{"id":76223,"date":"2025-09-17T17:00:08","date_gmt":"2025-09-17T15:00:08","guid":{"rendered":"https:\/\/www.embl.org\/news\/?p=76223"},"modified":"2025-09-17T17:00:11","modified_gmt":"2025-09-17T15:00:11","slug":"ai-model-forecasts-disease-risk-decades-in-advance","status":"publish","type":"post","link":"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/","title":{"rendered":"AI model forecasts disease risk decades in advance"},"content":{"rendered":"\n<article class=\"vf-card vf-card--brand vf-card--bordered vf-u-margin__bottom--800\" default>\n  <div class=\"vf-card__content | vf-stack vf-stack--400\">\n      <h3 class=\"vf-card__heading\">\n      Summary    <\/h3>\n                <p class=\"vf-card__text\"><ul>\r\n \t<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Researchers have developed an AI model that estimates long-term disease risk across more than 1,000 medical conditions<\/span><\/li>\r\n \t<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">The model, trained and tested on anonymised medical data from the UK and Denmark, can forecast health outcomes over a decade in advance<\/span><\/li>\r\n \t<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">While not ready for direct clinical use, the AI model offers new ways to study disease and inform healthcare strategies<\/span><\/li>\r\n<\/ul><\/p>\n      <\/div>\n<\/article>\n\n\n\n\n<p>Imagine a future where your medical history could help predict what health conditions you might face in the next two decades. Researchers have developed a generative AI model that uses large-scale health records to estimate how human health may change over time. It can forecast the risk and timing of over 1,000 diseases and predict health outcomes over a decade in advance.<\/p>\n\n\n\n<p>This new generative AI model was custom-built using algorithmic concepts similar to those used in large language models (LLMs). It was trained on anonymised patient data from 400,000 participants from the <a href=\"https:\/\/www.ukbiobank.ac.uk\/\">UK Biobank<\/a>. Researchers also successfully tested the model using data from 1.9 million patients in the Danish National Patient Registry. This approach is one of the most comprehensive demonstrations to date of how generative AI can model human disease progression at scale and was tested on data from two entirely separate healthcare systems.&nbsp;<\/p>\n\n\n\n<p>\u201cOur AI model is a proof of concept, showing that it\u2019s possible for AI to learn many of our long-term health patterns and use this information to generate meaningful predictions,\u201d said <a href=\"https:\/\/www.ebi.ac.uk\/people\/person\/ewan-birney\/\">Ewan Birney, Interim Executive Director at the <\/a><a href=\"https:\/\/www.embl.org\/\">European Molecular Biology Laboratory<\/a> (<a href=\"https:\/\/www.ebi.ac.uk\/people\/person\/ewan-birney\/\">EMBL<\/a>). \u201cBy modelling how illnesses develop over time, we can start to explore when certain risks emerge and how best to plan early interventions. It\u2019s a big step towards more personalised and preventive approaches to healthcare.\u201d<\/p>\n\n\n\n<p>This work, <a href=\"https:\/\/www.nature.com\/articles\/s41586-025-09529-3\">published in the journal <em>Nature<\/em><\/a>, was a collaboration between <a href=\"https:\/\/www.embl.org\/\">EMBL<\/a>, the <a href=\"https:\/\/www.dkfz.de\/en\/\">German Cancer Research Centre (DKFZ)<\/a>, and the <a href=\"https:\/\/www.ku.dk\/en\">University of Copenhagen<\/a>.&nbsp;<\/p>\n\n\n\n<style>\n  .video-container {\n    position: relative;\n    padding-bottom: 56.25%; \/* 16:9 aspect ratio *\/\n    height: 0;\n    overflow: hidden;\n  }\n  .video-container iframe {\n    position: absolute;\n    top: 0; left: 0;\n    width: 100%;\n    height: 100%;\n  }\n<\/style>\n\n<div class=\"video-container\">\n  <iframe\n    src=\"https:\/\/www.youtube.com\/embed\/82U65MrKm70\"\n    title=\"YouTube Shorts video player\"\n    frameborder=\"0\"\n    allow=\"accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture\"\n    allowfullscreen>\n  <\/iframe>\n<\/div>\n\n\n\n\n<div style=\"height:22px\" aria-hidden=\"true\" class=\"wp-block-spacer\"><\/div>\n\n\n\n<h2 class=\"wp-block-heading\">AI for health forecasting&nbsp;<\/h2>\n\n\n\n<p>Just as large language models can learn the structure of sentences, this AI model learns the &#8216;grammar&#8217; of health data to model medical histories as sequences of events unfolding over time. These events include medical diagnoses or lifestyle factors such as smoking. The model learns to forecast disease risk from the order in which such events happen and how much time passes between these events.&nbsp;<\/p>\n\n\n\n<p>\u201cMedical events often follow predictable patterns,\u201d said <a href=\"https:\/\/www.ebi.ac.uk\/people\/person\/tomas-fitzgerald\/\">Tom Fitzgerald, Staff Scientist at EMBL\u2019s European Bioinformatics Institute (EMBL-EBI<\/a>). \u201cOur AI model learns those patterns and can forecast future health outcomes. It gives us a way to explore what might happen based on a person\u2019s medical history and other key factors. Crucially, this is not a certainty, but an estimate of the potential risks.\u201d<\/p>\n\n\n\n<p>The model performs especially well for conditions with clear and consistent progression patterns, such as certain types of cancer, heart attacks, and septicaemia, which is a type of blood poisoning. However, the model is less reliable for more variable conditions, such as mental health disorders or pregnancy-related complications that depend on unpredictable life events.&nbsp;<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Future use and limitations&nbsp;<\/h2>\n\n\n\n<p>Like weather forecasts, this new AI model provides probabilities, not certainties. It doesn\u2019t predict exactly what will happen to an individual, but it offers well-calibrated estimates of how likely certain conditions are to occur over a given period. For example, it could predict the chance of developing heart disease within the next year. These risks are expressed as rates over time, similar to forecasting a 70% chance of rain tomorrow. Generally, forecasts over a shorter period of time have higher accuracy than long-range ones.<\/p>\n\n\n\n<p>For example the model predicts varying levels of risk for heart attacks. Taking the UK BioBank cohort at the age of 60\u201365, the risk of heart attack varies from a chance of 4 in 10,000 per year for some men to approximately 1 in 100 in other men, depending on their prior diagnoses and lifestyle. Women have a lower risk on average, but a similar spread of risk. Moreover, the risks increase, on average, as people age. A systematic assessment on data from the UK Biobank not used for training showed that these calculated risks correspond well to the observed number of cases across age and sex groups.<\/p>\n\n\n\n<p>The model is calibrated to produce accurate population-level risk estimates, forecasting how often certain conditions occur within groups of people. However, like any AI model, it has limitations. For example, because the model&#8217;s training data from the UK Biobank comes primarily from individuals aged 40\u201360, childhood and adolescent health events are underrepresented. The model also contains demographic biases due to gaps in the training data, including the underrepresentation of certain ethnic groups.<\/p>\n\n\n\n<p>While the model isn\u2019t ready for clinical use, it could already help researchers:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>understand how diseases develop and progress over time,<\/li>\n\n\n\n<li>explore how lifestyle and past illnesses affect long-term disease risk,<\/li>\n\n\n\n<li>simulate health outcomes using artificial patient data, in situations where real-world data are difficult to obtain or access.<\/li>\n<\/ul>\n\n\n\n<p>In the future, similar AI tools trained on more representative datasets could assist clinicians in identifying high-risk patients early. With ageing populations and rising rates of chronic illness, being able to forecast future health needs could help healthcare systems plan better and allocate resources more efficiently. But much more testing, consultation, and robust regulatory frameworks are needed before AI models can be deployed in a clinical setting.