Developing predictive diagnosis algorithms

Overview

NashBio utilized structured data and unstructured notes to develop a predictive NLP algorithm to detect bullous pemphigoid, performing better than existing algorithms. Our database, particularly with our extensive notes, can be utilized more broadly to develop algorithms to detect other diseases.

Challenge

Electronic health record (EHR) data can be leveraged to generate predictive algorithms that improve disease diagnoses by making them more accurate, faster, and cost-effective. Developing effective algorithms requires expansive real-world data alongside relevant clinical expertise, especially when analyzing diseases that aren’t easily characterized.

NashBio worked with a client to develop predictive algorithms to diagnose bullous pemphigoid (BP), an autoimmune condition that causes skin blistering. Predictive algorithms can help identify BP cases that may have gone undetected or been attributed to other conditions; given that BP is associated with increased mortality, its accurate diagnosis is important to ensure patients are treated effectively.

At a Glance

NLP

from unstructured notes

Detection

improved NLP algorithm for BP detection

Diagnosis

clinical expertise to BP diagnosis

Our Impact

In order to develop and test BP detection algorithms, NashBio’s clinical team manually defined a cohort of verified BP patients using structured and unstructured note EHR data from our database. The unstructured notes were critical for both the manual confirmation of diagnosis as well as for developing a new natural language processing (NLP) algorithm.

The NLP algorithm was then tested by NashBio and compared with another algorithm developed previously; our analysis demonstrated the NLP algorithm performed better and was able to correctly identify 94% of patients diagnosed with BP.

Working closely with our client, NashBio delivered an improved NLP algorithm that can be further refined to diagnose BP. Importantly, our enriched dataset is broadly applicable to develop algorithms that can better identify other diseases to improve patient health.