Reimbursement agencies require real-world evidence on the effectiveness and cost-effectiveness of new drugs and medical devices. In many settings, randomised controlled trial (RCT) data is unavailable or insufficient. Where non-randomised data is used to estimate treatment effectiveness and cost-effectiveness, the main methodological challenge is selection bias from confounding by indication. Standard regression or propensity score methods are frequently used to adjust for selection bias, but estimates of treatment effectiveness may be highly sensitive to the chosen parametric form of these models, and evidence that relies on such methods may be viewed as insufficient by reimbursement agencies. While new, more advanced methods for allowing for confounding cannot offer a panacea, they have been shown to provide estimates of treatment effectiveness that are relatively robust. This course offers an in-depth description of ‘cutting edge’ methods for addressing this form of selection bias. These methods include flexible regression which uses machine learning for model selection, propensity score matching with regression adjustment, and Genetic Matching, a recently developed approach that extends propensity score matching. The course introduces the participants to these methods using the R software, through a series of real world data examples. Faculty will also demonstrate sensitivity analyses that convey to decision makers the extent to which the estimates of effectiveness and cost-effectiveness are robust to that assumption of no unobserved confounding. Participants who wish to have hands-on experience must bring their personal laptops with appropriate software installed.