Article Text

Original article
Core groups, antimicrobial resistance and rebound in gonorrhoea in North America
  1. Christina H Chan1,2,
  2. Caitlin J McCabe1,2,
  3. David N Fisman1,2
  1. 1Division of Epidemiology, Dalla Lana School of Public Health, Faculty of Medicine, University of Toronto, Toronto, Ontario, Canada
  2. 2Department of Health Policy, Management and Evaluation and Medicine, Faculty of Medicine, University of Toronto, Toronto, Ontario, Canada
  1. Correspondence to Dr David N Fisman, Dalla Lana School of Public Health, Faculty of Medicine, University of Toronto, 155 College Street, Room 678, Toronto, Ontario, Canada M5T 3M7; david.fisman{at}


Background Genital tract infections caused by Neisseria gonorrhoeae are a major cause of sexually transmitted disease worldwide. Surveillance data suggest that incidence has increased in recent years after initially falling in the face of intensified control efforts.

Objectives The authors sought to evaluate the potential contribution of antimicrobial resistance to such rebound and to identify optimal treatment strategies in the face of resistance using a mathematical model of gonorrhoea.

Methods The authors built risk-structured ‘susceptible–infectious–susceptible’ models with and without the possibility of antibiotic resistance and used these models as a platform for the evaluation of competing plausible treatment strategies, including changing antimicrobial choice when resistance prevalence surpassed fixed thresholds, random assignment of treatment and use of combination antimicrobial therapy.

Results Absent antimicrobial resistance, strategies that focus on treatment of highest risk individuals (the so-called core group) result in collapse of disease transmission. When antimicrobial resistance exists, a focus on the core group causes rebound in incidence, with maximal dissemination of antibiotic resistance. Random assignment of antimicrobial treatment class outperformed the use of fixed resistance thresholds with respect to sustained reduction in gonorrhoea prevalence.

Conclusions Gonorrhoea control is achievable only when core groups are treated, but treatment of core groups maximises dissemination of antimicrobial-resistant strains. This paradox poses a great dilemma to the control and prevention of gonorrhoea and underlines the need for gonococcal vaccines.

  • Gonorrhea
  • drug resistance
  • microbial
  • mathematical model
  • infectious diseases
  • sexual health
  • epidemiology (general)
  • economic analysis

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


Genital tract infection due to Neisseria gonorrhoeae (NG) is an important sexually transmitted infection in North America, both in terms of the absolute number of individuals infected annually and in terms of the cost and burden of illness associated with disease sequelae, including pelvic inflammatory disease and its complications in women, epididymo-orchitis and prostatitis in men, ophthalmic disease in neonates, enhanced risk of HIV transmission and acquisition, and rarely disseminated infection.1 2

Starting from the late 1970s and early 1980s, the incidence of gonorrhoea had decreased sharply, and by an order of magnitude, in the USA and Canada.3 4 This represented an important public health success, and one that was predicted by mathematical models. Models demonstrate that a pathogen such as NG, characterised by a relatively brief duration of infectiousness, is strongly dependent on individuals with high rates of sex partner change or concurrency (the so-called core groups) for continued propagation such that antibiotic treatment of core groups precipitates a rapid decline in the incidence and prevalence of infection in the population as a whole.5

Cases of gonorrhoea persist, however, despite targeted treatment of core groups, which has been standard and largely effective for decades. Since 1997, incidence rates have remained stable but high in the USA, while rates have been increasing in Canada,6 7 nearly doubling in the Canadian province of Ontario, for example, where rates have risen from 15 to 25 cases per 100 000 population between 1997 and 2007.8 Given that the target goal for gonorrhoea incidence in the USA was a reduction to 19 cases in 100 000 people by 2010 and the actual incidence was 118.9 cases in 100 000 people in 2007,9 there is a clear cause for concern.

