IQVIA Virtual NLP Summit 2022 - Day 2

Healthcare systems are producing data at a rate that exceeds growth in any other industry. A huge amount (80%) of this data is unstructured.

Those who adopt technologies to untap the potential value of these data are best placed to understand their populations more. And are able to provide appropriate and tailored treatment and initiatives where they are needed.

The latest in AIMed’s webinar series will explore how leaders from Washington University School of Medicine, St Louis and Kaiser Permanente Northern California are using NLP to advance clinical care.

Data and AI now have leading roles to play in advancing precision medicine research by identifying early onset and treatments for disease. Dr Philip Payne will share how Washington University is building a set of NLP pipelines to extract high quality phenotypic data from the clinical narrative, to develop registries for patients with Alzheimer’s disease, breast cancer, diabetes and obesity. The use of natural language processing is key because close to 80% of the high value phenotypic data is encoded in the clinical narrative, not in structured or discrete fields in the electronic health record.

It has long been known that the unstructured data holds the answers to most questions in healthcare – but finding repeatable and scalable solutions to unlock this data across functions and applications has remained a challenge. In this session, we will learn how new components in the NLP toolkit can enable payers, providers and organizations processing medical records to transform processes, increase efficiencies and derive meaningful value to their organization.  

Artificial intelligence (AI) holds a promise of advancing our ability to improve patient care. Trained on routinely collected healthcare data, AI algorithms can identify patients at risk for a variety of medical events such as lingering misdiagnosis, disease progression, an upcoming severe adverse event, or non-adherence to treatment. Healthcare data can originate from structured fields (e.g., diagnoses captured via ICD-10) but 80% of data in the healthcare setting is unstructured, locked in sources such as radiology reports, clinical notes, discharge summaries and pathology notes.

Applying AI methods to these large data sets can provide a window of clarity for traditionally opaque patient predictors and support prioritization of patient engagement based on risk level and available clinical resources. This webinar will provide an overview of proven AI applications for medical event prediction, covering the lifecycle of solution development: data source selection, study design, training the AI algorithm, efficient integration into the clinical workflow, role of Social Determinants of Health as care gaps, etc. We will showcase real-world applications of such AI algorithms with measured impact to patient care across a variety of therapeutic areas.

The terminology used by pathologists to describe and grade dysplasia and premalignant changes of the cervical epithelium has evolved over time. Unfortunately, coexistence of different classification systems combined with non-standardized interpretive text has created multiple layers of interpretive ambiguity. We used natural language processing (NLP) to automate and expedite translation of interpretive text to a single most severe, and thus actionable, cervical intraepithelial neoplasia (CIN) diagnosis. The algorithms that were developed and then applied to 35,847 unstructured cervical pathology reports assessed NLP performance in identifying the severest diagnosis, compared to expert manual review. NLP performance was determined by calculating precision (0.957), recall (0.925), and F score (0.94). Using NLP also significantly reduced the time to evaluate each monthly biopsy file from 30 hours to 0.5 hours.

The use of NLP rapidly and efficiently assigned a discrete, actionable diagnosis using CIN classification that can assist with clinical management of cervical pathology and disease. Moreover, discrete diagnostic data encoded as CIN terminology can enhance the efficiency of clinical research.

NorthShore has established itself as a preeminent health system in building and implementing analytical tools at scale to improve the quality and efficiency of care. A current goal is to build a learning health system. A learning health system is a system where knowledge generation processes are embedded in daily practices to produce continual improvements in care. At NorthShore, we are at the nascent stages of building a learning health system, with Natural Language Processing (NLP) as a starting point. 

We partnered with Linguamatics to understand how to leverage vast unstructured language fields that are currently of limited practical use but pose tremendous potential value to our data assets. With the COVID pandemic and the drive to push into value-based care, it is more important than ever that healthcare providers need to understand Social Determinants of Health (SDOH) to effectively and equitably care for their populations. It is estimated that SDOH can account for 60-70% of health outcomes and at an urban health system on the east coast where NLP was used to identify SDOH, 31% of their entire population had at least one SDOH that could be affecting their health. These socioeconomic, behavioral, and environmental factors are not routinely and discretely documented and available in structured fields but rather are documented in unstructured text notes. Identifying and extracting these data elements at the patient level will provide a valuable path to intervene on the social determinants that impact care and outcomes.