There is a spectrum of automation in utilization management. The most common type of review is manual review—it involves no automation and plan UM staff manually review PA requests against clinical guidelines.

Two forms of automation that are growing in popularity are automated review and gold carding. With automated review, plans will pre-populate provider portals with algorithms that immediately approve or deny the documentation that providers submit based on the algorithm’s criteria. This method does not require AI, or specifically machine learning, if each algorithm is set up individually.

Many plans have also started gold carding pilots to automate prior authorizations at the provider level. With gold carding, providers with a high approval track record get their prior authorization request automatically approved. As one example, Texas passed a gold card bill in 2021 to mandate gold carding for providers who have a 90% or higher approval rating.

As mentioned above, nowhere near all PAs are processed through either of these automated methods, let alone the more advanced automation that is available through AI. AI-automated review is when computers use AI and machine learning to learn which PA requests to approve or deny based on past approvals. With machine learning, each algorithm does not need to be coded and updated individually.

There are two main reasons why we, as an industry, are progressing extremely slowly from the manual review side of the spectrum to middle and right side of the spectrum.

Read more on these reasons below.

Creating automation algorithms is not easy because there are a lot of codes in health care for each diagnosis and treatment. Often, the approval rules must be created for every possible scenario. For an overly simplified example, if a patient has X test with a diagnosis of Y, then they can get automatic approval for treatment Z. These approval rules need to be created for every possible scenario and there are a lot of diagnoses, treatments, and payers so creating these models is a giant feat. Then, imagine updating these algorithms throughout the year. It makes sense why some plans have started looking to AI-powered solutions instead, but of course that is not easy either.

Some plans, often large ones with strong tech development departments, can create their own machine learning tools. For example, we spoke to one plan that created a machine learning tool that pulls out key phrases from provider documents based on clinical decision-making criteria. The tool continues to learn which phrases to pull out, for the UM nurse to read, leading to faster (although not instantaneous) reviews.

Most plans, though, don’t have the time or resources to build these tools in-house, and are deciding to work with vendors instead. Below are four examples, in alphabetical order, that we have heard from plans:

• Apixio’s Apicare AuthAdvisor analyzes providers’ historical PA decisions with machine learning to automate PA decisions. They boast a 50% reduction in manual reviews.
• Cohere Health creates episode-specific care paths based on historical claims data, then physicians can request bundled authorizations for recommended care paths. Cohere claims 15% incremental medical servings on average and 63% reductions in denial rates.
• Machinify’s Patriot evaluates the PA request and medical record against the plan’s policies to reduce review time and increase accuracy. With one plan, Machinify automated the approval of over 50% of requests and decreased average turnaround time by 50%.
• OliveAI automates prior authorization approvals through AI-powered clinical reviews. In their partnership, Florida Blue has seen a 10-day reduction in time to decision and a 27% decrease in unnecessary PA requests.

So far, we have been talking about automating PA review for plans’ administrative sake, but we can’t make only plans’ lives easier, because providers also struggle with the administrative burden of PA submission. According to an American Medical Association survey, 88% of physicians describe the burden associated with PA as high or extremely high.

Like with PA review, there’s a spectrum of options with PA submission as well. And like with PA review, we as an industry are still mostly stuck between manual processes (fax and email) and basic technological processes (provider portal submissions). It’s not possible to automate PA review if 45% of providers say they still use fax always or often for medical PA submission.

Providers, and their staff, share that sometimes it’s easier to fax or phone in PA requests than to learn how to use 10+ different plan-provider portals and then spend time on data entry between patient visits. In response, most plans are investing in their provider portals to improve its user-friendliness. For example, select plans have created a single documents field in their PA request form so that providers can drag and drop all relevant documents at once rather than uploading them one by one into the correct fields. One plan we talked to this year used to be entirely fax- or phone-based for PAs but, within a year of implementing their portal, they now get 44% of all requests in the portal.

Moving further in the spectrum, plans and providers could use EHRs for PA submission rather than fax, phone, or provider portals. When providers submitted ePAs through their EHR in AHIP’s Fast PATH initiative, they saw a 9% reduction in time to PA decision.

On the far right of the spectrum, the plan could in theory pull utilization data from the provider’s EHR so the provider doesn’t need to submit anything at all. But not all providers are willing to open their clinical records to plans. There is skepticism amongst providers because they think plans might use this information to restrict care or decrease reimbursement rates. Often, plans are only able to create these bidirectional data partnerships with providers in a risk arrangement or ones that are integrated. This is one reason that interest in Health Information Exchanges (HIEs) is high—over 45 states have statewide HIEs, with California launching theirs this year.

We are still only touching the tip of the iceberg for what AI could do for prior authorizations and have high hopes for where the industry could head in the next decade. In an ideal world, health plans could access providers’ records in real-time directly through their EHR, so the provider never needs to submit paperwork—manually or digitally. Then, plans could use machine learning technology to automatically approve or deny requests without constantly re-writing algorithms.

Until this day comes, plans must continue to improve PA review and submission by automating reviews (in-house or through vendors), investing in provider portal improvements, and collaborating with providers through risk arrangements and HIEs.