From Years of Uncertainty to Answers: How AI Is Shortening the Rare Disease Diagnostic Odyssey
Key Takeaways:
- ThinkRare, an algorithm at the Children’s Hospital of Eastern Ontario, scans health records to flag children who may need a rare disease referral. It has led to 21 new diagnoses and a 70% success rate so far.
- At Boston Children’s Hospital, a reasoning model built by OpenAI solved all 20 known test cases and contributed to 18 diagnoses among 376 previously unresolved ones.
- Neither tool works alone. Data bias, hallucinations, and clinicians’ tendency to defer to AI mean a human in the loop remains essential.
Why a rare disease diagnosis can take half a decade
For people living with a rare disease, the route to a diagnosis can be long, circuitous, and occasionally never completed at all. Reaching an answer often demands specialist training and technical resources that many clinics simply do not have. In a News and Perspectives article for JMIR, correspondent Simon Spichak reports on how AI initiatives at one children’s hospital in the United States and another in Canada are working to close that gap, and how they could fundamentally reshape the diagnostic experience for children and families living with rare conditions.
For some people, the benefits are already tangible. AI algorithms have helped their doctors reach a diagnosis, and that diagnosis has in turn opened the door to accessibility accommodations, treatments, clinical trials, and further research into their condition. If these tools can be scaled, they could meaningfully shorten what is often described as the diagnostic odyssey – the almost five years, on average, of appointments, medical visits, and frustration that it typically takes to diagnose a rare disease.
Finding the children who are lost in the system
CHEO is helping to lead the way with ThinkRare, an algorithm that scans electronic health records and flags children who may need a referral to a rare disease specialist for further testing.
“The purpose of ThinkRare is to find the patients that aren’t coming to us, that are bouncing around, that are lost in the system,” says Ivan Terekhov, BCom, director of research informatics, AI, and technology at CHEO Research Institute.
The approach targets the very beginning of the diagnostic journey, where children are cycling through appointments without anyone joining the dots. ThinkRare has already prompted genetic sequencing and follow-up in a handful of children, leading to 21 new rare disease diagnoses and a 70% success rate to date. The researchers behind the algorithm are now working on rolling the model out to other hospitals.
Tackling the cases that leave clinicians stuck
BCH is addressing the opposite end of the odyssey, using a reasoning model built by OpenAI to help resolve cases that have defeated clinicians.
Catherine Brownstein, PhD, scientific director of the genetic investigations arm of the Manton Center for Orphan Disease Research at BCH, says she was sceptical at first. When the team fed the reasoning model rare disease cases for which the diagnosis was already known, it initially appeared to falter. Eventually it diagnosed 19 out of 20 cases correctly, and ultimately all 20.
“When we looked at the one that it got wrong,” she says, “we were wrong.” The AI had flagged a second diagnosis that clinicians had missed.
At that point, Brownstein and her team judged the model ready for the real test: helping to solve cases that remained undiagnosed. It proved remarkably successful, with its output leading to 18 diagnoses among 376 previously unresolved cases.
What does AI do that clinicians cannot?
According to a recent report, insufficient physician training may help to explain why people with rare diseases go undiagnosed for so long. ThinkRare addresses this gap by incorporating expert-curated criteria to flag rare diseases that the average doctor might miss, so that those children can be considered for genetic testing.
Even after sequencing, rare disease specialists can take a long time to arrive at a diagnosis, and complicated cases sometimes remain unsolved. Specialists have only a limited amount of time to spend with each family, and may be carrying many cases they are simply unable to revisit.
During her PhD, Brownstein spent six years working on a single family. “There’s just simply not enough geneticists in this world or scientists in this world to do the thorough, in-depth investigations that are necessary to make all the insights possible,” she says.
The reasoning model used at BCH, designed to reason carefully and show its work, was developed using o3 Deep Research, a publicly accessible model. When it identifies a potential genetic mutation that might be causing a child’s symptoms, it presents its chain of reasoning alongside the suggestion. Geneticists then take over, running the tests needed to confirm or rule out the diagnosis.
“But instead of having to go to the sequence and find that variant yourself, it’s presented to you in a way that’s much, much quicker,” says Brownstein.
The insights the model produces may also lead to discoveries that clinicians would not otherwise make. In one instance it surfaced a new genetic variant that might be implicated in vitiligo, an autoimmune condition in which the body attacks the pigment-producing cells in the skin.