<\/p>\n\n\n\n<p>\u201cThis is the beginning of a new way to understand human health and disease progression,\u201d said <a href=\"https:\/\/www.dkfz.de\/en\/employees\/moritz-gerstung\">Moritz Gerstung, Head of the Division of AI in Oncology at DKFZ<\/a> and former Group Leader at EMBL-EBI. \u201cGenerative models such as ours could one day help personalise care and anticipate healthcare needs at scale. By learning from large populations, these models offer a powerful lens into how diseases unfold, and could eventually support earlier, more tailored interventions.\u201d<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Data privacy and ethics<\/h3>\n\n\n\n<p>This AI model was trained using anonymised health data under strict ethical rules. UK Biobank participants gave informed consent, and Danish data were accessed in accordance with national regulations that require the data to remain within Denmark. Researchers used secure, virtual systems to analyse the data without moving them across borders. These safeguards help ensure that AI models are developed and used in ways that respect privacy and uphold ethical standards.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Funding<\/h3>\n\n\n\n<p>This work was funded by EMBL member state contributions, DKFZ funds and Novo Nordisk Foundation grant.&nbsp;<\/p>\n\n\n\n<hr class=\"vf-divider\"\/>\n\n\n\n<h1 class=\"wp-block-heading\" id=\"German\">KI-Modell prognostiziert Krankheitsrisiken Jahrzehnte im Voraus<\/h1>\n\n\n\n<p><strong>Wissenschaftlerinnen und Wissenschaftler vom European Molecular Biology Laboratory (EMBL) und vom Deutschen Krebsforschungszentrum (DKFZ) haben ein KI-Modell entwickelt, das das langfristige individuelle Risiko f\u00fcr mehr als 1.000 Erkrankungen einsch\u00e4tzt. Das Modell, das auf anonymisierten medizinischen Daten aus Gro\u00dfbritannien und D\u00e4nemark trainiert und getestet wurde, kann Gesundheitsereignisse f\u00fcr eine Zeitspanne von \u00fcber einem Jahrzehnt prognostizieren. Das in der Fachzeitschrift <em>Nature<\/em> vorgestellte Modell ist noch nicht f\u00fcr den klinischen Einsatz bereit, er\u00f6ffnet aber schon jetzt neue M\u00f6glichkeiten, um Gesundheitsstrategien zu entwickeln.<\/strong><\/p>\n\n\n\n<p>L\u00e4sst sich anhand Ihrer pers\u00f6nlichen Krankengeschichte vorhersagen, mit welchen Gesundheitsproblemen Sie in den n\u00e4chsten zwei Jahrzehnten konfrontiert sein k\u00f6nnten? Dass dies m\u00f6glich ist, zeigen nun Forschende vom EMBL, vom DKFZ und der Universit\u00e4t Kopenhagen. Siehaben ein generatives KI-Modell entwickelt, das auf der Basis umfangreicher Gesundheitsdaten absch\u00e4tzt, mit welchen gesundheitlichen Beeintr\u00e4chtigungen der oder die Einzelne im Laufe der Zeit rechnen muss. Es kann das Risiko und den Zeitpunkt von \u00fcber 1.000 Krankheiten prognostizieren und Gesundheitsentwicklungen \u00fcber einen Zeitraum von zehn Jahren vorhersagen.<\/p>\n\n\n\n<p>Die Algorithmen, auf deren Basis das neue generative KI-Modell entwickelt wurde, \u00e4hneln denen, die in gro\u00dfen Sprachmodellen (LLMs) verwendet werden. Das Modell wurde zun\u00e4chst an anonymisierten Patientendaten von 400.000 Teilnehmern aus der UK Biobanktrainiert. Anschlie\u00dfend pr\u00fcften die Forscher es erfolgreich mit Daten von 1,9 Millionen Personen aus dem d\u00e4nischen nationalen Patientenregister. Das Modell ist die bislang umfassendste Demonstration daf\u00fcr, wie generative KI den Verlauf menschlicher Krankheiten in gro\u00dfem Ma\u00dfstab modellieren kann, und wurde anhand von Daten aus zwei v\u00f6llig getrennten Gesundheitssystemen gepr\u00fcft.<\/p>\n\n\n\n<p>\u201eUnser KI-Modell ist ein Machbarkeitsnachweis, der zeigt, dass es m\u00f6glich ist, viele langfristige Gesundheitsmuster zu erkennen und diese Informationen zu nutzen, um aussagekr\u00e4ftige Vorhersagen zu generieren\u201c, sagt Ewan Birney vom EMBL. \u201eIndem wir modellieren, wie sich Krankheiten im Laufe der Zeit entwickeln, k\u00f6nnen wir untersuchen, wann bestimmte Risiken auftreten und wie fr\u00fchzeitige Interventionen am besten geplant werden k\u00f6nnen. Dass ist ein gro\u00dfer Schritt in Richtung personalisierter und pr\u00e4ventiverer Ans\u00e4tze in der Gesundheitsversorgung.\u201c<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Die \u201eGrammatik\u201c der Gesundheitsdaten&nbsp;<\/strong><\/h2>\n\n\n\n<p>\u201eSo wie gro\u00dfe Sprachmodelle aus der Abfolge von W\u00f6rtern in Texten die Grammatik unserer Sprache lernen k\u00f6nnen, lernt dieses KI-Modell die Logik der zeitlichen Abfolge von Ereignissen in Gesundheitsdaten, um ganze Krankengeschichten zu modellieren\u201c, erkl\u00e4rt Moritz Gerstung vom DKFZ. Zu diesen Ereignissen geh\u00f6ren medizinische Diagnosen oder auch Lebensstilfaktoren wie Rauchen. An der Reihenfolge, in der die Ereignisse eintreten, und der Zeit, die zwischen diesen Ereignissen vergeht, lernt das Modell, das Krankheitsrisiko vorherzusagen.<\/p>\n\n\n\n<p>\u201eMedizinische Ereignisse folgen oft vorhersehbaren Mustern\u201d, sagt Tom Fitzgerald vom Europ\u00e4ischen Bioinformatik-Institut des EMBL (EMBL-EBI). \u201eUnser KI-Modell lernt diese Muster und kann zuk\u00fcnftige Gesundheitsergebnisse prognostizieren. Es gibt uns die M\u00f6glichkeit, auf der Grundlage der Krankengeschichte einer Person und anderer wichtiger Faktoren zu untersuchen, was passieren k\u00f6nnte. Entscheidend ist, dass es sich dabei nicht um eine Gewissheit handelt, sondern um eine Einsch\u00e4tzung der potenziellen Risiken.\u201d<\/p>\n\n\n\n<p>Das Modell eignet sich besonders gut f\u00fcr Erkrankungen mit klaren und konsistenten Verlaufsmustern, wie bestimmte Krebsarten, Herzinfarkte oder Sepsis. Bei variableren Diagnosen, wie psychischen Erkrankungen oder Schwangerschaftskomplikationen, die von unvorhersehbaren Lebensereignissen abh\u00e4ngen, ist es jedoch weniger zuverl\u00e4ssig.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Wahrscheinlichkeiten, keine Gewissheiten<\/strong><\/h2>\n\n\n\n<p>Wie Wettervorhersagen liefert auch das neue KI-Modell Wahrscheinlichkeiten und keine Gewissheiten. Es kann das Schicksal einer bestimmten Person nicht genau vorhersagen, sondern bietet gut kalibrierte Sch\u00e4tzungen dar\u00fcber, wie wahrscheinlich bestimmte Erkrankungen in einem bestimmten Zeitraum auftreten werden. Zum Beispiel die Wahrscheinlichkeit, innerhalb des n\u00e4chsten Jahres eine Herzerkrankung zu entwickeln. Diese Risiken werden als Zeitraten ausgedr\u00fcckt, \u00e4hnlich wie bei der Vorhersage einer 70-prozentigen Regenwahrscheinlichkeit f\u00fcr morgen. Hier sind Vorhersagen \u00fcber einen k\u00fcrzeren Zeitraum f\u00fcr gew\u00f6hnlich pr\u00e4ziser als langfristige Prognosen.<\/p>\n\n\n\n<p>Die Forschenden konnten zeigen, dass die vom Modell berechneten Wahrscheinlichkeiten tats\u00e4chlich mit der erwarteten H\u00e4ufigkeit eintraten. Wie jedes KI-Modell hat es jedoch auch seine Grenzen. Da die Trainingsdaten aus der UK Biobank beispielsweise haupts\u00e4chlich von Personen im Alter von 40 bis 60 Jahren stammen, sind Gesundheitsereignisse im Kindes- und Jugendalter unterrepr\u00e4sentiert, das gilt auch f\u00fcr bestimmte ethische Gruppen.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Beispiel Herzinfarkt<\/strong><\/h2>\n\n\n\n<p>Das vom KI-Modell berechnete Risiko eines Herzinfarkts bei M\u00e4nnern im Alter zwischen 60 und 65 variiert zwischen einer Wahrscheinlichkeit von 4 pro 10.000\/Jahr und etwa 100 pro 10.000\/Jahr, abh\u00e4ngig von fr\u00fcheren Diagnosen und dem Lebensstil der M\u00e4nner. Frauen haben im Durchschnitt ein geringeres Herzinfarktrisiko, aber eine \u00e4hnlich breite Streuung.<\/p>\n\n\n\n<p>Dar\u00fcber hinaus steigt das Herzinfarkt-Risiko bei M\u00e4nnern und Frauen mit zunehmendem Alter. Eine systematische Bewertung dieser berechneten Risiken in verschiedenen Alters- und Geschlechtsgruppen zeigt, dass sie gut mit der Anzahl von F\u00e4llen \u00fcbereinstimmen, die in einem Teil der UK Biobank Kohorte, die nicht f\u00fcr das Training des Modells benutzt wurden, beobachtet wurden.<\/p>\n\n\n\n<p>Das Modell ist noch nicht f\u00fcr den klinischen Einsatz bereit, k\u00f6nnte aber bereits jetzt Forschern helfen\u2026<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>zu verstehen, wie Krankheiten sich im Laufe der Zeit entwickeln und fortschreiten.<\/li>\n\n\n\n<li>zu untersuchen, wie sich Lebensstil und fr\u00fchere Erkrankungen auf das langfristige Krankheitsrisiko auswirken.