Article Text

Download PDFPDF

Visceral fat area is an independent predictive biomarker of outcome after first-line bevacizumab-based treatment in metastatic colorectal cancer
  1. Boris Guiu1,2,3,
  2. Jean Michel Petit3,4,
  3. Franck Bonnetain5,6,
  4. Sylvain Ladoire4,7,
  5. Séverine Guiu4,7,
  6. Jean-Pierre Cercueil2,
  7. Denis Krausé2,
  8. Patrick Hillon8,
  9. Christophe Borg9,
  10. Bruno Chauffert4,7,
  11. François Ghiringhelli4,7
  1. 1Department of Radiology, Georges-Francois Leclerc Cancer Cente, Dijon, France
  2. 2Department of Radiology, Le Bocage University Hospital, Dijon, France
  3. 3Department of Endocrinology, Diabetology and Metabolic Diseases, Le Bocage University Hospital, Dijon, France
  4. 4AVENIR INSERM U866 Unit, School of Medicine, Dijon, France
  5. 5Biostatistics Unit, Georges-Francois Leclerc Cancer Center, Dijon, France
  6. 6EA 4184, School of Medicine, Dijon, France
  7. 7Department of Oncology, Georges-Francois Leclerc Cancer Centre, Dijon, France
  8. 8Department of Hepatology, Le Bocage University Hospital, Dijon, France
  9. 9Department of Oncology, University Hospital, Besançon, France
  1. Correspondence to Dr François Ghiringhelli, Georges-Francois Leclerc Cancer Centre, INSERM Avenir 866, 1 rue du Professeur Marion, Dijon 21000, France; fghiringhelli{at}dijon.fnclcc.fr

Abstract

Background Adipose tissue releases angiogenic factors that may promote tumour growth.

Objective To determine whether body mass index (BMI), subcutaneous fat area (SFA) and visceral fat area (VFA) are associated with outcomes in patients given first-line bevacizumab-based treatment for metastatic colorectal cancer (MCC).

Patients CT was used to measure SFA and VFA in 120 patients with MCC who received bevacizumab-based treatment (bevacizumab group, n=80) or chemotherapy alone (chemotherapy group, n=40) as first-line treatment. Associations linking BMI, SFA and VFA to tumour response, time-to-progression (TTP) and overall survival (OS) were evaluated.

Results In the bevacizumab group, median follow-up lasted for 24 months (3–70). BMI, SFA and VFA values above the median (ie, high BMI, high VFA and high SFA) were significantly associated with absence of a response. TTP was shorter in patients with high BMI (9 vs 12 months; p=0.01) or high VFA (9 vs 14 months; p=0.0008). High VFA was associated with shorter OS (p=0.0493). By multivariate analysis, high VFA was independently associated with response, TTP and OS (HR=7.18, p=0.008, HR=5.79, p=0.005 and HR=2.88, p=0.027, respectively). In the chemotherapy group, median follow-up lasted for 30 months (4–84). BMI, SFA and VFA were not associated with response, TTP or OS. In the whole population, interaction between VFA and bevacizumab administration was significant for response (OR=3.31, p=0.005) and TTP (HR=1.64, p=0.022), thereby confirming the results.

Conclusion This study provides the first evidence that high VFA independently predicts a poorer outcome in patients given first-line bevacizumab-based treatment for MCC. However, this predictive biomarker needs to be validated in a different dataset.

  • Colorectal cancer
  • bevacizumab
  • visceral fat
  • computed tomography
  • SFA
  • subcutaneous fat area
  • VFA
  • visceral fat area
  • CT
  • computed tomography
  • PS
  • performance status
  • CRC
  • colorectal cancer
  • CR
  • complete response
  • PR
  • partial response
  • SD
  • stable disease
  • TTP
  • time to progression
  • ROC
  • receiver-operating characteristic
  • AUROC
  • area under the ROC curve
View Full Text

Statistics from Altmetric.com

Introduction

Obesity is a risk factor for the development of colorectal cancer (CRC)1 2 and is associated with higher risks of recurrence and death after adjuvant chemotherapy.3 The mechanisms underlying the association between obesity and the subsequent development of CRC are incompletely understood but may involve insulin resistance4 and the production by adipose tissue of factors that may promote cancer growth, such as insulin-like growth factors4–9 (and vascular endothelial growth factor (VEGF).10–12

CT can be used to obtain an accurate assessment of intra-abdominal fat by measuring the visceral fat area (VFA) and subcutaneous fat area (SFA) on the same section.13–15 An increasing body of evidence indicates that the cytokine production profile differs between subcutaneous fat and visceral fat.16–21 These differences may explain why obesity-associated metabolic disorders22 23 and serum VEGF levels10 correlate positively with VFA but not SFA.

