Tumour–host divergence: a hypothesis for personalised blood-based early detection of hepatocellular carcinoma

Hepatocellular carcinoma (HCC) remains lethal because most tumours are found too late for cure, and the blood biomarkers used for surveillance — chiefly alpha-fetoprotein (AFP) — miss many early-stage and AFP-negative tumours; despite three decades of discovery, single-analyte and marker-plus-AFP panels have converged on a performance ceiling. We advance the tumour–host divergence (THD) hypothesis: that early malignant emergence is better detected as a personalised anomaly than as a population classification, and that pairing tumour-shed signals with the host’s immune–inflammatory response — read along each patient’s own trajectory and conditioned on aetiology — can reveal small tumours that any single marker, measured once against a population threshold, will miss. The hypothesis rests on three complementary axes: a temporal axis interpreting change from the individual’s baseline rather than a fixed cut-off; a compartmental axis treating the coupling between tumour-shed and host-response signals as the quantity of interest, on the premise that a small tumour may perturb the host before it sheds detectable marker; and a contextual axis conditioning interpretation on aetiology and fibrosis stage, including metabolic dysfunction-associated steatotic liver disease. These axes are integrated by a per-patient longitudinal risk model that flags malignant emergence as a departure from the patient’s stable equilibrium. To show that this gap is real rather than assumed, we map the field across a de-duplicated corpus of 8,069 PubMed and Embase records with an explicit, reproducible taxonomy, finding research effort concentrated on tumour-derived analytes — which outnumber host-response analytes by roughly six and a half to one — far more work on prognosis than early detection, and the temporal, paired-compartment and aetiology-conditioned dimensions almost unexplored. The hypothesis is deliberately falsifiable: its central predictions are benchmarked against GALAD and are testable in surveillance cohorts that already generate longitudinal data. Its contribution is integrative — reorganising existing biomarkers rather than requiring new ones — and is aimed at the early-stage and AFP-negative patients of greatest unmet need.