Extended adjuvant therapy: the role of subset analyses
Mené aux Pays-Bas sur 1 912 patientes atteintes d'un cancer du sein HR+ après la ménopause et ne présentant pas de signe de récidive après 2 à 3 ans de traitement adjuvant par tamoxifène ni après 3 à 6 ans de traitement par anastrozole, cet essai de phase III évalue l'efficacité, du point de vue de la survie sans maladie à 5 ans, de prolonger le traitement adjuvant par inhibiteur de l'aromatase
The report of the DATA trial1 by Vivianne Tjan-Heijnen and colleagues and published in The Lancet Oncology adds to the current body of information delineating the role of extended adjuvant aromatase inhibition in women already treated with tamoxifen followed by an aromatase inhibitor. Women were randomly assigned to treatment after 2–3 years of treatment with adjuvant tamoxifen to either 3 or 6 years of further anastrozole treatment with a primary study endpoint of disease-free survival. Since women in both study groups had a common initial 3-year period of anastrazole treatment, the analysis presented started 3 years after randomisation and is described by the authors as an adapted disease-free survival analysis.
The authors report a hazard ratio of 0·79 (95% CI 0·62–1·02; p=0·066) for 5-year adapted disease-free survival with the use of 3 additional years of anastrazole, which was technically—as they correctly report—a negative overall finding. They also report that two post-hoc subset analyses, one for patients with both positive oestrogen and positive progesterone receptors, and one for patients who received chemotherapy, might suggest efficacy of the extra 3-year treatment for these selected groups of patients.
In fact, the overall trial results, particularly in the setting of several other trials addressing the same question,2, 3, 4 could be interpreted as representing a positive signal. The DATA results are consistent with most, if not all, of the available published or presented literature, which includes results that have a significantly positive outcome,2 are indicative of a positive outcome,3 or are neutral.4 However, the interpretation of the two subset analyses in DATA with tests of interaction, which are also negative, is likely to be overly positive.
Great care must be taken when interpreting subset analyses in clinical trials. Although DATA is quite a large study with 1660 eligible patients and 261 recurrences, the number of events for any subgroup analysis becomes perilously small. It is most appropriate in such situations to prespecify what subset analyses will be done, to give priority to analyses regarding factors for which the randomisation is stratified, to test for interaction factors among subset outcomes, and to apply a stiff correction factor to the p values acquired. Additionally, these subset analyses should be interpreted as hypothesis generating, rather than definitive.
The DATA subset analysis of those who received or did not receive chemotherapy was not prespecified and not based on a stratification factor but might have some plausible rationale. Furthermore, similar results for this subset have been suggested by others.2 The subset analysis by oestrogen or progesterone receptor status is based on a stratification factor and one for which there is biological plausibility to postulate interaction with the randomly allocated therapies. These tests for interaction, which have produced p values of 0·077 (for the use of chemotherapy) and 0·078 (for oestrogen and progesterone receptor status) have, appropriately, been quite conservatively reported and discussed by the authors. However, they do mention in the discussion the possibility of analyses of even smaller subsets such as patients with positive oestrogen receptor status, positive progesterone receptor status, and node-positive disease, or with a large tumour size. Here, even more caution must be advised for fear of over-interpreting these tempting but perhaps misleading subset analyses.
And where do the results of this trial take us in terms of today's clinical practice? The DATA trial adds to an increasing body of information supporting the use of aromatase inhibitors beyond 2–5 years, much of it from trials including patients who have previously received 2–5 years of tamoxifen.2, 3, 4 There are some—but not as much—data available regarding longer aromatase inhibitor therapy in women who had not previously received tamoxifen,2, 3, 4, 5 but this too suggests that longer therapy is better. Of course, as briefly reported here, increasing length of treatment is also associated with increased side-effects, which include arthralgia or myalgia and osteoporosis or fractures.1, 2, 6 Furthermore, as has been seen in other trials,2, 3, 5 much of the effect of longer adjuvant aromatase inhibitor therapy is comprised of a reduction in the number of contralateral breast cancers, which is not likely to be as directly linked to overall survival as are reductions in distant or even loco-regional recurrences. (...)
The Lancet Oncology , commentaire en libre accès, 2016