Whether body mass index (BMI), SFA and/or VFA at treatment initiation predict outcomes in patients with metastatic CRC has not been investigated. Since most patients with metastatic CRC now receive anticancer chemotherapy plus the angiogenesis inhibitor bevacizumab, we designed a retrospective investigation into associations linking BMI, SFA and VFA to outcomes in patients given first-line bevacizumab-based treatment for metastatic CRC.

Patients and methods

Eligible patients

From April 2002 to June 2008, 120 patients with histologically proven metastatic colorectal adenocarcinoma received first-line treatment at the Georges-François Leclerc Cancer Centre (Dijon, France), which made them eligible for this retrospective study. As first-line treatment, 80 patients were given a bevacizumab-based chemotherapeutic regimen (bevacizumab group), while 40 patients received chemotherapy alone (chemotherapy group). Patients received one of the following treatment regimens: simplified LV5FU2 regimen (leucovorin 400 mg/m2 followed by 5-fluorouracil as a 400 mg/m2 bolus then a 2400 mg/m2 infusion over 46 h), modified FOLFIRI regimen (simplified LV5FU2 regimen plus oxaliplatin 85 mg/m2) or simplified LV5FU2 regimen plus irinotecan 180 mg/m2 every 2 weeks. Bevacizumab was given in a dosage of 5 mg/kg every 2 weeks. The tumour response was evaluated every 8 weeks7–9 by CT according to the Response Evaluation Criteria in Solid Tumours24 and classified as follows: complete response (CR), partial response (PR), stable disease (SD) or progressive disease (PD). For the statistical analysis, tumour response after 2 months of chemotherapy was selected. Patients with either CR or PR were classified as responders and patients with SD or PD as non-responders.

This retrospective study was approved by our institutional review board and all data were anonymised. Written informed consent was obtained from all patients.

Measurement of visceral and subcutaneous fat

VFA and SFA were measured retrospectively on the CT scans performed before chemotherapy initiation, as previously described15 at the level of the umbilicus with the patient in the supine position. Briefly, we used ImageJ software (http://rsb.info.nih.gov/ij/, accessed 10 January 2010) to measure pixels with densities in the −190 HU to −30 HU range in order to delineate the subcutaneous and visceral compartments and to compute the cross-sectional area of each in cm2 (figure 1). These measurements were performed by a radiologist blinded to patient information.

Figure 1

For each of the two groups—bevacizumab group (top) and chemotherapy group (bottom)—time to progression (TTP) was compared according to body mass index (BMI) (left), subcutaneous fat area (SFA) (middle) and visceral fat area (VFA) (right) dichotomised to the median (Kaplan–Meier estimates).

Statistical analysis

Since the primary aim of our study was to demonstrate the predictive value of BMI, SFA and VFA for response, time-to-progression (TTP) and overall survival (OS) in patients treated with bevacizumab, analyses were initially performed in the bevacizumab group. The analyses were then repeated among patients treated with chemotherapy to check that in this group, BMI, SFA and VFA were neither associated with response nor TTP or OS. Finally, interactions between bevacizumab administration and, respectively, BMI, SFA and VFA were tested in the whole population to confirm the results. For all analyses, BMI, VFA and SFA were dichotomised using the median as the cut-off point.

We looked for associations linking BMI, SFA and VFA to the response (CR–PR vs SD–PD) by comparing distributions using the Wilcoxon rank-sum test. To summarise predictive value for BMI, SFA and VFA, we computed the area under the receiver-operating characteristic (ROC) curve with its 95% CI. Univariate logistic regression was used to compute odds ratios for non-response (SD or PD) for BMI, SFA and VFA. Finally, a multivariate logistic regression model including VFA, SFA, WHO performance status (WHO PS), BMI, type of chemotherapy, liver-only metastases, age and sex was built and internally validated using bootstrapping (200 replications).

