"I Would Rather Speak in Private": Seven Patients Set Terms for an AI That Calls Them

US · CanadaAmérica Latina

Weekly AI Watch

Every week, what changed and what it means for you. No hurry, no alarm.

Week of 23 October 2026EN | ESPast editions18,276 papers found · 17,064 read · 5,349 belonged · 37 stories passed every check

Observatorio de la
Inteligencia Artificial

Cada semana, qué cambió y qué significa para usted. Sin prisa y sin alarma.

Semana del 23 de octubre de 2026EN | ESEdiciones anteriores18,276 estudios encontrados · 17,064 leídos · 5,349 correspondían · 37 notas pasaron todas las comprobaciones

weeklyAI · Week of 23 October 2026weeklyAI · Semana del 23 de octubre de 2026

← Your rights← Sus derechos

interviews · JMIR formative research · la publicación, 22 jul 2026 · gratis

«A veces no quiero decírselo al intérprete; preferiría hablar en privado»

Siete adultos turcohablantes en Londres pidieron una llamada automática en su idioma, pero solo con aviso previo, prueba de quién llama y la seguridad de que sus respuestas las ve su médico. Fue una sola conversación de dos horas y la herramienta solo se escuchó grabada.

Versión breve · la versión detallada sigue, unos 10 min

Pregunte al chat de weeklyAI

Pregúnteme por este estudio: a quiénes se estudió, qué encontró y qué no dice.

Las conversaciones se guardan mientras exista weeklyAI, para mejorar la publicación. Se responde en el idioma en que usted escribe.

El estudio, de un vistazo
Quiénes: adultos turcohablantes con poco inglés que usaron servicios de ojos en el Reino Unido
Cuántos: 7 adultos (5 mujeres y 2 hombres)
Dónde: Londres, Reino Unido
Cuándo: mayo de 2025
Tipo de estudio: interviews
Quién lo hizo: Institute of Ophthalmology, University College London, y NIHR Moorfields Biomedical Research Centre
El límite que importa: Los siete llegaron por una sola organización comunitaria; pueden no hablar por los turcohablantes más aislados.

La próxima vez que suene una llamada de su servicio de salud, usted ya sabe qué preguntar antes de dar su nombre: quién llama, cómo lo comprueba y quién más va a escuchar.
Lectura de weeklyAI

Ver · 0:32

La frase la dijo una persona de las siete que se sentaron en el Centro de Educación del hospital oftalmológico Moorfields, en Londres, en mayo de 2025. Todos habían pasado por la atención ocular del Reino Unido y decían hablar poco inglés. Los reunió la organización benéfica Derman, que apoya a la comunidad kurdo-turca.

Detrás de esa frase hay una historia repetida en el grupo. Contaron que les pidieron esperar a que la vista empeorara antes de operarlos, y cuatro viajaron a Turquía para operarse antes. Dijeron que los intérpretes del servicio de salud llegaban tarde, cortaban la llamada o hablaban con desprecio, y que por eso muchas veces traducían sus hijos adultos. Esa ayuda también pesa: una persona contó que ahora le dicen «resuélvelo tú, usa un intérprete».

Les pusieron entonces una grabación: una llamada de seguimiento tras la cirugía de cataratas hecha por Dora, un asistente de voz en turco. Es un programa que llama por teléfono, hace preguntas sobre dolor, enrojecimiento y visión, y pasa un informe a un médico, que decide los siguientes pasos. En el servicio de salud inglés, la versión que ya se usa sigue un guion fijo. La conversación abierta con un modelo de lenguaje se probará en el próximo ensayo multilingüe. Primero el grupo dijo qué le parecía la llamada, y después quien dirigía la sesión les contó que la había hecho un asistente automático.

La aceptaron, con condiciones. Una persona dijo que cuando la llamada es automática y en inglés, cuelga. Por eso pidieron que el hospital avise antes, que el asistente diga su nombre y número de hospital para probar que es real, y que solo su equipo médico vea lo que responden. También pidieron un turco estándar de Estambul, más lento, una voz cálida que suene humana, espacio para preguntar y paciencia con los acentos regionales.

Los investigadores resumieron esas peticiones en una lista de diez puntos para la versión turca. La lista no está ordenada por importancia y los propios autores dicen que es un punto de partida que se probará con otras comunidades.

Fueron siete personas de una sola organización, ya conectadas a una red de apoyo, así que las voces más aisladas pueden pedir otra cosa. Hablaron solo de atención ocular. La herramienta se escuchó como grabación, no en una llamada real, y nadie midió si funciona. Esto no dice qué exige ninguna ley.

Cuando reciba una llamada automática sobre su salud, pregunte primero quién avisó que llamarían, cómo puede comprobar que es de verdad su hospital y quién verá lo que usted conteste.

Qué significa para usted

Lo que se probó fue una grabación, no una llamada real, y nadie midió todavía si el asistente acierta. Si un hospital de su país empieza a usar algo parecido, lo razonable es preguntar quién revisa las respuestas y qué pasa cuando la conversación se pone difícil. La lista de diez puntos sigue siendo una propuesta de siete personas.

Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809

Quién pagó: La actividad de participación de pacientes y público fue financiada por una beca PPIE Starter Grant del Biomedical Research Centre de University College London Hospitals del National Institute for Health and Care Research, y el trabajo contribuye a un programa más amplio financiado por el NIHR adjudicado a Ufonia Ltd y Health Innovation Oxford; el artículo indica que los financiadores no tuvieron ningún papel en el diseño del estudio, la recogida, el análisis o la interpretación de los datos, ni en la redacción del manuscrito.

No tome esto como consejo médico profesional.

Versión detalladaLos pasajes copiados del artículo, las ilustraciones y cada fuente con cuánto leímos de ella · 2095 palabras · unos 10 minLeerla →Cerrar

Una voz automática que llama en turco: siete pacientes dicen qué le exigirían antes de contestar

Un grupo de discusión en Londres preguntó a personas con poco inglés qué necesita una llamada médica automática para merecer confianza. No probaron que la máquina funcione.

Imagine el teléfono sonando tres o cuatro semanas después de una cirugía de cataratas. Del otro lado, una voz habla en turco, pregunta por el ojo operado y anota las respuestas. Eso es Dora: un programa de cómputo registrado en el Reino Unido como dispositivo médico de clase I, que hace llamadas telefónicas automáticas y estructuradas a pacientes, a su celular o a su teléfono fijo¹. Un grupo de siete adultos turcohablantes, cinco mujeres y dos hombres, escuchó una grabación de esa llamada y dijo qué le pediría antes de confiar en ella. El ejercicio se hizo una sola vez, en una sesión de dos horas en mayo de 2025, en el Centro de Educación del Hospital Oftalmológico Moorfields, en Londres, con personas reclutadas por la organización comunitaria Derman.

