Machine learning IDs family history, carpal tunnel as ‘red flags’ in hATTR
AI model may help predict positive genetic tests in amyloidosis
Written by |
A family history of nerve disease, carpal tunnel syndrome in both hands, and balance problems are the “clinical ‘red flags'” most strongly linked to a positive genetic test for mutations associated with hereditary transthyretin amyloidosis (hATTR) — including hATTR with polyneuropathy (hATTR-PN), a disease type marked by damage to the peripheral nerves.
That’s according to a new study by researchers in Italy that employed a type of artificial intelligence (AI) known as machine learning to identify these relationships. In basic terms, machine learning is the process of training a piece of software, called a model, to make useful predictions.
In this case, the team was seeking to “refine diagnostic algorithms and prioritize predictive features” to aid in better diagnosing hATTR and hATTR-PN.
“These findings suggest that integrating [machine learning] with clinical red flags may support the diagnostic decision-making process in patients referred for suspected [hATTR],” the researchers wrote.
The team cautioned, however, that the “results should be considered exploratory and require validation in independent cohorts before clinical implementation.”
Interestingly, the researchers noted that other warning signs previously linked to a positive genetic test, such as unexplained weight loss and gastrointestinal symptoms, did not rank among the most important predictors in this analysis.
The study, “The role of ‘red flags’ in the diagnostic work-up of hereditary transthyretin amyloidosis: a study using a machine-learning approach,” was published in the journal Neurological Sciences. The research team comprised scientists from the University of Palermo and employees of Demetrix, an Italian digital technology company.
In hATTR, inherited mutations in the TTR gene disrupt the structure of the protein transthyretin, causing it to form clumps that accumulate in tissues and damage them. Also known as familial amyloid polyneuropathy (FAP), hATTR-PN mainly affects the nerves; hATTR cardiomyopathy (hATTR-CM) mainly affects the heart.
The disease can manifest differently from person to person, even among family members, and can mimic other neurological or cardiac conditions. As a result, a confirmed disease diagnosis is often delayed, especially in places where hATTR is less common.
Using AI to ID red flags that might prompt genetic testing
To help address this problem, the research team set out to determine which so-called red flags, or clinical warning signs, might prompt a doctor to suspect hATTR and recommend a genetic test.
The team used machine learning models and explainable AI to analyze the medical records of people who underwent genetic testing for TTR mutations. Together, these approaches help detect patterns in large amounts of data and identify the most influential clinical features.
The analysis involved medical records from 452 patients, ages 19-86, who underwent genetic testing between 2019 and 2024. Some were referred for testing because they had possible symptoms of hATTR, while others were referred due to a family history, the researchers noted.
Overall, 68 people (15%) carried a TTR mutation, the data showed. Among them, about half were symptomatic at the time of genetic testing.
According to the researchers, the most common clinical red flag observed in the study group was sensory neuropathy, a type of nerve damage that causes loss of sensation or abnormal sensations. This symptom was seen in 283 patients, or slightly more than 60% of those tested.
The second most common red flag was a family history of cardiomyopathy, seen for 52%. Cardiomyopathy is a disease of the heart muscle that makes it harder for the heart to pump blood to the rest of the body. That was followed by carpal tunnel syndrome — characterized by numbness, tingling, and weakness in the hands due to compressed nerves — in both hands, experienced by 46%.
A wide range of other clinical features was also reported. Signs that were each observed in approximately one-quarter to one-third of patients were a family history of neuropathy, autonomic dysfunction — problems with involuntary body functions — and difficulty with coordination and balance, known as ataxia. Gastrointestinal disturbances, unexplained weight loss, and a known family history of hATTR were also seen in about a quarter to a third of these individuals, but the researchers noted that these features were reported less often than expected.
Selected machine learning model had 89% accuracy
The scientists then evaluated various machine learning models to identify the one that best predicted which patients would be TTR mutation carriers based on their clinical red-flag profile.
The model the team ultimately selected, called Random Forest, delivered the most “balanced and stable performance,” according to the authors. Specifically, the model was about 82% accurate in making its predictions. It correctly identified about 89% of cases in which a TTR mutation was present.
Another algorithm called XGBoost, used in a previous hATTR-PN study by the same group, could correctly identify 100% of mutation carriers; however, it was also more likely to come up with false positives, that is, saying a patients had a positive genetic test when they did not.
The [selected] model may represent a promising tool to support the identification of patients who could benefit from confirmatory genetic testing.
Explainable AI tools revealed that a family history of neuropathy and carpal tunnel syndrome in both hands were the clinical red flags most strongly linked to a positive genetic test result, the researchers noted.
While ataxia was also a relevant predictor, it was associated with a lower likelihood of a positive result, which the authors suggested may reflect that gait problems are often a late and nonspecific sign that can appear in other conditions unrelated to hATTR.
Overall, “the model may represent a promising tool to support the identification of patients who could benefit from confirmatory genetic testing,” the researchers wrote.
Because the data were drawn from the same general region of Sicily, where participants likely had similar genetic backgrounds, the researchers cautioned that the findings cannot yet be applied to the general population. The team stressed that “multi-center validation is required before any clinical implementation.”
Leave a comment
Fill in the required fields to post. Your email address will not be published.