[Impact of predictive models and decision making in prostate cancer: An integral debate]

Abstract

Objectives: To review the various methods to predict the risk of having prostate cancer, or that localized disease may be cured or progress after a given treatment.

Methods: We performed a review of the various mathematic models known for the probability analysis of the event, with a critical analysis of weaknesses and strengths of each method. In a Medline update we review the most relevant papers referred to diagnosis and management of localized prostate cancer in its diagnosis and management sides, as well as the probability of developing metastatic disease and to die.

Results: There are multiple methods and models to predict the various events in a patient candidate to diagnosis of prostate cancer, as well as to analyze the possibilities of success of a specific treatment, in many cases with an important exactness. We emphasize the heterogeneity in the methods, data and variables used for the analysis, basically about retrospective studies. Many of the most sophisticated methods, Neural Network or cart, do not present greater exactness than classic methods like logistic regression.

Conclusions: Predictive models are an important element for decision making in usual clinical practice, favoring the decision of a diagnosis or certain treatment is not taken in a random manner and therefore it is taken following scientific criteria. Waiting for more precise methods, we have to know no method is perfect, and therefore it is an important tool, which should not by pass personal knowledge or the experience of a specific working group.