Performance of ChatGPT and Google Translate for Pediatric Discharge Instruction Translation.

BACKGROUND AND OBJECTIVES: Patients who speak languages other than English face barriers to equitable healthcare delivery. Machine translation systems, including emerging large language models, have the potential to expand access to translation services, but their merits and limitations in clinical practice remain poorly defined. We aimed to assess the performance of Google Translate and ChatGPT for multilingual translation of pediatric discharge instructions. METHODS: Twenty standardized discharge instructions for pediatric conditions were translated into Spanish, Brazilian Portuguese, and Haitian Creole by professional translation services, Google Translate and ChatGPT-4.0, and evaluated for adequacy (preserved information), fluency (grammatical correctness), meaning (preserved connotation), and severity (clinical harm), along with assessment of overall preference. Domain-level ratings and preferred translation source were summarized with descriptive statistics and compared with professional translations. RESULTS: Google Translate and ChatGPT demonstrated similar domain-level ratings to professional translations for Spanish and Portuguese. For Haitian Creole, compared with both Google Translate and ChatGPT, professional translations demonstrated significantly greater adequacy, fluency meaning, and severity scores. ChatGPT (33.3%, P < .001) and Google Translate (23.3%, P = .024) contained more potentially clinically significant errors (severity score ≤3) for Haitian Creole than professional translations (8.3%). Professional Haitian Creole (48.3%) and Portuguese (43.3%), but not Spanish (15%), translations were most frequently preferred among translation sources. CONCLUSIONS: Machine translation platforms have comparable performance to professional translations for Spanish and Portuguese but shortcomings in quality, accuracy, and preference persist for Haitian Creole. Diverse multilingual training data are needed, along with regulations ensuring safe and equitable applications of machine translation in clinical practice.

Creating Culturally Adapted Multilingual Materials for Research.

Patients who speak languages other than English are frequently excluded from research. This exclusion exacerbates inequities, biases results, and may violate federal regulations and research ethics. Language justice is the right to communicate in an individual's preferred language to address power imbalances and promote equity. To promote language justice in research, we propose a method to translate and culturally-adapt multifaceted research materials into multiple languages simultaneously. Our method involves a multistep approach, including professional translation, review by bilingual expert panels to refine and reach consensus, and piloting or cognitive interviews with patients and families. Key differences from other translation approaches (eg, the World Health Organization) include omitting back-translation, given its limited utility in identifying translation challenges, and limiting expert panelist and piloting-participant numbers for feasibility. We detail a step-by-step approach to operationalizing this method and outline key considerations learned after utilizing this method to translate materials into 8 languages other than English for an ongoing multicenter pediatric research study on family safety-reporting. Materials included family brochures, surveys, and intervention materials. This approach took ∼6 months overall at a cost of <$2000 per language (not including study personnel costs). Key themes across the project included (1) tailor scope to timeline, budget, and resources, (2) thoughtfully design English source materials, (3) identify and apply guiding principles throughout the translation and editing process, and (4) carefully review content and formatting to account for nuances across multiple languages. This method balances feasibility and rigor in translating participant-facing materials into multiple languages simultaneously, advancing language justice in research.

Validity of the reduced-sample insulin modified frequently-sampled intravenous glucose tolerance test using the nonlinear regression approach.

The disposition index, the product of the insulin sensitivity index (S(I)) and the acute insulin response to glucose, is linked in African Americans to chromosome 11q. This link was determined with S(I) calculated with the nonlinear regression approach to the minimal model and data from the reduced-sample insulin-modified frequently-sampled intravenous glucose tolerance test (Reduced-Sample-IM-FSIGT). However, the application of the nonlinear regression approach to calculate S(I) using data from the Reduced-Sample-IM-FSIGT has been challenged as being not only inaccurate but also having a high failure rate in insulin-resistant subjects. Our goal was to determine the accuracy and failure rate of the Reduced-Sample-IM-FSIGT using the nonlinear regression approach to the minimal model. With S(I) from the Full-Sample-IM-FSIGT considered the standard and using the nonlinear regression approach to the minimal model, we compared the agreement between S(I) from the Full- and Reduced-Sample-IM-FSIGT protocols. One hundred African Americans (body mass index, 31.3 +/- 7.6 kg/m(2) [mean +/- SD]; range, 19.0-56.9 kg/m(2)) had FSIGTs. Glucose (0.3 g/kg) was given at baseline. Insulin was infused from 20 to 25 minutes (total insulin dose, 0.02 U/kg). For the Full-Sample-IM-FSIGT, S(I) was calculated based on the glucose and insulin samples taken at -1, 1, 2, 3, 4, 5, 6, 7, 8,10, 12, 14, 16, 19, 22, 23, 24, 25, 27, 30, 40, 50, 60, 70, 80, 90, 100, 120, 150, and 180 minutes. For the Reduced-Sample-FSIGT, S(I) was calculated based on the time points that appear in bold. Agreement was determined by Spearman correlation, concordance, and the Bland-Altman method. In addition, for both protocols, the population was divided into tertiles of S(I). Insulin resistance was defined by the lowest tertile of S(I) from the Full-Sample-IM-FSIGT. The distribution of subjects across tertiles was compared by rank order and kappa statistic. We found that the rate of failure of resolution of S(I) by the Reduced-Sample-IM-FSIGT was 3% (3/100). For the remaining 97 subjects, S(I) for the Full- and Reduced-Sample-IM-FSIGTs were as follows: 3.76 +/- 2.41 L mU(-1) min(-1) (range, 0.58-14.50) and 4.29 +/- 2.89 L mU(-1) min(-1) (range, 0.52-14.42); relative error, 21% +/- 18%; Spearman r = 0.97; and concordance, 0.94 (both P < .001). After log transformation, the Bland-Altman limits of agreement were -0.29 and 0.53. The exact agreement for distribution of the population in the insulin-resistant tertile vs the insulin-sensitive tertiles was 92%, kappa of 0.82 +/- 0.06. Using the nonlinear regression approach and data from the Reduced-Sample-IM-FSIGT in subjects with a wide range of insulin sensitivity, failure to resolve S(I) occurred in only 3% of subjects. The agreement and maintenance of rank order of S(I) between protocols support the use of the nonlinear regression approach to the minimal model and the Reduced-Sample-IM-FSIGT in clinical studies.

Bacterial chemotaxis differences in Escherichia coli isolated from different hosts.

The mechanisms mediating the association between Escherichia coli and specific hosts are unknown. This study investigates the hypothesis that the host-specific associations of E. coli strains are mediated in part by differences in chemotaxis. To test this hypothesis, chemotactic responses of E. coli strains isolated from different host groups (carnivores, herbivores, and omnivores) were tested with various attractants. In low-density agar chemotaxis assays, the average motility of E. coli in response to aspartate, serine, and ribose among the different groups was not significantly different; however, strains from carnivores responded significantly more to aspartate, relative to their responses to serine, in comparison with strains from herbivores, which responded equally or better to serine than to aspartate. The relatively greater chemotactic response of strains from carnivores to aspartate than to serine was confirmed in a subset of strains by capillary chemotaxis assay. Differences in responses to serine and aspartate were not due to growth differences, as determined by comparison of 24 h growth curves with glycerol, aspartate, and serine carbon sources. The differences in chemotactic behavior of E. coli strains isolated from herbivores and carnivores support the hypothesis that host-specific associations of E. coli strains are mediated in part by differences in chemotactic behavior.