<\/li>\n\n\n\n<li>Gesundheitsergebnisse anhand k\u00fcnstlicher Patientendaten zu simulieren, wenn reale Daten schwer zu beschaffen oder zug\u00e4nglich sind.<\/li>\n<\/ul>\n\n\n\n<p>In Zukunft k\u00f6nnten KI-Tools, die auf repr\u00e4sentativeren Datens\u00e4tzen trainiert wurden, \u00c4rzten dabei helfen, Hochrisikopatienten fr\u00fchzeitig zu identifizieren. Angesichts der alternden Bev\u00f6lkerung und der steigenden Rate chronischer Erkrankungen k\u00f6nnte die F\u00e4higkeit, zuk\u00fcnftige Gesundheitsbed\u00fcrfnisse vorherzusagen, den Gesundheitssystemen helfen, besser zu planen und Ressourcen effizienter zuzuweisen. Bevor KI-Modelle jedoch in einer klinischen Umgebung eingesetzt werden k\u00f6nnen, sind noch viele weitere Tests sowie robuste regulatorische Rahmenbedingungen erforderlich.<\/p>\n\n\n\n<p>\u201eDas ist der Beginn einer neuen Art, die menschliche Gesundheit und den Verlauf von Krankheiten zu verstehen\u201c, prognostiziert Moritz Gerstung. \u201eSolche generativen Modelle k\u00f6nnten eines Tages dazu beitragen, die Versorgung zu personalisieren und den Bedarf an medizinischer Versorgung in gro\u00dfem Ma\u00dfstab zu antizipieren. Durch das Lernen aus gro\u00dfen Populationen bieten diese Modelle einen aussagekr\u00e4ftigen Einblick in den Verlauf von Krankheiten und k\u00f6nnten letztendlich fr\u00fchzeitigere, ma\u00dfgeschneiderte Interventionen unterst\u00fctzen.\u201c<\/p>\n\n\n\n<p>Das KI-Modell wurde unter strengen ethischen Regeln mit anonymisierten Gesundheitsdaten trainiert. Die Teilnehmer der UK Biobank gaben ihre Einwilligung, und auf die d\u00e4nischen Register wurde gem\u00e4\u00df den nationalen Vorschriften zugegriffen, die vorschreiben, dass die Daten innerhalb D\u00e4nemarks bleiben m\u00fcssen. Die Forscher verwendeten sichere, virtuelle Systeme, um die Daten zu analysieren, ohne sie \u00fcber Grenzen hinweg zu \u00fcbertragen. Diese Sicherheitsvorkehrungen tragen dazu bei, dass KI-Modelle unter Wahrung der Privatsph\u00e4re und unter Einhaltung ethischer Standards entwickelt und eingesetzt werden.<\/p>\n\n\n\n<p>Die Arbeit wurde durch Beitr\u00e4ge der EMBL-Mitgliedstaaten, Mittel des DKFZ und einen Zuschuss der Novo Nordisk Foundation finanziert.<\/p>\n\n\n\n<hr class=\"vf-divider\"\/>\n\n\n\n<h1 class=\"wp-block-heading\" id=\"Spanish\">Un nuevo modelo de IA predice el riesgo de padecer enfermedades con d\u00e9cadas de antelaci\u00f3n<\/h1>\n\n\n\n<p><strong><em>El modelo puede estimar el riesgo a largo plazo de m\u00e1s de 1000 enfermedades y predice los cambios en la salud humana con d\u00e9cadas de antelaci\u00f3n<\/em><\/strong><\/p>\n\n\n\n<article class=\"vf-card vf-card--brand vf-card--bordered vf-u-margin__bottom--800\" default>\n  <div class=\"vf-card__content | vf-stack vf-stack--400\">\n      <h3 class=\"vf-card__heading\">\n      Resumen    <\/h3>\n                <p class=\"vf-card__text\"><ul>\r\n \t<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Un equipo de cient\u00edficos desarrolla un modelo de IA que estima el riesgo a largo plazo de padecer m\u00e1s de 1000 enfermedades<\/span><\/li>\r\n \t<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">El modelo se ha entrenado con datos m\u00e9dicos an\u00f3nimos procedentes de Reino Unido y Dinamarca, y puede predecir resultados de salud con m\u00e1s de una d\u00e9cada de antelaci\u00f3n<\/span><\/li>\r\n \t<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">El modelo todav\u00eda no est\u00e1 listo para uso cl\u00ednico, pero ofrece nuevas v\u00edas para estudiar enfermedades y desarrollar estrategias para atenci\u00f3n sanitaria<\/span><\/li>\r\n<\/ul><\/p>\n      <\/div>\n<\/article>\n\n\n\n\n<p>Imagina un futuro donde tu historial m\u00e9dico pudiera ayudar a predecir qu\u00e9 enfermedades podr\u00edas padecer en las pr\u00f3ximas dos d\u00e9cadas. Cient\u00edficos del EMBL y DKFZ desarrollan un modelo de IA generativa que usa informes m\u00e9dicos a gran escala para estimar c\u00f3mo cambia la salud humana durante a\u00f1os. Este modelo puede predecir el riesgo y el momento concreto de m\u00e1s de 1000 enfermedades y predecir resultados de salud con m\u00e1s de una d\u00e9cada de antelaci\u00f3n.<\/p>\n\n\n\n<p>Este nuevo modelo de IA generativa ha sido construido a medida usando conceptos algor\u00edtmicos similares a los de modelos de lenguaje a gran escala (LLM por sus siglas en ingl\u00e9s). Los cient\u00edficos entrenaron el modelo con datos an\u00f3nimos de m\u00e1s de 400.000 pacientes del UK Biobank. El modelo se prob\u00f3 de manera exitosa usando datos de 1,9 millones de pacientes del Registro Nacional de Pacientes Daneses. Este m\u00e9todo es una de las demostraciones m\u00e1s completas hasta la fecha de c\u00f3mo la IA generativa puede modelar la progresi\u00f3n de enfermedades humanas a gran escala y se teste\u00f3 con datos de dos sistemas de atenci\u00f3n sanitaria completamente independientes.<\/p>\n\n\n\n<p>\u201cNuestro modelo de IA es una prueba de concepto: demuestra que es posible aprender de nuestros patrones de salud a largo plazo y usar esta informaci\u00f3n&nbsp; para generar predicciones valiosas,\u201d dice <a href=\"https:\/\/www.ebi.ac.uk\/people\/person\/ewan-birney\/\">Ewan Birney, Director General Interino del Laboratorio Europeo de Biolog\u00eda Molecular (EMBL)<\/a>. \u201cSi modelamos c\u00f3mo se desarrollan las enfermedades a lo largo del tiempo, podemos empezar a explorar cu\u00e1ndo empiezan a emerger ciertos riesgos y esto nos permite planificar intervenciones preventivas. Es un gran paso hacia un sistema de salud personalizado y hacia la medicina preventiva.\u201d<\/p>\n\n\n\n<p>Este trabajo se publica en <a href=\"https:\/\/www.nature.com\/articles\/s41586-025-09529-3\"><em>Nature<\/em><\/a> y es una colaboraci\u00f3n entre el <a href=\"https:\/\/www.embl.org\/\">EMBL<\/a>, la <a href=\"https:\/\/www.dkfz.de\/en\/\">DKFZ<\/a> y la <a href=\"https:\/\/www.ku.dk\/en\">Universidad de Copenhague<\/a>.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">IA para la predicci\u00f3n de la salud<\/h2>\n\n\n\n<p>As\u00ed como los modelos de lenguaje a gran escala pueden aprender la estructura de las oraciones, este modelo de IA aprende la \u2018gram\u00e1tica\u2019 de los datos de salud para modelar los historiales m\u00e9dicos como secuencias de eventos que se desarrollan a lo largo del tiempo. Estos eventos incluyen diagn\u00f3sticos m\u00e9dicos o factores de estilo de vida, como el tabaquismo. El modelo aprende a predecir el riesgo de enfermedad a partir del orden en que ocurren dichos eventos y del tiempo que transcurre entre ellos.<\/p>\n\n\n\n<p>\u201cLos eventos m\u00e9dicos a menudo siguen patrones predecibles\u201d. dice <a href=\"https:\/\/www.ebi.ac.uk\/people\/person\/tomas-fitzgerald\/\">Tom Fitzgerald, investigador del Instituto Europeo de Bioinform\u00e1tica del EMBL<\/a>. \u201cNuestro modelo de IA aprende esos patrones y puede predecir resultados de salud. Nos proporciona una v\u00eda para explorar lo que podr\u00eda pasarle a una persona bas\u00e1ndose en su historial m\u00e9dico y otros factores clave. Obviamente la predicci\u00f3n no es una certeza, si no una estimaci\u00f3n de los riesgos potenciales.\u201d<\/p>\n\n\n\n<p>El modelo funciona especialmente bien para condiciones con patrones de desarrollo claros y consistentes como por ejemplo ciertos tipos de c\u00e1ncer, infartos y sepsis en sangre.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Usos futuros y limitaciones<\/h2>\n\n\n\n<p>Como las predicciones del tiempo, este nuevo modelo de IA proporciona probabilidades, no certezas. No predice de manera exacta lo que le pasar\u00e1 a un individuo, pero proporciona estimaciones bien calibradas de c\u00f3mo ciertas condiciones m\u00e9dicas pueden ocurrir durante un periodo de tiempo. Por ejemplo, el modelo podr\u00eda predecir las probabilidades de desarrollar una enfermedad cardiovascular en el pr\u00f3ximo a\u00f1o. Estos riesgos vienen expresados como ratios o tasas a lo largo del tiempo, similar a prever un 70% de probabilidad de lluvia para ma\u00f1ana.<\/p>\n\n\n\n<p>Algunos sucesos, como el riesgo de ser hospitalizado por un evento m\u00e9dico importante \u2013 como un infarto \u2013 se pueden predecir con certeza, mientras que otros son m\u00e1s inciertos. As\u00ed mismo pasa con las predicciones a corto plazo, que son m\u00e1s exactas que aquellas que se hacen a largo plazo.