This talk will give an overview of learning and reasoning with Probabilistic Graphical Models by discussing their links with causality and Bayes' theorem. A case study of health outcome data collection with PGMs will demonstrate the use and potential benefits of PGMs in healthcare. Adequate availability and quality of outcome data are major bottlenecks for machine learning applications in healthcare. Outcome data for many medical conditions can only be collected by surveys called Patient Reported Outcome Measures (PROMs). It is challenging to collect large amounts of PROM data due to the question burden they create. PROMs require a large number of questions to make reasonable estimations. For example, pain has biopsychosocial dimensions requiring separate questions for each dimension. Moreover, more questions are needed to improve the precision of measurements. PGMs can enhance PROMs bu enabling dynamic estimation of the probability distribution of the measured dimensions and the mutual information of the unanswered questions. This enables us to select which question to ask next and when to stop asking questions in a personalized way based on previous responses. We show that the question burden can be decreased with PGMs by applying it to a selection of PROMs in MSK care.