Los siete habían usado servicios de ojos en el Reino Unido y dijeron tener poco inglés. La sesión se hizo en dos partes: primero contaron cómo viven la atención oftalmológica, y después oyeron una llamada pregrabada entre Dora, en turco, y un voluntario turcohablante. Al principio no sabían que la voz era de una máquina; se les reveló después. De ahí salieron cuatro temas y una lista de diez puntos que el equipo armó a partir de lo que dijeron los participantes, llamada "qué hace buena a una llamada": aviso previo, verificación, confidencialidad, claridad y ritmo, voz, empatía, interacción, manejo de acentos regionales, accesibilidad y ahorro de tiempo. La lista no está ordenada por importancia y los autores la presentan como punto de partida, no como algo ya validado².

Lo que este ejercicio no puede decir es igual de importante. No midió si la máquina acierta, si es segura o si mejora algo. Los siete participantes llegaron por una sola organización comunitaria, así que ya tenían una red de apoyo; los autores advierten que su opinión puede no reflejar la de turcohablantes más aislados³. Tampoco se buscó agotar los temas, y el trabajo se clasificó como participación de pacientes y público, no como investigación². Se trabajó con las grabaciones y las notas de campo, no con una transcripción palabra por palabra, y solo se consultó a pacientes: cuidadores, familiares e intérpretes quedaron fuera.

Piense en la escena que describieron: una sala de espera, un hijo al teléfono traduciendo lo que el médico acaba de decir. En el Reino Unido hay 5.1 millones de personas, cerca del 8% de la población, cuyo idioma principal no es el inglés⁴. Para ese grupo, las barreras de idioma están bien documentadas como factor de riesgo clínico y se les asocia con más diagnósticos equivocados, más errores de medicación y peores resultados de salud⁴. El turco está entre los diez idiomas principales de Inglaterra y Gales con menor proporción de hablantes que dicen hablar inglés "bien" o "muy bien": 69.9%, lo que deja casi a uno de cada tres sin hablarlo bien o sin hablarlo del todo⁵.

Así funciona el aparato, paso a paso. El sistema vive en la nube y marca él mismo al paciente¹. En las versiones que ya operan en inglés dentro del servicio de salud británico, la conversación sigue un guion fijo para preguntar por un conjunto predeterminado de síntomas: enrojecimiento, dolor, problemas de visión, luces centelleantes y cuerpos flotantes. Después, una regla determinada produce un resumen y una recomendación de "pasa" o "falla" para que un clínico la revise. Para el ensayo multilingüe que viene, la parte conversacional usará un modelo grande de lenguaje, capaz de sostener diálogo abierto en varios idiomas, pero conservando el mismo paso determinista de clasificación y la supervisión de un clínico⁶. La voz que escucharon los participantes era una grabación, no el sistema en vivo.

¿Y qué se sabe de esta clase de herramientas? El artículo describe un estudio en el que estos agentes pueden hacer evaluaciones de seguimiento con exactitud clínica y seguridad comparables a las de un oftalmólogo humano, con buena aceptación entre pacientes⁷. Es el mismo tipo de promesa que se repite en otros frentes: los asistentes conversacionales de salud están entrando en uso clínico más rápido de lo que la evidencia alcanza a caracterizar su seguridad en el mundo real, según el resumen de un estudio alemán sobre un sistema informativo para pacientes con párkinson; de ese trabajo pudimos leer solo el resumen, el texto completo está detrás de una suscripción⁸. Ese estudio siguió 2,035 conversaciones durante los primeros 129 días de operación de un sistema público en Alemania⁹.

Lo que encontró ese seguimiento alemán merece contarse con cuidado, porque habla de lo que puede esconderse detrás de una buena calificación automática. Una clasificación asistida por inteligencia artificial marcó como buenas 1,803 de esas 2,035 conversaciones, es decir 88.6%; 224, el 11.0%, quedaron como parcialmente adecuadas, y 8, el 0.4%, como inadecuadas¹⁰. Hasta ahí, el panorama se ve limpio. Pero una relectura independiente de 100 conversaciones elegidas al azar entre las calificadas como buenas encontró cuatro, el 4%, con errores clínicamente críticos que la clasificación automática no vio¹¹. Entre los eventos críticos confirmados hubo una recomendación inapropiada de un medicamento para demencia en párkinson y una ideación suicida explícita que no activó la respuesta de emergencia prevista¹². Los autores concluyen que un sistema acotado y con búsqueda documental puede lograr buenas clasificaciones automáticas y aun así albergar fallas clínicas críticas invisibles a esas métricas, y que detectarlas exigió revisión clínica experta conversación por conversación¹³. Conviene decir con claridad que nada de esto se refiere a la consulta de Londres ni a Dora.

El contexto de privacidad también pesa. En el resumen de un análisis sobre consentimiento en oncología de precisión, se reporta que muchos pacientes apoyan la promesa clínica de la inteligencia artificial pero expresan preocupación por el uso secundario de sus datos sin consentimiento significativo, por la comercialización y por las filtraciones de seguridad; de ese texto pudimos leer solo el resumen¹⁴. El mismo resumen señala que los marcos regulatorios actuales, como el reglamento europeo de protección de datos y la ley estadounidense de portabilidad y responsabilidad en seguros de salud, se centran sobre todo en la identificabilidad y en quitar los identificadores explícitos¹⁵. Los participantes de Londres pidieron justamente lo contrario de un trámite opaco: que se les asegure que su información no se comparte con "nadie más que nuestros doctores"¹⁶.

Hay un detalle que ayuda a entender por qué la lista de diez puntos empieza como empieza. Varios participantes dijeron que suelen colgar las llamadas automáticas: "Normalmente, cuando es una máquina y está en inglés, cuelgo", contó uno, y el grupo habló de su conciencia de las estafas telefónicas¹⁷. Sus propuestas para verificar la llamada fueron concretas: que la inteligencia artificial diga el nombre del paciente y su número de expediente hospitalario, o incluso un código de acceso compartido de antemano¹⁸. También pidieron un límite claro para el papel de la máquina: adecuada solo para tareas rutinarias e informativas, no para consultas complejas o emocionalmente delicadas¹⁶. Y sobre la voz: clara, con acento turco estándar de Estambul y a un ritmo más lento y pausado; "me pareció un poco rápida; cuando es demasiado rápida, algo se te puede perder", dijo una persona¹⁹.

