Electronic Health Records (EHR) systems, fueled by recent federal provisions in the HITECH Act, have been increasingly implemented at US hospitals. Huge amounts of longitudinal and detailed patient information, including lab tests, medications, disease status, and treatment outcome, have been accumulated and are available electronically. These large clinical databases are valuable and cost-effective data sources for clinical and translational research. Dense and irregularly recorded vital signs and lab measurements are great components of EHRs with potential error inputs. Manual evaluation is time- and cost-consuming and hence infeasible if not impossible. We proposed a residual-based statistical approach to dynamically identify potential error-prone measurements so that clinical practitioners or patients themselves can either confirm or correct the records. The key feature of the method is to learn and update the model from the interactive response from the practitioners or patients. We applied the proposed method to Growth Chart Study using EHR data.