Many factors likely contribute to the increasing numbers of infection in the last decade, including ‘risk compensation’ in sexual behaviour, due to perceptions of decreased risk and declining fear of HIV spread10 and the development of antimicrobial resistance in NG.11 Multidrug-resistant NG isolates are becoming increasingly common,8 and there have been reports of rapidly emerging resistance to ‘single-dose’ therapies (including fluoroquinolones and third-generation cephalosporins),12 suggesting that resistance might play a larger role in the dissemination of NG than previously acknowledged, specifically with regard to the rate of rebound infection.

While a focus on core groups might be the optimal means for controlling NG at the population level in the absence of antimicrobial resistance, few investigators have considered the impact that core groups could have in propagating resistant strains, thus paradoxically undermining disease control efforts. We sought to use a simple mathematical disease transmission model to evaluate the impact of core groups on the dissemination of antimicrobial resistance and to use modelling as a platform to explore the potential impact of competing antibiotic management strategies, with and without adjunctive laboratory testing, on the long-term control of NG.


The mathematical model

A schematic representation of the model used to represent the natural history of gonorrhoea is presented in figure 1. The model represents a modified version of the generic antimicrobial resistance model previously published by Bonhoeffer et al13 and is risk stratified.5 The model is described in further detail in online supplementary appendix 1. Berkeley Madonna (V. was used for all simulations.

Figure 1

Basic gonorrhoea transmission model, susceptible (S) and infected (I) individuals. Infected individuals may have a strain susceptible to both available antibiotics (IO), resistant to drug A or B (IA or IB, respectively) or resistant to both available antibiotics (IAB). Subscript i represents one of three possible sexual risk classes. Allowed transitions between classes are indicated by arrows; individuals may acquire resistant infection either through sexual contact or (uncommonly) as a result of antibiotic treatment while infected.

Parameterisation and population structure

Consistent with available data, we assumed that antimicrobial resistance in N. gonorrhoeae could be acquired as a result of genetic mutation in the face of selective antimicrobial pressure without a significant fitness cost11 12 14 such that the transmissibility and other characteristics of resistant strains were identical to their susceptible counterparts. We constructed a hypothetical population of 106 individuals, who were divided into three sexual risk groups of varying sizes such that each group contained an identical absolute number of infected individuals at baseline. Group construction was performed as follows: we first estimated the total number of NG infections that would occur in a given year in order to approximate observed trends in NG incidence. These cases were then divided into three risk tertiles, representing infections in individuals with high, low and intermediate levels of risk behaviour. Highest risk individuals were assigned a population denominator (eg, the size of the risk group that produced those cases) necessary to produce a prevalence of infection of 10%, which is similar to that seen in empirical studies of gonorrhoea in core groups.15 The intermediate-risk tertile was drawn from a population size sufficient to produce a prevalence of 1%; the size of the lowest risk group was determined by the total population size minus the size of the core and intermediate-risk groups. This population structure was chosen because (1) it approximates empirically observed right skewing of sexual risk behaviours in many populations16 and (2) because the structure permits comparison of treatment strategies that initially target identical absolute numbers of individuals in each sexual risk group (as the absolute number of cases in high-, intermediate- and low-risk tertiles is identical). Preference for sex partners from each sexual activity class was estimated using a mixing matrix using an approach similar to that described by Garnett et al17 in their work on partner change and HIV dynamics (details in online appendix 1). Parameter values used in the base case are presented in table 1.

Table 1

Model parameters and baseline values


The model was calibrated such that the baseline equilibrium incidence of recognised infection in the absence of treatment was 200 cases per 100 000 person-years, reflecting historical trends in NG incidence in Canada and the USA.18 19 Diagnosed cases of gonorrhoea likely represent only a subset of all cases in the community: a recent community-based probability sample in Baltimore, Maryland, estimated that the number of undiagnosed gonococcal infections likely exceeds diagnosed infections by a factor of 320; as such, our upper-bound estimate for all gonococcal infections was three times the number of observed infections. We estimated the total incidence of gonorrhoea (recognised and unrecognised) to be 600 cases per 100 000 person-years in the absence of treatment.

