In this systematic review, by pooling all the studies comparing survival of lung cancer patients according to the angiogenic activity of the tumour, as expressed by the MVC, we show that a high MVC is a poor prognosis factor for survival in surgical NSCLC whatever the antibody used for assessment of the vessel count. This observation is potentially important for prognostic reasons and treatment purposes. Angiogenesis assessment might be useful not only in stratifying patients for different (adjuvant) treatment regimens but also in predicting their response to chemotherapy (Koukourakis, 2001
; Mattern, 2001
), to anti-angiogenic therapies and identifying the precancerous lesions (Pazouki et al, 1997
It is also to be noted that the trials published on the topic concerned NSCLC treated, at least, by surgery; we could thus not extrapolate our results to metastatic NSCLC or small cell lung cancer.
To perform the meta-analysis, we have used a methodology similar to previous systematic reviews of our group on the treatment of lung cancer (Meert et al, 1999
) but adapted to the field of biological prognostic factors (Steels et al, 2001
). The absence of statistically significant difference in quality score between significant and non-significant publications allowed us to perform a quantitative aggregation of the individual trials results.
Our approach does not however prevent all potential biases. We restricted our review to articles published in French or English language because other languages such as Japanese were not accessible to the readers. This selection could favour the positive studies that are more often published in English while the negative ones tend to be more often reported in native languages (Egger et al, 1997
). Another possible source of confusion is the use of a same cohort of patients in different publications. It might be difficult to avoid the same patients being included more than once in the meta-analysis. We have excluded publications where it seemed to be the case after writing to some of the authors in order to have more information on patients' cohort, a procedure that was unsuccessful: we had only one partial response. Harris confirmed that there were an overlapping of the same patients in the series of Giatromanolaki, Koukourakis et al
The method of extrapolation of HRs also needs to be discussed. When HRs were not reported by the authors, they were calculated from the data reported in the article and, if not available, extrapolated from the survival curves, implying assumptions on the censoring process. This approach might also have been associated with errors due to imprecision in the reading, although three independent persons read the curves to reduce the reading variation.
Our review took into account only fully published studies. We did not look for unpublished trials and abstracts because our methodology required data available in full publications only. Meta-analysis based on individual data is considered by some authors as a gold standard (Stewart and Parmar, 1993
). Systematic reviews of the literature should not be confused with meta-analyses of individual patient data. The first approach is based on fully published studies and provides an exhaustive and critical analysis on the topic with an adequate methodology based on the criteria of Mulrow (1987)
. The second approach is, in fact, a new study taking into account all performed studies on the topic, published or not, requiring individual data update by the investigators and is much more time-consuming. Nevertheless, as shown by our meta-analysis on the role of prophylactic cerebral irradiation in small-cell lung cancer (Meert et al, 2001
), based on published data, we obtained the same results as in the meta-analysis based on individual data (Auperin et al, 1999
Variations in survival results among the studies could be explained by the heterogeneity in methodologies used to stain and count microvessels in the tumours in addition to variation in patients population. The estimated vascularity in tissues sections can be significantly affected by variations in the applied methodology including pre-treatment and antibody use. The vessels in tumour samples can be identified by some different endothelial cell-specific antibodies: most often recognising factor VIII, CD31 or CD34 related antigen. Factor VIII (Von Willebrand's factor) was one of the first marker used for staining microvessels but it may be imprecise to quantify microvasculature for various reasons. Firstly, factor VIII is not expressed in all endothelial cells. The microvessels endothelial cells are less rich in Weibel-Palade bodies, which store factor VIII, than the endothelial cells of macrovessels (and the endothelial cells of neocapillaries may be activated by cytokines releasing their factor VIII stores). Secondly, factor VIII is also present in lymphatic endothelium and in platelets leading to a cross-reactivity with megakaryocytes, platelets and lymphatic endothelial cells. CD34, a heavily glycosylated transmembrane protein, is expressed on immature human haematopoietic precursor cells and is progressively lost during maturation. It is also present in the luminal endothelial membrane. CD34 is more sensitive and specific than factor VIII for staining endothelial cells induced by tumour neovascularisation (Tanigawa et al, 1996
) but could also stain some lymphatic vessels. Only specific antibodies (i.e. LYVE 1, VEGF-C) can be used to detect lymphatics and not blood vessels. Anti-CD34 antibody seems to be more reliable in terms of specificity and reproducibility than monoclonal antibodies generated against other endothelial cell antigens (Tanigawa et al, 1996
