immunohistochemistry. Occult disease was defined according to the definition of the International Union Against Cancer (UICC; ie, as a single or small number of tumor cells ≤ 0.2 mm in dimension [ITCs] and as tumor deposits ≤ 2.0 mm but greater than 0.2 mm in dimension [MMs]).¹¹ If the UICC definition was not applied or mentioned explicitly, single or clustered cells detected on molecular analyses were also considered as occult disease. We excluded studies that enrolled less than 20 patients with node-negative CRC who underwent molecular analysis of regional lymph nodes as well as studies for which no hazard ratio (HR) could be calculated for any of the outcomes. Furthermore, we excluded studies that were published in a language other than English, German, or French, studies that were not published in peer-reviewed journals, and studies that did not apply the definitions of ITCs and MMs used in this study. In case of multiple publications from the same institution with identical or overlapping patient cohorts, the most informative report was included. Studies that were excluded for these reasons are listed in the Data Supplement as are identified review articles and relevant meeting abstracts.

Two authors (N.N.R. and U.B.) independently extracted the following data from eligible articles: first author, year of publication, study period, sample size, patient age, site of disease (colon and/or rectum), stage of disease (stage I and/or stage II), adjuvant chemotherapy, study type, rate of patients with positive lymph nodes on molecular analysis, definition of occult disease, histopathologic analysis, number of retrieved lymph nodes, methods of tumor-cell detection, target genes and antigens, duration of follow-up, prognostic outcomes, and use of multivariate models. Disagreements were resolved by discussion.

The study quality of included studies was assessed by using the modified risk of bias tool recommended by the Cochrane Collaboration as described previously.^12,13

Synchronized extraction results were pooled statistically as effect estimates in meta-analyses. HRs and their corresponding SEs were extracted for individual time-to-event outcome parameters of each primary study. If an HR and the associated SE or CIs were not reported, we approximated the HR by using the statistical data provided in the article (eg, individual patient data or survival plots).^14,15

Extracted HRs were pooled by using the generic inverse-variance method available in the Review Manager software (Version 5.0, The Nordic Cochrane Centre; The Cochrane Collaboration, Copenhagen, Denmark). An HR greater than 1 indicated a worse prognosis in the tumor cell–positive group. A minimum number of three studies was required to perform meta-analyses. A random-effect analysis model was applied because we expected interstudy heterogeneity with respect to, for example, study populations, detection assays, definitions of positive lymph nodes, and duration of follow-up¹⁶ Statistical heterogeneity was assessed with I² statistics. Reasons for statistical heterogeneity were explored by using sensitivity analyses (ie, the removal of certain studies from the analysis as suggested by the forest plot) and a priori subgroup analyses. We performed tests of interaction to test for differences between subgroups.¹⁷ To avoid double patient evaluation among studies that evaluated multiple detection assays and/or target genes, these parameters were combined when possible to keep a maximum of information. Otherwise, cytokeratins (CKs) were prioritized over alternative tumor-cell markers, and immunohistochemistry was prioritized over RT-PCR assays. Sensitivity analyses (by choosing the alternative study arm) were performed to assess the statistical impact of such prioritization. Funnel plot analyses were used to evaluate publication bias.¹⁸

We identified a total of 39 eligible studies that included 4,087 patients (Fig 1).^5,19–56 The median study sample size was 90 patients (range, 21 to 399 patients). All eligible studies were published between 1994 and 2011 (Table 1). The extracted variables related to the study design of the included studies are summarized in Table 2. The median number of lymph nodes that were retrieved, on average, was 12 (range, 4 to 48 lymph nodes), and the median duration of follow-up across all studies was 60 months (range, 28 to 128 months). The majority of studies used CKs for molecular tumor-cell detection. In five studies, additional markers were analyzed,^{22,26,35,43,46} and in five studies, only alternative genes/antigens were used.^{20,24,39,51,54} A polymerase chain reaction–based detection assay was used in seven studies.^{20,24,35,39,40,51,54} In two studies, lymph nodes were analyzed by RT-PCR and immunohistochemistry.^43,46 The remaining studies exclusively used immunohistochemistry for tumor-cell detection. Nine studies that used immunohistochemistry assays applied the UICC definition of ITCs and MMs.^{21,23,28,32,41,45,49,55,56} One or more oncologic outcome parameters were analyzed on multivariate analysis in 22 studies.^{5,19,21,23,25–28,30,32,35,38–41,43,45,46,51,54–56} A significant association between molecular tumor-cell detection and the long-term outcome of patients was reported in 33% of studies on OS,^{21,24,39,42,47} 58% of studies on DSS,^{5,20,23,26,33,39,41,46,53,56} and 33% of studies on DFS.^{24,28,35,51,52} Thirteen studies were graded with low risk of bias (Data Supplement).^{25,28,32,35,38,40,41,43,45,46,51,54,56} Funnel plot analyses did not indicate a significant publication bias for analyzed outcomes (Data Supplement).

