E-News Exclusive

Five Ways to Use NLP for Risk Adjustment Coding

By Emmy Johnson

Historically, natural language processing (NLP) platforms don’t produce reliable results for risk adjustment programs when compared with manual coding. Concerns with NLP center around invalid codes and resources needed to reject these or even manually recode charts. Early testers of NLP platforms also found it was more expensive than traditional manual coding.

However, NLP is turning a corner thanks to significant investments in platforms that address the complexities of risk adjustment. The clinical rule sets are much more refined, and machine learning and artificial intelligence components can process millions of charts to a higher level of specificity while also understanding context, making correlations, and refining data extractions. NLP vendors are developing business rules to meet the specific requirements of risk adjustment for Medicare Advantage and the Affordable Care Act. Implementing NLP can also lower coding costs per chart and accelerate the time from chart retrieval to final data output.

What’s the best way to embed an NLP-enabled process into a risk adjustment program? We often think the best application is within the first-pass coding process. There is nothing wrong with this approach, but there are other ways to access the benefits of NLP quickly and without having to embed it into the first-pass coding.

Here are five ways to make the most of NLP to run a more targeted, efficient, and accurate risk adjustment program.

No Hierarchical Condition Category Charts
Typically, you must code a chart to know whether there is a hierarchical condition category mapped diagnosis code in it. However, NLP can tell you before a coder spends time in that chart. You may find that you don’t need to code those charts after NLP has run, or you may decide to code a sample of the charts to feel comfortable with the results of the NLP run.

Chart Segmentation
With actionable data from an NLP run, you can think less about the traditional shotgun approach to coding and take more of a sharpshooter approach. For example, if you have coders who specialize in cancer coding, charts that have suspected cancer codes are funneled to those coders. Charts with more “vanilla” codes can go to your new or less-specialized coders.

Chart Value
Running NLP before coding also provides better information about the possible financial value of each chart, allowing you to decide whether you want to treat charts differently based on financial factors. High-value charts can be run through multiple passes to mitigate the increased regulatory risk associated with those charts. This information can also be used to sync the coding of your high- and low-value charts with submission deadlines.

These are just a few ways in which you can drive value with NLP without buying an instream solution.

Coding Accuracy Review
In a traditional accuracy review—otherwise known as a second-level review or an over-read—you have another coder or auditor take a fresh second pass at the chart. While this process can increase your overall chart accuracy, it also introduces a second opportunity for human error and leaves you with the problem of what to do with the unmatched codes from the various passes.

In an NLP-enabled accuracy review, the technology runs on the charts before the second manual validation. These NLP results can be matched against the first-pass results to find the variances. Now, you need to run manual coding validation only on the new codes from the NLP run and the codes the NLP didn’t uncover. This has several benefits, including the following:

• Including a coder review of the unmatched codes increases your accuracy and decreases compliance risk.
• The manual coding review is being done on fewer codes, making it more efficient.
• Usually coupled with an indexed and highlighted chart, the review of NLP codes is much faster than a traditional review.

The net effect results in higher coding accuracy rates at a lower cost than a traditional over-read.

Risk Adjustment Factor Accuracy Review
A coding accuracy review is at the chart level, while a risk adjustment factor accuracy review is at the member level. Risk adjustment factor involves adding data sources such as charts from other programs, a risk adjustment processing system, claims, or other data. This kind of review drives close to 100% accuracy at the member level and has deep benefits, including the following:

• Most codes found are validated from multiple sources.
• Member conditions are effectively matched to financial reimbursement.
• It can be executed at a meager cost per chart.

Even when used as a standalone process, NLP can enhance your risk adjustment programs by bringing significant efficiency and accuracy to your coding operations.

— Emmy Johnson is senior vice president of coding and abstraction at Ciox.