4 Ways Natural Language Processing Will Help Payers Manage the Clinical and Financial Challenges of Long COVID

As of early this year, as many as 23 million Americans may have developed long COVID, in which symptoms persist four or more weeks after first being infected with the virus. The condition is likely to have additional long-term effects that are not yet clear. However, the U.S. has begun to obtain a glimpse of long COVID’s far-reaching impact on those who suffer from it - and the picture is rather disturbing.

The healthcare system has only just begun reckoning with the challenges posed by “long COVID”, a largely mysterious complication that affects between 10% and 30% of COVID-19 survivors and in some cases, delivers devastating consequences to their health, lifestyle, and finances.

As of early this year, as many as 23 million Americans may have developed long COVID, in which symptoms persist four or more weeks after first being infected with the virus. The condition is likely to have additional long-term effects that are not yet clear, according to the U.S. Government Accountability Office (GAO). However, the U.S. has begun to obtain a glimpse of long COVID’s far-reaching impact on those who suffer from it - and the picture is rather disturbing.

For example, one analysis cited by the GAO suggests long COVID may be responsible for over 1 million workers being out of the labor force at any given time. Separately, another study of nearly 4,000 long COVID patients found that 45% reduced their work hours. One Swiss study of COVID-19 patients found that 26% were still symptomatic after six months, with 55% reporting symptoms of fatigue, 25% reporting some degree of shortness of breath, and 26% reporting symptoms of depression.

For payers, long COVID may represent a looming storm of clinical and financial headwinds, as even properly diagnosing those who have developed the condition remains a challenge today. In the near future, long COVID may create additional difficulties for health plans as organizations develop treatment protocols for suspected long COVID patients, forecast future rates, and work to detect instances of fraud, waste, and abuse.

For all these use cases, artificial-intelligence-driven technology can help payers analyze messy, unstructured patient data in electronic health records (EHRs) to yield insights that enable payers to better understand and predict the needs of patients experiencing long COVID. Specifically, natural language processing (NLP) enables computers to “read” and understand text by simulating humans’ ability to interpret language, helping payers extract key insights from colossal amounts of unstructured EHR data that can be used to surmount clinical and financial barriers.

The (not-so-great) state of long COVID knowledge today
The first thing to understand about long COVID is that, despite researchers’ best efforts, we currently lack sufficient data for an in-depth understanding of the condition. For example, researchers do not yet fully understand the risk factors, causes, and effects of long COVID, according to the GAO.

Long COVID symptoms may include the more common COVID-19 symptoms such as fatigue and shortness of breath, in addition to more serious complications such as brain fog, heart palpitations, pins-and-needles feelings, and sleep problems.

In contrast to other post-COVID conditions that tend only to occur in people who have had severe illness, long COVID symptoms can happen to anyone who has had COVID-19, even if the illness was mild, or if they had no initial symptoms, according to the U.S Centers for Disease Control and Prevention (CDC).

Unfortunately, the federal government’s progress in studying long COVID has remained slow, drawing criticism from patient advocates and healthcare policy experts. As of March, the biggest federal study into long COVID had only enrolled 3% of its recruitment goal, more than a year after the National Institutes of Health received $1.2 billion for the effort. In response, the Biden Administration has promised to accelerate enrollment in the study and create a new task force to coordinate research into long COVID across federal agencies.

How NLP can help
To help begin filling in the massive gaps in knowledge and understanding of long COVID, many payers will turn to AI-based technologies like NLP. NLP automates expensive manual chart reviews, which sometimes require clinicians to wade through thousands of pages of documentation to pick out the tiny pieces of data they seek.

In giving computers the ability to read, understand and interpret medical language, NLP does far more than merely identify the presence of words or elements within text. By leveraging NLP, payers can organize data from an individual’s health journey, an entire patient population, or an enterprise's data warehouse.

Following are four ways that NLP can help payers manage long COVID patients:

  • Diagnosis: Because there is no universally accepted definition of exactly what long COVID is, there may be some confusion or dispute over whether a patient who has been diagnosed by a provider actually has the condition. Many of the symptoms associated with long COVID are similar to other conditions, such as chronic fatigue syndrome, mononucleosis, and fibromyalgia. Ultimately, we need a way to precisely define long COVID and its symptoms - and much of the information we need to do that lives in patient records as unstructured text that NLP can help identify and interpret.
  • Treatment plan development: Payers will face challenges authorizing treatment plans for long COVID patients, because some of the treatment approaches for these patients are highly experimental and risky due to the novelty of the condition. Traditionally payers have been understandably reluctant to authorize experimental procedures but may soon need to for the sake of long COVID patients. NLP can help payers scour patient records for evidence of treatment approaches that have been effective for patients with various long COVID symptoms.
  • Forecasting future rates: Due to the uncertainty and novelty surrounding long COVID, payers will be confronted with difficulties in forecasting future care costs for these patients. NLP will help payers to more efficiently extract accurate data to feed the predictive models that are used to develop rate estimates.
  • Detecting fraud: Given the lack of training and best practices associated with coding for long COVID today, it is virtually inevitable that providers will “upcode” some visits with these patients, in most cases, unintentionally. NLP can help payers spot these anomalies in patient records and then verify the occurrence of improper payments, billing problems, or clerical errors.

Because we still have much to learn about long COVID, the condition will likely present ongoing challenges to payers and healthcare as a whole. With NLP, however, payers have a vital tool to help them manage complexities by enabling the efficient extraction of critical clinical and financial insights.