Since the first risk-modulated treatment strategy in oncology was proposed for children with acute lymphoblastic leukaemia based on their response to initial therapy (Lancet Oncol. 2013 March 1;14(3):199-209), de-escalation approaches for patients with indolent disease according to prognostic biomarkers have multiplied in other tumour settings—and created a heterogeneous research landscape marked by variability in trial methodology, studied endpoints and non-inferiority thresholds.
The newly published ESMO classification for the risk-adapted modulation through de-intensification of cancer treatments, developed by an expert subgroup within the ESMO Translational Research and Precision Medicine Working Group, aims to provide clarity on the levels of evidence supporting existing biomarkers for treatment modulation and guide future research in this field. The paper author Prof. Fabrice André, Gustave Roussy Cancer Campus, France, believes that standardising the criteria for showing non-inferiority in different clinical situations will accelerate progress in tailoring therapy for patients with a favourable prognosis while preventing undue harm from premature implementation of biomarkers in clinical practice.
How should the risk-benefit ratio be assessed in research on treatment intensity modulation?
In the metastatic setting, de-intensifying a palliative treatment should reduce acute side-effects and improve patients’ quality of life without compromising survival: for example, by giving medicines sequentially rather than in combination, or substituting an intravenous agent with an oral medication. In early-stage disease, the goal is to reduce treatment and its associated long-term toxicities in patients who are unlikely to benefit from it based on a prognostic or predictive biomarker exhibited by their disease—keeping in mind that the cost of being wrong is higher in this setting because if some patients end up being undertreated and relapse, they can be confronted with an incurable metastatic disease.
Importantly, we should not assume that achieving equivalent efficacy with less treatment is intrinsically better. Quantifying the improvement in terms of quality of life, decreased toxicity or cost-effectiveness is crucial to confirm that the residual risk of incurring a small loss in survival—which is inherent to the confidence intervals used in non-inferiority trials—is offset by an important benefit. Collaboration with social scientists and health economists is recommended for this purpose.
What are the challenges in producing high-quality evidence to support de-intensification approaches?
It can be difficult to recruit patients for de-escalation trials because they often fear that they will be undertreated and do not see the benefit of participating. This is critical in the current context where randomised controlled non-inferiority trials, recognised as the gold standard when it comes to levels of evidence, require very large sample sizes to be powered to measure minute differences that amount to equivalence between standard and de-escalated treatment strategies. Some recent studies enrolled more than 6,000 patients (Lancet Oncol. 2021 April;22(4):476-488), but they remain an exception. If we want to drive progress in a field where research is more often conducted by academic groups with public funding than by industry, we need to be able to design high-quality studies with fewer patients and shorter running times. The ESMO framework therefore suggests a stepwise approach to testing harm-minimisation regimens in phase II studies before proceeding with larger trials, and outlines clinical scenarios in which more flexible models for generating evidence may be employed.
The tool also proposes a three-tiered classification of evidence for treatment de-intensification, following a similar methodology to the ESMO-MCBS and ESCAT. What do you expect will be its impact on clinical practice?
This tool aims to help investigators better select trial design according to the type of biomarker and clinical situation they want to address, as well as define the conditions necessary for results to be considered valid at different levels of evidence. The greater the level of doubt about the impact that de-intensification will have on outcomes, the more stringent the conditions in which you measure that effect have to be. When, on the contrary, you expect outcomes to be very good for the near totality of the study population because you are targeting patients known to have an extremely favourable prognosis, you probably do not need a control group and could consider a single-arm design to show non-inferiority. Meanwhile, retrospective analyses of data collected in prospective trials may be sufficient to validate biomarkers if they include multiple cohorts, achieve large enough sample sizes, and concern a patient population in whom the risk of substandard treatment is very low.
By providing a clear structure and a common language for the interpretation of results in this field of research, the tool additionally addresses a wider need to reduce the uncertainty surrounding the translation of data from highly selected patient cohorts to the real-life population—not just among oncologists, but also for regulatory bodies making approval and reimbursement decisions.