Discipline: Social, Behavioral, and Economic Sciences
Subcategory: Social Sciences/Psychology/Economics
Samuel F. Rosenblatt - St. Mary's College of Maryland
The sharing of injection drug apparatus is one of the most prevalent disease vectors for HIV among people who inject drugs (PWID). Recent research suggests that when using interventions to prevent the spread of disease, targeting interventions toward individuals based on their position in the risk network is more effective than randomized intervention. However, targeting based on network properties requires network data, and it is expensive and often infeasible to get enough data to have an approximation of the full network. Furthermore, the network measure that much prior research has shown to be most effective for intervention assignment, betweenness centrality, is extremely sensitive to missing data. We propose the use of a different measure, called brokerage score, which is highly correlated to betweenness, but which remains fairly constant as we move from sample to a full network. We hypothesize that using brokerage to determine interventions is more efficient than current intervention strategies due both to its effectiveness in reducing overall network risk and its cost efficiency. We investigated a sample of a directed risk network of PWID in Puerto Rico and define brokerage score for an individual, B, to be the number of ordered pairs of PWID, A and C, where B uses needles after A, and C uses needles after B. We found that if this were the full network, and we were able to sever the risk edges of the 5 individuals with the highest brokerage via intervention, we would reduce the number of people that the average network member is downstream from by nearly 80%, much more than the reduction from a randomized intervention. Since brokerage score changes little relative to other nodes as the sample approaches the full network, this selection of the top 5 PWID is likely to remain consistent. However, even collecting a sample of a network is costly compared to conducting short surveys, and so we create a model for brokerage score based on survey variables. We used a zero-inflated negative binomial regression and found that brokerage has a quadratic relationship to daily drug expenditures (linear term p=0.01, negative quadratic term p=0.03). The zero-predictor was the number of ties with other PWID (p=0.05; whether risk ties or social ties), and we controlled for the percent of an individuals partners who were not interviewed (p=0.008). We are currently using simulation-based research to investigate the hypothesis that this type of quadratic relationship is typical in directed risk networks. If our hypotheses are supported, we can create a more generalizable model that will allow us to go to an area where a PWID network exists, use regular sampling methods to conduct a short survey, and target an intervention at those PWID who our model predicts to have the highest brokerage score. This should result in much more effective prevention than random intervention would, at a fraction of the cost of collecting full network data.
Funder Acknowledgement(s): This research was supported in part by NSF Grant SMA 1461132 - REU Site: Undergraduate Research Opportunities to Broaden Participation in Minority Health Research (PI: K Dombrowski, University of Nebraska Lincoln). I would like to thank Robin Gauthier, Patrick Habecker, Kirk Dombrowski and Amanda Nguyen for their help and advice with various parts of this process. I would also like to thank the University of Nebraska-Lincoln and the Minority Health Disparity Initiative.
Faculty Advisor: Kirk Dombrowski, kdombrowski2@unl.edu
Role: Besides data collection and converting raw data into network form, I did all of the research myself.