Treatment strategies

We assumed that individuals would receive treatment either as a result of clinical presentation with symptomatic gonorrhoea or as a result of testing. Treatment strategies, in the absence of testing for the presence of antimicrobial resistance, included (1) use of a single antimicrobial agent for all treated cases, (2) the random allocation of two available agents with fixed probability (eg, 50% chance of receiving antimicrobial A and 50% chance of receiving antimicrobial B), (3) combined treatment with antimicrobials A and B for all cases, (4) treatment of all cases with antimicrobial A until the probability of resistance to that agent exceeded a predefined threshold (eg, 5% of strains resistant as per WHO and Centers for Disease Control and Prevention recommendation11 12) at which time all individuals receive antimicrobial B and (5) differential treatment strategies for different risk groups based on population resistance levels. Strategy (5) is consistent with early US Centers for Disease Control and Prevention guidelines for avoidance of fluoroquinolones in men who have sex with men shortly after high rates of fluoroquinolone-resistant NG infection were observed in this group.21 In each of the model runs, a third of the total number of infected individuals in each risk group is treated such that downstream differences in disease incidence cannot be attributed to initial treatment of larger absolute numbers of individuals in a given risk group.

Rapid testing options for the presence of antimicrobial resistance in gonorrhoea are currently limited. However, the emergence of nanotechnology applications for microbial diagnosis suggests a relatively near future with the capacity for such swift assessment.22 Consequently, we conducted additional analyses in which hypothetical point-of-care testing for the presence of resistance to antimicrobial A or B was assumed to be available, and such testing information was integrated into clinical care such that the accurate identification of resistance phenotype resulted in targeted antimicrobial therapy.


Assumptions included (1) homogeneous mixing within risk groups and (2) resistance as an ‘all-or-none’ phenomenon, with partial resistance not included. It was also assumed that treatment effects were instantaneous and that treatment was 100% effective in individuals with susceptible infection and 0% effective in individuals with resistant infection. The population structure was highly simplified and structured only according to sexual activity level and not according to age or gender, and births, deaths and migrations were not included. As this is a susceptible–infected–susceptible system, the dynamical effects of births and deaths in reintroducing susceptible individuals would add little to the use of a fixed population.


Baseline point prevalence of NG increased with risk group, with the highest risk group having a point prevalence of 10% vs 23 per 100 000 in the low-risk group, 0.1% in the intermediate-risk group and 69 per 100 000 overall. In the absence of antimicrobial resistance, the most effective strategy, as expected, involved targeting treatment at the core group (figure 2). In the presence of antimicrobial resistance, a core group-focused strategy ceased to be optimal, ultimately leading to a greater increase in prevalence of infection in the overall population than did treatment targeted at low or intermediate groups or treatment distributed across population risk groups (figure 3A). When represented in terms of fraction of cases caused by resistant strains, a core group-focused strategy resulted in >90% of infections being caused by resistant strains within 16 years; this threshold was crossed 32 years more slowly when treatment was distributed across groups and was not crossed within a 500-year time span when treatment was focused on low-risk groups or on the intermediate-risk group.

Figure 2

The curves depict the prevalence of gonorrhoea infection in the simulated population over time, with treatment targeting different risk groups. Identical numbers of individuals are treated at baseline in each simulation, as described in the text. As expected, a strategy that focuses on high-risk core individuals results in elimination of gonorrhoea transmission.

Figure 3

(A) Prevalence of gonorrhoea infection in population over time, given different treatment target groups and in the presence of resistance; it can be seen that when resistance is present the core group-focused treatment results in rapid rebound of prevalence. (B) Change in point prevalence over time given (1) single treatment, (2) random drug allocation, (3) combination therapy, (4) point-of-care testing therapy and (5) switching from drug A to drug B for all groups after reaching population resistance of 5% to drug A. A third of all infected individuals in each tertile is treated under all scenarios. (C) Prevalence over time with different threshold level using the ‘threshold treatment strategy’, where a new antibiotic is used when the population-level resistance to the old antibiotic reaches a pre-set level, in comparison to random drug allocation. It can be seen that a commonly used 5% threshold results in the least durable reduction in prevalence. (D) Prevalence over time, with differential treatment strategies for different risk groups; differential allocation of antibiotics by risk group results in a more durable reduction in prevalence than does random allocation of antimicrobials.

