We used a previously described dynamic network model of sexually transmitted mpxv in GBMSM [3,5,6], with parameters updated based on national surveys of sexual behavior among GBMSM [7]. This model simulates a network of partnerships between the estimated 167,000 GBMSM living in NYC. We model three types of sexual partnerships: individuals can have 0 or 1 ‘Main’ partners (enduring sexual partners who took priority over others), between 0 and 3 ‘Casual’ partners (enduring sexual partners besides the main partner), and 0 or 1 ‘one-time’ partners per timestep (partners without repeat sexual contact). For each timestep, enduring partnerships have a probability of forming and dissolving, and individuals have a probability of engaging in one-time partnerships based on their assignment to one of 6 groups stratified by sexual partnership rates, with group 1 representing the lowest sexual activity level (40% of the population, < 0.1% chance of engaging in one-time partnerships) and group 6 representing the highest sexual activity level (1% of the population, 37% chance of engaging in a one-time partnership per day, Appendix S1, Table S1).

We used a modified SEIR (susceptible, exposed, infectious, resistant) model [8]. Individuals start in the susceptible (S) class. An infectious individual has a probability of infecting a susceptible partner given by the probability of sexual contact per timestep multiplied by the probability of infection per sexual contact (µ). Upon acquiring infection, individuals enter a pre-symptomatic non-infectious state (E), after which they have a probability per day of entering the pre-symptomatic and infectious state (P), then the symptomatic and infectious state (I), and finally the recovered and resistant state (R). Individuals can also be vaccinated with 1 (V1) or 2 (V2) doses of the vaccine, with effects described below. We assume symptom onset reduces sexual contact rate with ‘main’ and ‘casual’ partners by 50% and causes individuals to move to one lower sexual activity level until they recover. A proportion of individuals, (ρ), will seek treatment γ days after they develop symptoms, with γ based on the date-varying time between reported symptom onset and mpox tests in case data. We assume that among GBMSM in NYC, individuals seeking treatment for mpox symptoms would be tested for mpox and reported as a diagnosed cases, hereafter referred to as cases. After a positive mpox test, individuals abstain from sexual activity until recovery. Individuals who learn their infection status sustain contact with their ‘main’ partner such that they have a 10% chance of infecting their ‘main’ partner over the duration of the infection, reflecting estimates of household transmission [9].

We initiated the model by introducing 5 exposed individuals into the three groups representing the most active 15% of population on May 14^th, 2022. This produced a median of 5 cumulative cases in the model on May 31^st, matching case data from NYC.

Our network does not account for potential periods of increased transmission associated with GBMSM gatherings (e.g., Pride festivals) throughout the summer. Additionally, our model does not account for the impact of importation of mpox cases from outside NYC. Thus, we model a ‘surge period’ during the initial period of our simulations in which we add extra-network transmission to the model to represent transmission occurring during GBMSM gatherings, and import infections into the network to represent travel related infection. We assume that these extra infections occur among individuals in the three highest sexual activity groups. For extra-network transmission, we assume an additional number of infections per day equal to SHPH+IHNHβ

Where S_H is the number of high-activity individuals (activity groups 4–6), P_H and I_H are the number of pre-symptomatic and symptomatic infectious high-activity individuals, N_H is the total number of high activity individuals, and β is the extra-network transmission rate. For importation, we randomly select high-activity individuals to enter the E state at a constant rate, ε. β, ε, and the length of the surge period are determined by our fitting procedure, described below.

During each simulation, individuals decrease their probability of one-time partnerships and their rate of sexual contact with casual partners as a function of perceived mpox risk [10]. We parameterize this perceived risk based on frequency of mpox discussion on online forums over time, as discussed by Clay et al. [6], and fit the magnitude of this behavioral adaptation to incident case report data (fitting procedure).

Our fitting procedure selected for a distribution of surge duration with a median of 89 days [IQR: 69 – 109], a distribution of importation rate with a median of 1.7/day [IQR: 0.99 – 2.4], a distribution of surge transmission with a median of 18% [IQR: 12% – 24%], a distribution of transmission probability with a median of 67% [IQR: 40% – 81%], and a distribution of behavioral adaptation with a median of 49% [IQR: 16% – 63%] (Figure 2A, Table 1). The fit parameters are generally negatively correlated with one another (Appendix S3, Figure S2). Thus, these parameter estimates should not be used in other contexts.

