Are Individual patient data meta-analyses still needed today in oncology? A discussion focused on Head & Neck oncology

International audienceHead and neck cancers are a leading cause of cancer incidence and mortality, with over 800 000 new cases and 400 000 deaths estimated in 2018 by GLOBOCAN[1]. Over the years, a wide range of randomized controlled trials have been conducted to define the best treatment strategies for each disease site and tumor stage. These trials have evaluated the role of chemotherapy, altered fractionated radiotherapy, targeted therapy or radioprotectants. To help define treatment guidelines, individual patient data meta-analyses were conducted by collaborative groups led by the meta-analysis unit at Gustave Roussy Cancer Center, launched in 1994 by Jean Bourhis and Jean-Pierre Pignon. They have focused on the role of chemotherapy[2,3] or altered fractionation radiotherapy[4] in locally advanced head and neck squamous cell cancers (HNSCC), amifostine[5], and chemotherapy in nasopharyngeal cancer (NPC)[6]. Additional works have studied surrogate endpoints for HNSCC[7] and NPC[8], network meta-analyses for HNSCC[9] and NPC[10]; but also the impact of missing data on trial characteristics[11] or the use of alternative relative efficacy metrics such as the restricted mean survival time difference[12].The aim of this article is to focus on the relevance of meta-analyses today and tomorrow in a world where patients’ and tumors’ genomic profiling and tailored treatment could become the rule. We will concentrate on individual patient data meta-analyses, which is the gold standard for collecting and synthesizing evidence[13]. The medical literature is currently flooded with meta-analyses based on published data. A quick search on Pubmed performed on December 28th 2018 using the keywords “head neck cancer” and the built-in filter “meta-analysis” retrieved 2080 references, with more than 250 new “meta-analyses” performed each year since 2014. A minority of these is synthesizing comparative data prospectively collected, and only a handful is based on individual patient data