Skip to main content
Image of a young businesswoman explaining research results.

The unprecedented uncertainty of the COVID-19 pandemic has created significant pressure on public health leaders to rapidly respond to changing operational and clinical demands, typically without timely data or historical context to drive optimal decision-making.

Paired with operational programs such as a statewide COVID-19 information or vaccine contact center, the use of speech analytics and natural language processing (NLP) tools can supplement traditional population and survey analysis to measure and improve program outcomes and citizen support.

Machine learning approaches like topic mining and sentiment analysis of transcribed calls use each interaction to measure and contribute to a new category of actionable health statistics — despite increased social and physical distance.

The foundation of NLP and machine learning methods in the measurement of citizen interactions has facilitated collaboration between data scientists and public health professionals, enabling measurement of the magnitude and strength of vaccine hesitancy and other social determinants of health. 

Three initial insights below were uncovered by the team and are now the focus of further focused public health intervention and inquiry:

  1. Food Delivery & Insecurity — Call clustering using LDA identified a small portion of COVID Support calls suggesting the callers need for food delivery support due to food insecurity issues.
  2. Vaccine Hesitancy — Data mining has identified 4-6% of callers using language associated with vaccine hesitancy (i.e., “afraid,” “allergy,” “trust,” “religion,” “science”). Language indicative of fear, unwillingness, and indecision is observed throughout these calls.
  3. Canceled Appointments — Further mining into calls with topics of fear, unwillingness, and indecision identified an associated pattern of phrases indicating appointment cancellations or reschedules. The team is currently working to apply a Deep Learning technique called Transfer Learning to understand better this pattern and the implications for both public health and operations management.

Maximus’ ability to utilize unsupervised machine learning methods to monitor and identify clinical and public health trends has enabled clinical and operational leaders to implement important public health interventions and effectively disseminate information critical for public health. In this way, Maximus continues to find new ways to deliver government services directly related and responsive to the needs of the citizens they serve.  

About the authors and contributors:

These capabilities are driven by the Maximus Performance Analytics Organization (PAX): Megha Gupta is a Product Owner, David Salvador is Manager of Data Science, and Charles Mendoza is Sr. Director responsible for Data Science initiatives. Arnold Slabbekoorn, Product Director, Eric Stewart, Vice President and Oana Cheta, Senior Vice President work towards the development of advance analytics capabilities and customer experience improvement for our government partners.