AI Meets Frailty: Rethinking Hematology in Older Patients

Medscape

20 hours ago

MILAN — Conditions like leukemia and myelodysplastic syndromes become more prevalent with age, but older patients often present with multiple comorbidities, greater physical and cognitive frailty, and diminished tolerance to intensive therapies. This creates complex challenges for the hematologist.
Matteo Giovanni Della Porta, MD
'It is not enough to focus solely on the disease; the patient must be assessed as a whole, balancing treatment efficacy with quality of life,' Matteo Giovanni Della Porta, MD, head of the Leukemia Unit at Humanitas Cancer Center and professor of hematology at Humanitas University, Milan, Italy, told Medscape Medical News . 'This poses a significant clinical challenge that demands multidisciplinary expertise and increasingly personalized therapeutic strategies.'
Can artificial intelligence (AI) play a role in addressing this challenge? Experts discussed the pros and cons here on the opening day of the 2025 European Hematology Association (EHA 2025) Congress.
AI: Friend or Foe?
Polypharmacy is common in older patients, and it is simply not feasible for a single physician to know and recall all possible interactions between medications and the therapies required to treat a hematologic condition. 'Every clinical decision must be tailored to the patient's age, comorbidities, and pharmacologic complexity, where efficacy meets vulnerability,' said Torsten Haferlach, MD, co-founder of the MLL Munich Leukemia Laboratory in Germany, speaking at the session titled 'Aging and Hematology: Artificial Intelligence in Geriatric Hematology.'
AI can help address these limitations, provided that human oversight is maintained, he said.
Torsten Haferlach, MD
'AI can integrate heterogeneous clinical, functional, and social data to develop personalized risk profiles, predict treatment tolerability, and recommend more appropriate care pathways. In the future, AI-driven predictive models may support complex clinical decision-making, helping ensure that treatments for older patients are better balanced, more effective, and more sustainable,' Della Porta explained to Medscape Medical News .
Large language models are now being used to support automatic diagnosis, often generating clinically usable results. But Della Porta cautioned that most AI tools are not yet optimized for older populations. A recent scoping review of FDA-approved AI-based devices found that only 0.4% focused exclusively on geriatric health.
Reshaping the Doctor-Patient Relationship
Still, as AI increasingly takes over administrative tasks, such as note taking and report generation, it might give clinicians more time for meaningful patient interaction.
'AI does not replace physicians, but it can make their work more effective and help foster a closer connection with the patient, also through building trust,' said Esther Lueje, MD, a geriatrician at Fundación Jiménez Díaz University Hospital, Madrid, Spain. She explained that AI can free up time, reduce diagnostic uncertainty, and support more transparent decision-making.
However, Della Porta cautioned while speaking with Medscape Medical News , that '[i]f patients feel that technology is replacing empathy or meaningful dialogue, there is a risk of emotional detachment. The key point is that AI should serve the doctor–patient relationship, not hinder it.'
Lueje also warned that integrating large language models into daily practice comes with challenges: hallucinations, clinical errors, the need for medical oversight, and limited digital literacy among physicians.
'We have a stethoscope in one hand but no prompt — or the skill to write one — in the other,' Haferlach also noted.
Synthetic Patients, Virtual Trials
Another realm for AI integration is synthetic data, which could become as revolutionary as telecommunications and biotechnology. 'Data are the new oil,' said Alfonso Piciocchi, PhD, chief scientific officer and head of the Biostatistics Unit at Fondazione GIMEMA, Italy.
Alfonso Piciocchi, PhD
In his talk, Piciocchi explored the concept of synthetic patients: AI-generated models that mirror real patient populations, preserving key data correlations.
These are not mere simulations, as they preserve the same multivariate structure, correlations, and observed distributions as real-world data. They are primarily used to train AI algorithms while safeguarding patient privacy. 'The generation of synthetic patients is not without risks: if the initial database is not well-defined, the risk of failure can be significant," Piciocchi told Medscape Medical News .
However, current experiences indicate that these synthetic patient cohorts closely resemble actual human cohorts and can be effectively used to create control groups in so-called 'virtual' clinical trials or to enhance the representation of under-recruited populations, such as elderly patients, who are often difficult to enroll.
In addition to synthetic patients, there is increasing discussion around digital twins: virtual representations of real patients created by integrating biological, clinical, and environmental data. "They allow clinicians to simulate disease progression or predict therapeutic responses before initiating treatment, thereby enhancing personalization of care," Della Porta said, emphasizing that both technologies are already in use in experimental settings and are expected to become integral components of precision medicine in hematology in the coming years.