Triomics to attempt to revolutionize clinical trials for cancer patients

Founded by two college friends, Hrituraj Singh and Sarim Khan, Triomics is an artificial intelligence company centered around oncology that has generated buzz in the cancer clinical trial world due to its ability to analyze unstructured medical data, which can consequently help match patients to clinical trials and extract structured data from unstructured notes, such as a doctor’s free-text note. This is particularly notable since this type of data makes up around 80% of medical intelligence available.

Singh and Khan, who worked as an Adobe AI researcher and MIT biotech researcher respectively, tout their large language model technology, OncoLLM.

Only 20% of medical data exists and is stored in a structured, uniform manner, including information on patient demographics and laboratory values, and many software exists to quickly analyze this information. In contrast, generative AI that could analyze unstructured information did not exist prior to Triomics. The software achieves this by using institution-specific inputs that sift through the cumbersome details within patients’ medical records.

Additionally, the model can perform a wide range of tasks by utilizing specific use cases. There is gross inefficiency among cancer patients being assigned to clinical trials, as observed by statistics published in the Journal of Clinical Oncology. In its April issue, the journal shared that only about 20% of cancer patients are enrolled in a clinical trial. One important function that the model can perform is matching patients to relevant clinical trials. This helps manage new treatment options for cancer patients, test them out, as well as manage the side effects of existing treatments as identified by the National Cancer Institute.

In an example from the published study, it was observed that tasks that took medical practitioners hours of manual labor could be achieved within minutes through the use of the model. In the domain of matching patients to relevant clinical trials, healthcare staff has to devote hours upon hours to assign a single patient to a clinical trial because they have to be conscious of strict inclusion and exclusion criteria for trials. In comparison, the Triomics model took minutes to match 90% of the patients assigned to it to a clinical trial.

Executives at the company also pointed out the model’s competency when extracting structured data points from unstructured notes. Its accuracy in this regard was observed to be comparable to proprietary models like GPT-4 or Claude.

Despite being almost equally effective, there is a marked difference in the price points for both models. According to company representatives, Triomics is 40 times cheaper than these proprietary models. While the running costs for GPT-4 are over $6000 per hour, OncoLLM costs about $170 per hour.
An end-to-end large-scale empirical evaluation of the model’s abilities conducted in collaboration with the Medical College of Wisconsin led to Triomics setting up pilot systems across several cancer centers. Data from these centers regarding the model’s capabilities will be released by the end of this year.