Prognostic models in hepatocellular carcinoma (HCC) and statistical methodologies behind them

Hepatocellular carcinoma (HCC) is estimated to be responsible for 250,000 deaths worldwide yearly. Aggressive surgical resection or liver transplantation still remain the only viable curative options for patients suffering the disease despite the multitude of emerging therapies for HCC. However, even with the most aggressive surgical intervention, survival varies widely within each particular stage of HCC. In order to improve utilization of available therapeutic modalities, a number of outcome prognostic models have been developed. This manuscript reviews the prognostic models most commonly utilized in clinical practice and the statistical methodologies on which these models are based. A multitude of statistical and mathematical techniques can be used for prognostic model development. The most common methodologies used for HCC prognostic model development can be generally divided into four groups: survival, artificial neural networks, analysis of variance, and cluster analysis. Survival methodologies (such as Cox proportional hazard model) are commonly employed for estimation of relative significance of risk factors for patient survival or cancer recurrence. Artificial neural networks (such as back-propagation network) can be supreme approximation tools for any continuous or binary function, and as such can be employed for prognostication of HCC recurrence (death). Analysis of variance and cluster analysis are the most common statistical tools of recently evolved microarrays technology, which, in turn, is one of the most promising tools available to the cancer researcher.