Let us begin with a more general topic. How has AI transformed Electronic Medical Records (EMRs)?
EMRs have changed dramatically over the past 10 years - this is because they now belong in a digital era, and advancements in data and AI enable EMRs to have so many more uses. Initially, EMRs were predominantly used for billing and diagnostic purposes, focusing on patient interactions with healthcare systems. However, the most valuable medical information often resided in unstructured forms, like physician notes, which were not easily computable. Often, physician notes are dictated instead and written up by someone else. With the introduction of advanced computing and AI, we aim to make this vast wealth of biomedical knowledge computable.
Speaking to the broader question of what is happening to patient data - we need more AI because the explosion of medical knowledge, especially after the Human Genome Project, has made it almost impossible for professionals to analyze everything. EMRs are now more than just records; they are a comprehensive digital footprint of a patient’s lifetime of care, encompassing everything from genetic information to treatment outcomes, and this data can really help people, so we can’t let it be wasted.
In that context, what is nference’s value proposition?
EMRs have the potential to encapsulate a patient’s complete medical history in a digital format, making it accessible and analyzable. Our approach at nference is to harness this rich source of information from EMRs, especially from academic medical centers, since these are known for their meticulous record-keeping. By integrating this data with other forms of biomedical knowledge, we aim to extract meaningful insights while maintaining strict privacy standards. This integration allows us to explore complex associations between genes, diseases, and treatments, enhancing our understanding and opening new avenues for medical research and practice. Our focus is not just on structuring this information but also on preserving the privacy and integrity of the data, ensuring that it is used responsibly and only for advancing medical science.
How do you distinguish truly valuable knowledge from the rest of the information contained in EMRs?
Our mission at nference is to make this vast ocean of biomedical knowledge computable and accessible. This involves distinguishing valuable knowledge from mere background noise in the immense pool of data. To achieve this, we focus on identifying correlative signals quickly, acknowledging that establishing causation is a more complex process requiring deeper scientific experimentation.
Our software system, nSights, is designed to process diverse datasets, from published literature to real-world medical practice encoded in EMRs. nSights helps medical researchers hypothesize and validate new medical theories. We are essentially creating a platform where biomedical knowledge, in its various forms, is not only digitized but also made comprehensible and useful for medical advancement.
Can you provide some examples of how nference's data analysis has led to significant medical insights?
One recent case study involved the drug Semaglutide, of recent fame under the brand name Ozempic, for its weight loss benefits. Prior to official clinical trial results being published, we utilized our nSights platform to investigate Semaglutide’s potential cardiovascular benefits. Analyzing data from two leading academic medical centers serving diverse populations , we found that it not only reduced blood glucose and weight but also decreased the incidence of cardiovascular events. This finding was consistent across diverse patient groups, including a significant proportion of African-American patients, which means it could really help those at risk - and this has merited further study. Our systems can discover insights that are easily missed, and this evidence is a high-level example.
While our real-world data is most applicable from phase I to phase IV of clinical trials, it also offers glimpses into early drug discovery phases. These insights often stem from clinical data, suggesting hypotheses for preclinical experiments. For example, our platform can identify subtle benefits or side effects of a drug, which can then be explored in more depth through scientific experiments. This approach enriches early drug discovery by providing informed hypotheses based on real-world clinical data.
Beyond drug discovery, our platform significantly impacts patient care. For instance, we have developed AI algorithms to diagnose rare conditions like pulmonary hypertension earlier than traditional methods, which greatly improves patient outcomes. Additionally, our platform aids in detecting common conditions such as heart failure risk, using ejection fraction analysis (analyzing the total amount of blood in your heart that is pumped out with each heartbeat) from ECG data.
How extensive is the patient data that nference processes?
Since our collaboration with Mayo Clinic in 2019-2020, we have significantly expanded our data pool to include diverse patient demographics, working with various health systems across the US and internationally. Our dataset now encompasses approximately 50 million patients globally. This expansion is crucial to ensure our data represents humanity as a whole.
Ensuring privacy in handling EMRs is a cornerstone of our work at nference. We have developed a privacy-preserving architecture in collaboration with strategic partners like Mayo Clinic. This involves a meticulous process of anonymizing and de-identifying patient-related information, particularly in the unstructured domain of EMRs.
Our system allows researchers and medical professionals to access and learn from this data within a secure cloud environment, without the possibility of their raw data being extracted. This is crucial because even de-identified data, if mishandled, can sometimes be re-identified, so we stop that risk at its source. We adhere to stringent legal and ethical standards, requiring our partners and customers to observe high levels of privacy protection. Our approach ensures that the data is used exclusively for advancing medical knowledge and developing new diagnostics and treatments, without compromising patient privacy.
To ensure privacy, we have implemented a 'federated network' approach. Data remains within its original healthcare system, that means it is protected under strict privacy laws like GDPR, and is never transported across borders or even between health systems. Federated learning allows the analytical model to be transferred between data centers, and so we don’t need to transfer patient data out of these centers. We want patients to know that privacy is deeply important to us, and we will protect and respect their privacy, always.
