A considerable proportion of chronic diseases and long-term complications might be preventable if patients under risk are adequately identified. This is especially true for the elderly, who exhibit considerable prevalences of non-communicable diseases, e.g. diabetes mellitus and depression. Biomarkers from the peripheral blood could be helpful, as long as their predictions are reliable. However, a plethora of disturbance factors influence the reliability of a biomarker, and most of them occur not during blood analysis, but in the pre- or the post-analytical phase. The pre-analytical phase is particularly vulnerable to potential influence. As the existing literature reveals, those disturbance variables include not only technical, but also patient-related factors, e.g. the patients level of physical activity. The present thesis aims to extend the existing evidence to include genetic risk markers for depression (rs6265 within the Brain-derived neurotropic factor, BDNF; rs6295 within the 5-HT1A receptor) and circulating vitamin D3 as a predictive marker for diabetes mellitus. Moreover, if exercise-induced changes in blood composition occurred in a directed and dose-dependent manner, they might be exploitable in order to evaluate training efficacy and motivation. Hence the aim was to calculate a statistical model enabling prediction of future declines in fitness. To this end, 55 elderly marathon athletes above the age of 60 and a sedentary control group (N = 58), matched for age, sex and years of education, were enrolled and examined at baseline and after three years follow-up. Participants underwent a thorough medical and psychological check-up, including ergometry and blood examinations. Genetic variants within BDNF and 5-HT1A were determined by means of the 5-nuclease assay. Depressiveness was assessed using the Beck Depression Inventory (BDI) and the Geriatric Depression Scale (GDS). A binary logistic regression model allowing for future fitness prediction was compiled in two independent participant samples. The predictive capability of both genetic risk markers was different in athletes when compared to sedentary controls. In detail, physical activity seemed to ameliorate the correlation between the rs6265 [C];[C] genotype and depressiveness, resulting in a relative risk of 3.5 (95% confidence interval, CI: 1.3 9.8) for [C]; [C] carriers of the control group regarding the presence of a suspicious BDI score. In contrast, within rs6295 in the 5HT1A receptor, the [G];[G] genotype predicted depressiveness only in athletes, but not in controls. Regarding the prediction of future (pre-)diabetic states, a significant correlation between baseline vitamin D3 levels and a HbA1c level 5.6% was registered among controls. However, this relationship was not seen in athletes, where vitamin D3 correlated with weekly training amounts. In order to predict follow-up ergometry results from baseline blood markers, a binary logistic regression model with excellent statistical significance ( = 21.412, df = 5, p = 0.001) could be calculated. The model presented with high discriminatory power in both the model training sample (receiver-operator characteristics area under the curve, ROC AUC = 0.9510.050) and the test sample (ROC AUC = 0.7860.098). The present thesis revealed that the predictive capabilities of the genetic risk markers for depression in BDNF and 5HT1A, as well as the diabetes-associated risk marker vitamin D3, were influenced by individual life-style factors, i.e. engaging in endurance sports. These findings are largely novel and reinforce the need to consider patient-related pre-analytical confounders when interpreting disease markers. Moreover, it was demonstrated that physical activity seemed to influence blood components in a directed manner, thus making it possible to estimate training motivation and efficacy from easily assessable blood parameters.