Biological tumor markers are expected to improve outcome prediction and response to tailored therapies. However, complex effects could be underlying the dependence of the outcome from several variables measured on a continuous scale. Such a problem has been of increasing importance since the advent of high throughput genomic/ proteomic bioassay techniques. Linear and non-linear flexible regression analysis techniques, such as those based on splines and feed forward artificial neural networks (FFANN), are proposed for the statistical analysis of censored survival time data, to account for the presence of non-linear effects of predictors. Among survival functions, the hazard has a biological interest for the study of the disease dynamics. Moreover, it allows for the estimation of cumulative incidence functions for predicting outcome probabilities over follow-up. Therefore, specific error functions and data representation have been introduced for feed forward artificial neural networks (FFANN) extensions of generalized linear models, in the perspective of modeling the hazard function of censored survival data. These techniques can be exploited for the assessment of the prognostic contribution of new biological markers, investigated by means of genomic/proteomic techniques. The application of suitable measures of prognostic accuracy helps in the evaluation of the real improvement in outcome prediction due to the addition of the new molecular markers to the traditional clinical ones.
ASJC Scopus subject areas
- Biochemistry, Genetics and Molecular Biology(all)