OS was defined as the time from the first day of chemotherapy to death (all causes). Survivors were censored at last follow-up. TTP was defined as the time from the first day of chemotherapy to the first recorded evidence of progression. Survivors were censored at last follow-up and non-survivors without progression were censored at the date of death.

Median follow-up with its 95% CI was calculated using the reverse Kaplan–Meier method. Survival curves were estimated using the Kaplan–Meier method and compared using log-rank tests. Univariate Cox proportional-hazards models of all potential baseline predictors were built to compute the hazard ratios (HRs) with their 95% CIs. A multivariate Cox model for TTP was constructed including VFA, SFA, WHO PS, BMI, type of chemotherapy, liver-only metastases, age (continuous), sex and serum carcinoembryonic antigen (CEA) level (continuous). For overall survival, owing to the number of events, we included only the variables that were significant or nearly significant by univariate analysis and SFA.

From the final multivariate model, we computed the Akaike information criterion for goodness of fit and Harrell's C-statistic for discrimination (a Harrell's C index of 0.5 indicates no predictive discrimination and a Harrell's C index of 1.0 indicates perfect separation of patients). The multivariate models were internally validated using bootstrapping (200 replications). All analyses were performed using Stata V.10 software (StataCorp LP, College Station, Texas, USA). p Values were two tailed and considered significant when ≤0.05.

Results

Patient characteristics and clinical outcomes

The baseline characteristics of the 120 patients enrolled are listed in table 1.

Table 1

Main characteristics of the 120 study patients

There was no difference in the two groups for the main characteristics except for age, metastases and type of chemotherapy. Patients in the chemotherapy group were older (median 67.5 years) than in bevacizumab group (median 62 years). Only 23 (29%) patients in the bevacizumab group versus 26 (65%) patients in the chemotherapy group had potentially curative disease without extrahepatic metastases. Finally, first-line chemotherapy consisted of LV5FU2 (7.5%, n=6), FOLFOX (44%, n=35) or FOLFIRI (49%, n=39) in the bevacizumab group, whereas it consisted of FOLFOX (77.5%, n=31) and FOLFIRI (22.5%, n=9) in the chemotherapy group.

Median follow-up at the data cut-off point was 24 months (range 3–70) in the bevacizumab group and 30 months (range 4–84) in the chemotherapy group. No patients died before their first radiographic assessment.

Response to treatment

Bevacizumab group

Mean BMI, SFA and VFA values differed significantly between responders and non-responders in the bevacizumab group. Mean BMI was 23.1±4.3 in responders and 25.8±4.4 in non-responders (p=0.0022, Wilcoxon rank-sum test). Mean SFA was 175.18±92 cm2 in responders and 280.45±209.2 cm2 in non-responders (p=0.0023, Wilcoxon rank-sum test). Mean VFA was 85.39±55.31 cm2 in responders and 144.28±56.56 cm2 in non-responders (p=0.00001, Wilcoxon rank-sum test). The area under the ROC curve (AUROC) for BMI was 0.70 (95% CI 0.58 to 0.82). The AUROC for SFA was 0.70 (95% CI 0.58 to 0.81) and the AUROC for VFA was 0.77 (95% CI 0.67 to 0.88).

By univariate logistic regression analysis, factors significantly associated with non-response were high BMI, high SFA, high VFA, WHO PS of 2 and diffuse metastatic disease (online supplementary table 1). By multivariate analysis, the only factors that independently predicted non-response were WHO PS of 2 and high VFA (p=0.004 and p=0.008, respectively) (supplementary table 1). Compared with patients with low VFA values, those with high VFA values had a seven times greater risk of non-response (OR=7.18; 95% CI 1.69 to 30.6). The p value obtained using bootstrapping was not statistically significant.