<\/p>\n\n\n\n<p>Por ejemplo, cuando se usan nuevos datos que no se utilizaron para entrenar el modelo, \u00e9ste predice niveles de riesgo variables para un infarto. Si se toma el cohorte del UK BioBank para edades entre 50 y 55, el riesgo de infarto var\u00eda desde una probabilidad de 1 en 10.000 por a\u00f1o para algunos hombres hasta aproximadamente 1 en 100 para otros, dependiendo de sus diagn\u00f3sticos anteriores y su estilo de vida. Las mujeres tienen un riesgo promedio menor, pero una distribuci\u00f3n de riesgo similar. Adem\u00e1s, de media, el riesgo aumenta con la edad de los pacientes. Una evaluaci\u00f3n sistem\u00e1tica de estos riesgos calculados en distintos grupos de edad y sexo mostr\u00f3 que corresponden bien con el n\u00famero de casos observados.<\/p>\n\n\n\n<p>El modelo est\u00e1 calibrado para producir estimaciones precisas de riesgo a nivel poblacional, prediciendo con qu\u00e9 frecuencia ocurren ciertas condiciones en grupos de personas. No obstante, como cualquier modelo de IA, tiene ciertas limitaciones. Por ejemplo, como los datos que se usaron para entrenarlo son del UK Biobank y \u00e9ste principalmente contiene informaci\u00f3n de individuos entre 40 y 60 a\u00f1os, las condiciones m\u00e9dicas pedi\u00e1tricas y de adolescentes est\u00e1n subrepresentadas. El modelo tambi\u00e9n tiene sesgos demogr\u00e1ficos debido a la falta de datos para entrenarlo, incluyendo subrepresentaci\u00f3n de ciertos grupos \u00e9tnicos.<\/p>\n\n\n\n<p>Pese a que el modelo no est\u00e1 en la fase de uso cl\u00ednico, ya puede ayudar a investigadores a:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>entender c\u00f3mo las enfermedades se desarrollan a lo largo del tiempo,<\/li>\n\n\n\n<li>explorar c\u00f3mo el estilo de vida y enfermedades pasadas afectan al riesgo de enfermedad a largo plazo,<\/li>\n\n\n\n<li>simular resultados de salud usando datos de pacientes artificiales para situaciones en las que es dif\u00edcil tener o acceder a datos reales.<\/li>\n<\/ul>\n\n\n\n<p>En el futuro, modelos similares de IA entrenados con datos m\u00e1s representativos, podr\u00edan ayudar al personal sanitario a identificar de manera preventiva pacientes de alto riesgo. La poblaci\u00f3n envejece, las tasas de enfermedades cr\u00f3nicas aumentan, y modelos como estos pueden ayudar a predecir necesidades futuras en los sistemas de salud, as\u00ed como planificar mejor y destinar recursos de manera m\u00e1s eficiente. No obstante, antes de que modelos de IA como este puedan ser implementados en contextos cl\u00ednicos, se necesita mucho m\u00e1s testeo, asesoramiento y marcos regulatorios s\u00f3lidos.<\/p>\n\n\n\n<p>\u201cEste es el principio de una nueva manera de entender la salud humana y el desarrollo de enfermedades,\u201d dice <a href=\"https:\/\/www.dkfz.de\/en\/employees\/moritz-gerstung\">Moritz Gerstung, Director de la Divisi\u00f3n de IA en Oncolog\u00eda en DKFZ y ex-jefe de grupo en EMBL-EBI<\/a> \u201cAlg\u00fan d\u00eda, modelos generativos como el nuestro podr\u00edan ayudar a personalizar la asistencia y a anticipar necesidades sanitarias a gran escala. Al aprender de grandes poblaciones, estos modelos ofrecen una perspectiva poderosa sobre c\u00f3mo se desarrollan las enfermedades y, a la larga, podr\u00edan ayudar a hacer intervenciones preventivas y m\u00e1s personalizadas.\u201d<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Privacidad y \u00e9tica<\/h2>\n\n\n\n<p>Este modelo de IA fue entrenado utilizando datos sanitarios anonimizados bajo estrictas normas \u00e9ticas. Los participantes del UK Biobank dieron su consentimiento informado, y los datos daneses se accedieron de acuerdo con las regulaciones nacionales que exigen que los datos permanezcan dentro de Dinamarca. Los investigadores utilizaron sistemas virtuales seguros para analizar los datos sin moverlos a trav\u00e9s de fronteras. Estas medidas de seguridad ayudan a garantizar que los modelos de IA se desarrollen y utilicen de manera que respeten la privacidad y cumplan con los est\u00e1ndares \u00e9ticos.<\/p>\n\n\n\n<p><strong>Financiaci\u00f3n<\/strong><\/p>\n\n\n\n<p>Este trabajo fue financiado por las contribuciones de los Estados miembros del EMBL, fondos del DKFZ y una subvenci\u00f3n de la Fundaci\u00f3n Novo Nordisk.<\/p>\n\n\n\n<hr class=\"vf-divider\"\/>\n\n\n\n<h1 class=\"wp-block-heading\" id=\"French\">Un mod\u00e8le d\u2019IA pr\u00e9dit les risques de maladies plus d\u2019une d\u00e9cennie en avance<\/h1>\n\n\n\n<p><strong><em>Un nouveau mod\u00e8le d\u2019IA peut pr\u00e9dire votre sant\u00e9 plus d\u2019une d\u00e9cennie en avance, anticipant les risques de plus de 1 000 maladies<\/em><\/strong><\/p>\n\n\n\n<article class=\"vf-card vf-card--brand vf-card--bordered vf-u-margin__bottom--800\" default>\n  <div class=\"vf-card__content | vf-stack vf-stack--400\">\n      <h3 class=\"vf-card__heading\">\n      R\u00e9sum\u00e9    <\/h3>\n                <p class=\"vf-card__text\"><ul>\r\n \t<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Les chercheurs ont d\u00e9velopp\u00e9 un mod\u00e8le d\u2019IA qui pourrait estimer les risques de plus de 1 000 maladies\u00a0<\/span><\/li>\r\n \t<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Le mod\u00e8le, entra\u00een\u00e9 et test\u00e9 sur des donn\u00e9es m\u00e9dicales anonymis\u00e9es provenant du Royaume-Uni et du Danemark, peut pr\u00e9dire des r\u00e9sultats en mati\u00e8re de sant\u00e9 plus d\u2019une dizaine d\u2019ann\u00e9e en avance<\/span><\/li>\r\n \t<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Bien qu\u2019il ne soit pas encore pr\u00eat pour une utilisation clinique directe, le mod\u00e8le offre de nouvelles fa\u00e7ons d\u2019\u00e9tudier les maladies et de guider les strat\u00e9gies de soins de sant\u00e9.\u00a0<\/span><\/li>\r\n<\/ul><\/p>\n      <\/div>\n<\/article>\n\n\n\n\n<p>Imaginez un futur o\u00f9 votre historique de sant\u00e9 pourrait vous aider \u00e0 pr\u00e9dire les possibles probl\u00e8mes de sant\u00e9 que vous pourrez rencontrer les vingt prochaines ann\u00e9es. Des chercheurs ont d\u00e9velopp\u00e9 un nouveau mod\u00e8le d\u2019IA g\u00e9n\u00e9rative qui utilise des donn\u00e9es m\u00e9dicales \u00e0 grande \u00e9chelle pour estimer comment la sant\u00e9 de l\u2019Homme peut \u00e9voluer dans le temps. Il peut pr\u00e9voir le risque et le temps de survenue de plus de 1 000 maladies et pr\u00e9dire les r\u00e9sultats en mati\u00e8re de sant\u00e9 plus d&#8217;une d\u00e9cennie \u00e0 l&#8217;avance.<\/p>\n\n\n\n<p>Ce nouveau mod\u00e8le a \u00e9t\u00e9 fait sur-mesure utilisant les m\u00eames principes algorithmiques que ceux utilis\u00e9s pour les Grands mod\u00e8les de langage<em> <\/em>(abr\u00e9g\u00e9 <em>LLMs <\/em>de l\u2019anglais <em>Large Language Models<\/em>). Ce dernier fut entra\u00een\u00e9 sur des donn\u00e9es anonymis\u00e9es de plus de 4000 000 patients volontaires provenant de la <a href=\"https:\/\/www.ukbiobank.ac.uk\/\">UK Biobank<\/a>. Les chercheurs ont aussi r\u00e9ussi \u00e0 tester avec succ\u00e8s le mod\u00e8le sur plus d\u20191.9 millions de patients inscrits au <a href=\"https:\/\/ncrr.au.dk\/danish-registers\/the-national-patient-register\">Registre National Danois des Patients<\/a>. Cette approche, l&#8217;une des plus compl\u00e8tes \u00e0 ce jour, d\u00e9montre la capacit\u00e9 d&#8217;un mod\u00e8le d&#8217;IA \u00e0 pr\u00e9dire la progression des maladies \u00e0 grande \u00e9chelle et sur le long terme. Elle a de plus \u00e9t\u00e9 test\u00e9e avec succ\u00e8s sur deux syst\u00e8mes de sant\u00e9 diff\u00e9rents.<\/p>\n\n\n\n<p>\u201cNotre mod\u00e8le d\u2019IA repr\u00e9sente une preuve de concept, d\u00e9montrant qu\u2019il est possible pour une IA de mieux comprendre les tendances concernant notre sant\u00e9 et utiliser cette information pour r\u00e9aliser des pr\u00e9dictions qui ont du sens\u201d a partag\u00e9 <a href=\"https:\/\/www.embl.org\/\">Ewan Birney, Directeur G\u00e9n\u00e9ral par int\u00e9rim au Laboratoire Europ\u00e9en de Biologie Mol\u00e9culaire (EMBL)<\/a>. \u201cEn mod\u00e9lisant l\u2019\u00e9volution des maladies dans le temps, nous pouvons commencer \u00e0 explorer l&#8217;\u00e9mergence de certains risques et comment&nbsp; mieux pr\u00e9parer des interventions pr\u00e9liminaires. C\u2019est un grand pas vers une approche plus pr\u00e9ventive et personnalis\u00e9e des soins donn\u00e9s aux patients.