Pacific AB
Missing diagnoses, uncovering hidden groups, and going beyond ‘encounters’ to assess health (308133)
*Sherri Rose, Harvard Medical SchoolKeywords: Electronic health Records, Missingness, Fairness, Risk adjustment
Our knowledge about the health care system, including reliance on traditional sources of data, is limited by multiple forms of missingness. This talk highlights three types of missingness in billing claims that impact our understanding of health. The first is dataset shift resulting in part from a lack of persistent coding of chronic conditions and the implications for plan payment risk adjustment. The second also focuses on risk adjustment, but considers group fairness and automatically detecting previously unidentified (i.e., ‘hidden’) complex groups using machine learning. Lastly, individuals who do not have an encounter with the health care system will not be observed in billing claims. Leveraging multiple data sources, including news media reports and digital data, may provide improved estimates for certain conditions, such as infectious diseases. An overarching theme of the talk is that developing data science tools tailored to the specific health question being addressed and the electronic health data available is critical given the stakes involved.