Our model estimates that there would have been 10,600 cumulative cases [IQR: 6,280 – 16,800] by May 14, 2023 in the scenario with no vaccination. Our model further estimates that the baseline ‘first-dose priority’ strategy averted 67% [IQR:47% – 78%] of these cases (Figure 2B).

We estimate that the ‘first-dose priority’ and ‘intermediate’ strategies would have averted a similar number of cases, and that both strategies would have averted more cases than the ‘second-dose priority strategy’. We estimate that the ‘intermediate’ strategy would have averted 68% [IQR:46% – 78%] of cases and the ‘second-dose priority’ strategy would have averted 55% [IQR:38% – 68%] of cases. We further estimate that the baseline ‘first-dose priority’ strategy employed by NYC would have averted 0.61% [IQR:−11% – 9.8%] more cases than the counterfactual ‘intermediate’ strategy, and would have averted 17% [IQR:2.9% – 38%] more cases than the counterfactual ‘second-dose priority’ strategy (Figure 3A) after 1 year. Note that we calculate percent differences in cases averted for each fit parameter set and then calculate the median of those differences, rather than calculating the percent difference in median cases averted by each strategy.

Reducing the number of available doses does not change which vaccine administration strategies would have averted the most cases. We estimate that the baseline ‘first-dose priority’ strategy would have averted 7.0% [IQR:−12% – 24%] more cases than the counterfactual ‘intermediate’ strategy, and would have averted 33% [IQR:17% – 76%] more cases than the counterfactual ‘second-dose priority’ strategy (Figure 3B).

We find that neither increasing incremental vaccine efficacy, decreasing incremental vaccine efficacy, or increasing per-exposure probability of transmission changes our qualitative results (Appendix S4, Appendix S5, Figure S3, Figure S4).

During the 2022 mpox outbreak, demand for vaccines initially outpaced supply, preventing 2-dose vaccination for the entire population at increased risk of mpox [2,22]. Thus, NYC prioritized administration of first doses to reach a large portion of the impacted population as quickly as possible. In this analysis, we modeled the strategy focusing on widespread, 1-dose vaccination employed by NYC, compared to counterfactual strategies focusing on 2-dose vaccination in narrower portions of the population. We estimated that the NYC vaccine rollout prevented 66% of potential mpox cases. We further estimated that prioritizing narrow, 2-dose vaccination would have averted a similar number of cases as prioritizing widespread, 1-dose vaccination as long as doses were not pre-allocated (i.e., held in reserve to ensure that individuals received the full vaccine course on schedule).

Increasing vaccine coverage will decrease the probability and size of future mpox outbreaks [3]. Thus, CDC recommends continued vaccination for those who are at increased risk of mpox to prevent future outbreaks [23]. While current vaccine demand may be low due to low-perceived risk of mpox infection and the widespread availability of the JYNNEOS vaccine, demand for mpox vaccines could increase in the future, e.g., due to the increased activity of clade I mpxv in the Democratic Republic of the Congo and surrounding countries [4]. In this situation, if vaccine availability is limited, local and state health departments may consider whether to focus on prioritizing narrow, 2-dose vaccination, or widespread, 1-dose vaccination. Our analysis shows that in NYC, prioritizing narrow, 2-dose vaccination would have resulted in additional cases if vaccine doses were pre-allocated. However, prioritizing widespread, 1-dose vaccination or narrow, 2-dose vaccination without pre-allocation would have averted a similar number of cases. Thus, if needed, local and state health departments may be able to avert the most cases by choosing whichever strategy allows them to administer doses to impacted populations most quickly.

One strength of prioritizing widespread, 1-dose vaccination that our model does not consider is that this strategy can improve vaccine equity. Groups who have historically faced systematic barriers to accessing healthcare often receive few initial vaccine doses during outbreaks (for example, during the initial COVID-19 outbreak [24]). Narrow, 2-dose vaccination strategies, which focus on concentrating all vaccine doses in the individuals who were able to access vaccine doses first, may thus exacerbate health inequities. This is particularly problematic as individuals are most likely to have close sexual contact with individuals of the same race or socioeconomic status [25]. Narrow, 2-dose strategies may thus lead to uncontrolled transmission in groups already facing a disproportionate burden of most health problems, exacerbating health disparities.

