Unsupervised Machine Learning to Detect and Characterize Barriers to Pre-exposure Prophylaxis Therapy: Multiplatform Social Media Study

Abstract

Background: Among racial and ethnic minority groups, the risk of HIV infection is an ongoing public health challenge. Pre-exposure prophylaxis (PrEP) is highly effective for preventing HIV when taken as prescribed. However, there is a need to understand the experiences, attitudes, and barriers of PrEP for racial and ethnic minority populations and sexual minority groups.Objective: This infodemiology study aimed to leverage big data and unsupervised machine learning to identify, characterize, and elucidate experiences and attitudes regarding perceived barriers associated with the uptake and adherence to PrEP therapy. This study also specifically examined shared experiences from racial or ethnic populations and sexual minority groups. Methods:The study used data mining approaches to collect posts from popular social media platforms such as Twitter, YouTube, Tumblr, Instagram, and Reddit. Posts were selected by filtering for keywords associated with PrEP, HIV, and approved PrEP therapies. We analyzed data using unsupervised machine learning, followed by manual annotation using a deductive coding approach to characterize PrEP and other HIV prevention-related themes discussed by users. Results:We collected 522,430 posts over a 60-day period, including 408,637 (78.22%) tweets, 13,768 (2.63%) YouTube comments, 8728 (1.67%) Tumblr posts, 88,177 (16.88%) Instagram posts, and 3120 (0.6%) Reddit posts. After applying unsupervised machine learning and content analysis, 785 posts were identified that specifically related to barriers to PrEP, and they were grouped into three major thematic domains: provider level (13/785, 1.7%), patient level (570/785, 72.6%), and community level (166/785, 21.1%). The main barriers identified in these categories included those associated with knowledge (lack of knowledge about PrEP), access issues (lack of insurance coverage, no prescription, and impact of COVID-19 pandemic), and adherence (subjective reasons for why users terminated PrEP or decided not to start PrEP, such as side effects, alternative HIV prevention measures, and social stigma). Among the 785 PrEP posts, we identified 320 (40.8%) posts where users self-identified as racial or ethnic minority or as a sexual minority group with their specific PrEP barriers and concerns.Conclusions: Both objective and subjective reasons were identified as barriers reported by social media users when initiating, accessing, and adhering to PrEP. Though ample evidence supports PrEP as an effective HIV prevention strategy, user-generated posts nevertheless provide insights into what barriers are preventing people from broader adoption of PrEP, including topics that are specific to 2 different groups of sexual minority groups and racial and ethnic minority populations. Results have the potential to inform future health promotion and regulatory science approaches that can reach these HIV and AIDS communities that may benefit from PrEP.

Digital Epidemiological Approaches in HIV Research: a Scoping Methodological Review

“…Exceptions to this were two studies by Young et al [ 21 , 22 •] that drew exclusively on Facebook data collected with consent from a cohort of sexual minority men, and one study each that leveraged data from Reddit [ 23 ] and Baidu Tieba [ 24 ], a Chinese social media platform. Three more studies incorporated data from multiple platforms [ 25 – 27 ] (i.e., various combinations of data from Twitter, Reddit, Instagram, YouTube, and Tumblr). Irrespective of the platform, 20 of the 21 social media studies analyzed post content.…”

Public Figure Vaccination Rhetoric and Vaccine Hesitancy: Retrospective Twitter Analysis

“…We set a total number of 20 different clusters (ie, total number of topics for BTM to output: k=20), resulting in texts with similar themes put into the same clusters. To find the appropriate k value, we used a topic coherence score [ 21 , 24 ]. Coherence score is used to measure the performance of a topic model with different number of clusters and can help differentiate between topics that are semantically interpretable and topics that are artifacts of statistical inference [ 24 , 25 ].…”

“…To find the appropriate k value, we used a topic coherence score [ 21 , 24 ]. Coherence score is used to measure the performance of a topic model with different number of clusters and can help differentiate between topics that are semantically interpretable and topics that are artifacts of statistical inference [ 24 , 25 ]. We tested 5 different k values (k=10, 20, 30, 40, and 50) for each data set and found that when k=20, we generated the highest coherence score, and this score did not change significantly with an increase in the k value.…”

Impact of COVID-19 on HIV Prevention Access: A Multi-platform Social Media Infodemiology Study

“…BTM can be used to sort short text into highly prevalent themes without the need for predetermined training data and has been previously used for exploration of other public health topics [ 19 – 24 ]. The methodological approach of using BTM for detection of HIV and PrEP-related topics is also detailed in a separate published paper [ 25 ].…”

“…A general deductive coding schema using the socio-ecological perspective outline (SEPO) [ 26 ] that outlines three intervention levels for PrEP including the “Individual and Relationships Domains: Provider Level”, “Individual and Relationships Domains: Patient Level” and “Community Domains: Healthcare-System Level”, were selected as parent codes as used in a prior PrEP infodemiology study [ 25 ]. All posts were reviewed by first and second author, and notes were taken on general themes of posts from which an initial code list was created.…”