We propose a method that uses a patient's continuous biomarker to adaptively assign them a treatment within a clinical trial. This method has an initial burn-in period where each patient is assigned each treatment with equal probability. Data from previous patients are used together with the next patient's biomarker value to predict the best treatment for the next patient, using a regression procedure. The estimated best treatment is then assigned to each patient with high probability. This probability increases linearly as more patients enter the trial. This is done for each patient sequentially. This capsule compares 5 different regression methods to each other and a fixed randomisation design with equal allocation. This simulation is used to show that the method assigns more patients in the trial to the best treatment than the fixed design with equal allocation.