The users' mental state such as attention variations can have an effect on the brain-computer interface (BCI) performance. In this project, we implemented an adaptive online BCI system with alterations in the users' attention. Twelve electroencephalography (EEG) signals were obtained from six patients with Amyotrophic Lateral Sclerosis (ALS). Participants were asked to execute 40 trials of ankle dorsiflexion concurrently with an auditory oddball task. EEG channels, classifiers and features with superior offline performance in the training phase of the classification of attention level were selected to use in the online mode for prediction the attention status. A feedback was provided to the users to reduce the amount of attention diversion created by the oddball task. The findings revealed that the users' attention can control an online BCI system and real-time neurofeedback can be applied to focus the attention of the user back onto the main task.