&nbsp;<\/p>\n\n\n\n<p>Ce projet, publi\u00e9 dans le journal Nature, est le fruit d\u2019une collaboration entre <a href=\"https:\/\/www.embl.org\/\">L\u2019EMBL<\/a>, le <a href=\"https:\/\/www.dkfz.de\/en\/\">Centre Allemand de recherche sur le Cancer (DKFZ)<\/a> et <a href=\"https:\/\/www.ku.dk\/en\">l\u2019Universit\u00e9 de Copenhague.&nbsp;<\/a><\/p>\n\n\n\n<h2 class=\"wp-block-heading\">L\u2019IA pour les pr\u00e9visions en sant\u00e9<\/h2>\n\n\n\n<p>Tout comme les grands mod\u00e8les de langage peuvent apprendre la structure des phrases, ce mod\u00e8le d\u2019IA apprend la \u201cgrammaire\u201d des donn\u00e9es de sant\u00e9 afin de mod\u00e9liser les ant\u00e9c\u00e9dents m\u00e9dicaux sous forme de s\u00e9quences d&#8217;\u00e9v\u00e9nements se d\u00e9roulant dans le temps. Ces \u00e9v\u00e9nements incluent des diagnostics m\u00e9dicaux ou encore des facteurs li\u00e9s au mode de vie comme le tabagisme. Le mod\u00e8le apprend \u00e0 pr\u00e9voir le risque de maladie en se basant sur l&#8217;ordre dans lequel ces \u00e9v\u00e9nements se produisent et du temps qui s&#8217;\u00e9coule entre eux.<\/p>\n\n\n\n<p>\u201cDes changements li\u00e9s \u00e0 la sant\u00e9 suivent souvent des tendances pr\u00e9visibles\u201d ajoute <a href=\"https:\/\/www.ebi.ac.uk\/people\/person\/tomas-fitzgerald\/\">Tom Fitzgerald, Scientifique \u00e0 l\u2019Institut Europ\u00e9en de Bioinformatique de l\u2019EMBL (EMBL-EBI)<\/a>. \u201cNotre mod\u00e8le d\u2019IA apprend ces tendances et peut pr\u00e9voir les r\u00e9sultats futurs en mati\u00e8re de sant\u00e9. Il nous permet d&#8217;explorer ce qui pourrait arriver en fonction des ant\u00e9c\u00e9dents m\u00e9dicaux d&#8217;une personne et d&#8217;autres facteurs cl\u00e9s. Il est important de noter qu&#8217;il ne s&#8217;agit pas d&#8217;une certitude, mais d&#8217;une estimation des risques potentiels. \u00bb<\/p>\n\n\n\n<p>Le mod\u00e8le est particuli\u00e8rement efficace lorsque les sch\u00e9mas d\u2019\u00e9volutions sont clairs et coh\u00e9rents, comme pour certains types de cancer, les crises cardiaques, et les cas de septic\u00e9mie, qui sont un type d&#8217;empoisonnement du sang. Cependant, le mod\u00e8le l\u2019est moins pour des maladies et \u00e9v\u00e9nements de la vie qui repr\u00e9sentent une plus grande variabilit\u00e9, comme les troubles psychologiques et complications li\u00e9es \u00e0 la grossesse qui d\u00e9pendent d&#8217;\u00e9v\u00e9nements de vie plus impr\u00e9visibles.&nbsp;<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Usages futurs et limites<\/h2>\n\n\n\n<p>Tout comme les pr\u00e9visions m\u00e9t\u00e9o, ce nouveau mod\u00e8le nous donne des probabilit\u00e9s et non des certitudes. Il ne pr\u00e9dit pas avec exactitude ce qui pourrait arriver \u00e0 une individu, n\u00e9anmoins, il offre des estimations de la probabilit\u00e9 que certaines situations se produisent au cours d&#8217;une p\u00e9riode donn\u00e9e. Par exemple, il pourrait pr\u00e9dire les chances de d\u00e9velopper une maladie cardiaque dans l\u2019ann\u00e9e qui suit. Ces risques sont exprim\u00e9s comme tendances dans le temps, similaires \u00e0 une pr\u00e9vision de 70% de chance de pluie le jour suivant. En g\u00e9n\u00e9ral, les pr\u00e9visions \u00e0 court terme sont plus pr\u00e9cises que celles \u00e0 long terme.<\/p>\n\n\n\n<p>Prenons un autre exemple, le mod\u00e8le pr\u00e9dit diff\u00e9rents niveaux de risque de crise cardiaque. Si l&#8217;on prend la cohorte UK BioBank \u00e2g\u00e9e de 60 \u00e0 65 ans, le risque de crise cardiaque varie de 4 sur 10 000 par an pour certains hommes \u00e0 environ 1 sur 100 pour d&#8217;autres, en fonction de leurs diagnostics ant\u00e9rieurs et de leur mode de vie.Les femmes pr\u00e9sentent un risque moyen plus faible, mais une r\u00e9partition similaire du risque. De plus, les risques augmentent en moyenne avec l&#8217;\u00e2ge. Une \u00e9valuation syst\u00e9matique des donn\u00e9es de la UK Biobank non utilis\u00e9es lors de la phase d\u2019entra\u00eenement a montr\u00e9 que ces risques calcul\u00e9s correspondent bien au nombre de cas observ\u00e9s dans les diff\u00e9rents groupes d&#8217;\u00e2ge et de sexe.<\/p>\n\n\n\n<p>Le mod\u00e8le est calibr\u00e9 pour produire des estimations pr\u00e9cises du risque au niveau de la population. Cependant, comme n\u2019importe quel mod\u00e8le d\u2019IA, il a ses limites. Par exemple, puisque les donn\u00e9es d\u2019entra\u00eenement de la UK Biobank proviennent d&#8217;individus \u00e2g\u00e9s entre 40 et 60 ans, les situations li\u00e9es \u00e0 l\u2019enfance et l\u2019adolescence sont sous-repr\u00e9sent\u00e9es. Le mod\u00e8le poss\u00e8de aussi des biais d&#8217;\u00e9chantillonnages d\u00fbs \u00e0 des groupes d\u2019individus sous-repr\u00e9sent\u00e9s.&nbsp;<\/p>\n\n\n\n<p>M\u00eame si le mod\u00e8le n\u2019est pas pr\u00eat pour les essais cliniques, il pourrait d\u00e9j\u00e0 aider les chercheurs \u00e0:&nbsp;<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>mieux comprendre comment les maladies se d\u00e9veloppent et progressent dans le temps,<\/li>\n\n\n\n<li>Explorer comment le mode de vie et les maladies pr\u00e9c\u00e9dentes peuvent avoir un effets sur les risques de maladie longue dur\u00e9e,<\/li>\n\n\n\n<li>simuler les r\u00e9sultats cliniques \u00e0 l&#8217;aide de donn\u00e9es de patients artificiels, dans les situations o\u00f9 il est difficile d&#8217;obtenir ou d&#8217;acc\u00e9der \u00e0 des donn\u00e9es r\u00e9elles.<\/li>\n<\/ul>\n\n\n\n<p>Dans le futur, des outils IA similaires entra\u00een\u00e9s sur des bases de donn\u00e9es plus repr\u00e9sentatives pourront assister les cliniciens dans l\u2019identification pr\u00e9coce des patients \u00e0 haut risque. Avec le vieillissement de la population et l&#8217;augmentation des taux de maladies chroniques, la capacit\u00e9 \u00e0 pr\u00e9voir les besoins futurs en mati\u00e8re de sant\u00e9 pourrait aider les syst\u00e8mes de sant\u00e9 \u00e0 mieux planifier et \u00e0 allouer plus efficacement les ressources. Mais de nombreux tests, consultations et cadres r\u00e9glementaires solides sont encore n\u00e9cessaires avant que les mod\u00e8les d&#8217;IA puissent \u00eatre d\u00e9ploy\u00e9s dans un contexte clinique.<\/p>\n\n\n\n<p>\u00ab C&#8217;est le d\u00e9but d&#8217;une nouvelle mani\u00e8re de comprendre la sant\u00e9 humaine et la progression des maladies \u00bb, a d\u00e9clar\u00e9 <a href=\"https:\/\/www.dkfz.de\/en\/employees\/moritz-gerstung\">Moritz Gerstung, chef de la division IA en oncologie au DKFZ<\/a> et ancien chef de groupe \u00e0 l&#8217;EMBL-EBI. \u00ab Les mod\u00e8les g\u00e9n\u00e9ratifs tels que le n\u00f4tre pourraient un jour contribuer \u00e0 personnaliser les soins et \u00e0 anticiper les besoins en mati\u00e8re de sant\u00e9 \u00e0 grande \u00e9chelle. En s&#8217;appuyant sur des populations importantes, ces mod\u00e8les offrent un aper\u00e7u pr\u00e9cieux de l&#8217;\u00e9volution des maladies et pourraient \u00e0 terme permettre des interventions plus pr\u00e9coces et mieux adapt\u00e9es. \u00bb<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Confidentialit\u00e9 des donn\u00e9es et \u00e9thique<\/h2>\n\n\n\n<p>Ce mod\u00e8le d\u2019IA a \u00e9t\u00e9 entra\u00een\u00e9 \u00e0 partir de donn\u00e9es de sant\u00e9 anonymis\u00e9es, dans le respect des r\u00e8gles \u00e9thiques strictes. Les participants de la UK Biobank ont donn\u00e9 leur consentement \u00e9clair\u00e9, et les donn\u00e9es danoises ont \u00e9t\u00e9 consult\u00e9es conform\u00e9ment \u00e0 la r\u00e9glementation nationale qui exige que les donn\u00e9es restent au Danemark. Les chercheurs ont utilis\u00e9 des syst\u00e8mes virtuels s\u00e9curis\u00e9s pour analyser les donn\u00e9es sans les transf\u00e9rer \u00e0 l\u2019\u00e9tranger. Ces mesures de protection permettent de garantir que les mod\u00e8les d\u2019IA sont d\u00e9velopp\u00e9s et utilis\u00e9s dans le respect de la vie priv\u00e9e et des normes \u00e9thiques.&nbsp;<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Financements<\/h2>\n\n\n\n<p>Ce projet a \u00e9t\u00e9 financ\u00e9 par les \u00c9tats membres de l\u2019EMBL, les fonds DKFZ et une subvention de la fondation Novo Nordisk.