A 2006 U.S. National Institutes of Health study identified several major challenges in researching medical records, including assessing the quality of data and combining data from companies with dissimilar coding systems. (Medical coding entails extracting billable information from the medical record and accompanying clinical documentation.) But AI — specifically natural language processing (NLP), the subfield of AI concerned with language data — could offer a solution in its ability to automatically read, summarize, and analyze unstructured text, including medical records entries.
One of several vendors offering an AI platform for medical records analysis is DigitalOwl, which today announced that it raised $20 million in series A funding from Insight Partners. Based in Portland, Maine, DigitalOwl claims its software can help insurers, reinsures, and lawyers to better combat fraud, underwrite health plans, execute claims, and build stronger legal cases.
Analyzing medical records
The global market for big data analytics in health care was valued at $16.87 billion in 2017 and is projected to reach $67.82 billion by 2025, according to a recent report from Allied Market Research. It’s believed that health care organizations’ implementation of big data analytics might lead to a more than 25% reduction in annual costs in the coming years. Better diagnosis and disease predictions, enabled by AI and analytics, can lead to cost reduction by decreasing hospital readmission rates, among other factors.
DigitalOwl, which was founded in 2018 by brothers Amit Man and Yuval Man, leverages a proprietary engine to extract information from hundreds to thousands of pages of electronic medical records. The data is presented chronologically, allowing users to search and filter by condition, date, body part, body system, and provider. The complete history is contained within a PDF — every condition, date, and entry is clickable, taking users to the source of the information in the record.
“The need for medical document analysis has gone far beyond classical NLP. Detecting medical entities is not enough to provide solutions to our clients. There is a need to extract the narrative from the case, separate the key findings from the noise, find relationships, and align it with medical and insurance knowledge,” Man told VentureBeat via email. “As the pandemic has forced employees to work remotely, companies have had to adopt new technologies faster than ever just to keep their employees productive. This has generally had a positive impact on attitudes towards technology in some otherwise traditional business models, like insurance.”
Prior to cofounding DigitalOwl, Yuval Man was a part of Israeli law firm EKT. Amit Man led the algorithms and core technology group at Briefcam, a computer vision startup, before founding a firm called Takes and joining assistive device company OrCam.
By extracting medical information from the records submitted to the platform, the Man brothers say that DigitalOwl can provide a “focused summary” of data points with a streamlined navigation system. For one carrier, the company claims to have identified an illegitimate disability claim worth $150,000. For another, DigitalOwl reportedly saved $270,000.
“The DigitalOwl technology solution extracts … medical data points with extreme accuracy, including all types of cancer, heart disease, accidents, orthopedic surgeries, diabetes, arthritis, hypertension, brain injuries, and many more,” the company writes on its website. “DigitalOwl has recently reached a milestone of more than 30 million pages of medical records processed by [our platform.]”
DigitalOwl claims to serve “multiple” customers in Israel, Canada, and the U.S., including insurance carriers, third-party administrators, and record retrieval companies.
Potential drawbacks
It’s worth noting that health insurers are in the early days of adopting AI. In a 2017 survey of the German insurer market, McKinsey found that “only a few health insurers” were leveraging machine learning technologies, owing to uncertainty about practical use cases, gaps in technology expertise within organizations, and a lack of transparency regarding the available data.
And for all of NLP’s potential, the technology can be susceptible to the biases in the datasets used to “teach” it to find certain patterns in documents — including medical records. Studies have revealed a number of biases that can crop up in health records, including derogatory mentions of Black patients with “negative” characteristics and discrimination against those with sickle cell disease. Medical records also contain stigmatizing language, which can express approval, but also disapproval and stereotyping.
As Wired’s Tom Simonite pointed out in a recent piece, skewed datasets are the norm in health AI research due to historical and ongoing inequalities. A 2020 Stanford paper found that 71% of data used in studies that applied AI to U.S. medical data came from patients in California, Massachusetts, or New York. Another study, published last year, examined more than 150 systems using AI to predict diagnoses or courses of disease and found that that most “are at high risk of bias.”
The insurance industry isn’t immune. A 2019 study found that an algorithm used by insurers to identify which patients will benefit from “high-risk care management” programs selected fewer Black patients than white patients, denying Black patients access to specially-trained nursing staff and extra primary-care visits for closer monitoring.
“Machine learning algorithms have the potential to improve medical care by predicting a variety of different outcomes measured in the electronic health record and providing clinical decision support based on these predictions,” the coauthors of a 2018 article on AI bias in health care wrote in JAMA Internal Medicine. “However, attention should be paid to the data that are being used to produce these algorithms, including what and who may be missing from the data. Existing health care disparities should not be amplified by thoughtless or excessive reliance on machines.”
In response, Man said that DigitalOwl’s models have been “toughened” over time by ingesting “tens of millions” of medical documents. To teach the models the relationships within the documents, each page was labeled by a team of specialists “knowledgeable in healthcare and insurance business logic.”
“With the help of neural networks, understanding, summarization, and even prediction are becoming possible. Standardization and normalization of unstructured information are equally important to enterprises. Data from different vendors and providers can now be merged, compared, and used by the systems of an organization,” Man added. “Frankly, companies have underinvested in these areas while focusing their digital efforts on sales and customer service for many years. But the trifecta of staffing challenges, COVID, and escalating costs is now driving these organizations to adopt technology solutions at a faster pace than ever before.”
Beyond Ibex, Fusion LA, Menora Mivtachim, and Mobileye founder Amnon Shashua are among 50-employee Digital Owl’s backers. The company currently employs more than 40 people across offices in Israel and the U.S.; the latest funding round brings its total capital raised to over $26 million.
Source: Charles Taylor