La razón de fondo está en cómo consiguen ayuda hoy. Pedirle a un hijo que traduzca puede volverse una carga para los dos: el apoyo llega "después de mil quejas" y deja culpa, y la situación se complica a medida que los hijos adultos forman sus propias casas y carreras²⁰. Muchos preferirían hablar de su salud sin familiares ni intérpretes oficiales presentes, sobre todo en temas delicados: "A veces no quiero decírselo al intérprete; preferiría hablar en privado"²¹. Ahí se entiende por qué una voz en su idioma les pareció valiosa: no por gusto por la tecnología, sino por privacidad. Y también por algo tan simple como leer: "No sé leer ni escribir; por mis problemas de ojos, me es más fácil apretar, hablar y mandar"²².

Lo que sigue, según los autores, es un ensayo clínico con el sistema multilingüe, que se desarrolla y evalúa en diez idiomas, en un centro oftalmológico grande que atiende a una población diversa²³. Ese ensayo medirá su impacto en seguridad clínica, eficiencia del recorrido del paciente y equidad en salud²³. Todavía no ha reportado resultados, y nada de lo aquí contado debe leerse como evidencia de ese ensayo.

Sobre quién paga y quién tiene intereses: la actividad fue financiada con una beca del instituto de investigación biomédica de los hospitales universitarios del University College London y contribuye a un programa financiado por el Instituto Nacional de Investigación en Salud del Reino Unido otorgado a Ufonia Ltd y a Health Innovation Oxford. Dos autores son empleados de Ufonia Ltd, la empresa detrás del sistema, y otro autor ha actuado como consultor para la misma empresa²⁴. Eso no invalida lo que dijeron los participantes; sí conviene saberlo al leer las conclusiones.

Así lo leemos nosotros. Hay un patrón que se repite cuando un servicio llega por una vía rápida y directa: la llamada, la voz, la pantalla se sienten más cercanas que la ventanilla lenta, aunque esa cercanía también puede ocultar quién responde del otro lado. Esperaríamos que muchas personas acepten con gusto una llamada en su idioma justamente porque evita intérpretes y filas, y que esa misma comodidad las vuelva menos exigentes al verificar quién llama y qué hará con lo que cuenten. Nos equivocaríamos si, al ver a los pacientes valorando la rapidez y el idioma, los viéramos insistir por su cuenta en pedir el número de expediente o preguntar quién más escuchará la conversación antes de responder. Lo que sí puede hacer usted, desde hoy, es sencillo: antes de contestar una llamada médica automatizada, pida el nombre y el número de expediente que le dio su hospital, y pregunte quién más oirá la grabación. Si no puede confirmarlo, cuelgue y marque usted al número oficial de su proveedor.

También vemos una segunda cosa, y es la que más nos hace pensar. Quien representa a una comunidad actúa como delegado, no como dueño de la autoridad: su papel es escuchar las exigencias de la gente a la que sirve, y la autoridad última está en esa gente. Cabe esperar que los pacientes acepten una voz automática solo si sienten que su propio hospital la presenta y la respalda, y que traten a la máquina como un intermediario que rinde cuentas ante ellos y ante su médico, no como una autoridad por sí misma. Sabríamos que nos equivocamos si los pacientes aceptaran la llamada sin que su proveedor la anuncie, o si no les importara quién responde por los errores de la conversación. Usted puede exigir que sea su clínica quien le avise por escrito que esa llamada existirá y con qué número llegará, y preguntar ante quién se reclama si la llamada se equivoca o filtra información. Esa pregunta no la contestó este grupo, y es la que más va a importar cuando el sistema llegue a más idiomas.

Un tercer punto, sobre el idioma mismo. Traducir y transcribir automáticamente entre lenguas sigue siendo imperfecto, y el diálogo multilingüe amplía el alcance pero no garantiza que el mensaje llegue intacto. Esperaríamos que en una llamada automatizada en turco se cuelen malentendidos con acentos regionales, palabras médicas o instrucciones concretas, y que el paciente no siempre note que algo se entendió mal. Nos pondría en duda que los pacientes detectaran y corrigieran por sí mismos los errores de la máquina, o que la conversación registrara y aclarara cada instrucción sin una persona que la revise. Al terminar una llamada así, usted puede repetir en sus propias palabras lo que entendió sobre medicamentos, cuidado del ojo operado o cuándo volver, y pedir que un profesional se lo confirme por escrito. Si algo no cuadra, pida hablar con una persona del equipo médico.

Volvamos al teléfono que suena. Los siete participantes no pidieron que la máquina los reemplace ni que decida por ellos; pidieron saber quién llama, que se les hable despacio, que se les escuche y que haya un médico detrás. Esa lista es corta y se puede pedir en cualquier país, en cualquier clínica, sin esperar a que llegue la tecnología. La próxima vez que suene una llamada de su servicio de salud, usted ya sabe qué preguntar antes de dar su nombre: quién llama, cómo lo comprueba y quién más va a escuchar.

