Background

Carcinoembryonic antigen (CEA) is a glycoprotein involved in cell adhesion and is normally produced in gastrointestinal tissues during foetal development. Serum levels of CEA are usually low in the blood of healthy individuals (0–5 ng/ml), but may raise in some non-neoplastic conditions including inflammatory bowel disease, hepatitis, pancreatitis, pulmonary infections, in smokers (rarely over 10 ng/ml), and in patients affected by gastrointestinal, lung and breast tumours [1, 2].

Although CEA is elevated in approximately two-thirds of colorectal cancer (CRC) patients, international guidelines do not recommend its use as a screening or diagnostic tool due to the low sensitivity and specificity [3,4,5].

In metastatic CRC (mCRC), changes in CEA serum levels are used to monitor response to systemic therapies together with radiological imaging. In particular, a series of retrospective studies showed a correlation between CEA variation and response to chemotherapy, thus suggesting a role for CEA in avoiding radiological imaging [6,7,8,9]. A recent pooled analysis of mCRC patients with elevated baseline CEA from seven randomised clinical trials of first-line chemotherapy with or without targeted therapy collected in the ARCAD database identified a threshold for CEA decrease from baseline to the first and second CT scan reassessment to select patients with no progressive disease in order to avoid CT scan [10]. However, in this phase some information provided by imaging assessment highly affect subsequent treatment choices including the possibility to perform locoregional treatments, the choice to administer maintenance therapy or to recommend a treatment holiday. Therefore, radiological assessment during upfront chemotherapy may be highly informative in patients with decreased CEA levels.

Nowadays, after the induction phase, maintenance periods or treatment breaks are common choices in the treatment of mCRC patients [11]. Indeed after 4–6 months of combination therapy, no major improvement in tumour shrinkage is expected. Therefore, in this phase, the availability of a marker able to predict disease control status (progression disease (PD) versus no-PD) would be useful to avoid or defer CT scan. To the best of our knowledge, no specific data are available in this regard.

Drawing from these considerations, we assessed the role of CEA in predicting PD after induction therapy in mCRC patients randomised in two Phase III randomised trials by Gruppo Oncologico del Nord Ovest, TRIBE and TRIBE2, that compared FOLFOXIRI (5-fluorouracil, leucovorin, oxaliplatin and irinotecan) plus bevacizumab with FOLFIRI (5-fluorouracil, leucovorin and irinotecan) or FOLFOX (5-fluorouracil, leucovorin and oxaliplatin) plus bevacizumab, as first-line treatment. In both studies, maintenance therapy with 5-fluorouracil and bevacizumab until disease progression, unacceptable adverse events or consent withdrawal was planned following the upfront combination therapy [12,13,14].

Methods

Study design and procedures

TRIBE and TRIBE2 are two Phase III randomised, open-label, multicentre trials involving 1187 initially unresectable untreated mCRC patients. In the TRIBE study, 508 patients were randomised in a 1:1 ratio to receive FOLFIRI/bevacizumab or FOLFOXIRI/bevacizumab, while in the TRIBE2 trial, 679 patients were randomised in a 1:1 ratio to receive FOLFOX/bevacizumab followed by FOLFIRI/bevacizumab after disease progression (PD) or FOLFOXIRI/bevacizumab followed by the reintroduction of the same agents after PD. All treatments were administered up to 12 cycles in TRIBE and up to 8 cycles in TRIBE2, followed by 5-fluorouracil plus bevacizumab until PD, unacceptable adverse events, or consent withdrawal in both trials [12,13,14].

CEA and radiological disease assessments (contrast-enhanced chest and abdominal CT scan, or abdomen MRI and chest CT if contrast-enhanced CT scan was contraindicated) were performed every 8 weeks. PD assessment was based on investigator reported measurements according to RECIST version 1.0 and version 1.1 in TRIBE and TRIBE2, respectively.

In the present analysis, only patients with a baseline value of CEA ≥ 10 ng/mL prior to the beginning of the induction treatment, not progressing during induction therapy, and with at least one paired radiological evaluation and CEA assessment after the end of the induction therapy were included. CEA and CT scan assessments were considered paired if no more than 15 days between the two exams had elapsed.

Objectives

The primary aim was to investigate whether the increase of CEA from nadir (the lowest value of CEA after baseline) could predict progressive disease during maintenance or follow-up after the end of the induction therapy in mCRC patients enrolled in the TRIBE and TRIBE2 studies. In particular, we aimed at identifying a threshold for percent increase in CEA from nadir able to predict disease progression. To this purpose, all paired CEA and radiological assessments after the end of induction were analysed.

Statistics

The cut-off for percentage increase in CEA from nadir was selected using two approaches:

  • The optimal cut-off determined by ROC analysis to maximise the Youden index (i.e. sensitivity + specificity −1). Internal validation was conducted by calculating the bootstrap optimism-corrected AUC.

