In cancer clinical trials, overall survival (OS), defined as time from randomization to death from any cause, is considered as the gold standard endpoint for the demonstration of clinical benefit 1 . It is an objective, unambiguously defined endpoint with clear clinical meaning to the patient. However, the evaluation of OS may require extended follow-up and the effect of the experimental treatment on OS may be confounded by effective subsequent therapies, including cross-over to the experimental treatment for patients initially allocated to control therapy 2 .

The use of an earlier alternative endpoint would allow to shorten clinical trials and to evaluate treatment effect with better specificity. Such an endpoint can be used as definitive measure of treatment clinical efficacy instead of OS if it is surrogate for OS. A surrogate endpoint is defined as a biomarker "expected to predict clinical benefit (or harm or lack of benefit or harm) based on epidemiologic, therapeutic, pathophysiologic, or other scientific evidence" 3 . In addition to the biologic plausibility of the relationship between the surrogate and OS, evaluation of a surrogate endpoint requires statistical analyses using data from completed phase III clinical trials to demonstrate how reliably and in which proportion an effect on the surrogate can predict the effect on OS 4 5 .

In digestive oncology, few variables have undergone evaluation process 5 6 7 . To date, only disease-free survival (DFS) for the adjuvant treatment of colon cancer 8 9 and progression-free survival (PFS) in the metastatic setting 10 11 have been fully validated as surrogates for OS for the evaluation of first line chemotherapies. It is therefore of interest to identify putative surrogates, so that they can get priority statistical evaluation.

The main objective of this study was to draw up a list of potential surrogate endpoints for OS in digestive oncology trials, by way of a survey among clinicians and methodologists involved in the conduct of clinical trials. Secondary objective was to obtain their opinion on particular points concerning surrogacy and quality of life (QoL).

We polled a panel of French clinicians and methodologists involved in cancer trials by means of self administered sequential questionnaires in 2007 and 2008. The panel was composed of 14 methodologists or biostatisticians, and 66 clinicians working in the field of cancer clinical trials (gastroenterologists, oncologists, surgeons, radiotherapists). Clinicians were members of the scientific committee of the Fédération Francophone de Cancérologie Digestive (FFCD), to whom the protocol of the survey was presented during a scientific council sitting.

The design of our survey was based on Delphi technique guidelines 12 , which is a consensus method. However, in our survey, while participants were asked to classify the endpoints they proposed, they did not re-evaluate their opinion according to responses of other panellists and formal consensus was not searched for.

Two rounds were planned (Figure 1). In each round the questionnaire was sent by postal way and by e-mail. Reminder letter (e-mail) was sent about two months after the postal delivery. The second questionnaire was sent to all panellists, whether they had answered the first one or not.

First Round. The first questionnaire was composed of three parts (see additional file 1: Survey questionnaire No1):

- Part I: panellists were asked 1) to choose the most important characteristics defining a surrogate among six proposals; 2) to give advantages and drawbacks of the surrogates and 3) to answer questions about their validation and use in digestive oncology;

- Part II: they were asked to mention, if any, endpoints they thought worthwhile to be evaluated as surrogate for OS in each of the following tumour sites, according to the stage of the disease (e.g. locally advanced, metastatic): oesophagus, stomach, liver, pancreas, biliary tract, lymphoma, colon, rectum, and anus;

- Part III: they were asked to give their opinion about quality of life (QoL) as candidate surrogate.

At the end of the questionnaire, they were free to add comments.

Second Round (see additional file 2: Survey questionnaire No2):

A feedback report summarized the respondents' answers to parts I and III items and accompanied the second questionnaire. The second questionnaire tabulated all the surrogate endpoints proposed in part II of the first questionnaire. In each situation, participants were asked to classify the previously proposed surrogates, from the most promising (position #1) to the last promising in their opinion.

Descriptive statistics (histograms with frequencies) were used to summarize the responses of the first round (parts I and III). For each proposed surrogate, the frequency (%) at which it was chosen in the three first positions was calculated and served as numerical classification for the final ranking.

Participants of our survey highlighted key points of the definition, properties and use of surrogate endpoints for OS in digestive cancer clinical trials. Because other criteria than OS take part in their therapeutic decisions and practice, they agreed on the need for validation, to ensure that surrogate endpoints allow reliable prediction of survival benefit. A large amount of endpoints was proposed for surrogacy evaluation. Event-free survival endpoints, such as DFS and PFS, were preferred to early treatment effect evaluation, such as R0 resection and response. There was discrepancy about the relevance of studying QoL as surrogate for OS. However, QoL was often proposed in metastatic stages, in association or not with PFS.

Our panellists were active participants in the field of oncology trials (investigators, methodologists), therefore the results of this survey may not fully reflect knowledge and opinions of other professionals not directly involved in the conduct of a trial. Indeed, we limited our panel to FFCD scientific council members and methodologists in anticancer centres for practical reasons, but also under the assumption that they would be more committed and disposed to participate. However, a low proportion of them fulfilled the questionnaires. One can suppose that some non-respondents felt not sufficiently comfortable with this issue. Besides, the parts of the questionnaires that were asking participants to suggest and classify potential surrogate endpoints may have best suited clinicians than methodologists. Moreover, since a lot of digestive tumour sites and settings were dealt with, the questionnaires may have appeared heavy-going. A higher participation rate could have generated more propositions of potential surrogate endpoints and might have modified the final sorting. The survey was also performed with an educational goal, to lead people to question themselves on the relevance of the endpoints used in oncology trials. Thus we chose a self-administered questionnaires approach to allow participants to freely express their opinion, without being influenced by others.