Vaccination can yield greater health benefits when the selection of vaccine administration strategies considers heterogeneity in immune function. People with HIV (PWH) make up 19% of GBMSM [26,25], but 55% of clade II mpox cases in the United States among people with known HIV status. For these cases, coinfection between HIV and mpox can lead to severe, life-threatening outcomes in individuals not receiving antiretroviral therapy and who had low CD4 counts [26]. Further, JYNNEOS^® vaccination has reduced overall efficacy but higher incremental efficacy (increase in protection from 2-dose compared to 1-dose vaccination) in individuals living with immunocompromising conditions, including uncontrolled HIV [13]. Thus, because PWH may be more likely to acquire clade II mpox, are more likely to experience severe symptoms, and benefit more from receiving a second vaccine dose than others [13, 26–28], future vaccination strategies may benefit from giving PWH 2-dose vaccine courses as quickly as possible, even if the larger vaccine strategy is for widespread, 1-dose vaccination. This strategy could be paired with further efforts to connect people living with uncontrolled HIV to HIV services, given that increased CD4 counts may increase vaccine efficacy and reduce severity of mpox [29].

Our study is limited in that the strategies in our model may not reflect some of the practical considerations of vaccine logistics. First, we parameterized our ‘first-dose priority’ strategy using first and second doses administered by NYC, as this was the strategy explicitly used during the outbreak [2]. However, in early September, as demand decreased and vaccine supply improved, NYC began offering second doses to those who were eligible, based on time since vaccination (first to those whose first dose was administered more than 10 weeks previously, then moving to anyone who received their first dose at least 4 weeks previously). Thus, in practical terms, the ‘first-dose priority’ strategy in our model, which was parameterized via NYC vaccination records, was a sequential implementation of two strategies, and a more rigorous ‘first-dose priority’ strategy may have averted more cases. Second, we assume in our ‘intermediate’ and ‘second-dose priority’ scenarios that individuals will always choose to receive a second vaccine dose if available. However, jurisdictions will experience losses to follow up, with individuals choosing not to receive a second dose, or individuals may delay their second dose. In the ‘intermediate’ strategy, those doses would be used as first doses for eligible individuals, but in the ‘second-dose priority’ strategy, those doses would be wasted, leading to an increase in cases. Thus, our results must be interpreted while considering the logistic constraints of local jurisdiction and understanding that vaccination programs are not static and will change and adapt to the situation and realities on the ground. Our results are further limited by uncertainty around single dose vaccine effectiveness. We relied on modeled estimates of per-act efficacy for this study. However, empirical estimates of one dose vaccine effectiveness range from 33% to 86%, indicating large levels of uncertainty around these estimates [30]. Finally, our results are limited by the lack of available data on behavioral adaptations to mpox infection. We reasonably assumed that individuals would reduce their number of sexual partners upon developing mpox symptoms. However, we may have under or overestimated the magnitude of this behavioral adaptation, and thus may have under or overestimated the contact rate of infectious individuals.

Choosing whether to prioritize a multi-dose course in a narrow portion of the population or widespread, 1-dose vaccination will arise in future outbreaks of other diseases, whenever vaccination involves multiple doses [31]. Most notably, models employed during the COVID-19 pandemic lead to countries such as England delaying second-dose administration in favor of more widespread first-dose administration [32,33]. When choosing between vaccination strategies during an outbreak, public health agencies need to quickly make decisions with limited data. Mathematical modelers can conduct studies before those outbreaks occur, that will allow public health officials to make informed decisions during the initial stages of an outbreak. The optimal vaccine strategy outbreak will depend on the incremental effectiveness of first vs. second doses, the population contact network, vaccine supply and administration capacity, and the epidemic growth rate. Modeling studies can explore how these parameters shift the optimal vaccination strategy, while connecting these parameters to pre-existing data (e.g., the population contact network), or data that can be easily collected in the early stages of an outbreak (e.g., epidemic growth rate). Results from these studies can inform decisions for public health officials, leading to quick, evidence-based public health leadership.