May 2019

Industry Perspectives: Strategies to Survive the AI Revolution
By Dale Kivi, MBA
For The Record
Vol. 31 No. 5 P. 28

Clinical documentation and revenue cycle management (RCM) have always been achieved through an ever-shifting balance of human effort and technology. With the coming onslaught of artificial intelligence (AI) applications, however, that balance is about to dramatically tip, improving detailed data capture, consistency, and analytics while promising to reduce traditional labor efforts.

With that being the case, HIM professionals can—and need to—play a critical role in this transition to ensure process integrity, clinical and financial outcomes, and total process cost management. After all, it’s only in the HIM department where all these factors intersect.

AHIMA Focus Considerations
A few years ago, AHIMA predicted the future of the HIM profession would be in data analytics. As the industry makes a hard turn toward AI, that future is now.

Unfortunately, AHIMA’s Vision for Transformation and 2019 Strategic Plan announced efforts to discontinue products related to information governance, informatics, standards, and consumer engagement. Instead, it will focus on coding, clinical documentation improvement (CDI), the AHIMA World Congress, advocacy, certification/credentialing, privacy/security, and higher education.

Although this reimagined focus may better align AHIMA with the predominant (and growing) coding and CDI workforces that support its 103,000-member association, it is seen by some as short-sighted, given the inevitable market shift toward AI.

HIM directors have been known to lament being unable to participate in the technology decision-making process for their departments. Some have even been left out of workflow planning.

It can be argued that AHIMA’s shift in focus appears to throw in the towel on the possibility of HIM being involved in these decisions. Instead, it has opted to concentrate on coding and CDI workforce development. Unfortunately, it’s in those areas where AI promises to result in the greatest labor cuts.

As health care enters the AI transitional market, the most critical areas of need are informatics and standards, each of which was dropped in AHIMA’s strategic plan. Traditional HIM efforts in standards will need significant expansion in the AI world. As AI enables increased automation at the expense of traditional labor that considers more than just typical workflow routing, someone has to expand on quality expectations for each stage to prevent missed opportunities or downstream exception issues. Accordingly, HIM should be expected to define quality, not just fix what goes bad.

A recent example of improper informatics control and standards occurred at Providence Health of California, where an aggressive CDI effort led to a lack of discipline and a $188.1 million Medicare fine for filing unsupported secondary diagnoses.

If vendors’ automated programs do not offer enough billing recommendations, they show no value. However, if they offer too many choices, there’s the risk of following in Providence’s footsteps. AHIMA’s informatics discipline was dedicated to appropriately managing such issues.

AI’s Impact on Cost
The business process cost reduction expected from the AI movement can best be predicted by reviewing speech recognition’s cost impact on document creation.

Back-end speech recognition reduced the number of transcriptionists required and set expectations for lower total process costs. Outsourced vendors significantly undercut in-house costs. Once front-end speech recognition began to take hold, total costs for document creation as a business process fell to a fraction of what it had been. Consequently, suggesting a 60% labor force reduction over the past 15 years may be a low estimate; practitioners who remain earn much less.

AI vendors hope to have a parallel impact on the coding and CDI labor forces.

The vendor game plan is simple if not predictable. First, outsourced vendors chip away at in-house staffing models with lower labor costs (weaker wages, fewer benefits, and/or offshore teams). Then, technology vendors offer productivity improvement apps that promise to reduce total staffing levels but in reality may not reduce total process costs if additional quality assurance is required.

Once technology matures into true labor-eliminating apps, outsourced vendors bundle it to deliver dramatic savings that can’t be ignored. Vendor costs and labor rates are squeezed by leveraged technology that set expectations for lower total process costs. That’s what happened in transcription, and it’s about to unfold in coding and CDI.

In coding, production uncertainties during the ICD-10 transition created a seller’s market in which it was reported that nearly one-half of all hospitals signed with outsourced vendors. Enabled by that uncertainty, a few notable vendors signed contracts for $100 per-hour per-coder with no productivity guarantees, while technology vendors pushed their ICD-10 computer-assisted coding (CAC) wares to support productivity.

Labor firms, which offered higher wages and a $5,000 sign-on bonus, forced most in-house teams to raise their pay scales to retain staff while CAC technology—parallel to early back-end speech recognition offerings—improved productivity but didn’t necessarily lower total process costs due to performance limitations that necessitated increased quality assurance.

However, now that ICD-10 implementation is in the rearview mirror, production levels are back to normal and many facilities have eliminated the transition-inspired, artificially inflated outsourced “help” costs. As a result, vendors are forced to drop their rates to stay in business. And because in-house staff pay rates increased during the transition, entire departments are now ripe for the significantly less expensive outsourced vendor rates.

Outsourcing prices are falling fast, and formerly well-paid in-house production team members find themselves accepting lower-paying vendor jobs to stay employed. As AI apps prove to be true labor-eliminating tools, vendor pricing, pay rates, and total process costs will drop even further, just as they did when the transcription industry underwent the same transformation.

AI’s Impact on Quality
The most notable differences between productivity-enhancing tools (such as back-end speech recognition and CAC) and labor-eliminating AI (such as front-end speech recognition with NoteReader-type real-time coding queries) are the quality and consistency required of the applications.

