Machine Translation & Emergency Department Discharge Instructions

Patients who have limited proficiency with the English language often experience communication barriers when receiving healthcare in English-speaking institutions. “Nearly 10% of the American population has limited English proficiency,” says Elaine C. Khoong, MD, MS. “Research suggests that these patients receive suboptimal care and have worse outcomes than those who are proficient in English. Written communication can improve patients’ understanding of directions from their healthcare providers, but standard instructions that are pre-translated often cannot address many important patient-specific issues, like medication changes or upcoming appointments.”

New Data

Machine translation tools like Google Translate (GT) have the potential to improve communication with patients who have limited English proficiency, but some studies have questioned their accuracy. In 2017, GT changed its translation algorithm to utilize machine learning tools. In a study published in JAMA Internal Medicine, Dr. Khoong and colleagues assessed the use of GT to translate emergency department (ED) discharge instructions into Spanish and simplified Chinese. They abstracted 100 free-texted ED discharge instructions and oversampled for medication changes and common complaints. Each sentence was analyzed by content category, readability, use of medical jargon, and the presence of nonstandard English. Using GT, the study group translated instructions into Spanish and Chinese, and then bilingual translators translated the text back into English.

Of 647 sentences contained in the original 100 sets of instructions, researchers found that GT was 92% accurate for Spanish translations and 81% accurate for Chinese translations. A small minority of inaccurate translations—2% of Spanish and 8% of Chinese sentence translations—had the potential to cause clinically significant harm (Table). These were mostly due to grammar or typographical errors in original written English instructions, which would readily have been overlooked or understood by those who can read English text.

Important Caveats

“Machine translation errors were linked to the use of medical jargon, and long, complicated sentences,” says Dr. Khoong. According to the study, the GT algorithm was also less accurate when clinicians used colloquial English terms. Some of these mistranslations were deemed harmless, but others had more serious implications. For example, the phrase “hold the kidney medicine” was translated by the GT algorithm into Spanish as “keep the medication” and into Chinese as “keep taking” the medication. These types of mistranslations have the potential to be life threatening.

Some phrases and statements used by clinicians confused GT because the sentence structure was too complicated, according to Dr. Khoong. “When using GT, clinicians should aim to use short and simple sentences,” she says. “It also behooves us to have to ensure patients have an opportunity to read the written translations before discharge in addition to also receiving verbal instruction so that we may further reduce potential harm. In addition, clinicians should be vigilant about spelling and grammar and avoid using complex grammar and medical jargon whenever possible.”

Appropriate Use

The study noted that it may be best to use GT only in combination with human interpreters. If interpreters are not available onsite, remotely-located phone or video interpreters can still help patients understand verbal instructions while clinicians show patients the written translations of instructions. This approach can empower patients to flag errors or raise questions if there are confusing translations or conflicting messages between the translations and verbal instructions. Another strategy to consider is to give patients the English version of written instructions so that any English-speaking family members or friends can compare them to the machine translations.

“Our study shows that discharge instructions translated by the new GT algorithm were more accurate and associated with fewer seriously harmful inaccuracies than what has been seen previously,” says Dr. Khoong. “We were cautiously pleased with the results. GT should supplement—not replace—written English instructions. Machine translation is not perfect, but we believe it is better than no written instructions in their preferred language at all.” She adds that machine-translated instructions should include a warning about potentially inaccurate translations.