Computer-assisted expert case definition in electronic health records

Evaluating automated approaches to anaphylaxis case classification using unstructured data from the FDA Sentinel System

“…It is also possible that the BC case definition itself is not sufficiently precise to lend itself to automation and improvement in the case definition is needed first. An iterative approach to improving the case definition as described by Walker et al might be worthwhile to explore. Any improvement in the case definition itself would like likely need to be based on empirical evidence of what human experts include and do not include when classifying particular cases.…”

Pharmacoepidemiology in the Era of Real-World Evidence

“…The era of RWD and RWE calls for closer collaborations with experts from computer science, data science, informatics, genomic research, and other disciplines. Data-adaptive techniques (such as machine learning) combined with thoughtful human input are increasingly being used to mine electronic health record databases [34, 35] and improve analytic methods commonly used in pharmacoepidemiology [36, 37]. The ability to collect more data from mobile devices enables exploration of new issues, e.g., the relation between weather and joint pain in patients with rheumatoid arthritis [38].…”

Guidance to reinforce the credibility of health care database studies and ensure their appropriate impact

“…In addition to changes in the data landscape, changes in analytics approaches and methods also add challenges for transparency. While pharmacoepidemiological study designs and methodological approaches in general continue to evolve, we now see the application of machine learning and similar approaches more widely . These approaches, based on learning from training data, can generate inscrutable outputs.…”

“…While pharmacoepidemiological study designs and methodological approaches in general continue to evolve, we now see the application of machine learning and similar approaches more widely. 6 These approaches, based on learning from training data, can generate inscrutable outputs. Cabitza et al 7 illustrate the challenge by discussing an analysis that found hospitalised patients with pneumonia and asthma at lower risk of death than patients with pneumonia alone.…”