Why artificial intelligence and machine learning are the future of coding accuracy

By David Fletcher
Vice President – Innovations, Streamline Health, Inc.

When checking out of a hotel recently, as usual, I found a copy of my bill slid under my door from the night before. The charges were itemized by date and department for my review, which meant that all I had to do was review them for accuracy, then stop by the desk in the lobby to confirm payment on my way out.

Imagine a patient leaving the hospital after an inpatient procedure having a similar experience: an itemized bill produced for their consideration prior to discharge. Inconceivable, right? It would certainly be a challenge, given the amount of documentation across multiple systems and the various departments and processes involved. A hotel simply needs to compile and present information from its various operations; a healthcare provider must compile that information, check it for accuracy and then translate it into complex coding that accurately reflects the patient experience. And then there’s the myriad clinical measures that need to be documented and tracked—Quality Indicators, Patient Safety Indicators, HACs and HAIs and more. Even if all the information was compiled and coded, reviewing it for clinical accuracy and revenue compliance is too time consuming— and too important— to rush through.

Ensuring coding accuracy is crucial to the optimization of reimbursements and mitigates an organization’s exposure to compliance risks. However, increasing accuracy within the current revenue cycle paradigm can be difficult due to resource and access limitations, disparate processes/systems and time constraints. It’s not humanly possible to achieve this mission critical function through strictly manual means. But what if a new approach, powered by progressive technology, made this seemingly impossible scenario a real possibility?

Accurate coding is challenging

Accurate charts must capture a complicated story that begins the moment a patient enters the facility and ends when he or she is discharged. Creating an accurate document of this encounter necessitates several data capture methods, from the natural language input of a doctor’s notes to the data entered into the organization’s electronic health record system, while also including many variables that track quality measures.

The data behind a patient’s encounter is managed via multiple systems and processes across different departments. Data updated by a team in one department is often left unchanged in their counterparts’ department. This increases the challenge of accurately leveraging data and the subsequent metrics it impacts.

“Accurate charts must capture a complicated patient story.”

This complex network of clinical data can be difficult to follow, yet code staff are tasked with translating this data into coding that accurately reflects the patient’s treatment and myriad elements that impact quality measures and revenue capture. When you consider the constant pressure to complete cases on time to keep bills flowing, it’s easy to see where accuracy is sacrificed for expediency. If you play it safe and undercode, you leave money on the table. Conversely, an aggressive strategy of overcoding may yield greater revenue, but you also risk payor compliance audits, revenue clawbacks and penalties. With either approach, financial performance will ultimately suffer.

Additionally, the final DRG and supporting codes that are submitted for billing will impact how each encounter is reflected in numerous metrics that measure performance and drive revenue. Quality Indicators (e.g., PSI, PDI, and IDI), HACs, and HAIs are just a few measures impacted. In an outpatient setting, professional fee billing, ACO performance metrics, and MIPS scores for MACRA are all influenced by the coding of the encounter.

As you can see, there are seemingly endless opportunities for coding and quality measure accuracy to go off the rails due to poor communication, lack of integration and other issues.

Interdependent interests

An accurately coded chart represents a series of complex interdependencies among different stakeholders of mid-revenue cycle management. Like a row of dominoes, a change in one element can cascade into a series of others but the siloed approach to managing these changes reduces visibility and collaboration between teams. In an inpatient setting, CDI teams are usually the first to focus on coding accuracy by establishing a working set of codes and coordinating activities to refine the clinical documentation that support the codes. At the same time, QI teams are looking for potential patient safety issues such as HAIs, and HACs.

Additional teams, like Case Management, Utilization Review and various HIM roles, also leverage this data for various purposes. This influx of activity makes concurrent review of the data (and subsequent coding) a seemingly impossible task. As a result, many organizations wait until after the proverbial dust has settled, the case is coded and submitted for payment, then they perform coding audits.

While certainly easier, post billing audits come with their own pitfalls. After a bill has been sent to CMS for reimbursement, organizations only have 60 days to correct errors and resubmit for the additional revenue they deserve. However, when it comes to overcoding for services, CMS has no limitation on how long it has to review and request any overbilled funds be returned, often with penalties and the specter of future audits.

Coding audits require a great deal of expertise, which is often in short supply. As a result, many organizations can only audit a small percentage of cases, usually chosen at random. This means the majority of revenue leakage and compliance exposure goes undetected.

For greater accuracy, begin with the end in mind

To address these issues, begin with the end in mind and develop a technology-enabled, concurrent and repeatable auditing system that can assess each case in real time and prior to submission for billing.  Such a system would offer the best of both worlds: the complex understanding of interrelated variables that comprise accurate documentation and coding, while also providing the bandwidth to review 100 percent of cases, not just a small sample, and do so prior to billing when there’s maximum opportunity for optimization.

For example: for each day of an inpatient encounter, the system would review the available data and estimate all elements of the patient bill, such as diagnoses, procedures, charges, the DRG, and all relevant performance metrics to determine what may need more attention. If attention is warranted, routing criteria can be established to push the case to the appropriate resource.

How would you develop such auditing technology? The answer is Artificial Intelligence (AI) and Machine Learning (ML), both of which “learn” from a team of expert auditors. The accuracy AI journey begins with over a thousand expert rules that are defined and validated by veteran auditors with experience auditing cases that represent virtually every encounter possible. Encounters with potential accuracy problems are flagged by one or more rules, then returned for review along with detailed, proscriptive advice on how to correct and complete.

Imagine a repository of auditing data from nearly 100,000 cases, including the original coding, the optimized coding and the supporting rationale. ML can be leveraged to identify patterns in the coding data that either triggered a recommendation to change the codes or not. Once patterns are identified, a multivariate algorithm can augment the expert rules to estimate accuracy of a given DRG and/or other metrics driving revenue and reimbursement.

As additional data becomes available, all rules leveraging it can be automatically reassessed and adjusted accordingly. And this powerful analysis can drive accuracy earlier in your revenue cycle for greater financial and operational results.

Optimized efforts and results

With such a system in place, the seemingly impossible challenge of concurrent assessment of coding and documentation is now possible. All respective metrics are estimated simultaneously and the impact of any proposed changes are visible to all stakeholders. Organizations can then stratify issues and assign resources accordingly, increasing consistency and predictability in their mid-revenue cycles processes.

While the process required to discharge a patient will never be quite as simple as checking a guest out of a hotel, we can dramatically improve the process of preparing, reviewing and optimizing the supporting data beyond the current model.

Greater accuracy in coding and quality measures that reduce exposure to compliance risk and enhance financial performance to fund your mission. That’s what you’ll receive from Streamline Health® eValuator™.