De dónde sale cada dato de contexto, y cuánto leímos de cada documento

Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Dora is a UK Conformity Assessed class I software as a Medical Device used for autonomous, structured telephone conversations, such as follow-up after uncomplicated adult cataract surgery, typically 3 to 4 weeks postoperatively. The system runs as a cloud-hosted service and reaches patients via telephone, either on mobile or landline."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - el artículo del que trata esta nota — el artículo completo — el pasaje: "As is appropriate for a formative PPI activity rather than a qualitative research study, we did not seek thematic saturation, and the consultation domains should be read as a community-generated starting point for design rather than as generalizable findings about Turkish speaking patients in the United Kingdom [ 38 ]."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The contributors were, by definition, already connected to a community support network, so their input may not reflect the priorities of more isolated Turkish speakers—a gap that further consultation in the next phase of the design program will need to address."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - el artículo del que trata esta nota — el artículo completo — el pasaje: "In the United Kingdom, there are 5.1 million people (close to 8% of the population) whose main language is not English [ 9 ]. For this population, language barriers are a well-documented clinical risk factor, linked to a higher incidence of misdiagnosis, medication errors, and poorer health outcomes [ 10 , 11 ]."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Turkish is among the 10 main non-English languages in England and Wales with the lowest proportion of speakers who report speaking English “well” or “very well” (69.9%), leaving almost 1 in 3 who cannot speak English “well” or “at all” [ 19 ]."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - el artículo del que trata esta nota — el artículo completo — el pasaje: "For the forthcoming multilingual clinical trial, the conversational component will use a large language model (LLM) to support open-ended dialog in multiple languages while retaining the same deterministic triage step and clinician oversight ( Figure 1 )."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Initial evidence from English language deployments shows these AI agents can perform follow-up assessments with clinical accuracy and safety comparable to human ophthalmologists [ 5 ], while achieving high usability and patient acceptability [ 4 , 6 ]."
Lange F, Mardi S, Binder T, Reich MM, Odorfer T, Volkmann J. (2026). Real-world use and evaluation of a generative AI chatbot for Parkinson's disease information: a prospective observational study. The Lancet Regional Health - Europe. 10.1016/j.lanepe.2026.101866 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Patient-facing medical AI chatbots are entering clinical use faster than evidence can characterise their real-world safety. We evaluated one deployed system and introduce CARE-LLM, a proposed framework for conversation-level post-market surveillance."
Lange F, Mardi S, Binder T, Reich MM, Odorfer T, Volkmann J. (2026). Real-world use and evaluation of a generative AI chatbot for Parkinson's disease information: a prospective observational study. The Lancet Regional Health - Europe. 10.1016/j.lanepe.2026.101866 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "We conducted a prospective, observational, conversation-level evaluation of jAImes, a retrieval-augmented AI information system for Parkinson's disease deployed publicly in Germany, across its first 129 days of operation (Nov 11, 2025-Mar 20, 2026), analysing all 2035 conversations (6146 messages)."
Lange F, Mardi S, Binder T, Reich MM, Odorfer T, Volkmann J. (2026). Real-world use and evaluation of a generative AI chatbot for Parkinson's disease information: a prospective observational study. The Lancet Regional Health - Europe. 10.1016/j.lanepe.2026.101866 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "AI-assisted triage classified 1803/2035 conversations (88·6%) as good, 224/2035 (11·0%) as partially adequate, and 8/2035 (0·4%) as inadequate."
Lange F, Mardi S, Binder T, Reich MM, Odorfer T, Volkmann J. (2026). Real-world use and evaluation of a generative AI chatbot for Parkinson's disease information: a prospective observational study. The Lancet Regional Health - Europe. 10.1016/j.lanepe.2026.101866 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Independent re-review of 100 randomly sampled 'good'-rated conversations identified four (4/100, 4%; 95% CI 1·6-9·8) with clinically critical errors missed by the triage-a conditional false-negative rate within the 'good' stratum, not an overall critical-event rate."
Lange F, Mardi S, Binder T, Reich MM, Odorfer T, Volkmann J. (2026). Real-world use and evaluation of a generative AI chatbot for Parkinson's disease information: a prospective observational study. The Lancet Regional Health - Europe. 10.1016/j.lanepe.2026.101866 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Confirmed critical events-five flagged, four sampled-spanned three failure classes (knowledge boundary, robustness, and escalation failures) and included an inappropriate memantine recommendation in Parkinson's disease dementia and explicit suicidal ideation that did not trigger the intended emergency response."
Lange F, Mardi S, Binder T, Reich MM, Odorfer T, Volkmann J. (2026). Real-world use and evaluation of a generative AI chatbot for Parkinson's disease information: a prospective observational study. The Lancet Regional Health - Europe. 10.1016/j.lanepe.2026.101866 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Tightly scoped, retrieval-augmented, patient-facing AI achieved favourable automated triage classifications while still harbouring clinically critical failures invisible to automated quality metrics alone; detecting them required conversation-level evaluation with clinical expert adjudication."
Jung M. (2026). Protecting Informational Self-determination in AI-driven Precision Oncology: Privacy, Ethics, and Governance Challenges. Journal of Preventive Medicine and Public Health. 10.3961/jpmph.26.094 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Although many patients support the clinical promise of AI, they express concerns about secondary data use without meaningful consent, commercialization, and security breaches."
Jung M. (2026). Protecting Informational Self-determination in AI-driven Precision Oncology: Privacy, Ethics, and Governance Challenges. Journal of Preventive Medicine and Public Health. 10.3961/jpmph.26.094 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Current regulatory frameworks, including the General Data Protection Regulation and the Health Insurance Portability and Accountability Act, focus primarily on identifiability and the removal of explicit identifiers."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The group also recommended a clear boundary for the AI’s role, considering it appropriate for routine, informational tasks only and not for complex or emotionally sensitive consultations. Finally, they emphasized the need for explicit assurances of data confidentiality, expressing a strong preference that their information not be shared with “anyone other than our doctors.”"
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Contributors’ skepticism toward unsolicited automated calls was clear, with 1 stating, “Usually when it’s a machine and it’s in English, I just put the phone down.” This sentiment, along with an awareness of fraudulent scams, highlighted the need for strong verification measures."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Suggestions for this included the AI stating the patient’s name and hospital number, or even providing preshared “pin codes for access” to prove the call was authentic."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Speech should be clear, use a standard Istanbul Turkish accent, and be delivered at a slower, more deliberate pace. “It felt a bit fast. When it’s too fast you can lose something.”"
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Contributors described how requests for assistance could lead to frustration for both parties, with support often coming after “a thousand complaints,” generating feelings of guilt. This dynamic was reported to be growing more complex as adult children establish their own households and careers, diminishing their availability."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Many contributors expressed a preference to discuss health matters without relatives or even official interpreters present, particularly for sensitive topics. The sentiment, “Sometimes I don’t want to tell the interpreter; I would rather speak in private,” underscores a powerful desire for an autonomous space to manage their health care."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - el artículo del que trata esta nota — el artículo completo — el pasaje: "A voice-first interface suitable for users with low vision or limited literacy. “I can’t read or write. Because I have problems with my eyes, it’s easier for me to press, speak, and send.”"
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Following this comprehensive PPI phase, the multilingual system, which is being developed and evaluated in 10 different languages, will be assessed in a clinical trial at a large tertiary eye care center serving a diverse population [ 39 ]. This trial will measure its impact on clinical safety, pathway efficiency, and health equity outcomes."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - el artículo del que trata esta nota — el artículo completo — el pasaje: "AH and EL are employees of Ufonia Ltd. MS has acted as a consultant for Ufonia Ltd."

No tome esto como consejo médico profesional.

