The Benefits of Crowdsourcing to Seed and Align an Algorithm in an mHealth Intervention for African American and Hispanic Adults: Survey Study (Preprint)

Abstract

BACKGROUND The lack of publicly available, culturally relevant data sets on African American and bilingual/Spanish-speaking Hispanic adults’ disease prevention and health promotion priorities presents a major challenge to researchers and developers who want to create and test personalized tools for the preventive health behaviors intervention space. Personalization depends on prediction and performance data. To develop such a ‘recommender system’ (RecSys) that predicts the most culturally and personally relevant preventative health information and serve it to African American and Hispanic users of a novel smartphone application while also avoiding the ‘cold start’ problem, we needed population appropriate seed data that aligned with the app’s purposes of setting health goals and finding associated articles and topics in healthfinder.gov, a federally supported database of health conditions and disease prevention information. OBJECTIVE To address the lack of culturally specific preventive personal health data and sidestep the type of algorithmic bias inherent in a RecSys not trained in the target population, we created a novel dataset on prevention-focused health goals by collecting a large amount of data quickly and at low cost from members of the target population. We seeded our RecSys with data collected anonymously from self-identified Hispanic and self-identified non-Hispanic African American adult respondents utilizing Amazon Mechanical Turk. METHODS We developed an online survey in which respondents completed a personal profile, health literacy assessment, family health history, and personal health history. Respondents then selected their top three health goals related to preventable health conditions, and for each goal reviewed and rated the top three healthfinder.gov information returns by importance, personal utility, whether the item should be added to their personal health library, and their satisfaction with the quality of the information returned. RESULTS We collected data from 985 self-identified Hispanic (49%) and self-identified non-Hispanic African American (51%) adult respondents utilizing Amazon Mechanical Turk over only 64 days at a cost of $6.74 per respondent. Respondents rated 92 unique articles. Both African American and Hispanic groups noted physical fitness (62.9%), healthy eating (43.2%), and nutrition and weight (24.0%) as their most frequent personal goals for health. Both African American and Hispanic groups noted mental health issues (34.6%), hypertension (31.0%), and vision or hearing impairments (24.4%) as their most frequently experienced health conditions, and hypertension (55.0%), diabetes (46.1%), and obesity (39.6%) as their most frequent family health conditions, although there are statistically significant differences when considering prevalences of goals, personal health, and family health conditions. Though both groups note experiencing mental health issues more frequently than any other condition, neither respondent group identified mental health as a high priority personal health goal. Respondents’ personal goals align with potentially preventive conditions they report in their family health history. CONCLUSIONS Researchers have options, such as Amazon Mechanical Turk, for quick, low-cost means to avoid the ‘cold start’ problem for algorithms and sidestep bias and low relevance for an intended population of app users. Seeding a RecSys with responses from people like the intended users allows the development of a digital health tool that can recommend information to users based on similar demography, health goals, and health history. This approach minimizes potential initial gaps in algorithm performance, allows quicker algorithm refinement in use, and may deliver a better user experience to individuals seeking preventative health information to improve health and achieve health goals. Additionally, this approach allowed investigating the correlation between personal health goals and known health history in a sample of African American and Hispanic participants. Health goals for African American and Hispanic adults are more likely to reflect self-reported somatic health conditions, and less likely to reflect psychological health conditions, even when experiencing mental health issues.