The State of AI in Healthcare PART II
In Part I of AI in Healthcare, I provided a brief introduction to the growing popularity of AI in Healthcare, not just as an alternative but as more of an investment in AI and the realization that it can help healthcare entities grow and improve by leaps and bounds. I focused on some of the AI solutions that are newer in healthcare, such as Software as a Medical Device (SAMD), compared to the data-focused AI solutions that have been around longer in segments such as Real World Data.
In this post, I will discuss how AI has changed some of the more established segments of Healthcare technology: Provider and Payor Administration, Outcomes - Real World Data, and Outcomes - Real World Evidence.
Provider and Payor Administration
Providers and Payors each deserve their own section and I might cover these in more detail in another post sans a specific focus on AI. However, it is important to look at some of the new, exciting projects leveraging AI centrally that have come up in the last few years.
The Healthcare industry is worth about $800 billion. $600 billion of this amount was spent on administration and billing alone. This $600 billion breaks up as the majority of the costs borne by Payors and Providers. In both cases, inefficiencies are rampant in day to day to operations. On the Provider side, some of these inefficiencies include mismanagement of patients and procedures; incorrect billing, documentation, and inaccurately translated problem lists; scheduling conflicts for care professionals; and sub-par reimbursements. On the payor side, the incompetencies are burdening high-risk pools, inadequate plan information and visibility for employers, overpayments, and inefficient third-party broker arrangements.
One of the major issues that are a pain on both sides for payors and providers is inefficient billing. Taking on this problem is Anomaly, an AI-powered solution that tracks bills from providers before going to the payor. Although claims data suffers from inconsistencies and loss of information once it leaves the doctor’s office, extracting codes and other valuable information that determines payments is possible with AI and NLP technology available today. Anomaly’s platform seamlessly integrates with the claims workflows of providers and corrects billing errors at scale before they are forwarded to the payor. Since, claims are nearly universal across the states, regardless of the type of visit or treatment, technologies like Anomaly could have a big impact on savings for all entities, savings that are fair.
Another major pain point that often delays care and impacts the quality of treatment is managing prior authorization. Until recently, authorizations were limited by burdensome communication processes such as faxes and manual reviews. This tedious process fails to deliver faster care to the patient. Enter, Cohere Health and Machine Learning. Much of the important, critical clinical information that helps reviewers approve authorizations is hidden in the EHR data, often in unstructured data and physician notes. Unstructured data cannot be stored in a relational database and usually exists in text format, a common example is physician notes. With advanced ML and NLP techniques, it is much easier and faster to identify concepts in the text that help health plans understand the patient history on a longitudinal level and make decisions faster. In addition, Cohere’s platform allows the collection of data and grouping of patients into populations. This grouping makes it possible to design care for individual populations on a longitudinal scale and optimize these paths on an ongoing basis. NLP and Deep Learning models are slowly becoming the status quo across sectors (more on this in coming posts) as startup founders realize that it is time to go deeper into healthcare data with ML, an approach that will not only help their products stand out but more importantly, deliver health-tech solutions that have much more precision and impact overall.
Outcomes - Real-World Data
The use of AI in real-world data and real-world evidence companies varies. It also depends on how one defines AI. I like Apple’s definition in their latest WWDC event where they strayed away from using the term AI and instead decided to be more technical with the term Machine Learning (ML). I believe it is the same in healthcare. A lot can be achieved in the RWD space without much ML and instead spending effort and infrastructure building a solid database for analytics and insights/visualization platforms. Just look at the industry’s best-in-class healthcare map at Komodo Health which covers approximately the entire US population for the last 5-10 years. But even Komodo has started to spend significant effort digging deeper into its data by leveraging Machine Learning to enhance the entities in its data and get more clinically relevant and accurate assertions from its data.
Yet another company leveraging RWD and AI successfully is OM1. OM1 uses AI to make predictions and extract characteristics about populations and multiple aspects of the patient journey, with a focus on chronic diseases. The specific ML piece at OM1 is interesting in that it is actually able to make predictions about risk scores and re-admissions. OM1’s team has been able to make predictions about heart failure patients at risk for unplanned readmission in the future and even a score for these patient cohorts. Using random forests, a powerful machine learning model that incorporates shuffling records (ex: claims or encounters) as well as features (important attributes in a dataset, ex: age, gender, place of care, insurance plan) in the data and aggregates results across all the different combinations (I want to write more about these models in detail but we’ll stick to a high-level summary here), OM1 is able to predict re-admissions scores for these patients. Having foresight rather than hindsight, especially in critical cases such as heart failure, can greatly improve care pathways for these patients and ultimately save millions of lives.
Tempus is not a new name in the RWD space. With products and platforms for Life Sciences, Patients, and Providers, Tempus has its feet in all the different sections of the healthcare care cycle. However, Tempus’ scale has not stopped it from using AI. Only a company of Tempus’ size can conduct a study across clinical studies to predict the probability of success for clinical trials. Tempus’ AI models use both clinical and molecular data to identify the determinants of successful clinical trials in each phase and come up with a value for metric Tempus has defined in-house - PTRS (The probability of Technical and Regulatory Success). “The biggest value and acceleration come when we combine both by selecting a Phase 2 program to help make an informed go/no-go decision, and then quickly design the subsequent Phase 3.” These studies can vastly change the course of a trial and steer the research arms of companies in the right direction. This will not only mean financial savings but faster treatments delivered to patients on average. Some of the other companies mentioned in this space such as Concert AI are leveraging their clinical trial optimization platforms to do the same (more on Concert below).
Outcomes - Real-World Evidence
I wanted to separate Real World Evidence (RWE) from Real World Data (RWD) above because I believe the Real World Evidence space will be shaped by FDA regulations and a certain standard going forward. Not that some of the companies mentioned in the RWD section aren’t successfully entering this space, Real World Evidence will be continued to be measured by the high validity of the data and insights as well as successful deployments at scale. The fact that the FDA is keeping a close eye on and shaping the industry standard further reinforces this fact and rightly so, RWE needs to be a step above the rest if it is to serve as absolute proof in a clinical outcome or therapy development vs a commercial tool for market access.
An RWD/E behemoth, Concert AI, provides Data and SAAS solutions for Clinical Trial Optimization, Predictive analytics for Patient Populations and Oncology studies. What’s even more impressive is how Concert AI has constantly stayed close to utilizing Machine Learning in all its products. Continuing on the NLP and Deep Learning theme from Cohere, Concert is leveraging NLP to extract metastasis information in structured EHR data. With better visibility into and assertion of metastasis, cancer treatments can be better designed by pharma and biotech companies and treatments can be managed in a much more efficient manner. Just imagine the potential of combining NLP and Deep learning in non-structured datasets. The advancements in transformer models and NLP will make it easier for companies to further meet FDA standards and bring RWE to the forefront of Digital Health.
Summary
The provider and payor administration space is vastly bigger in digital health than the few companies discussed above. However, the use cases at Anomaly and Cohere exhibit the importance of AI in alleviating the various inefficiencies in the healthcare system and the patient journey. Similarly, in the Real World Data space, in addition to AI, companies embracing unstructured data and healthcare data sets beyond claims and EHR is becoming more common. As the commercial space gets crowded, the ongoing wave and future of Healthcare lies in Digital Health startups getting more out of data and pushing for more precise, assured outcomes. I am looking forward to covering both these spaces in more detail across the healthcare ecosystem in my future posts, stay tuned!