Los hallazgos de otros estudios que aquí se mencionan los conocemos por este documento, que fue el que leímos; no abrimos cada uno de esos estudios.

interviews · JMIR formative research · the paper, 22 Jul 2026 · free

"I Would Rather Speak in Private": Seven Patients Set Terms for an AI That Calls Them

In one two-hour conversation in London, seven Turkish-speaking adults said they would take a medical follow-up call from a voice assistant, provided the caller can prove who it is.

Short version · the longer version follows, about 10 min

Ask the weeklyAI chat

Ask me about this study: who was studied, what it found, and what it does not say.

Conversations are saved for as long as weeklyAI exists, to improve the publication. Answers come in the language you write in.

The study at a glance
Who: Turkish-speaking adults who had used UK eye services
How many: 7
Where: London, United Kingdom
When: May 2025
Kind of study: interviews
Who did it: University College London Institute of Ophthalmology and Moorfields Eye Hospital
The limit that matters: Seven people from one community charity; not meant to speak for all Turkish speakers

If the voice cannot tell you your own hospital number, treat it as unverified and ring the number you already have.
weeklyAI's reading

How it could look · illustration generated by weeklyAI.watch, not a photograph

"Sometimes I don't want to tell the interpreter; I would rather speak in private."

That sentence came from one of seven Turkish-speaking adults in London, gathered for two hours at Moorfields Eye Hospital in May 2025. They were recruited through Derman, a charity that serves the Kurdish-Turkish community. All had used eye services in Britain and described limited English.

Their complaints ran in a pattern. Care came late; one was told cataract surgery could wait until age 75. Interpreters arrived late, spoke down to them, or cut calls short. Four traveled to Turkey for procedures. Several relied on their grown children to translate, and felt guilty for asking. One said the children now answer, "Sort it out yourself, use an interpreter."

Asked about a Turkish-language phone assistant for routine cataract follow-up, the group accepted the idea only with conditions. They wanted word from their hospital before the call came. They wanted proof of who was calling, such as the assistant saying their name and hospital number, or a pin code shared in advance. They wanted a promise that only their doctors would see their answers. One put it simply: "If the machine gives me my hospital number, then I'll speak."

They also asked for clear, slower speech in standard Istanbul Turkish, a warm voice that sounds human, with a choice of male or female tone. They wanted room to ask questions back, and patience with regional accents. One said of the recorded call, "It felt a bit fast. When it's too fast you can lose something."

Consider what this was. Seven people, one charity, one conversation. They heard the assistant only as a recording of a sample call. The article says the group was already connected to a support network, so more isolated speakers may want different things, and its authors call the checklist a starting point to be tested with other language communities. Nothing here shows the tool works, or that any authority requires it.

For anyone who gets an automated health call, three questions are worth carrying: who told me this call was coming, how do I know it is really them, and who will see my answers.

What this means for you

What this means for you is limited: seven people in one London room, all reached through a single community charity, set conditions for a tool that is still being built. No law or regulator in your country has ruled on a call like this, and the study cannot show the assistant works. If one ever rings, ask who arranged the call, how its identity can be checked, and who sees the answers.

Who paid: The patient and public involvement activity was funded by a National Institute for Health and Care Research University College London Hospitals Biomedical Research Centre PPIE Starter Grant, and the work contributes to a larger NIHR-funded programme awarded to Ufonia Ltd and Health Innovation Oxford; the article states the funders had no role in study design, data collection, analysis or interpretation, or in writing the manuscript.

Do not take this as professional medical advice.

The longer versionThe passages copied from the paper, the pictures, and every source with how much of it we read · 2001 words · about 10 minRead it →Close

A Voice in Turkish Asks About Your Eye. Seven Patients Said: Prove It's the Hospital.

Before a cataract follow-up call in their own language could be trusted, a London focus group wanted a name, a hospital number and a promise of privacy. The tool itself has not yet been tested on patients.

Consider the phone ringing three weeks after cataract surgery. The caller speaks Turkish, your Turkish, and asks about redness, pain, flashing lights. For someone who has spent years asking a son or daughter to translate at the clinic, the call is a small liberation. It is also, as seven Londoners pointed out in one two-hour conversation, exactly what a scam sounds like.

The tool is called Dora, made by a company called Ufonia. It is a piece of software registered as a medical device in the United Kingdom, and it holds structured telephone conversations: it runs as a cloud service and reaches patients on a mobile or a landline, typically three to four weeks after uncomplicated adult cataract surgery¹. In its English-language deployments it asks about a fixed list of symptoms, applies a fixed set of rules, and produces a report that a clinician reviews. For the planned multilingual trial, the conversation itself will be handled by a large language model, able to follow open-ended talk in several languages, while the triage step and the clinician oversight stay the same².

The new paper, published in JMIR Formative Research, is not a test of how well that works. It is a record of what a community asked for before the Turkish version is built.

The authors are Sevgi, Ong, Lean and colleagues, working from the Institute of Ophthalmology at University College London and the Moorfields Eye Hospital biomedical research centre, with co-authors from Oxford and from Ufonia itself. The work was paid for by a patient-and-public-involvement starter grant from the National Institute for Health and Care Research, and contributes to a larger NIHR-funded programme awarded to Ufonia Ltd and Health Innovation Oxford. Two of the authors are Ufonia employees, and the lead moderator has acted as a consultant to the company.

Seven Turkish-speaking adults, five women and two men, took part: two married couples among them, all of whom had used eye services in Britain for conditions from cataracts to diabetic retinopathy, all reporting limited English. They were recruited through Derman, a charity serving the Kurdish-Turkish community, met at Moorfields in May 2025, and were compensated for their time. The session ran bilingually, with an interpreter from Derman, and was recorded.

The first half was about their care. Contributors described being told to wait until a cataract was "really bad," one of them told surgery could wait until age 75. Four had travelled to Turkey for operations they could get sooner; one put the contrast plainly: here they were told to wait, in Istanbul they operated the same week.

The second half was about the machine. They listened to a recorded Turkish follow-up call, gave their first reactions, were told it was an AI, argued about trust, then listened again with instructions to judge the accent, pace and tone.

What they said about language is the heart of the paper. Interpreters arrived late, behaved confrontationally, cut calls short, or left out details, contributors reported, and the gaps had consequences: confusion over driving at night after surgery, confusion over fasting before another procedure. So families filled in. That reliance carried a cost: help often came after "a thousand complaints," with guilt on both sides, and it was thinning as adult children moved away and built careers³. Several said they would rather discuss health without relatives or even interpreters present, particularly for sensitive matters: "Sometimes I don't want to tell the interpreter; I would rather speak in private"⁴.

