Data scientists and technologies have made notable progress towards overcoming the initial barriers to utilizing electronic medical record (EMR) data for the purposes of medical research. However, inherent features of EMR data provide further hurdles for researchers attempting to draw valid inferences. This presentation will describe barriers and solutions to generating coherent real world evidence from EMR data with an applied example assessing cardiovascular risks associated with hypoglycemia.
The discussion will focus on epidemiological methods to address ambiguous observation periods (duration and completeness of data), algorithm building (including special considerations for natural language processing of the free text clinical notes), and quantitative analysis (including analytic considerations for bias and confounding).