We ensure comprehensive stakeholder involvement from organizational, legal, and technical perspectives, prioritizing the inclusion of all relevant parties in an early stage, including care professionals who will use the AI solution. We place a strong emphasis on standardization, where data is prepared following a standardized format to enable the development and application of AI models.

We conduct workshops with care professionals to gain insights into and define patterns within patient records for each complication. These identified patterns are then translated into statements, which serve as our criteria. The criteria are utilized as a baseline model and to generate a weak labeled dataset. All uncertain cases within the weak labeled dataset undergo manual review to obtain a strong dataset for fine-tuning and validating our models.

We train and fine-tune our models using both structured data (medication, lab, measures, procedures) and unstructured data (notes, discharge letters). To ensure transparency and foster trust, we use explainable AI techniques and facilitate workshops with care professionals to discuss the results. Together we assess if the explained patterns align with the patient context. The final algorithm integrates the models and criteria and undergoes validation on a new prospective dataset.