Against that background, the group's verdict on a Turkish-language voice assistant was conditional acceptance, not enthusiasm. One contributor described the default response to an English machine: "Usually when it's a machine and it's in English, I just put the phone down"⁵. Awareness of fraudulent calls sharpened the point⁵. The fixes they proposed were specific: the hospital should announce the service in advance; the caller should prove itself by stating the patient's name and hospital number, or by a pre-agreed PIN⁶; the AI should stay inside routine, informational tasks and hand complex or emotional matters to a clinician; and it should say plainly that the information goes to the doctors and no one else⁷.

On the voice itself, they asked for clear speech in standard Istanbul Turkish, delivered more slowly: "It felt a bit fast. When it's too fast you can lose something"⁸. They wanted a human-sounding voice with a choice of male or female, warm phrasing, and the ability to ask a question back. They wanted it to understand regional accents while keeping its own clear. And they wanted it usable by people who cannot read: "I can't read or write. Because I have problems with my eyes, it's easier for me to press, speak, and send"⁹.

The authors turned this into a ten-point checklist — preparation, verification, confidentiality, clarity and pace, voice, empathy, interactivity, dialect handling, accessibility, efficiency — and state that it is not rank-ordered and will be validated with other language communities. They also state the limits: the contributors were already connected to a support charity, so more isolated speakers may want different things¹⁰. They had experience of eye care only. Saturation was not sought, and the findings are not generalizable to Turkish speakers in Britain¹¹. Caregivers, interpreters and community staff were not consulted. And the analysis worked from recordings and notes, not a verbatim transcript.

The scale behind the question is not small. In Britain, 5.1 million people, close to 8 percent of the population, have a main language other than English¹², and language barriers there are linked to more misdiagnosis, more medication errors and poorer outcomes¹². Turkish is among the ten main non-English languages in England and Wales with the lowest share of speakers reporting they speak English well or very well, at 69.9 percent, leaving almost one in three who cannot speak it well or at all¹³.

The technology has a track record, of a kind, in English. The article describes a study in which initial evidence from English-language deployments shows these agents performing follow-up assessments with clinical accuracy and safety comparable to human ophthalmologists, with high usability and patient acceptability¹⁴. That sentence covers accuracy and safety; it does not cover whether a Turkish-speaking patient trusts the caller.

Elsewhere, researchers watching a patient-facing medical chatbot in the wild have found what automated quality scores miss. In a study of jAImes, a chatbot for Parkinson's disease deployed publicly in Germany, the team analysed all 2,035 conversations from its first 129 days; automated triage rated 1,803 of them, 88.6 percent, as good, 224, or 11.0 percent, as partially adequate, and 8, or 0.4 percent, as inadequate¹⁵¹⁶. Expert review of 45 flagged conversations confirmed five critical events¹⁷. Then came the uncomfortable part: an independent re-review of 100 randomly sampled conversations rated "good" found four with clinically critical errors the triage had missed, a rate within that sample, not an overall rate¹⁸. The confirmed events spanned three failure types and included an inappropriate drug recommendation and a statement of suicidal ideation that did not trigger the intended emergency response¹⁹. The authors' conclusion: tightly scoped systems can score well automatically while still harbouring critical failures invisible to automated metrics²⁰. We could read only the summary; the full paper is behind a subscription. The study was funded by German research bodies and the technical development of the chatbot was commissioned and financed by a German Parkinson's foundation²¹.

Privacy runs alongside trust. A paper on AI in cancer care, again read only in summary, reports that many patients support the clinical promise of AI while worrying about secondary use of their data without meaningful consent, about commercialization and about breaches²², and that privacy violations threaten autonomy, confidentiality and trust and may widen existing inequities²³. It notes that current frameworks such as the European data protection regulation and the American health privacy law focus mainly on whether data can identify a person²⁴, while AI can infer new privacy-relevant facts, such as disease susceptibility or family risk, from data that looks anonymous²⁵. The author calls for layered governance: privacy-preserving technology, consent that is renewed over time, data minimization, transparency and stronger legal safeguards²⁶.

Consent paperwork has its own gap. A review of publicly available consent documents in clinical trial registries, read only in summary, found that 66 of 114, or 58 percent, failed to disclose the type of AI or its intended use, and 18.4 percent, 21 documents, omitted risks entirely²⁷; registry entries and consent forms disagreed about AI methods²⁸; only 11.4 percent, 13 documents, included visual aids²⁹; and handling of data after a participant withdrew was inconsistent, with 44.7 percent giving no information at all and only 3.5 percent offering a choice³⁰. The authors note the limits: these were public registry documents only³¹. Their proposal is a minimum set of requirements for consent in AI-related trials³².

Here is how we read it. Authority travels on small tokens. A badge, a title, a number read back to you: people grant a voice the power to make them act when it carries something only the real institution could know, and they hang up when it carries nothing. In this group, the token was the hospital number, spoken by the machine before any question. That is a testable expectation, and here is how you would know we are wrong: if patients stayed on the line and followed advice from a call that offered no hospital-specific detail, or if adding a name and number changed nothing about whether they engaged, the reading fails.

The second thing we noticed is that the group judged the call by whether it let them speak. People who have sat through appointments where an interpreter talked over them, or waited months to be told to wait longer, do not measure a call by its speed. They measure it by whether a question can be asked back and answered. A smooth one-way script may confirm the sense of being invisible; a machine that answers may be let in. Watch for it: when a health service near you announces an automated call, ask whether you can interrupt it and get an answer. If not, ask the clinic for a route to a person and say you need to respond, not only listen.

The accent decision is worth naming for what it is. "Standard Istanbul Turkish" is a choice, the way broadcast English is a choice, and speakers of regional forms will hear in it a ranking. When a service offers a voice in your language, ask which variety it uses and whether it can understand how you actually speak. If it understands you poorly, say so and ask that it be recorded, rather than assuming the fault is yours.

One caveat about carrying over: what these seven asked for has not been shown to hold for Turkish speakers who are not connected to a community charity, nor for other language groups, nor for other kinds of clinic. That is a question about whether the finding travels, not about whether it happened.

What happens next is a trial. The multilingual system is being developed and evaluated in 10 languages and will be assessed in a clinical trial at a large tertiary eye centre serving a diverse population, measuring clinical safety, pathway efficiency and health equity outcomes³³. Until that reports, no one can say whether a Turkish-language call is safe, accurate or effective; this paper says only what seven people asked of it.

