The Use of Twitter to Explore Trends in Attitudes Toward Contraceptive Methods

Abstract

Background: Unlike other medications, contraceptive methods are often chosen based on the experiences and opinions of individuals’ social networks. Though social media, including Twitter, increasingly influences reproductive-age people, discussion of contraception in this setting has yet to be characterized. Natural Language Processing (NLP), a type of machine learning in which computers analyze natural language data, enables this analysis. This study aims to use NLP to explore attitudes toward different contraceptive methods, including both Long- and Short-Acting Reversible Contraception (LARC and SARC), on Twitter since 2006. Methods: We collected English language tweets mentioning reversible, prescription contraceptive methods with typical-use Pearl Indices of <10 pregnancies per 100 woman-years, including prescription LARC (the intrauterine device (IUD) and the contraceptive implant) and SARC (oral contraceptive pills; the contraceptive patch; the vaginal ring; and the Depo-Provera shot) between March 2006 and December 2019. We used the Amazon Comprehend NLP Sentiment Analysis Application Programming Interface to determine the sentiment of all tweets mentioning a single contraceptive method and evaluated the NLP algorithm’s performance based on ten human reviewers’ manual sentiment analysis of a random sample of 1000 tweets. All data and code to reproduce this analysis are available at https://github.com/hms-dbmi/contraceptionOnTwitter, and the initial steps can be replicated by launching the code at https://tinyurl.com/cleanTweetsMyBinder. Results: The number of annual tweets mentioning contraception has increased nearly three hundred-fold since 2007. Out of 838,739 total filtered tweets mentioning at least one contraceptive method, the most commonly tweeted-about method was the IUD (45.9%). LARC methods were mentioned more than SARC methods (58% vs. 42%), and the proportion of LARC-related tweets increased over time. Out of 665,064 tweets mentioning a single contraceptive method, there were nearly twice as many positive tweets about LARC methods compared to SARC methods (19.65% vs. 10.21%, p<0.05), though the greatest proportion of all tweets was negative (40.66%). Observed trends in the number and sentiment of tweets about individual contraceptive methods may reflect their historical context including regulation, advertising, availability, and satisfaction, though we did not investigate causal relationships between historical events and tweet volume or content in this analysis. Implications: Twitter is a potentially valuable source of data for consumer-level discourse regarding contraception and how attitudes toward individual methods have evolved over the past 13 years. Tweets may improve our insight into the perspectives of a traditionally difficult-to-reach population, related to a topic that is often stigmatized. Recognizing the influence of social media on people’s lives, and potentially when considering initiation of a contraceptive method, this and other social media platforms may allow clinicians and researchers to gather and potentially disseminate accurate information about contraceptive options.