  • A clinically relevant cut-off to maximise sensitivity (i.e. to minimise the cases of radiological PD not associated with increasing of CEA).

Using each of the above criteria, the prediction performance of the selected cut-offs was evaluated by sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV).

The chi-square test, Fisher’s exact test and Mann–Whitney test were used, when appropriate, to compare clinical and biological features between patients with CEA value <10 ng/mL and ≥10 ng/mL, and between patients with radiological PD not associated with CEA increase and all other patients. Odds ratios (OR) and the corresponding 95% Confidence Intervals (CIs) were estimated using the logistic regression model. Progression-free survival (PFS) was determined according to the Kaplan–Meier estimates method.

All statistical tests were two-sided, and P values of 0.05 or less were deemed significant. Statistical analyses were performed using SAS version 9.4 (SAS Institute, Inc., Cary, NC). The data cut-off for the present analysis was July 31, 2014 and July 30, 2019 for TRIBE and TRIBE2, respectively.

Results

Among 1187 patients enrolled in TRIBE and TRIBE2 studies, 1117 had a baseline value of CEA available (733 (66%) with CEA ≥ 10 ng/mL and 384 (34%) with CEA < 10 ng/mL) (Fig. 1). As shown in Supplementary Table 1, patients with CEA < 10 ng/mL had more frequently an ECOG-PS of 0 (P < 0.01), metachronous disease (P < 0.01), previously resected primary tumour (P < 0.01), no liver metastases (P < 0.01), oligometastatic disease (P < 0.01) and low tumour burden (P < 0.01) as previously defined [15], BRAF mutated tumours (P = 0.01), and had previously received adjuvant therapy (P < 0.01).

Fig. 1: Consort diagram of the study.
figure 1

N   number, CEA   carcinoembryonic antigen.

Full size image

Overall, 434 patients fulfilled the inclusion criteria of this study (Fig. 1). Their clinicopathological baseline characteristics are listed in Table 1. Most of the patients were males (61%) and had ECOG-PS 0 (89%), left-sided primary tumour (67%), synchronous disease (87%), liver metastases (89%) and RAS mutated tumour (63%). Median CEA levels were 99 ng/ml and 5 ng/ml at baseline and at nadir, respectively.

Table 1 Patients’ characteristics.
Full size table

At the ROC analysis, the raise of CEA from nadir was significantly associated with PD with high predictive accuracy (AUC: 0.81, 95% CI: 0.79–0.83, P < 0.01) (Fig. 2). Internal validation of the model using the bootstrap optimism-corrected AUC showed similar results (AUC: 0.81, 95% CI: 0.78–0.84, P < 0.01). The optimal cut-off determined by the ROC analysis was 120%: a CEA increase of at least 120% from nadir differentiated between PD and no-PD with a sensitivity of 74% (95% CI: 68–79%) and a specificity of 78% (95% CI: 75–81%) (Table 2 and Fig. 3a). Using this cut-off, PD could be properly excluded in the 92% (95% CI: 90–93%) of radiological assessments and 67% of CT scans (i.e. cases with no CEA increase or increase <120%) could be avoided. Patients with an increase in CEA of at least 120% from nadir had a 48% probability of PD compared to 8% in patients with no increase or an increase of CEA < 120% (OR: 10.0, 95% CI: 7.29–13.82, P < 0.01). However, in 26% of PD cases, no CEA increase or a CEA increase <120% was observed (Fig. 3b). No specific characteristics able to distinguish these patients from those in which PD occurs with a CEA increase of at least 120% could be identified (Table 3). In patients with no radiological evidence of PD at the time of a CEA increase of at least 120% from nadir, such an increase anticipated the evidence of disease progression by a median time of 3.9 months. Of note, after a median follow-up of 26.3 months, PD never occurred in the 6% of cases (Supplementary Fig. 1A).

Fig. 2: ROC curves.
figure 2

AUC area under the curve.

Full size image
Table 2 Progression disease prediction by percent change in CEA levels from nadir.
Full size table
Fig. 3: Waterfall plots of CEA variation from nadir.
figure 3

Waterfall plots demonstrate predicted and observed response using a CEA cut-off identified by ROC analysis (Panel a and b) and a CEA cut-off to maximize sensitivity (Panel c and d).

Full size image
Table 3 Characteristics of patients based on CEA and progression disease.
Full size table

Based on the best sensitivity cut-off, any CEA increase from nadir differentiated between PD and no-PD with the sensitivity of 93% (95% CI: 89–95%) and specificity of 35% (95% CI: 32–39%) (Table 2 and Fig. 3c). Using this cut-off, PD could be excluded in 95% (95% CI: 92–96%) of patients and 29% of CT scans could be avoided. Patients with an increase in CEA from nadir had a 28% probability of PD compared to 5% in patients without CEA increase (OR: 6.8, 95% CI: 4.16–11.02, P < 0.01). Only in 7% of PD cases, the CEA levels did not increase from nadir (Fig. 3d). This group had more frequently oligometastatic disease at baseline with respect to other patients (P = 0.04) (Table 3). In patients with no radiological evidence of PD at the time of any CEA increase from nadir, such an increase anticipated the evidence of disease progression by a median time of 5.8 months. After a median follow-up of 28.3 months, 11% of these patients never experienced PD (Supplementary Fig. 1B).