The phone still rings three weeks after surgery. What the reader now knows is what to ask before answering: which hospital is calling, what will it say to prove it, and can I talk back. If the voice cannot tell you your own hospital number, treat it as unverified and ring the number you already have.

Where each piece of context comes from, and how much of it we read

Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - the article this story is about — the whole article — the passage: "Dora is a UK Conformity Assessed class I software as a Medical Device used for autonomous, structured telephone conversations, such as follow-up after uncomplicated adult cataract surgery, typically 3 to 4 weeks postoperatively. The system runs as a cloud-hosted service and reaches patients via telephone, either on mobile or landline."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - the article this story is about — the whole article — the passage: "For the forthcoming multilingual clinical trial, the conversational component will use a large language model (LLM) to support open-ended dialog in multiple languages while retaining the same deterministic triage step and clinician oversight ( Figure 1 )."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - the article this story is about — the whole article — the passage: "Contributors described how requests for assistance could lead to frustration for both parties, with support often coming after “a thousand complaints,” generating feelings of guilt. This dynamic was reported to be growing more complex as adult children establish their own households and careers, diminishing their availability."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - the article this story is about — the whole article — the passage: "Many contributors expressed a preference to discuss health matters without relatives or even official interpreters present, particularly for sensitive topics. The sentiment, “Sometimes I don’t want to tell the interpreter; I would rather speak in private,” underscores a powerful desire for an autonomous space to manage their health care."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - the article this story is about — the whole article — the passage: "Contributors’ skepticism toward unsolicited automated calls was clear, with 1 stating, “Usually when it’s a machine and it’s in English, I just put the phone down.” This sentiment, along with an awareness of fraudulent scams, highlighted the need for strong verification measures."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - the article this story is about — the whole article — the passage: "Suggestions for this included the AI stating the patient’s name and hospital number, or even providing preshared “pin codes for access” to prove the call was authentic."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - the article this story is about — the whole article — the passage: "The group also recommended a clear boundary for the AI’s role, considering it appropriate for routine, informational tasks only and not for complex or emotionally sensitive consultations. Finally, they emphasized the need for explicit assurances of data confidentiality, expressing a strong preference that their information not be shared with “anyone other than our doctors.”"
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - the article this story is about — the whole article — the passage: "Speech should be clear, use a standard Istanbul Turkish accent, and be delivered at a slower, more deliberate pace. “It felt a bit fast. When it’s too fast you can lose something.”"
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - the article this story is about — the whole article — the passage: "A voice-first interface suitable for users with low vision or limited literacy. “I can’t read or write. Because I have problems with my eyes, it’s easier for me to press, speak, and send.”"
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - the article this story is about — the whole article — the passage: "The contributors were, by definition, already connected to a community support network, so their input may not reflect the priorities of more isolated Turkish speakers—a gap that further consultation in the next phase of the design program will need to address."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - the article this story is about — the whole article — the passage: "As is appropriate for a formative PPI activity rather than a qualitative research study, we did not seek thematic saturation, and the consultation domains should be read as a community-generated starting point for design rather than as generalizable findings about Turkish speaking patients in the United Kingdom [ 38 ]."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - the article this story is about — the whole article — the passage: "In the United Kingdom, there are 5.1 million people (close to 8% of the population) whose main language is not English [ 9 ]. For this population, language barriers are a well-documented clinical risk factor, linked to a higher incidence of misdiagnosis, medication errors, and poorer health outcomes [ 10 , 11 ]."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - the article this story is about — the whole article — the passage: "Turkish is among the 10 main non-English languages in England and Wales with the lowest proportion of speakers who report speaking English “well” or “very well” (69.9%), leaving almost 1 in 3 who cannot speak English “well” or “at all” [ 19 ]."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - the article this story is about — the whole article — the passage: "Initial evidence from English language deployments shows these AI agents can perform follow-up assessments with clinical accuracy and safety comparable to human ophthalmologists [ 5 ], while achieving high usability and patient acceptability [ 4 , 6 ]."
Lange F, Mardi S, Binder T, Reich MM, Odorfer T, Volkmann J. (2026). Real-world use and evaluation of a generative AI chatbot for Parkinson's disease information: a prospective observational study. The Lancet Regional Health - Europe. 10.1016/j.lanepe.2026.101866 — only the abstract - the full text could not be fetched — the passage: "We conducted a prospective, observational, conversation-level evaluation of jAImes, a retrieval-augmented AI information system for Parkinson's disease deployed publicly in Germany, across its first 129 days of operation (Nov 11, 2025-Mar 20, 2026), analysing all 2035 conversations (6146 messages)."
Lange F, Mardi S, Binder T, Reich MM, Odorfer T, Volkmann J. (2026). Real-world use and evaluation of a generative AI chatbot for Parkinson's disease information: a prospective observational study. The Lancet Regional Health - Europe. 10.1016/j.lanepe.2026.101866 — only the abstract - the full text could not be fetched — the passage: "AI-assisted triage classified 1803/2035 conversations (88·6%) as good, 224/2035 (11·0%) as partially adequate, and 8/2035 (0·4%) as inadequate."
Lange F, Mardi S, Binder T, Reich MM, Odorfer T, Volkmann J. (2026). Real-world use and evaluation of a generative AI chatbot for Parkinson's disease information: a prospective observational study. The Lancet Regional Health - Europe. 10.1016/j.lanepe.2026.101866 — only the abstract - the full text could not be fetched — the passage: "Expert review of 45 flagged conversations confirmed five critical events."
Lange F, Mardi S, Binder T, Reich MM, Odorfer T, Volkmann J. (2026). Real-world use and evaluation of a generative AI chatbot for Parkinson's disease information: a prospective observational study. The Lancet Regional Health - Europe. 10.1016/j.lanepe.2026.101866 — only the abstract - the full text could not be fetched — the passage: "Independent re-review of 100 randomly sampled 'good'-rated conversations identified four (4/100, 4%; 95% CI 1·6-9·8) with clinically critical errors missed by the triage-a conditional false-negative rate within the 'good' stratum, not an overall critical-event rate."
