Biofeedback mechanisms affect the overall efficacy of the medical experience in different ways; while being often a suitable approach to improve and enable efficient treatments, they might also lead to unexpected, unplanned or even undesired effects. Motivated by these considerations, we discuss the problem of detecting whether specific designs of biofeedback mechanisms lead to unexpected, unplanned or undesired behaviors in the specific case of gamified pelvic floor muscles exercising. To solve the problem we use a control-theoretic framework, and ladder on the construction of a online estimator of the intentions of the patients in performing muscle activity. We propose thus to use a mixed model- and data-driven framework for which a parametric model is first opportunely identified and then used to forecast the behavior of the patient. This model, in conjunction with the data collected under the influence of biofeedback, is then used to investigate the resulting changes in patients’ intentions in real life settings, and report quantitative assessments on the performance of the overall scheme in field tests.