Respondent answers covered major issues surrounding surrogacy. The main point concerns the difference with a prognostic factor. Though most surrogate endpoints are also prognostic, a prognostic factor is not necessarily a surrogate 18 19 : "a correlate does not a surrogate make" 20 . Indeed, even if two endpoints are correlated, treatment effects on them are not necessarily correlated too and effect on an intermediate variable does not necessarily predict the effect on the final endpoint 20 . Half of our participants related surrogacy with outcomes accepted as primary efficacy endpoints for the evaluation of oncology products by medical agencies, such as the US Food and Drug Administration and the European Agency for the Evaluation of Medicinal Products. They actually both may rely on surrogate endpoints for marketing approvals 21 22 . A well-established (validated) surrogate would be able to support regular approval, whereas a surrogate "reasonably likely to predict clinical benefit" might only support accelerated/conditional approvals 23 24 . Because of the difficulty in establishing reliable surrogacy, debates still remain around the use of surrogate endpoints as definitive efficacy criteria 25 26 . Our participants underlined some risks and limitations. They reminded that surrogacy depends upon the disease setting and the class of therapeutic agents and that validation cannot be extrapolated to another situation without further investigation. They added that there may still remain a risk of erroneous conclusion on OS and that short follow-up does not allow to observe unintended late adverse events. The established surrogate association may also alter over time with diagnosis and therapeutic advances, as discussed for 3-year DFS in adjuvant colon cancer trials 8 9 27 28 .

DFS and PFS have been ones of the preferred candidate surrogate endpoints, whatever cancer site. This could be explained by the fact that both have been validated and used in colorectal cancer trials. DFS, defined as the time from randomization to the first event of either recurrent disease or death (second primary tumours not counted as events) has been validated as surrogate for OS in adjuvant colon cancer trials 8 9 . PFS, defined as the time from random assignment to progressive disease or death from any cause, has been validated in advanced colorectal cancer 10 11 . On the contrary, time-to-progression, which does not include death as an event contrary to PFS, was shown to be less reliable 10 16 . As for response rate, while allowing an early evaluation of treatment effect, it does not allow accurate prediction of OS in colorectal cancers 6 10 15 16 .

DFS and PFS were sometimes both proposed in the questionnaire for the same disease stage. In accordance with Delphi guidelines, which advise to keep all participant propositions from first to second round 12 , we let both endpoints in the second questionnaire. However, it is finally unclear regarding the events taken into account. This raises the preliminary question of the definition of the survival endpoints, for which a uniform terminology has been stressed as a main concern to compare trial results 29 . In digestive oncology, standardization of the definitions of such endpoints has been carried out for hepatocellular carcinoma trials 30 and colon cancer trials 31 32 . In order to reach a consensus in oncology trials, a European Delphi survey has been initiated by French methodologists and statisticians from oncology centres and from the FFCD, in collaboration with the European Organisation for Research and Treatment of Cancer. Secondary objectives of that study are to assess the influence of the definitions over trial results and to investigate their surrogacy.

In some situations, OS is no longer a reliable endpoint, e.g. in advanced colorectal cancer treated in the first line, mainly because of the increasing availability of effective agents in the subsequent lines 33 . In such cases, rather than or before dealing with surrogacy issue, one has to define the most appropriate clinical efficacy endpoint. Clinical value to the patient of endpoints other than OS is thus sometimes questioned. This has been discussed for PFS in advanced colorectal cancer 25 33 and for DFS in adjuvant trials, where disease recurrence has high subsequent costs, QoL impact and debilitating consequences 8 . Similarly, in rectal cancer, preoperative chemoradiation has become the standard neoadjuvant treatment because of a significant benefit with respect to local control compared with preoperative radiotherapy alone, though no impact on survival was demonstrated 34 35 .

As reminded by Fleming et al. 25 , "oncology patients are interested in achieving clinically meaningful beneficial effects on disease-related symptoms, on ability to carry out normal activities, and on OS". As a direct measure of how a patient feels, functions, or survives 3 , QoL is considered as a clinical endpoint, thus allowing QoL improvement alone to serve as a basis for marketing approval 36 . By definition, QoL would be additionally surrogate for OS if treatment effect on QoL predicted a survival benefit. While our participants were divided about studying QoL as surrogate for OS in the first round, QoL was generally well-rated in the second round for advanced stages. They put forward complementarity between PFS and QoL by proposing these criteria as a composite surrogate endpoint for OS. Two thirds of our participants declared they considered QoL as the secondary endpoint in cancer clinical trials. There is an increasing interest in endpoints such as time to symptomatic deterioration or time to QoL deterioration. However, the variability in the definition and analysis of QoL endpoints, including rules about censoring and missing data, cut-off for QoL decrease, and QoL dimensions, currently limit surrogacy validation. Nevertheless, simple but well-validated tools for the measurement of global QoL 37 could be incorporated into future trials with relative ease for surrogacy evaluation purpose.

In conclusion, the overall results of our study should help prioritise the endpoints to be statistically evaluated as surrogate for OS and highlight original potential alternative endpoint. As preferred potential surrogates in each digestive tumour site, event-free survival endpoints should be first evaluated, with particular interest for QoL or composite endpoints including QoL in the metastatic setting.

DFS: disease-free survival; FFCD: Fédération Francophone de Cancérologie Digestive; OS: overall survival, PFS: progression-free survival; QoL: quality of life.

The authors declare that they have no competing interests.

FB conceived of the study. NM, LB, FB participated in the design of the study. NM performed the statistical analysis and wrote the manuscript. FB, LB helped to draft the manuscript. All authors read and approved the final manuscript.