Discussion

The evaluation of response to systemic therapies is based on radiological imaging by means of a CT scan or MRI every two to three months, according to RECIST [16].

During the upfront induction treatment, other measures of response, such as early tumour shrinkage and depth of response, provide additional useful prognostic information [17]. Moreover, radiological imaging is also essential to determine eligibility for local treatment modalities, including surgery, ablation, or radiation especially in potentially resectable diseases [11].

On the other side, during maintenance or treatment breaks, the main objective of disease reassessment by radiological imaging is the detection of PD or disease relapse. To this end, monitoring tumour evolution through a simple and cheap serum marker may limit the use of CT scans, thus avoiding unnecessary and relatively expensive series of imaging assessments.

According to international guidelines, CEA is the marker of choice to monitor mCRC in addition to radiological imaging. However, ASCO recommendations also highlight that data are not sufficient to routinely use CEA serum levels as an alternative to imaging [18]. Indeed, CEA increase during chemotherapy may also be related to treatment-induced changes in liver function [19], and other non-cancer-related causes, including acute or chronic inflammatory conditions. Moreover, only retrospective studies based on small numbers of patients that did not lead to the identification of a cut-off value for CEA variation able to predict radiological PD are available [6,7,8,9,10].

Here, we attempted at evaluating the effectiveness of different cut-off values of CEA increase to be adopted in the daily practice to predict PD during maintenance or treatment breaks following the completion of the first-line induction therapy. The optimal cut-off identified by the ROC analysis, a CEA increase of at least 120% from nadir, showed a predictive accuracy >80%, with high sensitivity and specificity (74% and 78%, respectively) confirmed by bootstrap as internal validation. Due to the high NPV, PD could be excluded in more than 90% of CEA measurements. On the other hand, due to the lower PPV (48%), the need to confirm PD with a radiological assessment is corroborated by our data. As a consequence, around 70% of the radiological assessments could be avoided adopting this cut-off value. The main limitation is that the CEA cut-off of 120% was not able to detect PD in 26% of cases.

In order to mitigate this issue, a different clinically relevant cut-off was chosen. A previous study, evaluating the association between CEA increase and disease control in 66 mCRC patients enrolled in a Phase I–II study, showed that any CEA increase from nadir or baseline is the best threshold to predict radiological PD at different time points with sensitivity and NPV of 100% [20]. Using any increase of CEA from nadir, the sensitivity raised from 74% to 93% and PD was not detected only in 7% of cases. Patients with no CEA increase at the time of PD were more frequently oligometastatic at baseline. Due to the lower specificity (35% vs 78%) and PPV (28% vs 48%), the number of avoidable radiological assessments was reduced at around 30% of cases.

Clear limitations of our work are the lack of centralised radiological review of imaging, and the local assessment of CEA. Indeed, no specific recommendation was reported in the study protocols to evaluate CEA always in the same site for the same patient during follow-up. Due to variability of the upper limit of normal CEA across laboratories, and to exclude both patients with normal CEA value (usually < 5 ng/ml), and with increased CEA for non-neoplastic conditions (generally < 10 ng/ml), we included only patients with baseline CEA ≥ 10 ng/mL in our analysis. Another limitation is the retrospective nature of the study, even if the large sample size based on prospectively collected data from two large Phase III randomised trials may mitigate this issue. In addition, all the patients included in this series received chemotherapy plus bevacizumab, so that we cannot conclude about the opportunity to translate present results to patients receiving an anti-EGFR-based upfront treatment.

In conclusion, in mCRC patients with baseline CEA ≥ 10 ng/mL, after the end of induction therapy, a CEA increase from nadir of at least 120% is able to predict the disease control status (PD vs no-PD) with the best sensitivity and specificity, and with very high NPV, and allows avoiding imaging in around the 70% of cases. Therefore, this threshold could be used during follow-up in most patients not candidates to metastases resection after the end of induction chemotherapy thus sparing a relevant amount of radiological assessments. Using any increase of CEA as cut-off further increases the sensitivity of this assessment while reducing specificity. This threshold should be especially evaluated when missing PD may cause immediate deterioration of patients’ conditions due to a high risk of disease-related symptoms (i.e. liver failure due to multiple liver metastases, intestinal occlusion due to peritoneal carcinomatosis, uncontrolled pain due to pelvic relapse). While novel tumour monitoring approaches, including circulating tumour DNA [21,22,23], are currently under investigation to properly assess their added value in the therapeutic route of affected patients and their cost-effectiveness, present data require confirmation before being translated into clinical practice.