ARRS 2022 Abstracts


E1388. Understanding Changes in Public Opinion Towards Mammography After the COVID-19 Pandemic: A Twitter-Based Sentiment Analysis
  1. Jefferson Chen; Boston Medical Center; Boston University School of Medicine
  2. Donghoon Shin; Boston Medical Center; Boston University School of Medicine
  3. Michael Fishman; Boston Medical Center; Boston University School of Medicine
  4. Christina LeBedis; Boston Medical Center; Boston University School of Medicine
The COVID-19 pandemic led to unprecedented disruptions for breast cancer screening in the United States. The odds of a woman receiving breast cancer screening were 20% lower in 2020 relative to 2019. Although mammography volumes are recovering, there is continued concern about the impact of the pandemic for patients who missed or delayed their screening appointments. Prior focus group studies have indicated that women may face increased anxiety surrounding cancer screening during the COVID-19 pandemic. However, there has yet to be a study of the effects of the broader public discourse on the effects of COVID-19 on breast cancer screening. We utilize an informatics-based approach to analyze the sentiment on Twitter over the course of the pandemic.

Materials and Methods:
All English language tweets containing the key words “mammo,” “mammogram,” “mammography,” and “breast cancer screening” were collected from Twitter between March 1, 2019 and March 1, 2021 using the Snscrape python library. The tweets were sorted into pre- and post-COVID-19 pandemic categories based upon their timestamps. As an estimate for the start of the COVID-19 pandemic, March 1, 2020 was used. The text from each tweet was cleaned using the clean-text python library. The tweets were given sentiment scores using RoBERTA-sentiment, a state-of-the-art transformer neural network-based natural language processing model. To account for multiple tweets by the same user, the sentiment score was averaged for each username within a given time period. The sentiment scores were then used to classify tweets into positive, neutral, and negative categories. The number of negative, positive, and neutral tweets were tabulated. To compare the pre- and post-COVID-19 pandemic categories, the Pearson’s chi-square test was used.

A corpus of 151,785 tweets was compiled, 75,151 tweets from 41,692 users pre-pandemic and 76,634 tweets from 44,494 users post-pandemic. The volume of tweets in the first 7 months of 2020 were strongly correlated with the monthly mammogram volumes reported by the Breast Cancer Surveillance Consortium (r = 0.9107, p = 0.0044). Our results demonstrate that in the post-COVID period the proportion of negative tweets significantly increased from 24.3% to 28.1% (p < .00001) and the proportion of positive tweets significantly decreased from 26.1% to 23.0% (p < .00001).

Building a Twitter-based tool allows for the understanding of public opinion of breast cancer screening as it relates to the COVID-19 pandemic and future disruptions. We plan to extend this analysis to include geographic variables to better understand the impacts of local changes in the public health landscape including outbreaks, vaccinations, and changes in policy. Future analyses will also allow for the understanding of more granular emotions expressed through social media. This information will help guide interventions to lessen the impact of future disruptions to breast cancer care.