“I was speechless,” Brownstein recalls. Her team now has research under way to investigate whether the mutation is causative. If that work bears out, she hopes it could point towards a new treatment.
As tools like these move closer to everyday practice, the demand for clinicians who understand both their promise and their failure modes is rising. Short, focused CPD training such as the College of Contemporary Health’s AI Essentials for Primary Care course is designed for exactly that: helping practitioners get to grips with how these systems work, where they add value, and where human judgement has to stay firmly in charge.
Why there needs to be a human in the loop
Neither model removes the clinician from the process.
ThinkRare’s algorithm is passive, flagging a small number of cases out of several hundred thousand children attending its clinics. Those children still need testing and assessment by a specialist, because the algorithm does not actually predict which condition a child may have.
And because the algorithm is built on clinical expertise and trained on electronic medical records, bias is baked in. “I think we’d be naive to say that there’s not data quality issues or there’s worse data quality for those who are not White and those who do come from rural locations,” notes Alexandre White-Brown, MSc, a genetic counsellor and project manager of ThinkRare. Canada’s diversity is nonetheless well represented within the algorithm’s training dataset, which helps to offset some of these potential issues.
While Brownstein says the reasoning model used at BCH improved rapidly, errors and hallucinations are common with any large language model (LLM). OpenAI’s reasoning model still needs to be used in collaboration with a clinician rather than as a stand-alone tool.
The risk of trusting the machine too readily
Research into human-AI interaction has repeatedly shown that experts may defer to the judgement of an AI system, even when it is wrong.
“Previous studies show that learning from evidence is especially difficult when the evidence is poorer and uncertainty is higher, as in the case of patients with rare diseases,” says Aranzazu Viñas, PhD, an assistant professor at the University of the Basque Country. “Therefore, AI’s errors might be still more difficult to detect in the case of rare diseases.”
Her recent study suggests that doctors will trust an AI’s judgement even when they receive new clinical information showing it to be wrong. Viñas emphasises the importance of conducting “ecological research” to establish whether these errors also appear in real-world settings. Beyond that, she says it is “important to develop strategies and protocols that increase human critical thinking and detection of AI errors.”
The future of AI for rare diseases
BCH is now looking towards further study and model validation, measuring time and cost savings, clinician effort, false positives, and whether using the model actually changes the care people receive. The team is working with OpenAI to democratise access for other researchers and clinicians.
At CHEO, ThinkRare is currently able to spot the low-hanging fruit, but Terekhov points to the data silos created by information locked away in unstructured reports and doctors’ notes. The next iteration of ThinkRare will incorporate LLMs that can access and interpret this material, allowing the system to flag more cases. Scaling the model countrywide will require regulators to sign it off.
What a diagnosis changes for a family
White-Brown explains the impact ThinkRare is already having. Antony was diagnosed with an ultrarare condition called Chung-Jansen Syndrome at 10 years old, after being flagged by the algorithm. The diagnosis provided an explanation for his symptoms, and testing revealed that the same condition affected his brothers and his mother.
“The biggest thing for this family was actually being able to access resources in school that were previously unattainable,” says White-Brown.
CCH insight
AI-assisted triage, diagnostic support, and documentation tools are arriving in clinical settings faster than most training has kept pace with. The College of Contemporary Health’s AI Essentials for Primary Care CPD short course gives healthcare professionals a practical grounding in how these technologies work, how to appraise their output critically, and how to use them safely alongside clinical judgement.
Explore AI Essentials for Primary Care →
Source: JMIR Publications
Read More
When the Diagnosis Arrives by App: Why Most Patients Still Want a Human Voice
Key Takeaways:
- In a UT Southwestern survey of more than 2,400 people diagnosed with cancer, 75% said they would prefer to receive the news directly from their physician – in person or by telemedicine – rather than through an electronic patient portal.
- More than half of those who first learned of their diagnosis via the portal were alone at the time, often without support from a clinician or family member.
- Researchers are calling for a more personalised approach, including better portal notification settings, tiered or delayed release of sensitive findings, and plain-language summaries of radiology and pathology reports.
A digital convenience with an emotional cost
Electronic patient portals have transformed how quickly people can see their own test results. For routine bloodwork or a clear scan, near-instant access is a welcome convenience. But the same speed raises a difficult question for clinicians: how should a new cancer diagnosis be communicated when a patient can open the result on a phone before anyone has had the chance to talk it through?
