While the study of beneficial effects almost always necessitates randomized trials, adverse effects of treatment can often be effectively investigated in non-randomized studies (Miettinen 1983). Vandenbroucke has proposed that observational studies of adverse effects of medical interventions offer some of the best chances for unbiased observational studies (Vandenbroucke 2004). This idea was empirically verified by a comparison of randomized and observational studies of adverse effects, which found that, if anything, risk estimates from observational studies were lower (Papanikolaou 2006)). In some instances where observational studies showed markedly higher risks, they better reflected actual patient care (Vandenbroucke 2006). Like any study, case-control and cohort studies are potentially susceptible to bias, and any limitations of the data should therefore be critically discussed. See Chapter 13 (Section 13.5) for further discussion of assessing risk of bias in such studies. Jick has drafted a taxonomy of the type of study that is most likely to detect an adverse effect, as well as the type of study that is necessary for verification (Jick 1977).