Multi-wave Validation Sampling to Improve Estimates Derived from Electronic Health Record Data

Because of the error-prone nature of electronic health record (EHR) data, we recommend validating a subsample of records and incorporating the validation data into analyses. Estimates can be more efficient if researchers target more informative records for data validation. The presenter will describe their experience performing a multi-wave validation study to target informative records for validation and then combining the validated data (~1000 records) with error-prone EHR data (~10,000 records) to estimate the association between maternal weight change during pregnancy and the risk of her child developing obesity.