Over the course of a year, the average patient will generate roughly 80 megabytes of clinical data, primarily through medical imaging results and electronic health records. It is also estimated that around 2 million scientific papers are published annually, with researchers often struggling to keep up-to-date with the latest literature. In a single week last year, 4000 papers were published on Covid-19, each with potentially vital clues on how to control the virus.
While access to such vast amounts of data may seem like a good thing, it is of little use unless it can be properly analyzed to gain insights.
In recent years, natural language processing has been leveraged to great effect to help solve the problem of unstructured medical data, with NLP techniques being employed to parse information and extract critical insights from a variety of sources.
Here we take a look at just 5 of the use cases of NLP in Healthcare.
An Electronic Health Record (EHR) is a digital version of a patient’s medical history, maintained by their provider over time. It contains key administrative data such as the patient’s demographic and what medication they are on, as well and notes on their medical history; information that is vital at the point of care.
As it stands, Physicians spend large portions of their day populating EHR’s with information stored as free text. Despite the laborious nature of this task, the notes are not easily accessible or structured in a way that can be analyzed effectively by a computer, making it difficult to summarize a patient’s condition on arrival.
NLP is increasingly being leveraged to process unstructured data from medical records, allowing analytics systems to interpret them more easily. Once converted into structured data, health systems can classify patients and summarize their condition on arrival. Rather than wasting precious time reviewing complex EHR’s, NLP allows physicians to extract critical insight.
Platforms like SyTrue’s NLP Operating System enable Doctors to interact with medical records, removing the need to hunt for key observations and allowing them to focus on patient care.
Researchers are increasingly under strict time constraints when it comes to drug discovery, something that was evident in the race to find a vaccine for the novel coronavirus.
The first step in the drug discovery process is identifying the biological origin of a disease, a task requiring a comprehensive understanding of the genes involved, often starting with a detailed review of existing literature. A variety of inputs, from medical journals to patient records, has led to an abundance of data with processing tools struggling to keep up.
NLP is giving researchers a much-needed head-start in the drug discovery process, allowing them to quickly learn about similar diseases by extracting information from unstructured sources. In late 2019, AI-platform BlueDot identified a cluster of pneumonia-like cases in Wuhan, noticing similarities with the SARS virus. BlueDot uses NLP to cull data from thousands of disparate sources before alerting physicians to anomalies.
Software companies like Linguamatics employ NLP-based text mining solutions to rapidly access and analyze scientific papers, quickly identifying relevant information. Another such example is London-based BenevolentAI, an AI-platform that is fed data from clinical trial notes, patient records, research papers and patents. Based in the Cloud, the platform can be queried like a search engine, producing knowledge graphs on medical conditions and their associated genes, allowing researchers to identify potential candidate drugs.
NLP allows for the knowledge of the Worlds experts to be extracted at the click of a button, creating a knowledge base from which key insights can be extracted.
Clinical Trial Matching
A crucial stage in the drug development process is the clinical trial phase, where a new drug is tested on participants to assess its efficacy and monitor potential side effects. A critical bottleneck during this stage is the identification and recruitment of suitable candidates who meet the research criteria.
In 1994, researchers at Massachusetts General Hospital needed volunteers with a particular type of early-stage breast cancer to participate in a clinical trial for a new treatment. A pool of 40,000 potential candidates was identified, with 636 enrolling in the study. Despite the trial ultimately being heralded as a success, the selection process alone took 5 years.
Traditionally this process takes time for a number of reasons. Clinical research coordinators will trawl through medical files and Electronic Health Records, identify eligible candidates from the appropriate demographic and approach them about participating in the trial.
One of the most exciting applications of NLP is its ability to remove this bottleneck and drastically speed up the recruitment process.
Companies like Deep 6 AI are utilizing software to analyze vast amounts of fragmented medical data (including pathology reports, doctors’ notes & ICD-10 codes), employing NLP to extract tens of thousands of clinical data points to match complex clinical trial criteria. The use of rules-based triggers can identify appropriate patient cohorts and automatically match between the description of a clinical trial and the information contained in an EHR.
Cutting the lengthy recruitment process will lead to potentially life-saving treatments being approved at a rate far quicker than would have been possible without the adoption of such techniques.
A recent study in the US found that physicians spend an average of 16 minutes using Electronic Health Records for each patient they see. Not only does this take valuable time away from patient care, it is also a leading cause of burnout and dissatisfaction among medical practitioners.
Advancements in the field of natural language processing have allowed computers to understand speech with greater clarity and accuracy, something that healthcare organizations are leveraging.
Medical practitioners are increasingly using speech recognition systems with built-in NLP algorithms to dictate patient notes. Not only does this improve the quality of patient records, it also removes the administrative burden on physicians, reducing the risk of burnout and allowing them to more efficiently use their time.
One such example is Nuance’s Dragon Medical One, an AI-platform that transcribes doctors notes in real-time to the patients EHR. The system was deployed to great effect at Concord Hospital in New Hampshire, reducing the workload on physicians and saving more than $1 million.
Another exciting application of NLP in the Healthcare sector is predictive analytics, and how it can be employed to solve population health problems. Applying NLP techniques to electronic medical records can help to identify patients facing a greater risk of health disparities and provide an additional level of surveillance.
From 1999 to 2006, an NLP search approach was used to identify patients at risk of post-operative surgical complications at 6 Veteran Health Administration Centers. The study found that an automated search of medical records using NLP techniques correctly identified at-risk patients more effectively than traditional methods.
Another often quoted example of this use of natural language processing relates to the identification of patients at risk of Kawasaki disease, where early diagnosis is critical. A 2016 study found an NLP-based algorithm was able to identify high-risk patients with a sensitivity of 93.6% compared to notes manually reviewed by clinicians.
The Global NLP in Healthcare and Life Sciences market is expected to reach $3.7 Billion by 2025, more than double its current valuation. This is largely down to the adoption of just some of the techniques mentioned in this article. Going forward it is likely this trend will continue, having a great impact on everything from drug discovery to patient rehabilitation, revolutionizing how healthcare is practiced.