AI Tool Aladynoulli Predicts Risk of 348 Diseases from Routine Health Records, Outperforming Existing Risk Models
Key Insights
Dana-Farber and MGH researchers developed Aladynoulli (search), an AI algorithm that simultaneously predicts the likelihood of 348 distinct diseases using only routinely collected electronic health record data and genetic risk information.
The model outperformed existing cardiovascular risk calculators (PCE, QRISK3, PREVENT) for 10-year predictions and the GAIL breast cancer (search) model for 1-year predictions, with particularly high accuracy for imminent colorectal cancer (search).
Aladynoulli (search) uses 20 curated biological signatures to drive predictions in an interpretable way, avoiding the "black box" problem common to deep learning approaches.
A team of researchers from Dana-Farber Cancer Institute and Mass General Brigham (search) has developed a first-of-its-kind artificial intelligence algorithm capable of simultaneously predicting the likelihood of 348 distinct diseases for a given patient, using only routinely collected electronic health record data combined with knowledge about the patient's genetic risks. The algorithm, called Aladynoulli (search), is described in a paper published in Nature.
The model represents a significant advance in clinical risk assessment by making multi-disease predictions from data already present in patient records, without requiring specialized testing. "This is a novel algorithm that analyzes full patient clinical data trajectories to make very accurate predictions of disease transitions into the future," said co-senior author Alexander Gusev, PhD, a Dana-Farber scientist. "People are thinking about what their health is going to look like over the next few years, especially with increasing intervention options. This tool offers a path toward improving the prediction of future diseases so doctors and patients can take action to try to prevent them."
How Aladynoulli (search) Works
The research team, led by co-senior authors Gusev, Giovanni Parmigiani, PhD, a Dana-Farber researcher and associate director of the Division of Population Sciences, Pradeep Natarajan of Massachusetts General Hospital, and first-author Sarah Urbut, MD, PhD, an MGH cardiologist, combined probabilistic modeling with machine learning in a unique design that makes the tool's predictions both powerful and biologically interpretable.
The researchers employed probabilistic modeling to define 20 biological "signatures" — sets of biological trends that have a high probability of initiating certain diseases. For example, high cholesterol in a patient's history increases the probability of cardiovascular diseases. The signatures also factor in genetic variations known to increase the likelihood of disease. These signatures are complex and overlapping; the risk of colon cancer, for instance, is associated with multiple different signatures.
"We have painstakingly curated these signatures, which is a big differentiator. In contrast to deep learning approaches, which are typically 'black boxes,' our curated signatures capture the underlying biology in an interpretable way," said Parmigiani. "These signatures are then the drivers of the model's ability to make predictions."
The model was trained and validated using three large biobanks including a total of over 683,000 patient records. Many of the records came from the UK Biobank, which houses health-related data from consenting patients without identifying them and is headquartered in Greater Manchester, England.
Outperforming Existing Risk Models
Aladynoulli (search) demonstrated superior performance compared to established risk calculators. The model outperformed existing cardiovascular risk models — PCE, QRISK3, and PREVENT — with more accurate 10-year predictions. It also outperformed the GAIL breast cancer (search) risk model for 1-year breast cancer predictions.
The algorithm is particularly impressive in predicting which patients will develop colorectal cancer (search) in the coming year based on findings in electronic health records. There are 17 conditions in a cluster associated with colorectal cancer risk, including ulcerative colitis, general gastrointestinal complications, and hemorrhoids, Parmigiani noted. Flagging a high risk of imminent colorectal cancer could help a primary care physician refer a patient for a colonoscopy, even if that patient is not yet eligible for screening based on current age-based guidelines.
This capability carries particular urgency given the sharp increase in colorectal cancer (search) in younger adults in the United States. It is now the leading cause of cancer death in adults under 50, according to Emory Healthcare. "This type of risk assessment could help doctors intervene early, when the disease is still preventable," the researchers noted.
A Holistic View of the Patient
A key strength of Aladynoulli (search) is its ability to synthesize information across medical specialties and over time. "It sees the patient holistically, both across departments and over time," said Urbut, who has worked on the algorithm for about three years.
She offered a concrete example: a rheumatologist treating a patient for an inflammatory disease could make notes that might be relevant to a cardiologist, given links between some inflammatory conditions and risks for developing heart disease. "Physicians are extremely conscientious, and we do often read notes from many other specialties," Urbut said. "But an interview is limited, and if we had an algorithm that said which notes we should pay attention to and what we could learn from other departments, that's amazing."
The tool gets better at predicting diseases as a patient sees more doctors and generates more electronic health records, and its predictive power improves as the patient ages.
"In clinic, two patients with the same diagnosis are not the same patient," said Pradeep Natarajan, MD, director of Preventive Cardiology at Mass General Brigham (search) Heart and Vascular Institute and associate member at the Broad Institute of MIT and Harvard. "This model shows they often have different underlying signature profiles, which can translate into different progression patterns and different responses to the same treatment."
Clinical Implementation and Future Directions
Parmigiani said the technology is "ready for prime time" but acknowledged that AI tools face a complex journey to reach widespread use. The researchers hope Dana-Farber and MGH will incorporate the AI tool into their electronic health records in the near future, though they could not specify a timeline for when it could become available to patients elsewhere.
"This model is an innovative and potentially disruptive tool in the clinic because it could encourage more clinicians to think in a cross-disciplinary way," said Parmigiani. "It is important to try to understand a patient with a full 360-degree perspective on the data, as opposed to compartmentalizing the information by specialty. One of the biggest strengths of this model is that it knows that many diseases are driven by the same underlying biology."
Next steps for the team include using the model to investigate why and how melanoma (search) spreads to other organs. The team has partnered with Dana-Farber medical oncologist David Liu, MD, MPH, to train the model to identify different patterns of metastases, such as spreading early or late, or spreading widely or narrowly. "Aladynoulli (search) not only can help predict different trajectories of metastases, but it also can give us a biological understanding of why patients are progressing along these different trajectories," said Gusev. "From a better understanding of the biology, we have the potential to identify more beneficial therapeutics."
The team is simultaneously working to expand the signatures in the model to increase the accuracy and biological grounding of the risk predictions, and is looking for opportunities to move toward implementing the model in clinical practice and as a tool to improve the design of clinical trials.