Asset 1
Fast, Efficient, and Effective Utilization Management

Utilization Management Challenges: $10 billion opportunity 

According to the CAQH Index, the healthcare industry could save $10 billion through technology enabled claims processes with the largest opportunities being eligibility verification ($4.2 billion) and claims status ($2.1 billion). Among all utilization management activities, prior authorization is the most challenging to ensure speed and efficacy.

Challenge of Speed and Efficiency in Prior Authorization

Payors and providers recognize and attest to the need to improve prior authorization in recent joint consensus statements. 85% of physicians find prior authorization interferes with continuity of care — specifically creating significant delays in delivering necessary treatment and even causing patients to abandon necessary treatment (Figures 1 and 2; AMA 2019). 

Figure 1. Delays for Necessary Care from Prior Authorization

Figure 2. Abandoning Treatment from Prior Authorization 

Time to Treatment Matters – Compliance & Quality

Timeliness reviews increasingly factor into CMS program audits for Organization Determinations, Appeals and Grievances (ODAG) and Coverage Determinations, Appeals and Grievances (CDAG) for Medicare Part C and D, respectively. Almost every health plan passes CMS audits and meets state requirements. Many adhere to the National Committee for Quality Assurance (NCQA) Utilization Management Accreditation Standards for Timeliness. While certainly important, these benchmarks do not represent high performance in decision speed.

Time to treatment matters more than just compliance. Certain nonurgent treatments subject to preservice decisions frequently can show substantial effects in clinical outcomes – even mortality risk – when time-to-treatment initiation improves by even just 1 week.

Solutions in the Current Paradigm Are Neither Obvious nor Easy

Efforts to alleviate documentation burden have limited effects. Gold card programs – where clinicians who consistently achieve high standards in guideline adherence earn reduced documentation burden for a time period — don’t necessarily work as the variation in this performance is frequently random and unpredictable. 

“While deep learning thrives on inputs and outputs, most of medical practice defies straightforward algorithmic processing. For these ‘clinicians without patterns,’ AI presents adjunctive opportunities, offsetting certain functions that are more efficiently handled by machines.” – Dr. Eric Topol, Deep Medicine

Due to the tremendous complexities from case to case, solutions under the current paradigm are neither obvious nor easy. Simple rules-based prior authorization and UM was effective at first but as the cases and treatments involved get more complex, the greater manual effort required and less clear efficacy results. 

The industry is also challenged by a lack standardization in forms, formats and evaluation criteria. According to a McKesson survey of 23 health plans, there was commonality in approval standards for only 8% of over 1,300 procedures. In a recent study of cardiovascular medications, approval rates varied widely across pharmacy benefit managers for the same drugs and patient populations (Navar et. al. 2017). 

Traditional prior authorization processes aren’t always reliably effective in reducing costs – let alone the comprehensive value equation that considers clinical outcomes and long-term member health. While early efforts were successful (e.g., 2010-2015), recent studies indicate diminishing returns and a gap between the goals of cost efficiency and utilization management (Table 1).

Table 1. Prior Authorization Effectiveness – Examples 

StudyDisease AreaResults Unique Characteristics
Lee et. al. 2020Outpatient CardiovascularAdministrative costs exceed medical cost savingsTraditional 
Myers et. al. 2019Cardiovascular MedicationsPatients rejected or abandoning treatment during prior authorization had significantly higher rate of cardiovascular events Traditional 
Wallace et. al. 2019Infusible MedicationsIncreased wait times and denials, although eventually approved negatively affected patient safetyTraditional 
Keast et. al. 2018Hepatitis-CDecreased pharmacy-related costs Enhanced Care Management 
Newcomer et. al. 2017Chemotherapy$5 million in savings and increased adherence to national guidelines Technology-enabled decision support process
Margolis et. al. 2010 Medications for painful diabetic neuropathy or postherpetic neuralgia Utilization reduced but no effect on total healthcare costsTraditional
Levin et. al. 2010Diagnostic imaging Utilization and costs reduced Telephonic UM 

Significantly, the only recent studies showing substantial gains in prior authorization were coupled with technology-enhanced processes or robotic process automation (RPA). For example, Newcomer et. al. (2017) report deploying rules-based decision support forms that automated prior authorization review. 

Humana’s HealthHelp, CoverMyMeds and SureScripts are excellent examples of electronic-based prior authorization programs that can serve as a foundation for applying NLP and RPA to take these efficiencies to the next level. Nevertheless, only 12% of prior authorization transactions occur electronically end-to-end (CAQH 2019) and many payors may feel trapped by these established, legacy processes

Change the Game: AI & NLP to Drive UM Efficiency 

Data Transformation – AI & NLP Converts Manual to Electronic Without Changing the Process

With the right AI & NLP partner, payors can transform the supposedly limited, “manual” data from telephone calls and faxes into usable, machine-readable electronic data. At Digital Reasoning, our audio analytics automatically transcribes phone calls inexpensively and accurately while applying machine learning models to derive valuable insights. With this approach, a payor doesn’t have to convert their existing systems or try to change the behavior of providers in their networks.

Can You Predict Denials & Authorizations?

Human language provides rich data to infer and predict meaning. Beyond key words, machine learning models that understand the context of sentences and phrases can mimic the same review and decision-making performed by UR nurses.

Digital Reasoning has applied proprietary NLP and machine learning models to UM conversations and documentation in order to calculate a denial probability on member claims cases using a scale of 0.0 for denial to 1.0 for authorization with high confidence. Our dataset included thousands of claims and UM reviews in appeals from four large payors (United, Aetna, Humana and Cigna).

Figure 3 shows the data pipeline analytic approach. Rules models are important components of Payor Intelligence as they can classify and match the types of claims to the best models and downstream workflows such as guidelines and pathways.

Figure 3. Denial Probability Pipeline – Payor Intelligence AI 

Figure 4 illustrates the potential efficacy of predictive modeling through a t-sne scatterplot visualization of high dimensional data. Each dot represents a case with sometimes thousands of words. The t-sne transforms the collective difference in the case’s resulting prediction of denial vs. other (typically approval) into a two dimensional scatterplot. The result is a clearly distinguishable variation in the human language involved in a case with documentation that results in denial versus cases with documentation that result in another outcome. 

Figure 4. Language Features in Utilization Review Documentation – Denials vs. Approvals

As one might expect, as the UM process moves forward, cases in later stages have far greater documentation volume and therefore far higher confidence and reliability in the model’s prediction. As we scale this Payor Intelligence across the health plan, new models and predictive features specific to disease categories, conditions, symptoms and treatments will emerge that create further specificity, model efficacy and valuable efficiencies.

Outcomes that Matter 

Supercharging the prior authorization process and leveraging all data for a continuously adjusting predictor of approval or denial across the UM process delivers an exponential ROI for payors in the following ways:

  • Reduces friction between payor, physicians and members;
  • Improves provider and member experience and satisfaction (e.g., HEDIS / STARS);
  • Ensures compliance with regulatory requirements for timeliness; 
  • Decreases time spent on decision-making for clear cases, enabling greater attention to cases with clinical ambiguity; and
  • Decreases time spent on medical reviews while maintaining run rates 

Now that we’ve explained the ROI for AI & NLP in Utilization Management, the next article will explore in greater detail the substantial ROI opportunity in Care Management.

Written By
Paul Clark

Paul Clark is our Director of Healthcare Research. Prior to joining Digital Reasoning he served as Vice President of Research & Education at The Health Management Academy, leading the research agenda and educational development for C-Suite executives at America’s leading healthcare systems.