Chemotherapy group

Mean BMI, SFA and VFA values did not differ significantly between responders and non-responders in the chemotherapy group. Mean BMI was 25.0 (SD 4.3) in responders and 24.2 (SD 4.5) in non-responders (p=0.8360, Wilcoxon rank-sum test). Mean SFA was 212.71 cm2 (SD 110.47) in responders and 219.41 cm2 (SD 108.52) in non-responders (p=0.8684, Wilcoxon rank-sum test). Mean VFA was 118.66 cm2 (SD 49.41) in responders and 125.0 cm2 (SD 55.09) in non-responders (p=0.8684, Wilcoxon rank-sum test). The AUROC for BMI, SFA and VFA was, respectively, 0.48 (95% CI 0.32 to 0.64), 0.52 (95% CI 0.36 to 0.68) and 0.52 (95% CI 0.36 to 0.68).

In logistic univariate analyses, high BMI (OR=1 (0.28 to 3.54), p=1.0), high SFA (OR=0.54 (0.14 to 2.03), p=0.359) and high VFA (OR=1 (0.28 to 3.54), p=1.0) were not significantly associated with response. None of the other clinical factors were significantly associated with response. Thus, multivariate logistic regression model was not performed.

Whole population

Interaction tests between bevacizumab administration and, respectively, BMI, SFA and VFA using logistic regression in the whole population highlighted that interaction was significant for VFA (OR=3.31 (1.43 to 7.68), p=0.005) and BMI (OR=2.33 (1.04 to 5.22), p=0.039) while it was not for SFA (OR=1.98 (0.89 to 4.37), p=0.093).

Time-to-progression

Bevacizumab group

Disease progression occurred before the data cut-off point in 64 patients. As shown in figure 1, TTP was significantly shorter in patients with high BMI values than in patients with low BMI values (9 months vs 12 months; p=0.01, log-rank test). TTP was shorter in patients with high SFA than in patients with low SFA values (p=0.0453, log-rank; figure 1). Finally, TTP was shorter in patients with high VFA values than in patients with low VFA values (9 months vs 14 months; p=0.0008, log-rank test; figure 1). Factors that predicted a short TTP by univariate Cox regression were high BMI, high VFA, WHO PS of 1 or 2,and CEA level (table 2). Harrell's C-statistic values were 0.61, 0.57 and 0.65 for BMI, SFA and VFA, respectively. The multivariate Cox regression model included age, sex, WHO PS, type of chemotherapy, CEA level, liver-only metastases, BMI, SFA and VFA. Three factors were independently associated with a shorter TTP: WHO PS of 2 (HR=5.49 (2.07 to 14.5)), high VFA (HR=2.80 (1.35 to 5.79)) and CEA level (HR=1.00006 (1.000002 to 1.0001)) (table 2). Harrell's C-statistic was 0.78, indicating good discrimination by the multivariate model. Moreover, the Harrell's C-statistic of 0.65 for high VFA in the univariate analysis indicated that high VFA was one of the major predictors of TTP in the multivariate model.

Table 2

Results of the univariate and multivariate analyses for factors associated with time to progression

To assess the validity of the multivariate Cox model, bootstrapping was performed using 200 replications generated from the original sample. Only WHO PS of 2 (p=0.004) independently predicted TTP, although high VFA was close to statistical significance (p=0.075).

Chemotherapy group

Disease progression occurred before the data cut-off point in 28 patients. TTP did not differ between patients with high BMI values and those with low BMI values (6 months vs 8 months; p=0.75, log-rank test; figure 1), or between patients with high SFA and those with low SFA values (6 months vs 8 months; p=0.35, log-rank test; figure 1) or between patients with high VFA values and those with low VFA values (6 months vs 10 months; p=0.65, log-rank test; figure 1). Factors that predicted a short TTP by univariate Cox regression were WHO PS of 1 (HR=3.24 (1.25 to 8.40), p=0.016) or 2 (HR=3.86 (1.33 to 11.14), p=0.013) and CEA level (HR=1.0002 (1.00002 to 1.0003), p=0.028). Given the few events and because BMI, SFA and VFA had no significant effect on TTP, no multivariate model was applied.

Whole population

Interaction tests between bevacizumab administration and, respectively, BMI, SFA and VFA and using Cox regression in the whole population highlighted that interaction was significant for VFA (HR=1.64 (1.07 to 2.52), p=0.022) while it was not for BMI (HR=1.42 (0.92 to 2.19), p=0.117) and SFA (HR=1.29 (0.84 to 1.98), p=0.236).

