The Medical Characteristics of an Auto Accident Injury Claim

DigitalOwl’s proprietary Natural Language Processing (NLP) platform has processed tens of millions of pages of medical records. The extracted medical data points from attending physician statements (APSs) and medical records provide us with a unique view of the medical characteristics of these claims.

What have we learned from analyzing and summarizing millions of pages of medical documents? In this article, we shed some light on the medical characteristics of the average auto accident injury claim.

Medical Extractions: The specific medical information contained in records

What is a medical extraction?

A medical extraction is a single data point of medical information—each appearance of a medical condition, procedure, or medication in the medical record is an extraction.

‍

But, a medical data point is more than a plain medical term. One data point can have multiple meanings based on the surrounding text.  For example, a condition can be a past condition, negated condition, or a family condition. A procedure can be a past procedure, referral, or something that has only been considered.

The average claim holds 1,299 medical condition extractions (like Fracture, Dizziness, Hearing Loss, etc.), 1,153 procedure extractions (X-ray, MRI, Spine Surgery), and 499 medication extractions.  The average time range of medical information that has been summarized is 18 years and four months per claim.

The average claim holds almost 3,000 medical data points spread along more than 18 years of medical history!  Reading and understanding this much data is very difficult without the assistance of technology.

‍

Unique Medical Extractions: The medical story of a claimant

What is a unique medical extraction?

A unique medical extraction consolidates medical extractions with essentially the same meaning. A unique medical extraction can be a group of multiple dates/appearances of the same extraction, a group of various extractions that mean the same thing, or a combination of the two.

For example, a patient went to his physician three times. On the first visit, the physician mentioned a “limitation of movement,” on the second a “limited range of motion,” and on the third, a “problem moving freely” = three extractions (limitation of movement, limited range of motion, problem moving freely) = one unique medical extraction (movement limitation).

The average claim holds 139 unique condition extractions, 76 unique procedure extractions, and 50 unique medication extractions.

Of the unique conditions, on average, 10.25 are “high severity” risk factors, and of the unique procedures, 2.31 are “high severity.”

‍

The Medical Categories: The high-level summary

What is a medical category?

‍

A medical category is a higher-level summary than a unique medical extraction. A category holds a group of medically-related unique extractions. For example, Rib fracture, Knee sprain, and Arthritis would be grouped as “Musculoskeletal conditions.”

The average claim holds 48 categories, of which 7 are considered “high severity.”

‍

‍

Most common medical categories by severity:

High: Psychiatric conditions, Joint disease, Spinal procedures

Medium: Physical therapy, Hospitalization, Joint pain

Low: Limitation of movement, Abnormal breathing, Dizziness

The number of medical data points of the average auto injury claim

Other Statistics

On average, a medical condition will appear 9 times along the medical record.

On average, a medication will appear 10 times along the medical record.

34% of the medical data points can dramatically affect the claim (all high and medium severity extractions).

‍

88% of claimants had at least one physical therapy session.

63% of claimants had a psychiatric problem related to the accident.

43% of claimants refused at least one recommended treatment related to the accident.

28% of claimants suffered a whiplash injury.

18% of claimants were diagnosed with a herniated disc.

‍

Summary

Auto injury claims and their associated medical records contain thousands of medical data points.  Some are important and relevant, and some are not.  To understand this data and make the best decision possible, claim analysts should lean heavily on technology to quickly and easily sort the data and highlight the most important information.

Humans are good at making decisions, but only when the right data is made readily available.

Leveraging advanced Natural Language Processing (NLP), DigitalOwl has built a unique, proprietary engine that addresses a long-pressing need from insurers to better predict risk and claim outcomes. Using a state-of-the-art NLP engine developed specifically for medical records, the system extracts relevant medical information that is necessary for assessing risk or managing a claim in a fraction of the time of manual review by a human. A deep understanding of the extracted text creates a focused set of medical data points in a meaningfully summarized, easy-to-navigate format, freeing up valuable resources to focus on making decisions and other critical business activities.

The high level of accuracy (trained to be 97% plus) leads to more precise risk selection and claim management and thus better risk results, all while reducing costs and improving productivity, cycle times, and consistency.

‍