GNS Healthcare – Using AI to ID novel drivers of disease

Machiene Learning is unlocking tremendous value for pharma and payers.

GNS (Gene Network Sciences) Healthcare uses novel AI technology to comb through longitudinal patient data to identify novel drivers of disease, simulate head to head trials and help payers identify high performing physicians (among other applications). The AI that powers this learning engine is called causal hypothesis free machine learning. Causal refers to the platforms ability to go beyond “black box” solutions and point to the “why” behind a finding. In the context of GNS Healthcare, this means for a provider project identifying which MDs provide the best quality care at low cost, the models can not only identify which providers perform better but can look at the inputs and determine “why” the performance improved. This enables users of the platform to figure out what practices lead to better performance and potentially spread them to others.

Hypothesis free means that as opposed to other hypothesis driven approaches, the data is input into a model with a specific research question. Relationships and pathways are surfaced organically rather than being derived from a researcher question. This expands the bounds of the types of findings possible.

The technology is deployed for both pharma companies and payers/providers in the healthcare world. Pharma companies, for example, can bring their clinical trial data often in combination with molecular level data from companies such as Flatiron to GNS and, dependent on quality and quantity of the data, GNS can apply its technology to answer major questions for pharma companies. The data includes a spectrum of genomic data, clinical data, EMR data and claims data brought together to create a more holistic picture of patient health.

An example of this may be a pharma company wanting to simulate a head to head trial for a drug before going to market. Head to head trials in real life are huge costly and time consuming. They also entail a great deal of risk – if the drug performs slightly worse that the drug they are comparing to then they have a poor result after tremendous money and time spent. Running a simulation through GNS is a fraction of cost and time. If the head to head goes well, they can go forward and recreate it in real life to submit to the FDA. However, if the simulation looks like it might produce a poor result for the drug, they can save themselves the time and money.

A key risk with GNS is their ability to explain the technology to clients. This is novel technology based on math by Judea Pearl only published in the last 20 years. Gaining buy in is challenging. Additionally, they need to find talent that can both execute the math and also explain how it works. Secondly, in healthcare, it is critical to have subject matter experts and clinical expertise informing the simulations. Sometimes there is a lot of noise in the models and a clinical eye can help weed out signals that don’t make biological sense.

 

Previous:

Scale: Labeling the Future of Data

Leave a comment