Andrew Ng
April 1, 2023
Use of Deep Learning to Predict Final Ischemic Stroke Lesions from Initial Magnetic Resonance Imaging JAMA Network Open, Neurology
April 1, 2023GNS Healthcare
GNS Healthcare accelerates medication research and growth to increase medical quality and lower overall healthcare costs. Gemini – The in silico Patient™ shows the dynamic system of relationships driving disease development and drug reaction, which incorporates and converts a wide range of patient data types. In collaboration with the world’s largest biopharma companies and health plans, the in silico patients allow drug response modeling at the individual patient level through oncology, autoimmunity, neurology, and cardio-metabolic disease.
AI Aspects
GNS Healthcare’s causal AI technology combines and converts a broad range of patient data types into Gemini The in silico Patient™
For exploration and translational study, a novel simulation has been created: silico simulations are run to discover biomarkers of disease development and drug resistance, as well as to assess the effects of interventions.
Clinical experiments that are more efficient and well-designed: choose the best patient subpopulations for new drugs and goals. Reduce costly trial and error by identifying and evaluating the most promising ideas before research design.
Business entries have been sped up: produce comparative efficacy data later in the drug discovery process to justify value-based arrangements, line-of-therapy placement, and product extension.
Advanced payer analytics: applies AI-driven models across business lines to better predict rising patient risk and identify patients who are unlikely to benefit from interventions or treatment.(1)
GNS Healthcare’s IT Overview
Gemini, the in silico Patient, is a data-driven ensemble of computer models connecting drug treatment to patient features to the complex molecular mechanisms and pathways driving clinical outcomes. Gemini simulates disease progression and drug response at the individual patient level to identify responder vs. non-responder subpopulations and the underlying mechanisms of response. This process enables the running of virtual head-to-head clinical trials that optimize inclusion/exclusion criteria in trial design and accelerates the generation of comparative effectiveness evidence.
PathAI selected Digital Realty due to the company’s scalable and flexible design options, unparalleled footprint in the data-dense Northern Virginia metropolitan area, availability of renewable energy solutions, and ability to rapidly deploy high-density, AI-ready infrastructure.
The REFS Platform:
Reverse Engineering Forward Simulation (REFS) is GNS Healthcare’s machine learning and simulation platform. This platform works beyond data correlations of predictive analytics and deep learning to identify the drivers of disease progression and individual patient response to drugs. The unique simulation capabilities of REFS allow scientists to implement new insights and make better decisions by answering questions further beyond the scope of conventional AI.
The platform is fast, scalable, and highly interactive. REFS includes built-in biopharma functions found in no other platform. Data-agnostic REFS is compatible with virtually any data type like EHRS, clinical trials, and claims, and it can work with diverse-sized data sets.
GNS Healthcare’s Financials Aspects
GNS Healthcare announced a $23 million series D fundraising led by Cigna Venture, a strategic corporate venture capital partner and a wholly-owned indirect subsidiary of Cigna Corporation. Joining the Cigna Venture-led funding round were Amgen Ventures, Celgene, Echo Health Ventures, Alexandria Venture Investments, as well as former Caesar’s CEO and Aetna Division. With this announcement, Cigna accompanies other leading carriers, health plans, and GNS investors, including Cambia Health Solutions, Echo Health Venture, Horizon Blue Cross of NJ, and Heritage Provider Networks.(3)
Machine Learning Estimates Healthcare Costs in Patients with Critical Limb Ischemia
Healthcare resource utilization among CLI patients is high, has increased in recent years, and is associated with substantial direct and indirect medical costs. Over one year, the hospitalizations cost for all CLI exceeded $4.2 billion, and healthcare resources and associated medical expenses considerably increased. Machine learning is able to identify patterns and uses algorithms from large clinical datasets and ultimately can increase diagnosis accuracy; thus, early and targeted management will carry out cost-effective medical services and promote a desirable healthcare economy. It has been proposed that comorbid health conditions that demand costly healthcare services be identified to work on the prevention and early intervention to ease the economic cost of CLI.(4)
Machine Learning Estimates Healthcare Costs in Patients with Critical Limb Ischemia
Approximately 20% of multiple myeloma (MM) patients have an aggressive disease course with rapid deterioration and are poorly controlled despite invasive treatment.
PHF19 is a histone methyltransferase associated with myeloma progression. A model using age, ISS, and PHF19 and MMSET expression showed that PHF19 eradication leads to myeloma regression. Specifically, PHF19 elimination leads to cell growth restriction cycle arrest. PHF19 and MMSET were found to be easily analyzed through PCR.(5)
Application of ML Models to Evaluate Hypoglycemia Risk in Type 2 Diabetes
Type 2 Diabetes Mellitus (T2D) is a rising health problem, affecting more than 24 million Americans with an estimated burden of $327 million in medical costs and reduced productivity in 2017. Introducing analytic systems offers new opportunities to overcome these practical barriers by precisely estimating both clinical and economic values. For this study, a retrospective cohort design was applied to a sample of 453,487 patients diagnosed with T2D between 2014 and 2017. A Bayesian ML analytics platform was used to predict six outcomes (Hypoglycemia, antidiabetic class persistence, HbA1c target attainment, HbA1c change, diabetes-related inpatient admissions, and diabetes-related medical costs) defined in patients’ 1-year post-index claims history. Results showed that patients with comorbidities had the highest risk of hypoglycemia. Other risk factors included insulin and sulfonylurea use and hyperglycemia.(6)
Causal Modeling of TCGA NSCLC and HNSCC Data Identifies Network Drivers of Tumor Immune Subtypes
Immune checkpoint inhibitors (ICI) have been used successfully in some types of cancers, but only a fraction of patients achieve clinical benefit. The basis for ICI response is the immunogenicity of the tumor. Understanding the tumor immunogenicity and molecular drivers is necessary to improve clinical outcomes in ICI.
Reverse Engineering Forward Simulation platform; ensembles were built on molecular and clinical data from primary tumor samples of 681 NSCLC, 328 Lung Adeno (LUAD), 353 Lung Squamous Cell Carcinoma (LUSC), and 413 HNSCC patients in TCGA to identify causal drivers of tumor immunogenicity (i.e., wound-healing, IFN Gamma dominant, inflammatory, lymphocyte depleted, immunologically quiet, and TGF Beta dominant. The models showed impressive k-fold validation. Predictive performance for the most prevalent immune subtypes in TCGA: wound-healing and IFN Gamma dominant in both LUAD and HNSCC, as well as inflammatory in LUAD.(7)
References