The impact of effective HIV prevention tools is limited because many people do not know that they are at risk for HIV acquisition, despite the availability of various risk assessment scores and criteria. This proposal aims to use a novel data science approach to assessing HIV prevention needs among 400 young women in Kisumu, Kenya- namely, topic modeling and network analysis of text and/or social media messages (e.g., WhatsApp, Instagram, Twitter). The study will involve in-depth assessment of relevant ethical and logistical factors to ensure appropriate and optimized use of a sentiment analysis tool for implementation in routine clinical care.

In the Social Media as a Risk Tool (SMaaRT) Study, the investigators hypothesize that topic modeling of SMS/social media data combined with network analysis among young women in Kenya will correlate well with existing HIV risk scales and ultimately yield a better understanding of HIV prevention needs. The investigators propose the following aims:

1. Explore ethical factors that may influence analysis of SMS and social media messages. Research assistants will conduct individual qualitative interviews with up to 32 young women (16 who would and 16 who would not provide SMS/social media data, stratified among four clinic sites) and one focus group of five Kenyan bioethicists. Questions will explore ethical concerns from individual and bystander (e.g., contacts involved in SMS/social media) perspectives and differences in ethical issues by type of social media (e.g., conversations vs posts). Follow-up interviews will be conducted with the women who provide SMS and/or social media data (in Aim 2).
2. Conduct topic modeling and network analysis of SMS and social media messages to predict HIV prevention needs among young women in Kenya. Working with four clinical sites in Kisumu, study staff will ask approximately 400 women (ages 18-24) seeking HIV testing, PrEP, and other health services to download six months of SMS/social media messages (e.g., WhatsApp, Instagram, Twitter) as a one-time procedure. For those providing data, research assistants will assess social networks engaged via SMS/social media (e.g., anonymously labeled as peers, sexual partners), administer multiple HIV risk assessments (e.g., VOICE, Wand risk scores), and obtain HIV test results. Data analysts will use automated structural topic modelling to determine "topics" (word clusters) and assess for association with other risk assessments (primary outcome) and HIV test results (exploratory outcome), and will also evaluate the impact of social networks, SMS/social media type, data volume, and language type on outcomes. Data collection and analysis will conform to Aim 1 findings.
3. Assess practical factors that may influence use of a sentiment analysis tool in routine care. In a needs assessment based on Implementation Mapping, research assistants will conduct four focus groups with five staff per clinic and two focus groups with five young women each to explore staffing best suited to implement a sentiment analysis tool and how it could be best integrated into routine care. The investigators will also assess available resources to determine optimal efficiency in developing a preliminary implementation strategy.

Inclusion Criteria:

• Identifying as a young woman (age 18-24 years)
• Attending clinic for any health services, including PrEP and HIV testing
• Smart phone ownership
• Ability to understand Kiswahili, DhoLuo, and/or English
• Use of SMS, WhatsApp, and/or other types of social media

Exclusion Criteria:

• Inability to provide informed consent (e.g., intoxication, developmental delay)

Research assistants will recruit young Kenyan women (ages 18-24), attending one of four clinics for any health services. Smart phone ownership and use of SMS, WhatsApp, or other social media is required.

This project will release and share final de-identified research data and materials from NIH-supported research for use by other researchers in a timely manner. Due to the sensitive nature of the SMS/social media messages, that data will be deleted at the conclusion of the study.

We will make this data available after publishing our findings.

We will post a de-identified dataset to the Harvard Dataverse, a datasharing platform.