DTR workflows that work: The case against clinical questionnaires

Do you ask the auth submitter a series of clinical questions, or hope their EMR supports CQL for adjudication? Neither achieves the actual goals of CMS-0057-F… but a better option exists.

Years before CMS-0057-F taught the industry to speak regarding CRD, DTR, and PAS, Cohere was already digitizing the prior auth process to embrace the benefits of interoperability and automation. A key portion of that digital workflow – now referred to as DTR (Documentation, Templates, and Rules) – was ensuring the medical service being authorized was evidence-based in accordance with the health plan policy. To accomplish this, Cohere started down the path that the industry is following today: converting those policies into a set of clinical questions for the prior auth submitter to answer.

On the surface, this is the most direct way to ensure a service meets clinical policy requirements, especially since many EMRs are not sophisticated enough to support CQL (Clinical Quality Language) for prior authorization support. (To date, CQL has been leveraged mostly for risk adjustment or quality measure use cases.) You can leverage FHIR to ask the auth submitter to verify each necessary condition is met, perhaps using some automation or even AI to knock out the easy ones. Then, you can automate the approval on the spot. You’re off to the races.

Then we started auditing the results – what we discovered shocked us.

Over 30% of clinical questionnaires in the prior auth workflow yielded inaccurate responses.¹

Just let that sink in. Nearly one in three decisions is based on potentially incorrect information. It was a wake-up call and a hard pivot point. We no longer promote this method of verifying medical necessity (although we can support it with our solutions, as many health plan workflows are built on these questionnaires). Years later, with the DTR requirement of CMS-0057-F adding urgency to the work, we see much of the industry starting down this path. It’s time to raise the alarm.

The case against clinical questionnaires

While FHIR-based questionnaires seem to be a direct method of obtaining the exact clinical information needed to satisfy health plan medical policy, they can undermine the process and introduce critical flaws into the workflow:

They increase provider admin burden
The goal of CMS-0057-F and similar legislation is to improve the provider experience by reducing administrative burden, yet these questionnaires do the opposite. They force the auth submitter to look through a case and respond to a variety of medical necessity questions, taking up valuable time. If we aim to reduce the 13 hours per week that the average provider spends on prior auth,² this is not the way to accomplish it.

They incentivize “gaming” the system
Who submits these auth requests? Generally speaking, it is not the clinician who meets with your member. In many situations, these are non-clinical staff working in a back office. In other words, it’s an unfair position to place non-clinically trained staff in, as they’re often not equipped with the context or expertise to answer these questions accurately.

40% of providers hire administrative staff who work exclusively on PA.³ According to the AMA, even with full-time employees dedicated to the task, the volume is overwhelming, averaging 39 requests per physician per week.⁴ Faced with this pressure, it’s understandable that staff might make mistakes or provide inaccurate responses, even when they do their best to ensure approval.

Here’s where we should be concerned. When incorrect responses enter a member’s health record, the risks go far beyond a single prior auth. These responses can lead to care decisions that are not evidence-based and potentially unsafe. 

For example, spinal fusion surgery might be approved because a clinical assessment question was completed “successfully,” but if the member is a smoker (and that was inaccurately reported), that omission could lead to serious patient complications. Smoking significantly increases the risk of pseudoarthrosis – a condition where the bones fail to fuse properly after a surgery, resulting in a non-union, which can in turn result in pain, reduced mobility, and additional surgeries. Literature recommends that patients go through smoking cessation and be nicotine-free for 6 weeks before a spinal fusion surgery. 

The above is just an example. Once inaccuracies resulting from the clinical questionnaire workflow are in the record, they’re often treated as fact. This can follow the member for the rest of their life, contributing to various downstream problems in their lifelong treatment and care management.   

After seeing our audit results, Cohere shifted away from using questionnaires whenever possible, investing heavily in a more reliable alternative approach. CMS-0057-F’s DTR requirement doesn’t mandate that payers surface clinical questions in every case—nor does it require an impractical, broad-scale CQL configuration across your entire clinical scope. You do need to demonstrate the ability to surface clinical questions and show that you can support pre-filling them with CQL (which Cohere can help you do). But there’s a better, more efficient way.

There is a better option

The DTR API doesn’t have to be used to serve up a series of clinical questions about the member (or rely on CQL). Instead, it could ask for the relevant member’s health records, clinical notes, and other attachments relevant to the auth. Once submitted through the API, the health plan can deploy AI and ML technology to extract clinical information directly from those attachments – the source of truth.

Doing so has several advantages:

⏱️ Time savings. We’ve seen that removing clinical assessment questions from the prior authorization process reduces submission time for providers by 21%.⁵

📈 Improved quality: By extracting clinical data directly from the member’s health records with highly trained machine learning models, we significantly improve the accuracy of the submission.

❌ No room for gaming: Bypassing clinical questionnaires eliminates the potential for altering the record to obtain prompt approval.

Here’s what it takes

The DTR API can be configured to solicit an attachment, much like it can serve up a clinical question. The difficult portion of the task is leveraging AI to comb through the attachments, clinicians’ notes, and structured EMR data (if available) for clinical data that maps to indications in the medical policy.

This requires:

• Converting health plan medical policy to code – not just chunking out a PDF into a set of questions. This also requires expertise in mapping procedure codes to policies, automating policy hierarchy, ongoing management, keeping up with CMS and organizational policy updates, and more.
• Extensive training and refinement of ML models – validated by expert clinicians to ensure high-quality, evidence-based standards are upheld. NLP tools will also be needed to read and digitize specific documentation.
• Advanced decisioning capabilities – able to configure business rules and health plan medical policy to automate as many approvals as possible and support quick access to care.

It sounds like a lot – and it is – but Cohere was built for this challenge. As a unique organization with automation experts and clinicians running UM operations in-house, we have a unique advantage when training our approximately 350 ML models to validate clinical indications. Our team has spent years codifying health plans and CMS policies, building an extensive content library and management structure to drive evidence-based automation. We’ve leveraged DTR to collect clinical information to support over 2,700 policies for over 6 million PAS submissions.

And getting back to the goal of CMS-0057-F, we’ve made a measurable positive impact on the provider experience. Our solution has earned a 93% provider satisfaction rating and a 67 NPS among providers. Removing or limiting the use of clinical questions has helped to build that positive experience.

Final thoughts

We’ve said it before: building prior authorization APIs is not the core of the challenge – the real challenge lurks beneath the surface. Health plans must configure the logic and content to be compliant and support effective day-to-day prior auth operations. CQL is only supported by EHRs in limited circumstances. While clinical questions may seem like a viable path to compliance with CMS-0057-F, they have the potential to undermine care quality and impede ongoing care management. We know – we’ve been there. We don’t recommend it.

Cohere has technology and clinical experts working with health plans to deliver a compliant prior auth solution supporting high-quality, efficient day-to-day operations. To learn more, contact us for a demo.

¹Cohere internal quality audits, 2021

^2-42024 AMA prior authorization  physician survey

⁵Cohere clinical questionnaire analysis, July 2024.