Second, a clear understanding of the way in which diagnostic accuracy may be affected by the population to which a screening test is applied is required. As noted earlier, the prevalence of a cannabis-related disorder or problem will affect the ability of any screening test to detect it accurately. Predictive power will be low when the prevalence is low, even if sensitivity and specificity of the instrument are high [27,35]. On population level, this effect may lead to a non-negligible overestimation of the prevalence. For example, positive predictive values of SDS and CUDIT tested in truly probabilistic samples [9,24] were rather low with 14.1% to 49.2%. Other studies used targeted samples of preselected cannabis users (e.g. recruited through magazines, or snowball sampling) [14,22,23] which resulted in higher positive predictive values up to 94.1%.
