Abstract 13177: Performance of Google Translate for Spanish Translations of Cardiac Care Unit Discharge Instructions

• Published in: Circulation (New York, N.Y.) Vol. 144; no. Suppl_1; p. A13177
• Main Authors: Duvalyan, Angela, Subramanian, Vinayak, Morales-Oyarvide, Vicente, Álvarez Concejo, Bruno, Sumarsono, Andrew
• Format: Journal Article
• Language: English
• Published: Lippincott Williams & Wilkins 16-11-2021
• ISSN: 0009-7322; 1524-4539
• DOI: 10.1161/circ.144.suppl_1.13177

IntroductionProviding written discharge instructions (DIs) to non-English speaking patients is often challenging. Providers often must decide whether to write DIs in English or use online or in-person translation services. Prior studies evaluating the overall efficacy of Google Translate (GT) in the emergency department setting demonstrated mixed results, however the accuracy of GT for cardiology-specific diagnoses has not been studied. We evaluated the accuracy of GT for DIs for patients discharged from the inpatient cardiac care unit (CCU). MethodsWe conducted a single-center, retrospective study of patients discharged from the CCU at Parkland Hospital. We extracted DIs along with the number of medication changes and procedures performed. We analyzed each sentence by content, Flesch-Kincaid Grade Level and Flesch-Kincaid Reading Ease formula. Using GT, we translated each sentence to Spanish and had two translators back-translate to English. The back-translated English was compared to the original English and graded for accuracy, meaning, fluency adequacy, and severity using a 5-point Likert Model. All scores were summed, with a maximum score of 20. We used univariate and multivariate logistic regressions to evaluate factors associated with perfect translation (score = 20). ResultsOf the 44 DIs, 859 sentences were extracted. 769 (90%) of sentences were translated perfectly (score = 20) and 817 (95%) scored greater than or equal to 18. Thirty four sentences received a score of 3 or lower in any category. 2.6% (22 of 859) had severity scores of 3 or lower. In a multivariate logistic regression, sentence content of follow-up instructions was the only factor associated with higher risk of incorrect translation (OR 2.25, 95% CI1.05-4.84, P = 0.038), while Flesch-Kincaid Grade Level (95% CI0.97-1.05), Flesch-Kincaid Reading Ease (95% CI0.996 - 1.02), medication changes (95% CI0.93-1.04), and procedures performed (95% CI0.89-1.26) were not. ConclusionGT perfectly translated 90% of DIs, but incorrectly translated 2.6% of instructions to a degree that may impair patient care. GT may be a useful tool to provide written DIs for cardiac care patients, but should still be used cautiously and in conjunction with verbal instructions.