Overall survival

Bevacizumab group

At the data cut-off point, 22 patients had died and all of them had experienced a progression. Although the median was not reached, overall survival was shorter in the patients with high BMI values than in patients with low BMI values, but the difference was not statistically significant (p=0.2232, log-rank test). Shorter overall survival of borderline significance was noted in the patients with high VFA values compared with those with low VFA values (p=0.0493, log-rank test). SFA was not significantly associated with overall survival (p=0.7206, log-rank test). By univariate Cox regression, factors predicting shorter overall survival were high VFA, WHO PS of 1 or 2 and CEA level (supplementary table 2). Harrell's C-statistic was 0.64, 0.57 and 0.62 for high VFA, high SFA and high BMI, respectively.

The multivariate Cox regression model included WHO PS, CEA, high SFA and high VFA. Factors that independently predicted shorter survival were WHO PS of 1 (HR=8.48 (2.35 to 30.6)), WHO PS of 2 (HR=10.9 (2.94 to 40.4)) and high VFA (HR=2.88 (1.13 to 7.32)) (supplementary table 2). Harrell's C-statistic was 0.84, indicating good discrimination of the multivariate model for predicting overall survival.

When bootstrapping was performed to check the validity of the multivariate Cox model, the association between high VFA and shorter overall survival was close to statistical significance (p=0.085).

Chemotherapy group

Death occurred before the data cut-off point in 22 patients. OS differed neither in patients with high BMI values versus low BMI values (17 months vs 15 months; p=0.54, log-rank test), nor in patients with high SFA versus low SFA values (17 months vs 15 months; p=0.23, log-rank test) and nor in patients with high VFA values versus low VFA values (17 months vs 15 months; p=0.86, log-rank test). Factors that predicted a short OS by univariate Cox regression were WHO PS of 2 (HR=5.54 (1.81 to 16.9), p=0.003), and CEA level (HR=1.0003 (1.00004 to 1.0005), p=0.021). Given the few events and because BMI, SFA and VFA had no significant effect on TTP, no multivariate model was applied.

Whole population

Interaction tests between bevacizumab administration and, respectively, BMI, SFA and VFA using Cox regression in the whole population highlighted that interaction was not significant for VFA (HR=1.01 (0.53 to 1.93), p=0.983), for BMI (HR=0.87 (0.44 to 1.72), p=0.687) and for SFA (HR=0.69 (0.36 to 1.35), p=0.284).

Discussion

Obesity is a well-established risk factor for developing CRC1 2 and is associated with increased mortality from colon cancer.25 26 Moreover, obesity, a sedentary lifestyle, and a typical Western diet are associated with increased rates of cancer recurrence and death among patients having a history of curative surgical resection for CRC.3 27–31 However, the definition of obesity is controversial and it is unclear whether BMI is the most appropriate measure of obesity.32 Several studies suggest that waist circumference or the waist–hip circumference ratio may perform better than BMI as predictors of the risk of developing colon or prostate cancer.33–35 These circumferences are crude measures of body fat distribution that fail to distinguish between deep abdominal or visceral fat and subcutaneous fat.23 36 37 CT can be used to accurately assess intra-abdominal fat13–15 via measurements of SFA and VFA at the level of the umbilicus.

Adipose tissue is now recognised as an endocrine and paracrine organ that releases cytokine-like polypeptides responsible for widespread biological effects.23 36 In particular, adipocytes produce insulin-like growth factor and multiple angiogenic factors including VEGF and leptin.9 Leptin exerts direct angiogenic effects,9 36 upregulates VEGF mRNA expression38 and induces VEGF-driven angiogenesis by colon epithelial cells.12 Inflammatory cells infiltrating the adipose tissue and adipose stromal cells also contribute significantly to VEGF production.9 39 40 Thus, elevated serum VEGF levels have been found in overweight or obese patients.10 11

The cytokine production profile differs between subcutaneous fat and visceral fat.16–21 For instance, the level of VEGF production is higher in omental fat than in fat located at any other site in the body.9 These functional differences may explain why obesity-associated metabolic disorders22 23 and serum VEGF levels10 correlated positively with visceral fat but not subcutaneous fat. Furthermore, visceral fat accumulation has been shown to be associated with colorectal adenoma formation,41 colon carcinogenesis42 and the risk for developing colon cancer43 44; and the amount of visceral fat significantly predicted disease-free survival in patients with resectable CRC.45 These data strongly suggest that visceral fat may induce the accumulation of protumourigenic factors46 and may be associated with poorer outcomes in patients with CRC. However, the prognostic significance of visceral fat in patients with metastatic CRC had not been studied previously.