That tension has grown sharper since 2021, when a provision of the 21st Century Cures Act came into force in the United States. The regulation requires that patients have timely, unrestricted access to their electronic health information – which, in practice, means a growing number of people are discovering a new or recurrent cancer diagnosis through their portal, sometimes with no clinician present to interpret it or answer the questions that immediately follow.
What the UT Southwestern survey found
A new survey carried out at UT Southwestern Medical Center suggests that, for most people facing a cancer diagnosis, faster is not better. The findings, published in JAMA Network Open, show that 75% of respondents would prefer to learn about a cancer diagnosis directly from their physician, whether in person or through a telemedicine appointment.
The 2025 survey gathered responses from more than 2,400 people who were diagnosed with cancer at the Harold C. Simmons Comprehensive Cancer Center between 2019 and 2023, giving the researchers a substantial real-world picture of how patients want sensitive results delivered.
According to the study’s lead author, Sheena Bhalla, M.D., Assistant Professor of Internal Medicine in the Division of Hematology and Oncology and a medical oncologist at the Simmons Cancer Center, the broad enthusiasm for digital access does not extend neatly to oncology. “While most patients in the general population appreciate rapid electronic access to test results, the situation for patients with cancer is much more nuanced,” she said. “Learning about a cancer diagnosis without the ability to immediately ask questions or discuss next steps with a trusted clinician can add to the significant stress, uncertainty, and fear that patients experience.”
Preferences are not one-size-fits-all
The survey also revealed that there is no single right way to share a result. Preferences varied according to people’s prior experiences, how frequently they used their portal, and their demographic characteristics. Men, for instance, were more likely than women to prefer learning of a diagnosis through the portal.
For senior author David Gerber, M.D., Professor of Internal Medicine in the Division of Hematology and Oncology and of Epidemiology in the Peter O’Donnell Jr. School of Public Health, and co-Director of the Simmons Cancer Center Office of Education and Training, that variation is precisely the point. “These findings highlight the need for a more personalized, tailored approach to communicating sensitive and life-changing results,” he said. “Moving beyond a one-size-fits-all approach can help clinicians provide a more thoughtful, compassionate patient experience.”
The hidden consequence: facing the news alone
Perhaps the most striking insight concerns the circumstances in which people are receiving these results. Among those who learned of their diagnosis through the portal, more than half reported that they were alone when they read it.
Dr Bhalla described this as one of the most troubling side effects of real-time access. “That’s one of the most unintended consequences of real-time access,” she said. “Patients are often alone without support from their physician or family at one of their most vulnerable moments.”
Possible solutions for clinicians and health systems
The researchers are clear that the answer is not to roll back access, but to design around it more thoughtfully. They point to several potential measures, including raising awareness among both clinicians and patients of the portal notification settings already available; developing tiered or delayed-release approaches for particularly sensitive findings; and integrating supportive digital tools such as plain-language summaries for radiology and pathology reports.
Policy is beginning to catch up. Since the Cures Act took effect, three states – including Texas – have enacted laws permitting the delayed portal release of cancer-related and other sensitive results, giving care teams a window to reach out before a patient is left to interpret difficult news on their own.
Looking ahead
For the study’s authors, the work is a starting point rather than a conclusion. “Further study and increased interdisciplinary collaboration among oncology clinicians, health services researchers, and digital health experts can help us better understand how patients receive and react to cancer diagnoses,” Dr Bhalla said. “Our goal is to increase awareness of this issue and help drive innovative approaches to patient-centered communication.”
Source: UT Southwestern Medical Center
Read More
AI Tools Show Promise for Improving Diagnostics and Outcome Prediction in Resource-Limited Health Care Settings
Key Takeaways:
- Transfer learning enables AI models trained in data-rich settings to be safely adapted for use in health care systems with limited local data, improving predictive performance without rebuilding models from scratch.
- In a Vietnam case study, an adapted AI model substantially improved prediction of neurological recovery after cardiac arrest compared with the unadapted model.
- Wider adoption of AI in low- and middle-income countries will require targeted skills development, infrastructure support and robust global governance frameworks.
Reducing uncertainty after cardiac arrest in constrained settings
After a cardiac arrest, families and clinicians are often confronted with profound uncertainty about a person’s chances of neurological recovery. This uncertainty is particularly acute in hospitals with limited resources, where access to advanced diagnostics and large datasets is constrained.
Researchers from Duke-NUS Medical School and collaborators have demonstrated how artificial intelligence can help address this challenge. By adapting an advanced AI model, the team improved the accuracy of neurological outcome prediction following cardiac arrest in a resource-limited setting.
