EHR Data Often 'Inaccurate,' Columbia University Research Finds

Data from electronic health records is often unreliable or incomplete, Columbia researchers wrote in an article published in the Journal of the American Medical Informatics Association.

Electronic health records data is often "inaccurate" or "missing," Columbia University researchers have found, according to an article published in the Journal of the Medical Informatics Association.

"Significant bias" could result from use of EHRs if they're used "naively," according to the report, which was published online Sept. 6. The authors include Dr. George Hripcsak, a professor of biomedical informatics at Columbia, and David J. Albers, associate research scientist of biomedical informatics at the university.

EHRs hold a lot of promise for record keeping in the health care industry and increasing patient outcomes. Federal "meaningful use" incentives on EHR use as well as the rise of cloud computing and mobility will lead 80 percent of health care organizations to adopt EHRs by 2016, according to a May 30 report by research firm IDC.

Data from EHRs can be used to boost population health as doctors are able to see patterns among a population, but if data is missing, this could hamper data research efforts, according to the report.

"EHRs have incredibly valuable information in them, and clinicians use them successfully to treat patients," Columbia's Hripcsak told eWEEK in an email. "Nevertheless, if they are used naively for research, they may produce distorted results."

The study discusses how data can be extracted from EHRs so that it's clinically relevant. New models of learning from data could lead the health care industry to use EHRs more efficiently, according to Hripcsak.

"We need models of how the health care process works to understand how to use the data properly," said Hripcsak. Data would need to account for information recorded at night vs. day and the fact that the patients at night are sicker, he added.

Collaboration among researchers in informatics, computer science, statistics, physics, mathematics, epidemiology and philosophy will also be necessary to better comprehend EHR data, according to Hripcsak.

Current methodologies for EHR use are inadequate because patient symptoms are poorly documented before death, he said. In fact, "naive" use of EHRs can lead people to believe that healthier people are at greater risk for death than those that are ill, Hripcsak suggested.

The report cited an EHR study of patients suffering from community-acquired pneumonia. Those patients that entered the emergency department and died quickly didn't have their symptoms entered into EHRs beforehand. At least in their medical records, they appeared healthy until they died, Hripcsak noted.

To better understand EHR data and avoid inaccuracies, researchers need to examine biases in data sets, said Hripcsak.

Partnering with clinicians to understand data accuracy issues and informatics literature and meetings can help, he said.

Doctors can improve their EHR workflows by relying on data from medical sensors, focusing on a patient's condition, and engaging the patient rather than only concentrating on meeting regulatory requirements, he said.

"The EHR is not a direct reflection of the patient and physiology, but a reflection of the recording process inherent in health care with noise and feedback loops," the report stated.

In a feedback loop a patient's documented condition leads to tests, then a diagnosis, which leads to additional tests and treatment.

"But there is a path forward, and it involves studying EHRs, understanding how the data are recorded, and using both common sense and sophisticated techniques to account for the biases," said Hripcsak.

A better understanding of EHRs will allow researchers to conduct research based on broad health outcomes and physiological research, according to the report.