With AI, any set of if/then human decisions should be programmed for faster and more consistent results. To achieve these goals, proactive quality measures such as NoteReader-type physician coding queries must be forced forward in the workflow process to eliminate downstream manual efforts and exception processing. Just like Six Sigma and Lean training suggest, quality verifications imbedded earlier in a workflow process improve output consistency and eliminate exception processing to improve quality and reduce costs.

Give the same five cases to five different inpatient coders and it’s more likely to result in different results than a complete consensus. This type of uncertainty highlights a problem. It used to be that manual transcription was considered a craft more than a production process. Slight variances were expected depending on who created the document. Today, coding is viewed much the same way, where even the slightest coding differences can have a significant impact on revenue.

In labor-centric document creation and RCM workflow schemes, there are a number of inherently manual steps that offer opportunities for variances that can impact clinical and financial outcomes. However, by properly automating those steps, potential discrepancies can be eliminated.

Automated solutions employing AI may help organizations monitor and improve key performance indicators (KPIs). It is in this realm that the true value of improved quality and consistency from AI will be measured.

AI’s Impact on Time Management
One of the most promising benefits of AI is how it can reduce the clinical documentation burden for physicians. A recent study supported by the American Medical Association found that primary care physicians spend nearly six hours a day on EHR data entry during a typical 11.4-hour workday. This extra time has often been criticized for contributing to physician burnout and a work-life imbalance.

AI tools for clinical documentation and RCM efforts instantaneously process countless if/then decisions exactly the same way, every time. Speedy resolution of if/then decisions—and, more importantly, the added coding and CDI questions they generate from incomplete or conflicting input—ensure improved quality and consistency.

By prompting the resolution of missing information or potential conflicts at the time of document creation, physicians resolve issues while the patient encounter is still fresh in their minds and discharged not final billed (DNFB) lags for what previously would have been coding department queries are eliminated.

These queries, which can add 10 to 15 days to DNFB for cases that, more often than not, are for the more complex, higher-fee encounters or stays, are where current workflow time difficulties create the greatest pain for both physicians and the RCM cycle.

Accelerating cycle times with AI relieves these pains and, as an important added bonus, will inherently deliver a measurable positive impact on cash flow.

AI Business Model Considerations
Labor costs typically eat up more than half of a hospital’s operating revenue, an allotment that is poised to increase as organizations face continued pressure to raise salaries. With current revenue cycle efforts being labor intensive, new technologies such as AI that promise to reduce staffing levels will inevitably prompt organizations to reevaluate their business models.

For AI solutions to be effective, they must be directly integrated into the EHR. The broader the scope of the chosen product, the greater the value it can deliver. Consequently, to maximize cost savings, the only AI-driven business model decisions will be if or when to outsource the labor portion of the equation and how to measure and control quality through the AI transition and beyond.

If coding and RCM efforts follow the transcription industry experience, the estimated in-house over outsource market preference ratio of 70/30 can be expected to flip within the next two to three years. The savings will be too big to ignore, especially after the wage inflation that occurred during the ICD-10 transition.

Also in parallel to what occurred in transcription, the greatest savings from outsourced coding and RCM service firms can be expected from vendors directly affiliated with the strongest technology. Because AI will continue to evolve, the amount of labor required and the role of the labor force will also change.

Because any savings they enable will reduce potential billable hours, firms not directly affiliated with AI technology can be expected to drag their heels on further adoption. Those who are closely affiliated with the technology will see those added savings as competitive differentiators.

The other business model decision to consider when outsourcing labor is deciding between onshore, offshore, or blended service offerings. Without getting into the typical emotional (keep it all in the United States) or financial (send it all offshore) perspectives of the debate, the emergence of AI—and the inherent quality and consistency advantages it promises—introduces a new variable into the discussion.

Recent industry news highlights both sides of the debate. One vendor has been getting a lot of mileage from a study it produced suggesting the extra cost to manage quality through its offshore partner actually increased its business costs.

At the same time, respected industry giants such as Tenet are aggressively moving to offshore more than 1,000 positions.

Both positions suggest that the biggest key to success when outsourcing is choosing the right partner, regardless of where they are located.

What’s Next?
Over the next few years, the AI revolution promises to change how coding and RCM efforts are achieved. How the transition rolls out and how workflow results are validated against current and future KPI targets are where HIM directors must be brought in to manage expectations for hospital leadership.

To survive the revolution and mitigate risks along the way, HIM needs to ensure step-by-step quality metrics are in place and AI-generated results are properly validated department by department before being rolled out to the entire organization. If not, everyday exceptions that could negate cost savings or overly aggressive system calibrations such as those experienced at Providence Health could make the investment problematic.

At the same time, it’s just as important to accept that the AI movement will reduce the labor required for coding and CDI, increasing the pressure to outsource those efforts. By staying focused on managing process integrity, clinical and financial outcomes, and total process costs and not just protecting the existing labor force, HIM is best suited to cost-effectively manage the change.

If it appears the primary objective is to protect staff, HIM won’t stand a chance. The profession needs to accept that the revolution is coming and assert itself as the ideal resource to navigate through it with the least amount of pain. In the end, the job is the same: the management of health information.

— Dale Kivi, MBA, is senior director of communications for Aquity Solutions.