Assessing Uncertainty in an Anatomical Site-Specific Gonorrhea Transmission Model of Men Who Have Sex With Men

Abstract

Background: Increased gonorrhea detection highlights the need for additional prevention efforts. Gonorrhea may only be acquired when there is contact between infected and uninfected anatomical sites. With 3 sites of infection, this leads to 7 plausible routes of men who have sex with men (MSM) transmission: urethra-to-rectum, rectum-to-urethra, urethra-to-oropharynx, rectum-to-oropharynx, oropharynx-to-urethra, oropharynx-to-rectum, and oropharynx-to-oropharynx. We characterize the uncertainty and potential importance of transmission from each anatomical site using a deterministic compartmental mathematical model.

Methods: We developed a model of site-specific gonococcal infection, where individuals are infected at 0, 1, 2, or all 3 sites. Sexual behavior and infection duration parameters were fixed similar to a recent model analysis of Australian MSM. Markov chain Monte Carlo methods were used to sample the posterior distribution of transmission probabilities that were consistent with site-specific prevalence in American MSM populations under specific scenarios. Scenarios were defined by whether transmission routes may or may not transmit by constraining specific transmission probabilities to zero rather than fitting them.

Results: Transmission contributions from each site have greater uncertainty when more routes may transmit; in the most extreme case, when all routes may transmit, the oropharynx can contribute 0% to 100% of all transmissions. In contrast, when only anal or oral sex may transmit, transmission from the oropharynx can account for only 0% to 25% of transmission. Intervention effectiveness against transmission from each site also has greater uncertainty when more routes may transmit.

Conclusions: Even under ideal conditions (ie, when site-specific gonococcal prevalence, relative rates of specific sex acts, and duration of infection at each anatomical site are known and do not vary), the relative importance of different anatomical sites for gonococcal infection transmission cannot be inferred with precision. Additional data informing per act transmissibility are needed to understand site-specific gonococcal infection transmission. This understanding is essential for predicting population-specific intervention effectiveness.

Figure 1.

(A) Simplified schematic of gonococcal infection and recovery paths where infection is possible at multiple anatomical sites. Individuals may be infected at zero (susceptible state), 1, 2, or all 3 anatomical sites. Infection is acquired 1 site at a time. Recovery from multiple infections also occurs one at a time except for treatment induced recovery either from symptomatic urethral infection or general screening. Model equations available in supplemental materials. Symptomatic and asymptomatic urethral infection is modeled separately upon acquisition of urethral infection (not shown). (B) Scenario definitions based on which routes may transmit.

Figure 2.

Simulated distribution of gonococcal prevalence and observed values (*) by scenario. Values are summarized marginally (left) where anyone with an infection is binned together by site (eg, prevalence of rectal infection regardless of infection status at other anatomical sites), or jointly (right) where each possible category of infection by anatomical site is shown separately (eg, prevalence of infection in the urethra and rectum but not the oropharynx). Model was fit to the observed marginal distributions of prevalence. Scenarios defined in Figure 1. Each scenario consists of 10,000 parameter sets sampled using approximate Bayesian computation based on Markov chain Monte Carlo methods.

Figure 3.

Distribution of fitted per act gonococcal infection transmission probabilities between each anatomical site by scenario. Fitting assumes static rates of sexual behavior (Table 1) in terms of rates of oral sex, anal sex, rimming, and kissing. Scenarios defined in Figure 1. Each scenario consists of 10,000 parameter sets sampled using approximate Bayesian computation based on Markov chain Monte Carlo methods.

Figure 4.

Distribution of gonococcal infection transmission contributions from each anatomical site by scenario. Distributions are either shown marginally (top) as separate boxplots or jointly (bottom) as ternary diagrams (3-dimensional scatterplots) where each point represents the joint classification of 1 parameterization based on the contributions from each site; points closer to a corner represent greater transmission contribution by the anatomical site represented in that corner. Scenarios defined in Figure 1. Each scenario consists of 10,000 parameter sets sampled using approximate Bayesian computation based on Markov chain Monte Carlo methods.

Figure 5.

Distribution of gonorrhea intervention effectiveness (percent reduction in equilibrium prevalence) for 3 separate site-specific interventions by scenario. Distributions are either shown marginally (top) as separate boxplots or jointly (bottom) as ternary diagrams (3-dimensional scatterplots) where each point represents the joint classification of 1 parameterization based on the relative effectiveness of each intervention. Scenarios defined in Figure 1. Each scenario consists of 10,000 parameter sets sampled using approximate Bayesian computation based on Markov chain Monte Carlo methods.