Abstract C-B3-02: Algorithm for Real-Time Identification of Blood Pressure Data Entry Errors Using Longitudinal Data From the Electronic Health Records

Abstract

Background/Aims: Data from electronic health records (EHR) are prone to error due to collection methods or human entry. However, historical measures can be used to validate subsequent measures. Real-time algorithms can be created that utilize longitudinal data to identify and require confirmation of questionable data during the patient encounter.

Methods: We developed algorithms for identifying potential errors in the Geisinger Health System EHR. These were developed with two key principals in mind:

1. the algorithms should be relatively simple to program and implement, and

2. the identification of erroneous values should depend, to varying degrees, on both a patient’s historical data and on data from the overall population.

The more encounters the patient has had in the past, the more heavily the outlier cutoff values should depend on that patient’s historical data. The algorithms were tested and refined on a sample of over 2 million systolic blood pressures from over 212,000 individuals.

Results: The selected algorithm utilized prior measurements in the EHR and allowed for the possibility that the stability of the measurements might vary across subjects (e.g., the standard deviation might vary with the mean). The properties of the algorithm are described and are used to identify cutoffs for defining 5%, 1%, and 0.1% of the measurements as questionable data. Future studies are planned to test the usefulness of the algorithm in real time. In addition, these studies can be used to identify a cutoff threshold that results in an appropriate balance between flagging too many accurate values, and not flagging enough inaccurate values.

Conclusions: An algorithm that uses historical blood pressure data can be used to identify potential errors during the patient encounter. The relatively simple algorithm does not require complex programming/software and likely can be tailored to optimize use by clinicians. The implementation of such an approach has the potential to improve the quality of EHR data.