Health tech startup Dandelion Health launched a clinical artificial intelligence marketplace that merges third-party algorithms and Dandelion's data platform and model validation capabilities to supplement clinical trials.
The company, a real-world data platform focused on clinical analytics and precision medicine, expects the marketplace to exponentially grow the business and possibly transform the industry.
Dandelion’s core competency is validating AI/machine learning algorithms. The startup’s aim is to use the highest-quality data to build out the next phase of AI in healthcare.
To show proof of concept for the clinical AI marketplace, Dandelion released a study, performed on its data, on the efficacy of GLP-1s to reduce cardiovascular disease risk on a previously unstudied population. The results were in line with Novo Nordisk’s clinical trial results.
In May, the company launched a GLP-1 data library as a multimodal real-world clinical data set built to surface insights and opportunities related to the GLP-1 receptor agonist drug class. They worked with Sharp HealthCare; Sanford Health in Sioux Falls, South Dakota; and Texas Health Resources to collect structured and unstructured data for more than 10 million patients.
“We took an algorithm from a partner in Germany, from Pheiron, and we ran it through a data set that we already have, and we compared it then to a clinical trial to show in really practical detail, how could AI just really impact clinical research in a way that people haven't foreseen?” Elliott Green, co-founder and CEO of Dandelion Health, told Fierce Healthcare in an interview.
In the marketplace, there are AI algorithms from various developers and structured and unstructured real-world data. Dandelion says life sciences companies, particularly drug developers, will be able to run studies on the validated algorithms in a fraction of the time it takes to conduct a randomized controlled trial (RCT).
For example, Dandelion’s proof-of-concept study, which included 4 million patients, was conducted in six weeks.
“It's the raw waveforms, it's the ECGs, which is what allowed us to do this study. It's all of those things that you kind of wonder, where do they store this and where does it go? And the answer is, in very weird places it take an incredibly long time to find and so that's what we've spent three to four years doing, is building the pipes to extract that information," Green said.
The primary use case Green foresees in the near term is for biopharma companies to understand other uses for their drugs, like if a heart failure drug could impact hypertension. Elaborating on this example, Green said that Dandelion has multiple algorithms that do ECG to hypertension predictions or diagnoses. Dandelion can then run the company’s question against the patient data it has, using an algorithm developed by a third party.
“Clinical trials cost billions upon billions of dollars and take many, many, many years. And AI is one of the short, most surefire ways they have of attempting to shorten and cheapen that process," he said.
Being able to use this AI experiment could save time and money for life sciences companies. While an AI-generated study can’t yet be used for purposes of FDA clearance, Green said that it could be used alongside a RCT.
Eventually, Dandelion Health thinks that AI-generated biomarkers and endpoints could be incorporated into RCTs and even approved by the FDA.
The clinical AI marketplace could ultimately engender more trust in AI for healthcare delivery organizations to use AI in clinical practice once the product is more built out.
“An algorithm that spots hypertension for a life sciences company could also be a diagnostic tool or clinical decision support tool for a physician within a hospital system,” Green said.
He also noted the results of AI clinical studies could be valuable for payers to understand the benefit of a treatment to a broader patient population.
By being involved in the clinical AI marketplace, AI developers have the opportunity to have their model validated, trusted and utilized.
“For AI developers, it's incredibly important because it gives you an ability to use your algorithm in a way that hasn't historically been possible, because without validation, you can't prove that it does what it says. So now we validate it, we use our data, we make it plausible,” Green said.