In their commentary ¿Fascin-1 is a novel biomarker of aggressiveness in some carcinomas¿ (BMC Medicine 2013, 11:53), Kulasingam and Diamandis raise several questions about outcomes or interpretations of our meta-analysis ¿Association of fascin-1 with Mortality, Disease Progression and Metastasis in carcinomas: a systematic review and meta-analysis¿ (Tan et al., BMC Medicine 2013, 11:52). We would like to respond to these questions.
1. Kulasingam and Diamandis question how fascin-1 could be associated with increased risk of lymph node and distant metastases in gastric cancer, yet was not found to be associated with mortality in gastric cancer. It would be premature and not correct to say there is no association of fascin-1 with mortality in this form of carcinoma. Rather, on the basis of the data published to date, there is a lack of statistical evidence to support an association with gastric cancer mortality, which may simply be an issue of sample size and statistical power (see Sterne and Davey Smith, 2001, British Medical Journal 322: 226-231). Two of the three published studies on which our meta-analysis was based examined relatively small numbers of tumour specimens, thus the overall dataset for analysis included 237 cancer cases vs 513 controls (Fig. 2 of Tan et al., 2013). The observed effect was a hazard ratio of 1.16 i.e. a positive association in line with the positive associations with lymph node and distant metastases. However, the 95% confidence interval in the meta-analysis was wide, 0.86 to 1.56 (Fig. 2 of Tan et al., 2013). This suggests that more data are required to be able to make a robust inference about whether or not there is an association with mortality. We cannot say at this stage that there is no association and, in view of the wide confidence interval, we did not state a positive association with mortality in gastric cancer as a conclusion of the meta-analysis.
2. Kulasingam and Diamandis question why we reported no association of fascin-1 with lymph node metastasis of lung carcinomas ¿when Fig. 4 of Tan et al., shows an increased risk¿. With reference to the lung cancer metastasis data in Fig. 4A, a fixed effects model gave strong evidence of heterogeneity between studies (I square = 63.3%). Thus, we could not assume a fixed effect model and our inference was based on a random effects model, which allows a distribution of true effect-sizes. Although a positive risk ratio of 3.11 was obtained, the random effects model yielded a wide 95% confidence interval that spanned the null value (0.64-15.25; Fig.4A). On the basis of the data available at present, we cannot say there is no association of fascin-1 with lymph node metastasis in lung cancer, only that there is little statistical evidence to support an association (Sterne and Davey Smith, 2001). Clearly, more data are required for a definitive answer and we welcome the suggestion of Kulasingam and Diamandis that well-designed prospective studies of fascin-1 are now needed.
3. Kulasingam and Diamandis also query whether fascin-1 has the potential to be a good biomarker because it is expressed in carcinomas of many tissue types. Although possible, it is unlikely that many patients would present with primary tumours of more than one tissue origin. It is also the case that molecules over-expressed in more than one carcinoma type, for example the epidermal growth factor receptor, have already achieved clinical utility.
In our view, the possibility of a biomarker for aggressiveness that could be applied to more than one carcinoma type, combined with the absence of fascin-1 expression in many normal epithelia, adds to the interest of fascin-1¿s potential utility as a biomarker.
From Vanessa Tan(1,2) Sarah J. Lewis(1), Josephine C. Adams(2) and Richard M. Martin(1)
1School of Social and Community Medicine, University of Bristol, Bristol, UK
2School of Biochemistry, University of Bristol, Bristol BS8 1TD, UK
jo.adams@bristol.ac.uk, richard.martin@bristol.ac.uk
Response to Kulasingam and Diamandis' commentary
29 May 2013
In their commentary ¿Fascin-1 is a novel biomarker of aggressiveness in some carcinomas¿ (BMC Medicine 2013, 11:53), Kulasingam and Diamandis raise several questions about outcomes or interpretations of our meta-analysis ¿Association of fascin-1 with Mortality, Disease Progression and Metastasis in carcinomas: a systematic review and meta-analysis¿ (Tan et al., BMC Medicine 2013, 11:52). We would like to respond to these questions.
1. Kulasingam and Diamandis question how fascin-1 could be associated with increased risk of lymph node and distant metastases in gastric cancer, yet was not found to be associated with mortality in gastric cancer. It would be premature and not correct to say there is no association of fascin-1 with mortality in this form of carcinoma. Rather, on the basis of the data published to date, there is a lack of statistical evidence to support an association with gastric cancer mortality, which may simply be an issue of sample size and statistical power (see Sterne and Davey Smith, 2001, British Medical Journal 322: 226-231). Two of the three published studies on which our meta-analysis was based examined relatively small numbers of tumour specimens, thus the overall dataset for analysis included 237 cancer cases vs 513 controls (Fig. 2 of Tan et al., 2013). The observed effect was a hazard ratio of 1.16 i.e. a positive association in line with the positive associations with lymph node and distant metastases. However, the 95% confidence interval in the meta-analysis was wide, 0.86 to 1.56 (Fig. 2 of Tan et al., 2013). This suggests that more data are required to be able to make a robust inference about whether or not there is an association with mortality. We cannot say at this stage that there is no association and, in view of the wide confidence interval, we did not state a positive association with mortality in gastric cancer as a conclusion of the meta-analysis.
2. Kulasingam and Diamandis question why we reported no association of fascin-1 with lymph node metastasis of lung carcinomas ¿when Fig. 4 of Tan et al., shows an increased risk¿. With reference to the lung cancer metastasis data in Fig. 4A, a fixed effects model gave strong evidence of heterogeneity between studies (I square = 63.3%). Thus, we could not assume a fixed effect model and our inference was based on a random effects model, which allows a distribution of true effect-sizes. Although a positive risk ratio of 3.11 was obtained, the random effects model yielded a wide 95% confidence interval that spanned the null value (0.64-15.25; Fig.4A). On the basis of the data available at present, we cannot say there is no association of fascin-1 with lymph node metastasis in lung cancer, only that there is little statistical evidence to support an association (Sterne and Davey Smith, 2001). Clearly, more data are required for a definitive answer and we welcome the suggestion of Kulasingam and Diamandis that well-designed prospective studies of fascin-1 are now needed.
3. Kulasingam and Diamandis also query whether fascin-1 has the potential to be a good biomarker because it is expressed in carcinomas of many tissue types. Although possible, it is unlikely that many patients would present with primary tumours of more than one tissue origin. It is also the case that molecules over-expressed in more than one carcinoma type, for example the epidermal growth factor receptor, have already achieved clinical utility.
In our view, the possibility of a biomarker for aggressiveness that could be applied to more than one carcinoma type, combined with the absence of fascin-1 expression in many normal epithelia, adds to the interest of fascin-1¿s potential utility as a biomarker.
From Vanessa Tan(1,2) Sarah J. Lewis(1), Josephine C. Adams(2) and Richard M. Martin(1)
1School of Social and Community Medicine, University of Bristol, Bristol, UK
2School of Biochemistry, University of Bristol, Bristol BS8 1TD, UK
jo.adams@bristol.ac.uk, richard.martin@bristol.ac.uk
Competing interests
None declared