Discovery of COVID-19 Symptomatic Experience Reported by Twitter Users

Area of Science:

• Social Media Analysis
• Infectious Disease Epidemiology
• Natural Language Processing

Background:

• The COVID-19 pandemic prompted widespread patient sharing of personal health experiences on social media.
• Analyzing these online narratives can supplement clinical data and enhance understanding of viral infections.
• Twitter serves as a valuable platform for collecting real-time patient-reported outcomes.

Purpose of the Study:

• To identify and analyze personal COVID-19 infection experiences shared on Twitter.
• To extract key information including infection status, symptoms, and symptom onset timing.
• To compare social media-reported symptoms with established public health guidelines.

Main Methods:

• Utilized a pre-trained and fine-tuned language model to identify COVID-19-related tweets.
• Annotated machine-identified tweets to extract infection status and symptom data.
• Recorded the day of symptom occurrence for each reported experience.

Main Results:

• The top 10 most frequently reported symptoms align with those documented by the World Health Organization (WHO) and Centers for Disease Control and Prevention (CDC).
• Extracted symptom data, coupled with temporal information, offers insights into the progression of COVID-19 infection.
• Machine learning effectively identified and categorized patient-reported symptoms from social media.

Conclusions:

• Social media, specifically Twitter, provides a rich source of patient-reported data for understanding infectious diseases like COVID-19.
• Analysis of patient tweets can corroborate and expand upon official health information.
• The temporal data of symptoms reported on Twitter aids in understanding disease progression patterns.