Recent phase III trials showed that adding bevacizumab to a first-line conventional chemotherapeutic regimen improved progression-free survival and overall survival in patients with metastatic CRC.47–49 Although VEGF expression is associated with advanced tumour progression and a poor prognosis in patients with colon cancer,50 no factors that predict the response to bevacizumab or conventional chemotherapy have been validated to date. In patients with breast cancer, high tumour levels of VEGF were associated with greater resistance to conventional chemotherapy.51 Interestingly, obese animals proved resistant to anti-VEGF treatment.52 These data could support the hypothesis that a large amount of visceral fat may be associated with high VEGF levels and therefore with resistance to bevacizumab-based regimens in patients with metastatic CRC.

Until now and despite extensive investigation, there are no known predictive biomarkers of VEGF-targeted treatment.53 Our study provides the first evidence that a large amount of visceral fat is independently associated with worse outcomes in patients given first-line bevacizumab for metastatic CRC. This well-defined population was chosen for homogeneity reasons and because bevacizumab-based treatment is currently a standard treatment in metastatic CRC. Given the absence of normative data on VFA or SFA in the literature, we used the median as the cut-off point. Both TTP and overall survival were shorter in patients with VFA values above the median than in patients with VFA values below the median. In contrast, BMI was not independently associated with the response rate, TTP or overall survival. Thus, VFA may be more accurate than BMI for predicting the treatment response and TTP. In the phase III TREE study,54 the median TTPs (8.3–10.3 months) were consistent with the 9-month TTP in our patients with high VFA values but were considerably shorter than the 14-month TTP in our patients with low VFA values.

Limitations of our study include the small number of patients, limited follow-up duration, single-centre patient recruitment and retrospective design. However, VFA was determined by a radiologist who had no information about the patients. A small sample size contributes to overfitting: a phenomenon in which a predictive model may well describe the relationship between predictors and outcome in the patients used to develop the model, but may subsequently fail to provide valid predictions in new patients.55 Then, bootstrapping was performed to internally validate the results and prevent overfitting. The results obtained by bootstrapping highlighted that high VFA remains a major independent predictor of short survival with a p value close to statistical significance. Further studies are continuing to validate our findings in a different dataset and to determine the optimal cut-off point for VFA. Finally, the study design does not allow exclusion of the possibility that the reason for the adverse correlation between fat and response to bevacizumab is the result of the distribution of the drug into the fat.

In conclusion, our results provide the first evidence that VFA measured before starting first-line bevacizumab-based treatment is likely to be a useful predictive biomarker in metastatic CRC. Further studies may help us to determine whether the predictive effect of high VFA is related to (a) a larger volume of distribution of bevacizumab; (b) the production of high levels of VEGF by visceral fat or (c) both preceding hypotheses. Consequently, patients with high VFA might either not benefit from bevacizumab or require a higher dosage, such as that used in breast cancer or in clear cell renal cell cancer. If further validation studies corroborate our results, the measurement of VFA will have to be included in clinical trials for metastatic CRC, thereby taking into account tumour parameters and also host parameters.

Acknowledgments

The authors thank Sandrine Guiu for revising the English.

References

View Abstract

Supplementary materials

  • Correction gut.2009.188946

    Files in this Data Supplement:

Footnotes

  • Linked articles 197210.

  • Funding ARC (Association Recherche Cancer), 9 rue Guy Moquet, 94800 Villejuif, France Ligue Nationale contre le Cancer, Comité de Côte d'Or, 38B rue Tivoli, 21000 Dijon, France.

  • Competing interests None.

  • Ethics approval This study was conducted with the approval of the Comité de Protection des Personnes (CPP) Est I - Pr B. Blettery.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Linked Articles