Published in npj Digital Medicine, the study focused on the use of transfer learning, an AI technique that adapts models pre-trained on large datasets to new environments with limited local data. This approach can enhance performance in new contexts without requiring extensive and costly data collection, making it particularly relevant for low- and middle-income countries.
Adapting a high-income model for local use
The researchers began with a brain-recovery prediction model developed in Japan using data from 46,918 people who experienced out-of-hospital cardiac arrest. They then adapted this model for use in Vietnam, where it was tested on a smaller cohort of 243 patients.
The adapted model performed markedly better than the original model when applied directly to the Vietnamese data. It correctly distinguished between people at higher and lower risk of poor neurological outcomes approximately 80 percent of the time, compared with around 46 percent accuracy when the original, unadapted model was used.
Senior author Associate Professor Liu Nan, from Duke-NUS’ Center for Biomedical Data Science and Director of the Duke-NUS AI + Medical Sciences Initiative, said,
“The study shows AI models do not need to be rebuilt from scratch for every new setting. By adapting existing tools safely and effectively, transfer learning can lower costs, reduce development time and help extend the benefits of AI to health care systems with fewer resources.”
Expanding AI’s role beyond outcome prediction
Beyond predicting outcomes after cardiac arrest, AI has potential applications across a broad range of health care needs in low- and middle-income countries. In a separate study published in Nature Health, Duke-NUS researchers and collaborators, including colleagues from University College London, explored how large language models, trained on extensive text data to understand and generate human language, could support global health.
In resource-constrained environments, such tools may improve access to care, diagnostics and clinical decision-making. Examples highlighted by the researchers included a chatbot providing pregnancy-related information to expectant mothers in South Africa and smartphone-based applications used by community health workers in Sierra Leone to detect malaria infections from blood smear samples. These approaches offer more cost-efficient alternatives to conventional microscope-based systems.
Despite these advances, the researchers noted that AI development and deployment remain concentrated in high-income and upper-middle-income settings. While 63 percent of surveyed researchers, clinicians and service providers reported actively using AI tools, many low- and middle-income countries continue to face significant barriers, including limited infrastructure, insufficient technical expertise and a lack of locally generated evidence on how best to address these gaps.
Co-author Siegfried Wagner, from UCL Institute of Ophthalmology and Moorfields Eye Hospital NHS Foundation Trust, said,
“LLMs have the greatest opportunity to transform health care in settings where specialist physicians are scarcest, but the global health community needs to work together with some urgency to ensure the implementation of LLMs is supported in regions where adoption is most challenging.”
Dr Ning Yilin, Senior Research Fellow at Duke-NUS’ Center for Biomedical Data Science and a co-first author of the study, emphasised the importance of prioritising people when integrating AI into health care:
“Strengthening digital literacy and building confidence in using these tools will ensure AI supports, rather than disrupts, the workforce. Tailored skills-development pathways can help under-resourced workers adapt and thrive, allowing AI to uplift and add value to clinical and administrative roles.”
Charting the path forward: Governance and guardrails
While AI tools have clear potential to improve health care delivery, the researchers stressed that appropriate governance frameworks are essential to ensure safe and ethical implementation. Existing regulations for medical technologies often do not adequately address AI-specific risks, such as data privacy concerns, model hallucinations or unclear accountability for deployment and oversight.
To help close these gaps, researchers led by Duke-NUS have proposed the creation of an international consortium known as the Partnership for Oversight, Leadership, and Accountability in Regulating Intelligent Systems-Generative Models in Medicine, or POLARIS-GM.
The consortium aims to develop actionable best-practice guidance for regulating emerging AI tools, monitoring their impact, establishing safety guardrails and adapting them for use in resource-limited settings. By bringing together health care leaders, regulators, ethicists and patient groups from around the world, POLARIS-GM plans to adopt a phased approach, beginning with a review of existing research before working towards global consensus on AI governance in health care.
Dr Jasmine Ong, from the Duke-NUS AI + Medical Sciences Initiative and a Principal Clinical Pharmacist at Singapore General Hospital, and first author of the correspondence published in Nature Medicine, said,
“With clear oversight and clearly defined guidelines, health care systems can confidently leverage AI’s many strengths to improve health outcomes while steering clear of potential pitfalls.
From policymakers to patient groups, all stakeholders have a crucial role to play in making this goal a reality.”
Read More