To develop a treatment recommendation we must first critically appraise and summarize the evidence, then decide how to weigh the necessary trade-offs between the benefits and harms. GRADE, which is becoming the standard for developing treatment recommendations, recommends that patients’ preferences should be explicitly considered when making these trade-offs, but methods for doing this quantitatively are lacking. The aim of this thesis was to develop a quantitative approach for combining comparative effectiveness research and patients’ preferences, and apply it to inform treatment recommendations for early rheumatoid arthritis. In the first study, we performed a Bayesian network meta-analysis of 150 randomized controlled trials to compare the benefits and harms of treatment options for early rheumatoid arthritis. In the second paper we used a discrete-choice experiment to measure the relative importance of the major outcomes from the network meta-analysis as well as other issues relevant to the decision (dosing, rare risks) in patients with early rheumatoid arthritis. For the final paper we used a Bayesian framework to combine the data from the network-meta-analysis and the discrete-choice experiment to determine the preferred treatments. Importantly, we modeled the uncertainty in the evidence and patients’ preferences, as well as individual variability in patients' preferences, which are three key considerations in the GRADE approach. Overall, we predicted that most patients in our population (78%) would prefer triple therapy (the combination of methotrexate, sulphasalazine and hydroxychloroquine) to methotrexate alone as initial therapy, suggesting a relatively strong recommendation could be made. In patients who have had an inadequate response to methotrexate, the preferred treatment was more split between triple therapy (62%) and methotrexate + anti-TNF (biologic) therapy. The patients who we predicted would prefer other treatments, had important differences in preferences that could be highlighted in a preference-sensitive treatment algorithm. The thesis therefore presents a novel method for developing evidence-based, preference-sensitive treatment recommendations. Applied to early rheumatoid arthritis, our findings have important policy implications, suggesting that patients would prefer triple therapy to other treatments, including costly biologic therapy.