5  Analysis

5.1 Descriptive Statistics

We first summarized participant characteristics in relation to group contact and visualized how categories of sexual orientation were distributed across levels of the race–gender variable. We then cross-tabulated event type by both sexual contact and distance from home to describe how types of gatherings related to sexual activity and travel patterns.

To explore spatial behavior, we compared the home community district of each participant to the districts where gatherings occurred. This produced a matrix showing patterns of movement across New York City. To identify which districts tended to receive participants from outside their home areas and which tended to send participants elsewhere, we calculated the difference between the proportion of gatherings and the proportion of residences in each district. These results were visualized as a bar graph.

Finally, we computed the spatial mixing coefficient for subgroups defined by age, sexual orientation, and race–gender to quantify how these groups interacted across space. Together, these descriptive analyses provided the empirical foundation for a subsequent simulation of outbreak control strategies.

5.2 Outbreak Control Strategies

Building on the spatial patterns observed in the descriptive analysis, we modeled a hypothetical intervention in which community districts were immunized sequentially according to priority rankings. To immunize a district was to immunize all individuals who had either a residence or a gathering in that district. Once immunized, individuals were assumed to be fully protected from acquiring or transmitting infection, and each person could only be immunized once.

We evaluated two alternative strategies for prioritizing districts: a contact-neutralizing strategy, which targets areas of high contact density, and a movement-neutralizing strategy, which targets districts that are highly connected across space.

Contact-Neutralizing Strategy

In the contact-neutralizing strategy, community districts were ranked by the number of gatherings and residences they contained, counting only residences of participants who reported group contact and had not yet been immunized. This approach prioritizes districts with the highest concentration of sexual or physical contact, aiming to immunize as many individuals as possible while focusing on a minimal number of locations.

Movement-Neutralizing Strategy

In the movement-neutralizing strategy, community districts were ranked according to their spatial centrality—a measure of how strongly each district is connected to others across the city. This strategy prioritizes districts that function as spatial hubs for participant movement, targeting them early in the immunization sequence to contain potential city-wide spread.

5.3 Outbreak Control Outcomes

After each community district was immunized, the MPX NYC Participant Network was updated by removing nodes representing immunized individuals. Following each update, we recalculated two key outcome measures: spatial efficiency and spatial network impact, each reflecting a different dimension of intervention performance.

Spatial Efficiency

Spatial efficiency was defined as the total number of immunized individuals divided by the number of community districts immunized. This measure captures how effectively an intervention converts each immunized district into individual-level protection. To compare the two intervention strategies, we visualized the number of community districts required to reach approximately 33% and 66% of the total population.

Spatial Network Impact

Spatial network impact was defined as the proportion of individuals belonging to the largest connected component of the MPX NYC Participant Network. This measure reflects the extent to which unimmunized individuals remain embedded within the same transmission-relevant network after each intervention step. To compare strategies, we plotted, for each additional community district immunized, the proportions of individuals who were immunized, unimmunized and still within the largest component, and unimmunized but outside of it.

5.4 Statistical Inference

number_reps <- targets::tar_read(config_list)[[1]][["settings"]][["n_reps_graph"]]

To quantify uncertainty around our network-based estimates, we used the bootstrap to calculate confidence intervals, assuming that the underlying data are a realization of a Bipartite Random Network Finest Fully Randomized Causally Interpretable Structured Tree Graph (BRNFFRCISTG) model (see Section C.3).

Recalling that the MPX NYC person-place graph includes participant observations and community district observations as nodes, we resampled participant observations 1000 times with replacement. An observation was defined as the complement of measures on a particular participant as well as all the edges that connect that participant to their community districts of residence and gatherings.

This approach provides robust, nonparametric inference by repeatedly drawing samples from the observed data to approximate sampling variability.

plot_icon(icon_name = "droplets", color = "dark_pink", shape = 16, alpha = 0.1)