<\/p>\n\n\n\n<hr class=\"vf-divider\"\/>\n\n\n\n<h1 class=\"wp-block-heading\" id=\"Italian\">Un modello di IA prevede il rischio di malattia con decenni di anticipo<\/h1>\n\n\n\n<p><em><strong>Un nuovo modello di intelligenza artificiale \u00e8 in grado di stimare il rischio a lungo termine per oltre 1.000 condizioni mediche e prevedere i cambiamenti nella salute umana con oltre un decennio di anticipo<\/strong><\/em><\/p>\n\n\n\n<article class=\"vf-card vf-card--brand vf-card--bordered vf-u-margin__bottom--800\" default>\n  <div class=\"vf-card__content | vf-stack vf-stack--400\">\n      <h3 class=\"vf-card__heading\">\n      Sintesi    <\/h3>\n                <p class=\"vf-card__text\"><ul>\r\n \t<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Il nuovo modello di IA \u00e8 capace di prevedere il rischio di insorgenza di pi\u00f9 di\u00a0 1.000 malattie<\/span><\/li>\r\n \t<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Addestrato e testato su dati clinici anonimizzati provenienti dal Regno Unito e dalla Danimarca, il modello \u00e8 in grado di formulare previsioni sanitarie su un orizzonte temporale di oltre dieci anni\u00a0<\/span><\/li>\r\n \t<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Sebbene non sia ancora pronto per l&#8217;applicazione clinica diretta, il modello di IA rappresenta un&#8217;importante innovazione nella ricerca medica, offrendo nuove opportunit\u00e0 per comprendere le malattie e definire strategie di prevenzione e intervento<\/span><\/li>\r\n<\/ul><\/p>\n      <\/div>\n<\/article>\n\n\n\n\n<p>Immaginate un futuro in cui la vostra storia clinica possa anticipare le sfide sanitarie che potreste affrontare nei prossimi vent&#8217;anni. Oggi quel futuro \u00e8 un po\u2019 pi\u00f9 vicino grazie ad un modello di intelligenza artificiale generativa che utilizza cartelle cliniche su larga scala per stimare come la salute di una persona potrebbe evolvere nel tempo. Il modello \u00e8 in grado di prevedere il rischio e la tempistica di oltre 1.000 malattie e di prevedere gli esiti di salute con pi\u00f9 di un decennio di anticipo.<\/p>\n\n\n\n<p>Basato su algoritmi simili a quelli utilizzati nei grandi modelli linguistici (LLM), il modello \u00e8 stato addestrato su dati anonimizzati relativi a 400.000 individui provenienti dalla <a href=\"https:\/\/www.ukbiobank.ac.uk\/\">UK Biobank<\/a>. La sua efficacia \u00e8 stata poi validata utilizzando i dati di 1,9 milioni di pazienti raccolti nel Registro Nazionale Danese dei Pazienti. Questo approccio rappresenta una delle applicazioni pi\u00f9 avanzate della IA generativa nella capacit\u00e0 di modellare la progressione delle malattie umane su larga scala ed \u00e8 stato testato su dati provenienti da due sistemi sanitari completamente separati.&nbsp;<\/p>\n\n\n\n<p>\u201cIl nostro modello di intelligenza artificiale \u00e8 una prova di concetto che dimostra come sia possibile apprendere molti dei modelli di salute a lungo termine e utilizzarli per generare previsioni significative\u201d, ha dichiarato <a href=\"https:\/\/www.ebi.ac.uk\/people\/person\/ewan-birney\/\">Ewan Birney, direttore esecutivo ad interim del Laboratorio Europeo di Biologia Molecolare (EMBL)<\/a>. \u201cModellando lo sviluppo delle malattie nel tempo, possiamo iniziare a comprendere quando emergono determinati rischi e come pianificare al meglio gli interventi precoci. Si tratta di un grande passo avanti verso un\u2019assistenza sanitaria pi\u00f9 personalizzata e preventiva\u201d.<\/p>\n\n\n\n<p>Lo studio, pubblicato sulla rivista Nature, \u00e8 frutto della collaborazione tra l&#8217;<a href=\"https:\/\/www.embl.org\/\">EMBL<\/a>, il <a href=\"https:\/\/www.dkfz.de\/en\/\">Centro tedesco di ricerca sul cancro (DKFZ)<\/a> e l&#8217;<a href=\"https:\/\/www.ku.dk\/en\">Universit\u00e0 di Copenaghen<\/a>.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">L&#8217;IA al servizio delle previsioni sanitarie&nbsp;<\/h2>\n\n\n\n<p>Proprio come i modelli linguistici di grandi dimensioni sono in grado di apprendere la struttura delle frasi, questo modello di IA apprende la \u201cgrammatica\u201d dei dati sanitari. Analizza le storie cliniche come sequenze temporali di eventi \u2014 come diagnosi mediche o fattori legati allo stile di vita, come il fumo \u2014 e impara a prevedere il rischio di malattia sulla base dell&#8217;ordine e dell\u2019intervallo con cui questi eventi si verificano.&nbsp;<\/p>\n\n\n\n<p>\u201cGli eventi medici seguono spesso schemi prevedibili\u201d, spiega <a href=\"https:\/\/www.ebi.ac.uk\/people\/person\/tomas-fitzgerald\/\">Tom Fitzgerald, ricercatore presso l&#8217;Istituto europeo di bioinformatica dell&#8217;EMBL (EMBL-EBI)<\/a>. \u201cIl nostro modello \u00e8 in grado di apprendere questi schemi e prevedere potenziali esiti futuri. Ci offre un nuovo modo per esplorare ci\u00f2 che potrebbe accadere sulla base della storia clinica di una persona e di altri fattori rilevanti. \u00c8 fondamentale sottolineare che non si tratta di una certezza, ma di una stima dei potenziali rischi\u201d.<\/p>\n\n\n\n<p>Il modello risulta particolarmente efficace per condizioni che presentano una progressione clinica ben definita, come alcuni tipi di cancro, infarti e setticemia (una grave infezione sistemica del&nbsp; sangue). Tuttavia, le sue previsioni sono meno affidabili per patologie pi\u00f9 variabili e influenzate da fattori esterni difficilmente prevedibili, come i disturbi mentali o le complicazioni legate alla gravidanza.&nbsp;<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Utilizzi futuri e limiti del modello<\/h2>\n\n\n\n<p>Come accade per le previsioni meteorologiche, questo nuovo modello di IA fornisce stime probabilistiche, non certezze. Non prevede esattamente cosa accadr\u00e0 a un individuo, ma offre stime ben calibrate della probabilit\u00e0 che determinate condizioni si verifichino in un dato periodo. Ad esempio, pu\u00f2 stimare la probabilit\u00e0 di sviluppare una malattia cardiaca entro l\u2019anno successivo, esprimendo questo rischio come una percentuale \u2014 proprio come si prevede una probabilit\u00e0 del 70% di pioggia per il giorno seguente.<\/p>\n\n\n\n<p>Ad esempio, il modello prevede livelli variabili di rischio di infarto. Prendendo in considerazione il gruppo della UK Biobank nella fascia di et\u00e0 tra i 60 e i 65 anni, il rischio di infarto varia da una probabilit\u00e0 di 4 su 10.000 all&#8217;anno per alcuni uomini fino a circa 1 su 100 per altri uomini, a seconda delle diagnosi precedenti e dello stile di vita. Le donne presentano in media un rischio inferiore, ma con una distribuzione del rischio simile. Inoltre, i rischi aumentano in media con l\u2019avanzare dell\u2019et\u00e0. Una valutazione sistematica dei dati della UK Biobank non utilizzati per l\u2019addestramento ha mostrato che questi rischi calcolati corrispondono bene al numero osservato di casi nei diversi gruppi di et\u00e0 e sesso.<\/p>\n\n\n\n<p>Il modello \u00e8 calibrato per produrre stime accurate del rischio a livello di popolazione, prevedendo la frequenza con cui determinate condizioni si verificano all&#8217;interno di gruppi di persone. Tuttavia, come qualsiasi modello di IA, presenta dei limiti. Ad esempio, poich\u00e9 i dati di addestramento del modello provenienti dalla UK Biobank provengono principalmente da individui di et\u00e0 compresa tra i 40 e i 60 anni, gli eventi sanitari relativi all&#8217;infanzia e all&#8217;adolescenza sono sottorappresentati. Il modello contiene anche distorsioni demografiche dovute a lacune nei dati di addestramento, tra cui la sottorappresentazione di alcuni gruppi etnici.<\/p>\n\n\n\n<p>Sebbene il modello non sia ancora pronto per l&#8217;uso clinico, potrebbe gi\u00e0 aiutare i ricercatori a:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Comprendere come le malattie si sviluppano e progrediscono nel tempo.<\/li>\n\n\n\n<li>Esplorare come lo stile di vita e le malattie pregresse influenzano il rischio di malattie a lungo termine.<\/li>\n\n\n\n<li>Simulare gli esiti sanitari utilizzando dati artificiali sui pazienti, in situazioni in cui \u00e8 difficile ottenere o accedere a dati reali.<\/li>\n<\/ul>\n\n\n\n<p>In futuro, strumenti di IA come questo, addestrati su set di dati pi\u00f9 ampi e rappresentativi, potrebbero aiutare i medici a identificare precocemente i pazienti ad alto rischio. Con l&#8217;invecchiamento della popolazione e l&#8217;aumento dei tassi di malattie croniche, la capacit\u00e0 di anticipare le esigenze sanitarie potrebbe aiutare i sistemi sanitari a pianificare meglio i servizi e ad allocare risorse in modo pi\u00f9 efficiente. Tuttavia, prima che i modelli di IA possano essere implementati in ambito clinico, sono necessari ulteriori studi di validazione, consultazioni interdisciplinari e un solido quadro normativo.