Getty Images

Implementing an Automatic Travel History EHR Extraction System

The spread of COVID-19 gave health IT experts another reason to implement patient travel history into the EHR.

An EHR extraction system could be the key for translating unstructured text about patient travel history into actionable health data, according to a study published in JMIR Publications.

Without an automatic extraction system, clinicians would have to manually review travel charts, utilize a specific EHR system that imposes travel history documentation, or ignore travel history completely.

The spread of COVID-19 provided urgency to integrate travel history information into the EHR. Implementing travel history into the EHR can help put infectious symptoms into context for clinicians.

If implemented as a vital sign, along with temperature, heart rate, respiratory rate, and blood pressure, travel history can add to detailed patient data, prompt further testing, and spark protective measures for individuals who come into contact with the patient.

EHRs can also integrate with travel history to customize immediate diagnosis for returning travelers, similar to how cardiovascular risk calculators can show the patient a personalized list of potential lifestyle changes.

Although the Department of Veterans Affairs (VA) currently integrates travel history into patient EHRs, the research team evaluated the feasibility of annotating and automatically extracting travel history mentions in clinician notes, which present as unstructured text, across disparate healthcare facilities to respond to public health emergencies.

The researchers created a standard for patient travel history EHR detection through manual patient chart abstraction and developed an automatic text extraction pipeline.

Out of over 4,500 annotated EHRs, 58 percent contained travel history mentions, 34.4 did not contain travel history, and the remaining were undetermined. The research team said automated text processing accuracy and clinician burden levels were acceptable enough to provide rapid screening in the future.

Travel history varied from semi-structured questionnaires, such as “Have you visited a region known for Zika transmission?” to “Has the patient recently returned from Brazil, Mexico, or Miami” to “Went to Europe.”

Several researcher disagreements stemmed from differing attribution of past affirmed travel as opposed to future or hypothetical travel.

For example, one researcher marked “Traveling to visit sister in Hungary in May” as future travel, while another marked this example as past affirmed travel.

Additionally, the study authors expected military deployment locations, but the patient did not always deploy. Some EHRs would display “Service Era: Vietnam” but that does not mean the patient traveled to Vietnam.

“Location agreement was calculated for all annotated location text spans and required an exact match of text offset and negation status,” researchers explained. “Any difference in status was counted as a disagreement and any difference in text span was considered as a separate annotation element. Record agreement combined any annotated location status so that each snippet would be assigned a class of either no travel mentioned, negated locations, positive locations, or mixed.”

The research team identified 561 distinct locations over 8,127 location spans.

“Our findings demonstrate that training an accurate model to extract travel mentions is feasible in an automated system,” wrote the study authors. “Both labeled sets and the modeling approaches were chosen to minimize development time and computational resources necessary to continue surveillance in day-to-day operations. The baseline comparison presented here is a simplified evaluation, but it demonstrates that general-purpose geoparsing solutions alone result in lower precision.”

Because the research team developed the technology three years before COVID-19, its use during the spread of the coronavirus was limited because travel was only a relevant risk factor during the early phases of transmission, the study authors wrote. When researchers developed the tool, its capabilities were primarily concerned with individuals bringing infectious disease into the United States.

“The Centers for Disease Control and Prevention (CDC) guidance for Persons Under Investigation on February 12, 2020, included explicit mention for travel to Wuhan or Hubei Province,” the study authors explained. “By March 4, the CDC removed these criteria and instead encouraged clinicians to use best judgment for virus testing. In some surveillance efforts, travel history was deemed to be less important in risk assessment once community acquisition increased.”

Researchers could leverage the method in the future to prevent and contain another COVID-19 spread and the spread of other infectious diseases.

Next Steps

Dig Deeper on Health IT optimization

Cloud Computing
Mobile Computing