Healthcare dialogue systems are developed to automate and simplify routine tasks such as collecting patient information or making an appointment. Often, these models are trained to mimic doctor-patient interaction as their constant availability is a key feature for patients, in particular with chronic conditions. Chronic patients regularly visit their doctor and are asked to repeatedly fill in standardized questionnaires, which may trigger repetitive, incorrect input. In collaboration with the ALIAE company, we focus on developing novel dialogue models which can maintain a conversation with the patient while collecting both specific answers to a set of pre-defined questions and serendipitous information about the patient condition that we conjecture, may be useful for the patient treatment. Specifically, we propose a dialogue system and automatic questionnaire filling pipeline that should complement existing routine between doctors and patients.This thesis makes three main contributions. We first propose an approach to flexibly guide the user through a pre-defined medical decision tree using naturally written user input -- this allows the health-bot to collect answers to a set of pre-defined questions. To improve user engagement, increase the probability of all required medical topics being addressed and allow for further information about the patient condition to be collected, we then extend this initial model by integrating additional bots designed to handle health-related follow-up questions and maintain small talk. Finally, we introduce novel zero-shot Question Answering models and pre-processing techniques so that standard, clinical questionnaires can be automatically filled in based on the content of collected human-bot dialogues.