The present systematic review and meta-analysis indicates that the addition of Bevacizumab to chemotherapy in patients with metastatic solid tumors resulted in a statistically significant improvement of OS and PFS. The OS benefit was significant and homogenous, with a HR 0.89 (95%CI; 0.84–0.93 P<0.00001, I2-4%) The effect on OS was similar for all malignancies, except for breast cancer.
The PFS benefit was significant with a HR 0.71 (95%CI 0.68–0.74 P<0.00001 I2-54%) but with high heterogeneity attributed to the NSCLC, colon and RCC trials.
There may be several potential explanations for the heterogeneity in PFS.
First, cessation of Bevacizumab treatment may lead to a rebound effect of angiogenesis, 
. This effect might explain the different results and heterogeneity of the breast cancer trials. The E2100 breast cancer trial continued Bevacizumab until progression as compared with the AVADO breast cancer trial that gave only nine cycles of Bevacizumab. This hypothesis may justify the continuation of Bevacizumab beyond progression 
that it is currently under evaluation (Ribbon 3).
Second is the lack of predictive markers for Bevacizumab treatment. Hypertension (HTN) is known to develop in up to 30% of patients treated with Bevacizumab. Retrospective subgroup analyses have appraised the predictive value of HTN as a biomarker for response to Bevacizumab in breast and colon cancer, and have shown correlation with response but not with survival benefit 
. A subgroup analysis of the RCC CALGB trial demonstrated a significant median OS benefit in patients with HTN≥grade 2 (according to the CTCAE 3) of 41.2 months versus 16.2 months in patients without HTN P<0.001 
. This putative correlation has not been evaluated in the large colon cancer studies. Other predictive parameters such as VEGF levels, vascular density or VEGF polymorphisms and molecular markers could also serve as markers of response to Bevacizumab. In the E2100 Breast cancer trial, VEGF polymorphisms (VEGF-2578 AA and VEGF-1154 AA) genotypes were associated with a superior median OS in the Bevacizumab arm compared with other genotypes (37 months and 46 months respectively compared to 25 months in the control arm) 
Third, the difference in the results may lie in the chemotherapy agents combined with Bevacizumab: It has been postulated that Paclitaxel given on a frequent basis also exhibits antiangiogenic and pro-apoptotic effects (partially by down-regulation of VEGF), thereby enhancing efficacy. 
. The different result of the E2100 and the AVADO trials (breast cancer) may also be a result of the Taxane protocol used – weekly Paclitaxel as opposed to three weekly Docetaxel, an effective but toxic regimen.
In the lung cancer trials Bevacizumab again demonstrated superior OS when combined with a taxane based regimen as opposed to a gemcitabine based regimen 
In colon cancer trials the heterogeneity could also be explained by the different chemotherapy regimens. In Hurwitz et al; 
the chemotherapy regimen used was IFL, known to be a toxic regimen and now rarely used whereas Saltz et al; 
added Bevacizumab to FOLFOX.
Regarding colon cancer our results are supported by other meta-analyses evaluating the addition of Bevacizumab to chemotherapy in the metastatic setting 
These should be interpreted in the light of the disease specific survival of the different maligancies. Therefore 3 months of survival benefit in metastatic colon cancer where the expected OS is more than 20 months differs from a three months benefit in a patient with metastatic lung or pancreatic cancer in which the median survival is less than a year 
The role of PFS as a surrogate for overall survival has been greatly debated in metastatic cancer. The multiple effective treatment options may explain the lack of OS advantage. Broglio et al assumed that OS represented the sum of PFS and survival post progression (SPP). He assumed that SPP contributes more to OS than PFS, therefore when SPP is 2 months there is a greater than 90% chance of detecting a statistically significant OS benefit but when SPP is 24 months there is a less than 20% chance of detecting a statistically significant OS benefit. Accordingly, breast cancer serves as a protoype for a long SPP and therefore no OS benefit was seen in these studies as compared with NSCLC with a short SPP 
Another important issue is that when improvements in OS are observed, their impact should be assessed considering the effect of treatment on the quality of life of the patient. Our toxicity analysis () shows an increase in adverse events >G2 RR-1.26, and fatal adverse events RR-1.47 with and increased risk for fatal pulmonary hemorrhage in the lung cancer trials with a RR-5.65. Our analysis also suggested that the most significant risk of FAEs was in patients with prostate and lung cancer, as reported by others 
. Our analysis showed no increased risk of venous thromboembolic events, in concordance with another recently published meta-analysis 
Moreover our analysis showed that the different serious adverse events seemed to be more prevalent in certain tumors compared to others: G3–4 adverse events were more common with NSCLC, colon and gastric cancer. HTN was significant in all cancer types but was most significant in breast cancer with a RR-17.63, proteinuria was also significant in all types but not in gastric and colon cancer, hemorrhage was significant only in RCC, and GI perforation was significant only in colon cancer RR3.99 (95%CI1.34–11.85 p
0.01). There was an increased risk of arterial thromboembolic events in breast cancer RR-5.97 (95% CI 1.07–33.22), and in RCC RR 6.55 (95%CI 1.5–28.59).
The calculated number needed to harm upon the incidence of fatal events is 117, i.e., for one fatal event in the bevacizumab arm 117 patients would need to be treated (RR 1.4).
The results of our analysis of Bevacizumab as second line therapy concurrently with chemotherapy in breast cancer patients showed the same trend of improved PFS and lack OS benefit, a fact that may support the use of Bevacizumab in advanced lines and could be more cost effective.
Limitations of our analysis should be acknowledged. The dissimilarity in the chemotherapy regimens, dosing and schedules between the various studies confound the analysis. The heterogeneous length of treatment and follow up contribute to the asymmetry as well. Also, the small number of included trials for each disease makes the outcomes more prone to be influenced by a potential publication bias. We attempted to avoid such bias by searching and including conference proceedings, databases of ongoing trials and unpublished data. The biological rationale to combine all studies in order to assess Bevacizumab universal effect might offset this limitation.
Implications for clinical practice
This analysis showed an overall survival benefit. It reinforces the use of Bevacizumab in colon and lung cancer.
Regarding breast cancer, the FDA recently revoked its approval of the use of Bevacizumab. Our data supports this decision at this time. Promising data supporting the use of Bevacizumab in the neoadjuvant treatment of locally advanced breast cancer 
emphasizes the need for further studies in the search of predictive markers.
Analysis of databases should be attempted in order to estimate the effect of Bevacizumab on non trial populations, as was recently published by Zhu et al 
Implications for research
Many questions still remain regarding the effect of Bevacizumab as maintenance therapy compared with Bevacizumab at disease progression and the optimal schedule and type of chemotherapy in each disease. Also, Bevacizumab is a nonspecific agent without any well defined predictive markers.
Data supporting the use of Bevacizumab as neoadjuvant therapy in breast cancer patients have been published recently 
and therefore further studies should be considered in the search for predictive markers.
Analysis of the trials already published in search of predictive markers through a patient based metaanalysis could help define possible predictive markers for future validation studies.
Finally, randomized controlled trials should have longer follow-up to appraise the long-term toxicity of Bevacizumab.
In conclusion, our results suggest that adding Bevacizumab to chemotherapy results in a small but significant effect on OS and a significant PFS advantage in the advanced solid tumors included in this analysis.