The pooled analysis of all studies on OS showed a significant prognostic value of tumor-cell detection in regional lymph nodes (HR, 2.20; 95% CI, 1.43 to 3.40; n = 15; I² = 66%; Fig 2A). Sensitivity analyses revealed that heterogeneity was not caused by a certain study. Meta-analyses confirmed the prognostic significance of molecular tumor-cell detection for the outcomes of DSS (HR, 3.37; 95% CI, 2.31 to 4.93; n = 17; I² = 44%) and DFS (HR, 2.24; 95% CI, 1.57-3.20; n = 18; I² = 44%) with moderate heterogeneity (Figs 2B and 3). Although the analysis of recurrence-free survival also showed less favorable outcome for the group of tumor cell–positive patients (HR, 2.83; 95% CI, 1.68 to 4.77; n = 9; I² = 61%), heterogeneity could not be reduced substantially by exclusion of a specific study.

Despite the limited number of included studies, the subgroup analyses on the definition of ITCs/MMs showed that detection of MMs according to the UICC definition was associated with poor OS (HR, 3.62; 95% CI, 1.34 to 9.80; n = 3; I² = 20%), DSS (HR, 2.07; 95% CI, 1.11 to 3.86; n = 3; I² = 26%) and DFS (HR, 2.81; 95% CI, 1.32 to 5.97; n = 5; I² = 53%), whereas no pooled analyses on ITC detection could be performed because of the lack of studies. Meta-analyses of studies that did not explicitly use the UICC definition of ITCs/MMs revealed a strong prognostic value of molecular tumor-cell detection regarding the oncological outcomes evaluated.

The prognostic impact of molecular tumor-cell detection was independent of the applied detection methods (ie, immunohistochemistry and RT-PCR), although we observed a more pronounced prognostic value for the analyses of studies using RT-PCR as indicated by the test of interaction on DFS (P < .001). The majority of studies used CKs as molecular targets, and meta-analyses that included these studies revealed a strong prognostic significance with respect to all three outcomes. Despite the low number of studies, there was a significant association of tumor-cell detection in regional lymph nodes by using alternative targets and OS (HR, 3.54; 95% CI, 1.23 to 10.16; n = 3; I² = 61%), DSS (HR, 3.59; 95% CI, 2.16 to 5.97; n = 5; I² = 0%), and DFS (HR, 5.13; 95% CI, 2.17 to 12.12; n = 4; I² = 38%).

Subgroup analyses of studies that enrolled patients with stage II CRC only showed that tumor-cell detection was associated with poor OS, DSS, and DFS. However, analyses of studies with patients with stages I and II CRC favored the group of patients with negative lymph nodes regarding DSS (HR, 2.10; 95% CI,1.17 to 3.76; n = 5; I² = 27%) and DFS (HR, 2.17; 95% CI,1.42 to 3.33; n = 10; I² = 13%), whereas the association for OS did not reach statistical significance (HR, 1.58; 95% CI, 0.77 to 3.24; n = 4; I² = 58%).

Subgroup analyses indicated molecular tumor-cell detection to be associated with poor OS (HR, 2.96; 95% CI, 1.29 to 6.79; n = 7; I² = 72), DSS (HR, 4.08; 95% CI, 2.20 to 7.56; n = 9; I² = 66%), and DFS (HR, 2.43; 95% CI, 1.27 to 4.68; n = 4; I² = 5%) even in studies with an inadequate average lymph node count. In studies with at least 12 resected lymph nodes, there was a similar association of tumor-cell detection with poor outcome in all three end points evaluated.

Finally, meta-analyses of studies with a high risk of bias indicated a prognostic impact of molecular tumor-cell detection regarding OS (HR, 2.72; 95% CI, 1.49 to 4.97; n = 12; I² = 72%), DSS (HR, 3.05; 95% CI,1.96 to 4.75; n = 13; I² = 45%), and DFS (HR, 2.45; 95% CI,1.32 to 4.55; n = 9; I² = 55%). Similarly, in studies with a low risk of bias, there was a significantly worse DSS (HR, 3.51; 95% CI, 1.75 to 7.02; n = 6; I² = 50%) and DFS (HR, 2.05; 95% CI,1.34 to 3.14; n = 9; I² = 33%) for patients with a positive molecular analysis of regional lymph nodes.

DFS was the only outcome measure with sufficient data to evaluate the prognostic impact of tumor-cell detection depending on the administration of adjuvant therapy. These subgroup analyses showed tumor cell–positive lymph nodes to be associated with less favorable DFS in patients who did (HR, 3.12; 95% CI,1.48 to 6.59; n = 7; ² = 64%) and did not (HR, 1.79; 95% CI,1.05 to 3.05; I² = 37%) receive adjuvant chemotherapy. The results of all subgroup analyses are summarized in Table 3.