). In invasive breast cancer, CD34 has been shown to yield higher microvessel values than CD31 or factor VIII (Martin et al, 1997
) and does not stain any tumour or inflammatory cells as CD31 or factor VIII. CD31 is a transmembrane glycoprotein highly expressed in mature and immature endothelium and its localisation at the endothelial cell junctions suggests an important role in transendothelial migration. CD31 is expressed during myelomonocytic cellular differentiation and consequently may cross-react with plasma cells, platelets, neutrophils, peripheral T cells and mantle zone B cells; endothelial staining can be easily differentiated on the basis of the morphological differences. JC70 antibody stains also CD31 positive lymphocytes that could be a prognostically important inflammatory component in lung cancer (Giatromanolaki et al, 1997
). For some authors, CD31 seems to be the most sensitive marker for the endothelial cells and consistently stains more vessels than did factor VIII (Horak et al, 1992
). An international consensus on the methodology and criteria of evaluation of microvessel density proposed that anti-CD31 monoclonal antibody should be the standard for microvessel assessment (Vermeulen et al, 1996
) as it is superior on paraffin sections. But as CD34 has been shown to yield higher microvessel values than CD31 or factor VIII in breast cancer (Martin et al, 1997
), it might be useful to combine CD34 and CD31 antibodies. In lung cancer, Offersen et al (2001)
compared the staining with these three antibodies. He found that CD34 showed the best labelling of the endothelial cells and no background staining (data not shown). Yano et al (2000)
found that correlation between factor VIII and CD34 staining for MVC was not strong and that staining for CD 34 significantly correlated with survival in adenocarcinoma but staining for factor VIII did not. Duarte et al (1998)
reported that CD 31 did not predict survival in stage I NSCLC and did not correlate strongly with factor VIII which is correlated with lung cancer death. Giatromanolaki et al (1997)
concluded that CD31 is sensitive for highlighting small, immature microvessels and is better correlated with nodal involvement and overall survival than factor VIII. Unfortunately, data were not sufficient to compare the three antibodies by a meta-analysis methodology.
Contradictory results in the literature may also be explained by variations in vascularity between areas in different sections from the same block or among blocks taken from the same tumour (de Jong et al, 1995
) and by the methods used to measure vascularity (Pazouki et al, 1997
). In large tumours, it could be necessary to examine multiple blocks in order to determine the overall vascularity of the tumour. Identifying the area of maximal microvessel density seems to be an important step in the counting method (Vermeulen et al, 1997
) as tumour dissemination is more likely to occur at sites of high microvessel density. In lung cancer, the border between malignant and benign tissues is often blurred by atelectases, fibrosis and inflammatory cells, making the problem more difficult. The difficulty in recognising the vascular ‘hotspots’ may account for studies that failed to find an association between MVC and poor patients survival.
The technique used to count the microvessels is also different among the articles. Most of the studies used a technique similar to that proposed by Weidner et al (1991)
. The areas of highest neovascularization (‘hotspots’) is found by scanning the tumour sections at low power (40× and 100×) and then individual microvessels are counted on a 200× and 400× field. Each count is expressed as the highest number of microvessels identified within any 200× or 400× field. This technique is slow and laborious. A eye piece graticule (as a 25-point Chalkey graticule) has also been applied for vascular scoring in patients with NSCLC (Giatromanolaki et al, 1996a
). In breast cancer, Fox et al (1995)
showed that Chalkey counting is a rapid and objective method of quantifying tumour angiogenesis and gives independent prognostic information. A proposition of consensus identified the Chalkey method as slightly more objective (Vermeulen et al, 1996
). We did not perform aggregation of the results in term of microvessel counting technique because the techniques were too heterogeneous.
Moreover, there is no standardised cut-off used for stratifying patients into high and low vascular groups. Some authors used the MVC median or the MVC mean and others the ‘best cut-off’, which is often arbitrary defined or chosen using multiple tests with a corresponding increase in the probability of founding a false positive results. The selection of the median value of the expression levels is a standard approach to analyse new prognostic factors, even if it may lead to some loss of information.
Assessment of tumour vascularity by immunohistochemistry on paraffin-embedded tissues can be easily performed in laboratory but standardisation of angiogenesis quantification is necessary in order to better define its prognostic value (Vermeulen et al, 1996
) and to facilitate a routine use.
In conclusion, a high MVC, reflecting tumour neoangiogenesis, is a poor survival prognostic factor for NSCLC surgically treated patients. These results were based on an aggregation of data obtained by univariate survival analysis in retrospective trials. In order to become an useful prognostic factor, a standardisation of angiogenesis quantification is necessary and the present results need to be confirmed by an adequately designed prospective study with an appropriate multivariate analysis taking into account the classical well defined prognostic factors for lung cancer.