JMIR Publications

ABSTRACT

With the amount of available medical data in Electronic Health Records (EHRs) rapidly increasing, perspectives of learning from this data to allow for better personalized care are growing. As an example, health professionals can be supported in the decision-making process, when presented with outcomes of past treatment options that are representative for the current patient as learned from EHRs. The feasibility of such an approach, in the Dutch primary care setting was investigated by adapting and applying a previously published personalized treatment data analysis workflow to data from the Nivel Primary Care Database (Nivel-PCD). Type-2 Diabetes Mellitus (T2DM) was chosen as the use case. The workflow consists of various steps: extracting patient data from the Nivel-PCD; training a similarity model; generating a precision cohort of the most similar patients; and analyzing treatment options. This analysis shows which treatment options have led to a better outcome for the precision cohort in terms of clinical readouts for glycemic control. The results carry a statistical significance value that should weigh in when a health professional is using this technology in the decision-making process. When physicians are considering treatment options for a specific patient, this retrospective analysis allows them to integrate and visualize real evidence of clinically similar patients and the set of treatment decisions that conduct to a better health outcome. The added value and usefulness of these models are highly dependent on the quality and quantity of the available data. For instance, it is hard to recommend treatment options for patients with clinically odd profiles since this method requires a large patient pool to recommend statistically significant treatment options. The workflow developed in this study demonstrated which treatment option has an associated better outcome for different clinical situations.