<\/p>\n\n\n\n<p>\u201cQuesto \u00e8 solo l&#8217;inizio di un nuovo modo di comprendere la salute umana e la progressione delle malattie\u201d, ha dichiarato <a href=\"https:\/\/www.dkfz.de\/en\/employees\/moritz-gerstung\">Moritz Gerstung, capo della divisione di IA in oncologia presso il DKFZ<\/a> ed ex capo gruppo presso l&#8217;EMBL-EBI. \u201cModelli generativi come il nostro potrebbero un giorno aiutare a personalizzare le cure e anticipare le esigenze sanitarie su larga scala. Imparando da grandi popolazioni, questi modelli offrono una potente lente d&#8217;ingrandimento sui meccanismi di sviluppo delle malattie, aprendo la strada ad interventi pi\u00f9 tempestivi e personalizzati\u201d.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Privacy dei dati ed etica<\/h2>\n\n\n\n<p>Questo modello di IA \u00e8 stato addestrato utilizzando dati sanitari anonimizzati, nel pieno rispetto di rigorosi standard etici e normativi. I partecipanti alla UK Biobank hanno fornito il loro consenso informato per l\u2019utilizzo dei propri dati a fini di ricerca. Per quanto riguarda i dati danesi, l\u2019accesso \u00e8 avvenuto in conformit\u00e0 con le normative nazionali che richiedono che i dati rimangano all&#8217;interno del territorio danese. I ricercatori hanno utilizzato sistemi virtuali sicuri per analizzare i dati senza trasferirli oltre confine. Queste misure di sicurezza contribuiscono a garantire che i modelli di IA siano sviluppati e utilizzati nel rispetto della privacy e degli standard etici condivisi a livello internazionale.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Finanziamenti<\/h3>\n\n\n\n<p>Questo lavoro \u00e8 stato sostenuto dai contributi degli Stati membri dell&#8217;EMBL, dai fondi DKFZ e da una sovvenzione della Fondazione Novo Nordisk.&nbsp;<\/p>\n","protected":false},"excerpt":{"rendered":"<p>New AI model can estimate the long-term risk of over 1,000 diseases and forecast human health changes over a decade in advance.<\/p>\n","protected":false},"author":77,"featured_media":76225,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[11060,17591],"tags":[4718,28,36,4760,1748,539],"embl_taxonomy":[2906,18907],"class_list":["post-76223","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-research-highlights","category-science-technology","tag-artificial-intelligence","tag-bioinformatics","tag-embl-ebi","tag-precision-medicine","tag-press-release","tag-research-highlight","embl_taxonomy-embl-ebi","embl_taxonomy-ewan-birney"],"acf":{"vfwp-news_embl_taxonomy":[2906,18907],"featured":true,"show_featured_image":false,"field_target_display":"both","field_article_language":{"value":"english","label":"English"},"article_intro":"<p>New AI model can estimate the long-term risk of over 1,000 diseases and forecast human health changes over a decade in advance<\/p>\n","related_links":false,"source_article":[{"publication_title":"Learning the natural history of human disease with generative transformers","publication_link":{"title":"","url":"https:\/\/www.nature.com\/articles\/s41586-025-09529-3","target":""},"publication_authors":"Shmatko A., et al","publication_source":"Nature","publication_date":"17 September 2025","publication_doi":"10.1038\/s41586-025-09529-3"}],"in_this_article":false,"press_contact":"EMBL-EBI Generic","article_translations":[{"translation_language":"Deutsch","translation_anchor":"#German"},{"translation_language":"Espa\u00f1ol","translation_anchor":"#Spanish"},{"translation_language":"Fran\u00e7ais","translation_anchor":"#French"},{"translation_language":"Italiano","translation_anchor":"#Italian"}],"languages":""},"embl_taxonomy_terms":[{"uuid":"a:3:{i:0;s:36:\"b14d3f13-5670-44fb-8970-e54dfd9c921a\";i:1;s:36:\"89e00fee-87f4-482e-a801-4c3548bb6a58\";i:2;s:36:\"a99d1a7c-ca83-4c00-ab61-d082d3e41ce3\";}","parents":[],"name":["EMBL-EBI"],"slug":"embl-ebi","description":"Where &gt; All EMBL sites &gt; EMBL-EBI"},{"uuid":"a:2:{i:0;s:36:\"4428d1fd-441a-4d6d-a1c5-5dcf5665f213\";i:1;s:36:\"8ea61d5f-8e18-4875-b043-3ace7c26c561\";}","parents":[],"name":["Ewan Birney"],"slug":"ewan-birney","description":"Who &gt; Ewan Birney"}],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v26.2 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>AI model forecasts disease risk decades in advance | EMBL<\/title>\n<meta name=\"description\" content=\"New AI model can estimate the long-term risk of over 1,000 diseases and forecast human health changes over a decade in advance.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"AI model forecasts disease risk decades in advance | EMBL\" \/>\n<meta property=\"og:description\" content=\"New AI model can estimate the long-term risk of over 1,000 diseases and forecast human health changes over a decade in advance.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/\" \/>\n<meta property=\"og:site_name\" content=\"EMBL\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/embl.org\/\" \/>\n<meta property=\"article:published_time\" content=\"2025-09-17T15:00:08+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2025-09-17T15:00:11+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.embl.org\/news\/wp-content\/uploads\/2025\/09\/2025-BIRNEY-Delphi-2M-1000x600-1.png\" \/>\n\t<meta property=\"og:image:width\" content=\"1000\" \/>\n\t<meta property=\"og:image:height\" content=\"600\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"Vicky Hatch\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@embl\" \/>\n<meta name=\"twitter:site\" content=\"@embl\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Vicky Hatch\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"26 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"NewsArticle\",\"@id\":\"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/#article\",\"isPartOf\":{\"@id\":\"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/\"},\"author\":{\"name\":\"Vicky Hatch\",\"@id\":\"https:\/\/www.embl.org\/news\/#\/schema\/person\/d8477ba2d7a6164b141a3872a25ee982\"},\"headline\":\"AI model forecasts disease risk decades in advance\",\"datePublished\":\"2025-09-17T15:00:08+00:00\",\"dateModified\":\"2025-09-17T15:00:11+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/\"},\"wordCount\":6096,\"publisher\":{\"@id\":\"https:\/\/www.embl.org\/news\/#organization\"},\"image\":{\"@id\":\"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/www.embl.org\/news\/wp-content\/uploads\/2025\/09\/2025-BIRNEY-Delphi-2M-1000x600-1.png\",\"keywords\":[\"artificial intelligence\",\"bioinformatics\",\"embl-ebi\",\"precision medicine\",\"press release\",\"research highlight\"],\"articleSection\":[\"Research highlights\",\"Science &amp; Technology\"],\"inLanguage\":\"en-US\"},{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/\",\"url\":\"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/\",\"name\":\"AI model forecasts disease risk decades in advance | EMBL\",\"isPartOf\":{\"@id\":\"https:\/\/www.embl.org\/news\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/#primaryimage\"},\"image\":{\"@id\":\"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/www.embl.org\/news\/wp-content\/uploads\/2025\/09\/2025-BIRNEY-Delphi-2M-1000x600-1.png\",\"datePublished\":\"2025-09-17T15:00:08+00:00\",\"dateModified\":\"2025-09-17T15:00:11+00:00\",\"description\":\"New AI model can estimate the long-term risk of over 1,000 diseases and forecast human health changes over a decade in advance.