Use of classical and novel biomarkers as prognostic risk factors for localised prostate cancer: a systematic review

Objectives: To provide an evidence-based perspective on the prognostic value of novel markers in localised prostate cancer and to identify the best prognostic model including the three classical markers and investigate whether models incorporating novel markers are better. Data sources: Eight electronic bibliographic databases were searched during March-April 2007. The reference lists of relevant articles were checked and various health services research-related resources consulted via the internet. The search was restricted to publications from 1970 onwards in the English language. Methods: Selected studies were assessed, data extracted using a standard template, and quality assessed using an adaptation of published criteria. Because of the heterogeneity regarding populations, outcomes and study type, meta-analyses were not undertaken and the results are presented in tabulated format with a narrative synthesis of the results. Results: In total 30 papers met the inclusion criteria, of which 28 reported on prognostic novel markers and five on prognostic models. A total of 21 novel markers were identified from the 28 novel marker studies. There was considerable variability in the results reported, the quality of the studies was generally poor and there was a shortage of studies in some categories. The marker with the strongest evidence for its prognostic significance was prostate-specific antigen (PSA) velocity (or doubling time). There was a particularly strong association between PSA velocity and prostate cancer death in both clinical and pathological models. In the clinical model the hazard ratio for death from prostate cancer was 9.8 (95% CI 2.8-34.3, p < 0.001) in men with an annual PSA velocity of more than 2 ng/ml versus an annual PSA velocity of 2 ng/ml or less; similarly, the hazard ratio was 12.8 (95% CI 3.7-43.7, p < 0.001) in the pathological model. The quality of the prognostic model studies was adequate and overall better than the quality of the prognostic marker studies. Two issues were poorly dealt with in most or all of the prognostic model studies: inclusion of established markers and consideration of the possible biases from study attrition. Given the heterogeneity of the models, they cannot be considered comparable. Only two models did not include a novel marker, and one of these included several demographic and co-morbidity variables to predict all-cause mortality. Only two models reported a measure of model performance, the C-statistic, and for neither was it calculated in an external data set. It was not possible to assess whether the models that included novel markers performed better than those without. Conclusions: This review highlighted the poor quality and heterogeneity of studies, which render much of the results inconclusive. It also pinpointed the small proportion of models reported in the literature that are based on patient cohorts with a mean or median follow-up of at least 5 years, thus making long-term predictions unreliable. PSA velocity, however, stood out in terms of the strength of the evidence supporting its prognostic value and the relatively high hazard ratios. There is great interest in PSA velocity as a monitoring tool for active surveillance but there is as yet no consensus on how it should be used and, in particular, what threshold should indicate the need for radical treatment.