ARPA-H Is Building Shared AI Infrastructure for Rare Disease Diagnosis and Treatment

ARPA-H is committing up to $98,5 million to a shared data and AI evaluation system for rare diseases

On 31 August, the US health research agency ARPA-H announced the four teams awarded contracts under the RAPID program. Over 4,5 years, the program will receive up to $98,5 million to develop secure data resources, identify rare diseases, and compare algorithms.

For a person with a rare disease, diagnosis takes an average of six years. The symptoms may resemble those of common conditions, and finding the right specialist can be difficult. To interpret a genetic test result, a physician needs the patient’s medical history and clinical observations. The prolonged search for a diagnosis also complicates drug trials because researchers must first identify enough people with the same disease. Approved therapies are available for about 5% of the more than 10 thousand rare diseases.

Each hospital sees few such cases, while medical records remain distributed across different institutions. The University of North Carolina will combine medical histories collected over years of observation with genomic data and information reported by patients themselves. This shared dataset should give algorithms material for training and consistent evaluation.

Sage Bionetworks will develop a secure platform that standardizes different types of data and establishes common criteria for evaluating algorithms. RAPID will organize Rare Challenges, open competitions in which different teams will address the same tasks using this infrastructure. This will allow developers to compare results under the same rules.

FDNA will collect photographs, videos, voice recordings, and patient descriptions of their symptoms. Probably Genetic will add wearable device data, information about what each person can do in daily life, medical records, and genomic data. Together, these data should help identify diseases whose signs change over time or affect several organ systems at once.

A drug trial first requires a group of patients whose disease has been accurately characterized, along with data on how it progresses. According to the RAPID program description, the program connects early disease identification with drug development, clinical trial design, and the selection of outcome measures used to evaluate trial results.