Lange F, Mardi S, Binder T, Reich MM, Odorfer T, Volkmann J. (2026). Real-world use and evaluation of a generative AI chatbot for Parkinson's disease information: a prospective observational study. The Lancet Regional Health - Europe. 10.1016/j.lanepe.2026.101866 — only the abstract - the full text could not be fetched — the passage: "Confirmed critical events-five flagged, four sampled-spanned three failure classes (knowledge boundary, robustness, and escalation failures) and included an inappropriate memantine recommendation in Parkinson's disease dementia and explicit suicidal ideation that did not trigger the intended emergency response."
Lange F, Mardi S, Binder T, Reich MM, Odorfer T, Volkmann J. (2026). Real-world use and evaluation of a generative AI chatbot for Parkinson's disease information: a prospective observational study. The Lancet Regional Health - Europe. 10.1016/j.lanepe.2026.101866 — only the abstract - the full text could not be fetched — the passage: "Tightly scoped, retrieval-augmented, patient-facing AI achieved favourable automated triage classifications while still harbouring clinically critical failures invisible to automated quality metrics alone; detecting them required conversation-level evaluation with clinical expert adjudication."
Lange F, Mardi S, Binder T, Reich MM, Odorfer T, Volkmann J. (2026). Real-world use and evaluation of a generative AI chatbot for Parkinson's disease information: a prospective observational study. The Lancet Regional Health - Europe. 10.1016/j.lanepe.2026.101866 — only the abstract - the full text could not be fetched — the passage: "Funding Deutsche Forschungsgemeinschaft, Interdisciplinary Center for Clinical Research Würzburg, and German Ministry of Education and Research (BMBF); technical development of jAImes was commissioned and financed by Parkinson Stiftung Deutschland."
Jung M. (2026). Protecting Informational Self-determination in AI-driven Precision Oncology: Privacy, Ethics, and Governance Challenges. Journal of Preventive Medicine and Public Health. 10.3961/jpmph.26.094 — only the abstract - the full text could not be fetched — the passage: "Although many patients support the clinical promise of AI, they express concerns about secondary data use without meaningful consent, commercialization, and security breaches."
Jung M. (2026). Protecting Informational Self-determination in AI-driven Precision Oncology: Privacy, Ethics, and Governance Challenges. Journal of Preventive Medicine and Public Health. 10.3961/jpmph.26.094 — only the abstract - the full text could not be fetched — the passage: "Privacy violations threaten autonomy, confidentiality, and trust and may exacerbate existing inequities."
Jung M. (2026). Protecting Informational Self-determination in AI-driven Precision Oncology: Privacy, Ethics, and Governance Challenges. Journal of Preventive Medicine and Public Health. 10.3961/jpmph.26.094 — only the abstract - the full text could not be fetched — the passage: "Current regulatory frameworks, including the General Data Protection Regulation and the Health Insurance Portability and Accountability Act, focus primarily on identifiability and the removal of explicit identifiers."
Jung M. (2026). Protecting Informational Self-determination in AI-driven Precision Oncology: Privacy, Ethics, and Governance Challenges. Journal of Preventive Medicine and Public Health. 10.3961/jpmph.26.094 — only the abstract - the full text could not be fetched — the passage: "However, AI systems generate risks not only through re-identification but also through inferential analytics that derive new privacy-relevant information, such as disease susceptibility or familial risk, from ostensibly anonymized data."
Jung M. (2026). Protecting Informational Self-determination in AI-driven Precision Oncology: Privacy, Ethics, and Governance Challenges. Journal of Preventive Medicine and Public Health. 10.3961/jpmph.26.094 — only the abstract - the full text could not be fetched — the passage: "Addressing these gaps requires multilayered governance that combines privacy-preserving technologies, dynamic consent, data minimization, transparency, and strengthened legal safeguards to sustain public trust."
Su H, Xiao F, Chau H, Tong Y, Han S, Cheng X, et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. 10.2196/94504 — only the abstract - the full text could not be fetched — the passage: "More than half (66/114, 58%) of ICs failed to disclose the AI type or its intended use, and 18.4% (n=21) omitted risks entirely."
Su H, Xiao F, Chau H, Tong Y, Han S, Cheng X, et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. 10.2196/94504 — only the abstract - the full text could not be fetched — the passage: "Discrepancy was observed between trial registry entries and IC reporting of AI methods."
Su H, Xiao F, Chau H, Tong Y, Han S, Cheng X, et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. 10.2196/94504 — only the abstract - the full text could not be fetched — the passage: "Only 11.4% (n=13) of ICs included visual aids, and their inclusion was not correlated with lower reading difficulty."
Su H, Xiao F, Chau H, Tong Y, Han S, Cheng X, et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. 10.2196/94504 — only the abstract - the full text could not be fetched — the passage: "Data handling protocols after participant withdrawal were inconsistent: 51 (44.7%) ICs provided no information, 30 (26.3%) specified data destruction, 29 (25.4%) allowed continued use, and only 4 (3.5%) offered participants a choice."
Su H, Xiao F, Chau H, Tong Y, Han S, Cheng X, et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. 10.2196/94504 — only the abstract - the full text could not be fetched — the passage: "However, it should be noted that these findings are limited to publicly available consent documents in the registry a"
Su H, Xiao F, Chau H, Tong Y, Han S, Cheng X, et al. (2026). Informed Consent Disclosures and Minimum Requirements in AI Clinical Trials: Cross-Sectional Analysis. Journal of Medical Internet Research. 10.2196/94504 — only the abstract - the full text could not be fetched — the passage: "We propose the "Minimum Requirements for Informed Consent in AI‑related Clinical Trials" as a possible framework to improve consent quality."
Sevgi, M., Ong, A. Y., Lean, K. et al. (2026). Multilingual Voice AI for Postoperative Cataract Follow-Up in Turkish Speaking Patients in the United Kingdom: Patient and Public Involvement Focus Group Study. JMIR Formative Research. https://doi.org/10.2196/90809 - the article this story is about — the whole article — the passage: "Following this comprehensive PPI phase, the multilingual system, which is being developed and evaluated in 10 different languages, will be assessed in a clinical trial at a large tertiary eye care center serving a diverse population [ 39 ]. This trial will measure its impact on clinical safety, pathway efficiency, and health equity outcomes."

Do not take this as professional medical advice.

The findings of other studies mentioned here are known to us through this document, which is the one we read; we did not open each of those studies.

The free weekly drip by email

One email a week: the free story of every section, with its picture, and the headlines of the rest.

No payment, nothing to cancel. Unsubscribe from any issue.

El goteo semanal gratis por correo

Un correo por semana: la nota gratuita de cada sección, con su imagen, y los titulares de las demás.

Sin pago y sin nada que cancelar. Puede darse de baja desde cualquier envío.