The opioid epidemic has coincided with the increasing use of social media platforms such as Twitter to discuss SUD-related topics, including first and second-hand experiences with overdose [18]. Our study was able to identify and characterize 58 tweets specifically posted from Tribal lands from 2014 to 2015, where users actively discuss struggles with SUD, overdose, and mental health challenges. Most coded posts (70.6%) described a primary account of an overdose, also annotated for co-occurring substance use or suicidal intent, suggesting that Twitter may be a useful platform to understand drivers of SUD in this population.
Substance use content was detected from over two dozen AI/AN communities; concerningly, almost half of this content originated from Tribal lands in Oklahoma. Tribes in Oklahoma, along with others across the country, have actively litigated against the pharmaceutical industry for their role in the spread of opioid-related health burden across Tribal lands. The Cherokee Nation, one of the first Tribes to sue the pharmaceutical industry over the opioid epidemic, reported in legislation that “there were so many opioids within the Tribe's 14-county reservation in 2015 that it amounted to 107 pills for every adult resident. [19]” In 2022, hundreds of AI/AN Tribes reached a settlement on National Prescription Opiate Multi-District Litigation, leading to more than half a billion dollars in funds that will be used to address AI/AN opioid addiction and long-term treatment.
This study represents a first step in determining whether social media data can be used to explore SUD health-related disparities among the AI/AN population living on Tribal lands. Unlike other social media platforms, Twitter generally provides the most robust location metadata (i.e., latitude and longitude). However, only a small percent of Tweets have available local metadata, having the potential to underrepresent the magnitude of a particular health topic-of-interest. While previous Twitter research has explored engagement with AI/AN-focused hashtags and topics relevant to AI/AN youth, this study explicitly limited data collection to areas where virtually every resident is AI/AN, increasing the sensitivity of our data collection [20].
Increasing the responsiveness and sensitivity of traditional mortality-based surveillance systems, which only capture limited dimensions of data (e.g., case rates, mortality rates), may help address the national substance use emergency faced by metropolitan and non-metropolitan-residing AI/ANs and other minorities in the US. Future research should use other platforms, explore other health topics, and consider comparing social media-based data to relevant electronic health record systems, such as large-scale data warehouses maintained by Regional Health Information Organizations, and other region-specific databases (e.g., behavioral risk factor surveillance surveys), including those maintained by local and state public health departments, Tribes, and Tribal Epidemiology Centers.
Additionally, attention must be dedicated to individual and community privacy in the conduct of this type of research due to known stigma associated with substance use and the potential that disparities evident in one AI/AN community may be erroneously or maliciously generalized to produce stereotypes about the entire AI/AN population [21]. To address this in our study, displayed Tweets were paraphrased and identifying metadata and account mentions were removed. We present paraphrased Tweets and an example of the inductive coding scheme we applied to the 58 Tweets. Therefore, risk of re-identification by public users is minimal.
Overall, study results may also be of interest to community advocates, Tribal leaders, and public health researchers interested in addressing the growing burden of substance use in the AI/AN population, especially as it pertains to AI/ANs residing in non-metropolitan areas such as Tribal lands. Results presented in this study stem from an exploratory analysis and should be further validated with confirmatory community-based participatory approaches and other data sources (e.g., surveys, focus groups). Further, this infodemiology approach, which may have the potential to be used at scale with residents on Tribal lands, may represent a useful adjunct for existing Tribal public health surveillance systems seeking to detect emerging Tribal public health challenges in near real-time using techniques such as supervised machine learning.
For example, our analysis discovered a tweet from Nebraska Tribal lands reporting increased use of K2-Spice (synthetic cannabinoid) in the community. De-identified information generated from these results can be routed to the proper public health authorities and acted upon to protect the public health. In addition, the development of comparative infodemiology studies that examine different demographic groups may reveal between-group differences and behaviors that can inform the development of policies and programs addressing minority substance use for specific and distinct communities.
In conclusion, overdoses resulting from substance use in AI/AN and other predominantly minority-residing communities may be largely preventable if actionable social media data is responsibly collected and results made available to key stakeholders. Depending on the scope of data collection and reasonable expectations for user privacy, this may require data-sharing agreements between social media platforms, public health authorities, and researchers, ensuring data is only accessed to protect communities at heightened risk of substance use and other harms.
