Endpoints in advanced breast cancer: methodological aspects & clinical implications

Primary and secondary endpoints

Clinical trials often have one primary and a few secondary endpoints, all of which are objective parameters that represent study results regarding efficacy or safety variables of interest and whose differential change after treatment allows for comparisons between groups of patients. Mostly due to statistical concerns against the practice of multiple testing, there needs to be a hierarchy of endpoints within a given clinical trial. According to this hierarchy, efficacy endpoints are ranked according to their perceived importance, reflection of treatment effect upon the natural history of the disease, historical practice within the field, and regulatory constraints. In practice, primary endpoints serve two very important functions: to allow estimation of the sample size and to ascertain whether a given trial yields positive results (Fig.). Following the usual conceptual framework that underlies statistical hypothesis-testing, studies are considered positive when the P value for the primary endpoint is below a prespecified value of interest in the main population for analysis (it should be noted that this view has been questioned with the argument that not only statistical significance, but also the magnitude of benefit, should be used for declaring a phase III trial positive). From a statistical standpoint, secondary endpoints have an exploratory nature; moreover, they should be limited in number and should work as supportive measurements regarding the primary endpoint5.

Close

Fig.

Suggested algorithm for interpretation of a phase III clinical trial.

Export to PPT

Time-to-event endpoints

As a general rule, efficacy endpoints in oncology represent variables that may be of three types: time-to-event, categorical and continuous. Time-to-event endpoints provide information on the timing of occurrence of events of interest in oncology, such as disease progression and death. Given the importance of such events in oncology, time-to-event endpoints are usually the most important in later phases of drug development. The analysis of such events is most frequently done using the Kaplan-Meier method,6 for which one or more events of interest and one or more reasons for censoring must be defined in advance. The events of interest and reasons for censoring in the most commonly used endpoints are shown in the Table. In the case of overall survival (OS), the event of interest is death, and patients are censored when they are last seen alive or when they are lost to follow up. Over the past decade or so, OS has been used only rarely as a primary endpoint in advanced breast cancer, but has been the most frequently used secondary endpoint7. In the original definition of progression-free survival (PFS), the events of interest are tumour progression and death from any cause, with censoring of patients who are lost to follow up89. On the other hand, for time to tumour progression (TTP) the event of interest is only disease progression, with censoring of patients who die from any cause or who are lost to follow up9. In recent studies, however, many investigators have used PFS and TTP interchangeably7. As a result, either PFS or TTP have been the most frequently used endpoint in advanced breast cancer in recent studies710. Other time-to-event endpoints sometimes used in breast cancer include time to treatment failure, rarely used as primary endpoint or for regulatory purposes11, and duration of response, for which only responding patients constitute the denominator (Table).

Table Events and reasons for censoring in time-to-event dependent endpoints most frequently used in advanced breast cancer trials

Censoring, the key feature of Kaplan-Meier analysis, consists in excluding from the denominator of the analysis those patients who have not had one of the events of interest at the time they are last known to be at risk for having such events, provided no further information is available for those patients thenceforth. The Kaplan-Meier method is only valid if there is reason to believe that the probability of being censored is randomly distributed among patients in the study and is not related to the probability of having the events of interest (in statistical jargon, censoring must be non-informative). When this is not the case, Kaplan-Meier analysis may be biased. The number of patients censored in the analysis of a given endpoint of a particular study can often, but unfortunately not always, be known directly by numerical information provided by authors or indirectly by analyzing the tick marks on survival curves. Two or more survival curves may be compared using non parametric tests, the most common being the logrank test. Furthermore, multivariate analysis of time-to-event endpoints may be conducted using various methods, with the Cox proportional hazards model being the most frequent.

Other efficacy endpoints

The categorical endpoint most frequently used in medical oncology is the objective response rate (ORR), currently defined according to the Response Criteria in Solid Tumors guidelines1213. Although typically considered a phase II endpoint, ORR has been used as the primary endpoint in 40 per cent of recent phase III trials in advanced breast cancer10. A variant of ORR is the clinical benefit rate, often defined as the proportion of patients with no disease progression after 6 months of therapy. Categorical endpoints may be compared between groups using various tests, such as Fisher's exact test and Pearson's chi-square test, among others. Multivariate analyses of such endpoints may be undertaken with logistic regression models. In medical oncology in general, and breast cancer in particular, continuous endpoints are rarely used, with the possible exception of quality of life variables measured on numerical scales.

Overall survival

In view of its objective measurement and the unquestionable benefit derived by patients, OS has been historically considered the most important endpoint in medical oncology14. The US Food and Drug Administration considers OS as a direct measure of treatment benefit; according to that agency, OS is usually the preferred endpoint when studies can be conducted to adequately assess survival9. However, OS is increasingly an elusive endpoint, mostly because it may be confounded by the use of treatments administered to patients after participation in a given trial, including post-progression cross-over to the experimental arm4. As a result, many randomized trials in breast cancer are underpowered to detect significant OS differences; notably, only approximately 7 per cent of recent phase III trials in advanced breast cancer have used OS as their primary endpoint10. Nevertheless, nearly 20 per cent of such trials demonstrated a significant survival improvement, most frequently in association with an accompanying gain in PFS/TTP and in trials involving patients in second- and third-line therapies10. From a strictly methodological standpoint, it is arguable that the extent to which survival gain in those trials - especially in the first line - was due to trial therapy is unknown, given the effect of post-trial interventions. As a corollary to the prior statement, the expectation of OS gain in trials for which PFS was the primary endpoint may be elusive.

PFS and TTP

Efficacy endpoints based on tumour assessments have been increasingly used in drug development, and both have been accepted as markers of clinical benefit for drug approval111516. For regulatory purposes, PFS seems preferable to TTP in so far as it captures fatal toxicities11. PFS is an attractive endpoint because it is available earlier than OS, is less likely than OS to be influenced by competing causes of death, and is not influenced by treatments administered after progression in a given trial. On the other hand, PFS is subject to measurement error and bias. Measurement error may stem from inconsistent use of definitions and standards among investigators17, whereas bias may results from unblinded ascertainment of progression and from the fact that the date at which progression is confirmed radiographically is a proxy for the true progression date, which lies somewhere within two successive assessments18. This overestimation of PFS does not raise serious methodological issues in randomized trials in which the same evaluation schedule is used for all arms, but may compromise comparisons across trials if different schedules have been used.

ORR

As stated previously, ORR is a frequent endpoint in phase III trials on advanced breast cancer. However, there are known limitations of ORR as an indicator of treatment benefit in oncology19, and patients with breast cancer and stable disease after therapy may also accrue benefit. In advanced, hormone-receptor-positive breast cancer, for example, the clinical benefit rate is often used because the survival experience of patients with stable disease after hormone therapy is commonly similar to that of patients with an objective response20. Indeed, the same has been found on occasion for patients treated with chemotherapy21. Thus, ORR is probably weaker than PFS as an efficacy parameter in advanced breast cancer, despite the fact that observing a response to treatment may be the only reliable indicator of treatment benefit in individual patients with cancer22.