\",\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/#primaryimage\",\"url\":\"https:\/\/www.embl.org\/news\/wp-content\/uploads\/2025\/09\/2025-BIRNEY-Delphi-2M-1000x600-1.png\",\"contentUrl\":\"https:\/\/www.embl.org\/news\/wp-content\/uploads\/2025\/09\/2025-BIRNEY-Delphi-2M-1000x600-1.png\",\"width\":1000,\"height\":600,\"caption\":\"AI model forecasts disease risk decades in advance. Karen Arnott\/EMBL-EBI\"},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.embl.org\/news\/#website\",\"url\":\"https:\/\/www.embl.org\/news\/\",\"name\":\"European Molecular Biology Laboratory News\",\"description\":\"News from the European Molecular Biology Laboratory\",\"publisher\":{\"@id\":\"https:\/\/www.embl.org\/news\/#organization\"},\"alternateName\":\"EMBL News\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.embl.org\/news\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\/\/www.embl.org\/news\/#organization\",\"name\":\"European Molecular Biology Laboratory\",\"alternateName\":\"EMBL\",\"url\":\"https:\/\/www.embl.org\/news\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/www.embl.org\/news\/#\/schema\/logo\/image\/\",\"url\":\"https:\/\/www.embl.org\/news\/wp-content\/uploads\/2025\/09\/EMBL_logo_colour-1-300x144-1.png\",\"contentUrl\":\"https:\/\/www.embl.org\/news\/wp-content\/uploads\/2025\/09\/EMBL_logo_colour-1-300x144-1.png\",\"width\":300,\"height\":144,\"caption\":\"European Molecular Biology Laboratory\"},\"image\":{\"@id\":\"https:\/\/www.embl.org\/news\/#\/schema\/logo\/image\/\"},\"sameAs\":[\"https:\/\/www.facebook.com\/embl.org\/\",\"https:\/\/x.com\/embl\",\"https:\/\/www.instagram.com\/embl_org\/\",\"https:\/\/www.linkedin.com\/company\/15813\/\",\"https:\/\/www.youtube.com\/user\/emblmedia\/\"]},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.embl.org\/news\/#\/schema\/person\/d8477ba2d7a6164b141a3872a25ee982\",\"name\":\"Vicky Hatch\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/www.embl.org\/news\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/secure.gravatar.com\/avatar\/6d864d58088d9e60f42c501aa30714d303efc1ca5aed268210905409910b90d5?s=96&d=mm&r=g\",\"contentUrl\":\"https:\/\/secure.gravatar.com\/avatar\/6d864d58088d9e60f42c501aa30714d303efc1ca5aed268210905409910b90d5?s=96&d=mm&r=g\",\"caption\":\"Vicky Hatch\"},\"url\":\"https:\/\/www.embl.org\/news\/author\/vicky-hatch-2-2-2-2-2-2-2-2-2-2-2-2-2-2-2-2-2-2--2\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"AI model forecasts disease risk decades in advance | EMBL","description":"New AI model can estimate the long-term risk of over 1,000 diseases and forecast human health changes over a decade in advance.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/","og_locale":"en_US","og_type":"article","og_title":"AI model forecasts disease risk decades in advance | EMBL","og_description":"New AI model can estimate the long-term risk of over 1,000 diseases and forecast human health changes over a decade in advance.","og_url":"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/","og_site_name":"EMBL","article_publisher":"https:\/\/www.facebook.com\/embl.org\/","article_published_time":"2025-09-17T15:00:08+00:00","article_modified_time":"2025-09-17T15:00:11+00:00","og_image":[{"width":1000,"height":600,"url":"https:\/\/www.embl.org\/news\/wp-content\/uploads\/2025\/09\/2025-BIRNEY-Delphi-2M-1000x600-1.png","type":"image\/png"}],"author":"Vicky Hatch","twitter_card":"summary_large_image","twitter_creator":"@embl","twitter_site":"@embl","twitter_misc":{"Written by":"Vicky Hatch","Est. reading time":"26 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"NewsArticle","@id":"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/#article","isPartOf":{"@id":"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/"},"author":{"name":"Vicky Hatch","@id":"https:\/\/www.embl.org\/news\/#\/schema\/person\/d8477ba2d7a6164b141a3872a25ee982"},"headline":"AI model forecasts disease risk decades in advance","datePublished":"2025-09-17T15:00:08+00:00","dateModified":"2025-09-17T15:00:11+00:00","mainEntityOfPage":{"@id":"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/"},"wordCount":6096,"publisher":{"@id":"https:\/\/www.embl.org\/news\/#organization"},"image":{"@id":"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/#primaryimage"},"thumbnailUrl":"https:\/\/www.embl.org\/news\/wp-content\/uploads\/2025\/09\/2025-BIRNEY-Delphi-2M-1000x600-1.png","keywords":["artificial intelligence","bioinformatics","embl-ebi","precision medicine","press release","research highlight"],"articleSection":["Research highlights","Science &amp; Technology"],"inLanguage":"en-US"},{"@type":"WebPage","@id":"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/","url":"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/","name":"AI model forecasts disease risk decades in advance | EMBL","isPartOf":{"@id":"https:\/\/www.embl.org\/news\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/#primaryimage"},"image":{"@id":"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/#primaryimage"},"thumbnailUrl":"https:\/\/www.embl.org\/news\/wp-content\/uploads\/2025\/09\/2025-BIRNEY-Delphi-2M-1000x600-1.png","datePublished":"2025-09-17T15:00:08+00:00","dateModified":"2025-09-17T15:00:11+00:00","description":"New AI model can estimate the long-term risk of over 1,000 diseases and forecast human health changes over a decade in advance.","inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.embl.org\/news\/science-technology\/ai-model-forecasts-disease-risk-decades-in-advance\/#primaryimage","url":"https:\/\/www.embl.org\/news\/wp-content\/uploads\/2025\/09\/2025-BIRNEY-Delphi-2M-1000x600-1.png","contentUrl":"https:\/\/www.embl.org\/news\/wp-content\/uploads\/2025\/09\/2025-BIRNEY-Delphi-2M-1000x600-1.png","width":1000,"height":600,"caption":"AI model forecasts disease risk decades in advance. Karen Arnott\/EMBL-EBI"},{"@type":"WebSite","@id":"https:\/\/www.embl.org\/news\/#website","url":"https:\/\/www.embl.org\/news\/","name":"European Molecular Biology Laboratory News","description":"News from the European Molecular Biology Laboratory","publisher":{"@id":"https:\/\/www.embl.org\/news\/#organization"},"alternateName":"EMBL News","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.embl.org\/news\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/www.embl.org\/news\/#organization","name":"European Molecular Biology Laboratory","alternateName":"EMBL","url":"https:\/\/www.embl.org\/news\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.embl.org\/news\/#\/schema\/logo\/image\/","url":"https:\/\/www.embl.org\/news\/wp-content\/uploads\/2025\/09\/EMBL_logo_colour-1-300x144-1.png","contentUrl":"https:\/\/www.embl.org\/news\/wp-content\/uploads\/2025\/09\/EMBL_logo_colour-1-300x144-1.png","width":300,"height":144,"caption":"European Molecular Biology Laboratory"},"image":{"@id":"https:\/\/www.embl.org\/news\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.facebook.com\/embl.org\/","https:\/\/x.com\/embl","https:\/\/www.instagram.com\/embl_org\/","https:\/\/www.linkedin.com\/company\/15813\/","https:\/\/www.youtube.com\/user\/emblmedia\/"]},{"@type":"Person","@id":"https:\/\/www.embl.org\/news\/#\/schema\/person\/d8477ba2d7a6164b141a3872a25ee982","name":"Vicky Hatch","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.embl.org\/news\/#\/schema\/person\/image\/","url":"https:\/\/secure.gravatar.com\/avatar\/6d864d58088d9e60f42c501aa30714d303efc1ca5aed268210905409910b90d5?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/6d864d58088d9e60f42c501aa30714d303efc1ca5aed268210905409910b90d5?s=96&d=mm&r=g","caption":"Vicky Hatch"},"url":"https:\/\/www.embl.org\/news\/author\/vicky-hatch-2-2-2-2-2-2-2-2-2-2-2-2-2-2-2-2-2-2--2\/"}]}},"field_target_display":"both","field_article_language":{"value":"english","label":"English"},"fimg_url":"https:\/\/www.embl.org\/news\/wp-content\/uploads\/2025\/09\/2025-BIRNEY-Delphi-2M-1000x600-1.png","featured_image_src":"https:\/\/www.embl.org\/news\/wp-content\/uploads\/2025\/09\/2025-BIRNEY-Delphi-2M-1000x600-1.png","_links":{"self":[{"href":"https:\/\/www.embl.org\/news\/wp-json\/wp\/v2\/posts\/76223","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.embl.org\/news\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.embl.org\/news\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.embl.org\/news\/wp-json\/wp\/v2\/users\/77"}],"replies":[{"embeddable":true,"href":"https:\/\/www.embl.org\/news\/wp-json\/wp\/v2\/comments?post=76223"}],"version-history":[{"count":10,"href":"https:\/\/www.embl.org\/news\/wp-json\/wp\/v2\/posts\/76223\/revisions"}],"predecessor-version":[{"id":77997,"href":"https:\/\/www.embl.org\/news\/wp-json\/wp\/v2\/posts\/76223\/revisions\/77997"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.embl.org\/news\/wp-json\/wp\/v2\/media\/76225"}],"wp:attachment":[{"href":"https:\/\/www.embl.org\/news\/wp-json\/wp\/v2\/media?parent=76223"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.embl.org\/news\/wp-json\/wp\/v2\/categories?post=76223"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.embl.org\/news\/wp-json\/wp\/v2\/tags?post=76223"},{"taxonomy":"embl_taxonomy","embeddable":true,"href":"https:\/\/www.embl.org\/news\/wp-json\/wp\/v2\/embl_taxonomy?post=76223"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}