Pelvic floor dysfunction can seriously affect a woman’s quality of life, both physically and mentally. Different types of surgeries can be performed in order to improve the patient’s conditions, with the challenge to sort out which surgery works best for which patient. Quite some data are available related to pelvic floor surgeries done in the past. These data contain information of the patient’s conditions, both pre- and post-surgery, and about both the patient’s physical state (as determined by the doctor) and the perceived physical state (through questionnaires). We aim to develop models that can predict both the post-surgery physical state as well as the post-surgery perceived state based on the pre-surgery information.
Machine learning methods can be expected to yield more accurate predictions than the more conventional statistical approaches, but have the disadvantage that they are often harder to interpret. We will design and apply novel methods for explainability, for example based on so-called shap values, and study to what extent such methods for explainability can indeed increase the usefulness of these models and improve their applicability in a clinical setting.