Innovating Prior Authorization With Responsible AI

Sam Roushan, Senior Vice President of Clinical Transformation at Cohere Health, shares how responsible AI is informed with sound clinical knowledge and can stay within ethical and legal boundaries and benefit payers, providers, and patients. 

Please share a little about yourself and your background. 

I am Sam Roushan, Senior Vice President of Clinical Transformation at Cohere Health. I've spent my entire career using business and human-centered design approaches to solve really big problems in health care, and I've worked all across the health care ecosystem with med tech companies, pharma, providers, and payers. I joined Cohere Health earlier this year because I'm incredibly passionate about utilizing technology to improve patient outcomes and create a provider experience that's free from burnout and enables them to thrive in their careers. Cohere does just that, which is what really attracted me to the company.

We provide artificially intelligent prior authorization as a springboard to better quality outcomes. The ways in which we do that is by aligning physicians and health plans on evidence-based care paths to reduce administrative expenses and burden while improving patient outcomes. The solutions that we have reduce costs, reduce provider abrasion, help patients receive care that's truly optimized for their needs, and we do it using AI that we have responsibly applied to more than 5 and a half million prior authorization requests today. We do that through Cohere's Unify AI Platform. The platform has nearly 130 machine learning models in production, and those models have saved over 11,000 clinical review hours and have 2 and a quarter million unique findings each day. We're a care transformation company, and this is really what makes us different from our competitors and what made me so excited about joining the company.

Can you explain how AI-driven technology can help health care professionals overcome the challenges posed by prior authorization?

First, let me just say that when we talk about artificial intelligence at Cohere we mean responsible AI. We rely on responsible AI to search, extract, and transform information but we don't ask our models to reason. I want to be very clear, we never deny using AI. We also have physicians that are front and center in building and training our models as well as compliance experts.

In terms of challenges that are posed by prior authorization, delays are a huge pain point in the prior authorization process and Cohere is using AI responsibly to minimize delays. We do that in a few ways. First, almost all prior authorization requests come in through our portal and we use AI to auto approve as much as possible. As I mentioned, we're continuously refining these models and rules with the latest clinical guidance and evidence.

There are a fraction of prior authorization requests that come in through fax. There are some providers that just can't let go of that fax machine and we use AI in both cases to digitize the incoming fax request by auto-filling fields and attaching clinical notes, which reduces the time required for human touch by almost half. And then, for prior authorization requests that can't be auto-approved, Cohere's AI solutions extract the relevant information from clinical notes and surface it to the clinical reviewers. That saves them, in our experience, over 13% of their review time. That's how we're using AI responsibly in terms of the delays pain point. There's another pain point in the prior authorization process, and that's ensuring that patients are aligned to evidence-based care paths that are matched to their unique needs. Through our AI solutions, we guide clinical choices to help providers and patients get to a yes for their authorization request without that unnecessary and an abrasive back and forth.

First, our AI reduces provider burden by leveraging patient history to reduce the required clinical assessment questions and attachments that are asked of providers in the prior authorization process. This saves providers 28%, in our experience, of their time for submitting prior authorizations. It also ensures better accuracy and safety because we're not asking providers to recall on the spot something about the patient but we actually can leverage the patient history, and it also reduces providers gaming the system, which is another pain point. And then, we also leverage AI to guide beyond this initial prior authorization request. Again, using what we know about patients we can suggest optimal evidence-based care paths and even provide an approval for the entire care episode, that contains all the next best actions following that initial author request. So lots and lots of opportunity for AI to address many prior authorization challenges.

What kind of communication improvements can be expected among payers, providers, and patients with the implementation of AI in prior authorization processes?

Yeah, and thinking about your question, it dawned on me that with auto authorizations we're actually eliminating the need to communicate on those prior requests, and that's really how it should be. Let's let providers provide and let patients get the care they need without this administrative process getting in the way. Auto authorization definitely solves that challenge. We also give guidance to providers in our portal with transparency for what's clinically appropriate and why, to help those providers get to a yes and, again, avoid those delays in care. That transparency that's coming from Cohere Health where it's representing the payer really helps that communication between payer and provider. And then, we also append only cases that truly need to be pended, meaning that communication between providers in a peer-to-peer discussion happens for only the most needed cases. This supports clinicians working at the top of their license and supporting the education of providers, so I think that there's a number of communication improvements here.

How is the integration of multiple clinical data sources important to AI’s decision-making and disease management skills?

This is what gets me really excited. At Cohere, we seek to leverage the prior auth moment to personalize care for patients, and the more we know about a patient the more personalized our recommendations to providers can be. We already have proprietary patient clinical data and provider performance data from the unique and valuable position that we're in between payers and providers. We can combine this with additional clinical data sources like HIEs and EMRs, health information exchanges, and electronic medical records to build an even fuller patient history and perspective on provider quality. The prior authorization moment serves as a signal that a patient's care episode is commencing, and then we can then use what we know about the patient and providers with our AI technology to recommend a personalized care path for downstream care.

This moves us away from one size fits all care to personalized care. In terms of disease management, we can use the early signals from prior authorization and clinical data we have to identify at-risk patients much earlier than waiting for claims data, and we can alert our payer clients so they can initiate their disease management and care navigation programs even earlier for greater impact. And then lastly, our intelligence on providers also helps our clients ensure that their provider networks are optimized for cost and quality because we can help them see at the earliest point who are the providers that are performing well and who are the providers that are underperforming. We also have provider improvement solutions that we can implement to support that. Our stories as patients are not static and every interaction that we have with the health care system is another piece of the puzzle that helps us personalize the care that we each need.

What measures can be taken to ensure patient safety while integrating AI technology? 

We are relying on AI responsibly applied for searching, extracting, and transforming information. We're not asking the technology to do any reasoning and, as I mentioned, we never deny any prior authorization requests using AI. The other thing that we do is we're always including our physicians who are on staff as key collaborators and building and training our models, and we have a compliance team to ensure that everything we're doing is compliant as well. As I mentioned earlier, our AI leverages patient history to reduce the required clinical assessment questions and attachments that prior authorization requests ask of providers. This not only saves providers that 28% of their time, as I mentioned, but it also ensures better accuracy and safety and can minimize gaming of the system.

Sometimes we see provider behavior in our data where it's clear that they're trying a number of combinations to answer these questions to get to a yes but, with our AI, we can remove the need for these questions because we're able to extract relevant information from patient's clinical histories, and so we get the truth through that AI versus relying on any type of gaming risk.

Is there anything else you would like to add that we have not discussed?

The only other thing I'll mention is that Cohere's solutions are geared for anyone in the health care system who is at risk. There's a number of primary care models that we see in the market where those providers take full risk for their patient populations and our solutions could be valuable to them as well.