Risk factors for prostate cancer : analysis of primary data, pooling, and related methodological aspects

Abstract

Prostate cancer is the second most common cancer among men worldwide, yet its etiology remains poorly understood. Obesity, on the other hand, is a prevalent but preventable medical condition that is associated with hormonal and metabolic changes. Since prostate cancer is a hormone-related cancer, the hypothesis of a link between body fatness and prostate cancer risk has been formulated. Furthermore, the considerable biologic heterogeneity of prostate cancer warrants analyses to be carried out separately by aggressiveness of the disease, differentiating indolent from potentially lethal tumors.

This thesis has two aims. First, to elucidate the association between obesity, as measured by body mass index (BMI), and the risk of localized, advanced, and fatal prostate cancer. This is done using both primary data (Paper I) and aggregated data extracted from published epidemiological studies (Paper IV). Second, to deal with some methodological aspects related to the analysis of primary and aggregated data (Paper II; Paper III; Paper V).

In Paper I, we used primary data from the Cohort of Swedish Men to examine the association of BMI during early adulthood (30 years of age) and middle-late adulthood (45–79 years of age) with the incidence of localized and advanced prostate cancer and with prostate cancer mortality. BMI during middle-late adulthood was observed to be inversely associated with the incidence of localized prostate cancer, while it was directly associated with the incidence of advanced prostate cancer and with prostate cancer mortality. At the same time, we observed limited evidence of an inverse association between BMI during early-adulthood and the risk of advanced and fatal prostate cancer.

In Paper II, we extended the use of quantile regression for censored data to those situations where the time scale of interest is attained age at the event instead of follow-up time. In particular, we described how to use Laplace regression to model percentiles of age at the event in the presence of delayed entries, by conditioning on age at entry. In Paper III, we identified three major misinterpretations of risk and rate advancement periods (RAP): first, equating RAP with the difference in mean survival times; second, interpreting RAP as the time by which the survival curve for the exposed individuals is shifted compared with that for the unexposed; third, equating the RAP to a simple ratio of two log–relative risks. Furthermore, we showed how RAP estimation is sensitive to the specification of the age-disease association.

In Paper IV, we carried out a dose–response meta-analysis to summarize the available evidence on the association between BMI during middle-late adulthood and the incidence of localized and advanced prostate cancer. Based on aggregated data extracted from 13 prospective studies, we observed that BMI was inversely associated with the incidence of localized prostate cancer, while it was directly associated with the incidence of advanced prostate cancer.

In Paper V, we stressed the importance of assessing the goodness of fit of dose–response metaanalysis models. We presented and discussed three tools (deviance, coefficient of determination, and decorrelated-residuals–versus–exposure plot) that are useful to test, quantify, and visually display the fit of dose–response meta-analysis models, while taking into account the correlation structure of the study-specific log–relative risks.

In conclusion, Paper I and Paper IV supported the hypothesis of etiological heterogeneity of prostate cancer in relation to obesity during middle-late adulthood. In particular, BMI was observed to be directly associated with advanced prostate cancer and with prostate cancer mortality. Paper II extended the use of quantile regression for censored data to those situations where attained age is the time scale of interest, Paper III clarified the appropriate use and interpretation of RAP, and Paper V proposed useful and relevant methods for assessing the goodness of fit of dose–response models in research synthesis.