OBJECTIVES: To develop a method which increases the potential to find statistically significant differences in costs and effects when a trial is powered using a dichotomous outcome. METHODS: An example is used of a trial assessing an intervention to prevent late pain. Treatment is expected to increase the percentage of pain-free patients from 85% to 92%, giving a power of 80% with 500 patients. Using EQ-5D as outcome decreases the power to 40%. We improve on this by deriving T-tests in which the following assumptions are taken into consideration: 1. quality of life with pain (8% vs 15%) is identical in both arms 2. quality of life without pain (85% vs 92%) is identical in both arms Alternatively, we use a Bayesian approach assuming that the differences between arms follow normal distributions with mean zero and varying precision. Using simulations the frequentist and Bayesian approach are linked and it is analysed to what extent the results depend on the base line probabilities. RESULTS: Making both assumptions increases the power to 80% as in the binary assessment. Applying assumption 1 increases the power with only 2%, applying assumption 2 increases it to almost 80%. When assuming that the outcome is 44% versus 56% instead of 85% vs 92% both assumptions contribute to the power approximately equally. The Bayesian model coincides with the assumptions from the frequentist approach when the precision is set to the extremes (zero or infinity). Between these it offers a flexible approach where the road from one extreme to another is defined by cumulative normal distributions on the log of the squared root of the precision. CONCLUSIONS: Traditional approaches may disregard common sense. Building this into the analysis and the assessment of the data will decrease suggested uncertainty and may decrease the need for large patients numbers.