Limitations
Several limitations must be considered in addition to those already discussed. Firstly, diagnostic and treatment protocols varied considerably both within and between studies. In some, treatment approaches varied upon likely putative primary site (61), and others reported a change in practice during the study period (60). Whilst some studies identified which patients were not treated ‘per protocol’ (53,54) (usually due to patient refusal of adjuvant treatment), the majority gave no explanation for decision making regarding treatments given, leading to potential selection bias. The lack of clarity regarding patients excluded, small sample sizes, and attrition rates in several studies leads to potential concerns for reporting bias within the studies. The geographical differences in incidence of disease (65) and prognostic impact of p16 (66) mean that the tumour biology both within and between these studies is likely to be variable, impacting on the results and interpretability.