When we examined the effect of using multiple antibiotic classes for treatment, we found that treating with a random combination of antibiotics delayed rebound in prevalence, especially when compared with treating with a single antibiotic, and also resulted in a slower increase in rebound prevalence compared with the treatment of all infections with combination treatment (figure 3B). Currently, NG treatment is based on the administration of specific antibiotics according to thresholds of resistance, so that patients will receive antibiotic A until the probability of resistance to A exceeds a certain threshold (typically >5%), after which treatment will be switched to antibiotic B11; when we compared the efficacy of treatment with random or combination antibiotics with switching treatment based on a range of thresholds, random antibiotic administration was again preferred over the long term (figure 3C).

The effect of a perfect ‘point-of-care’ test for accurate identification of NG resistance on the prevalence trend is nearly identical to that of combination therapy, offering no clear advantage as a treatment strategy (figure 3B). When treatment for the high-risk group was switched to antibiotic B after the population resistance to antibiotic A reached 5%, prevalence rebounded sooner in comparison to random drug allocation. The reverse (ie, only switching intermediate- and low-risk groups to a new drug) resulted in an even earlier rebound in prevalence, but both provided lower equilibrium prevalence over the long term compared with strategies tested (figure 3D).


The ecology of gonorrhoea, and in particular the short duration of infection relative to the average rate of sex partner acquisition in populations, suggests that treatment focused on core groups is the optimal means of controlling epidemic NG, and the marked declines in NG incidence in Canada and the USA since the 1970s provide a degree of validation for the expected impact of the core group-focused strategies. Using a simple mathematical model, we reconfirm this well-known finding but provide a novel and paradoxical insight that may help explain recent increases in NG incidence: when antimicrobial resistance is present, focusing antimicrobial treatment efforts on core groups serves to accelerate the spread of resistant NG strains to intermediate- and low-risk populations, ultimately undermining disease control efforts and precipitating rebound of infection rates to baseline levels.

The situation is further complicated by changes in diagnostic testing practice, which has been shifting from culture to nucleic acid amplification testing in recent years.8 As nucleic acid amplification testing replaces culture-based methods, both the ability to monitor resistance trends and the ability to identify discordant therapy and to retreat ineffectively treated individuals decline. If disease control strategies are based solely on targeting segments of the population according to their risk status, this observation would create a conundrum for public health agencies: should the aim be short-term disease control (with loss of antibiotic susceptibility and rebound to baseline over time, as would occur with a core group-focused strategy) or more limited reduction in disease prevalence, with more durable effects and maintenance of antimicrobial susceptibility (as would occur with treatment focused on lower risk individuals and groups). We do not claim to have an answer to this question: an optimal formulation would have to take into account the rights of individual patients and their sex partners, the rights of the community as a whole, the degree to which future health events are ‘discounted’ as a result of time preference23 and expectations around the likelihood of emergence of novel, safe, well-tolerated and easy to use drug regimens for the treatment of NG.12 Unfortunately, there is no reason to be particularly optimistic with regard to this last dimension. Few novel antimicrobial classes have been developed in recent decades, and most novel agents are parenterally administered drugs targeted at multiresistant pathogens likely to be encountered in the healthcare setting.24

Of course, drug treatment strategies are not restricted to a single agent. We explored competing multidrug strategies, including the use of resistance prevalence thresholds for changing preferred initial drug treatment, a strategy which is commonly used in North America12 as well as differential drug class recommendations for population risk groups according to prevalence levels21 and combination therapy, as has been used with success in the management of tuberculosis and HIV infection. We found that random allocation of available drug classes is, in fact, a strategy preferable to the adoption of thresholds across a fairly long time horizon (figure 3B); while this observation may be surprising to some and this strategy would be difficult to implement in practice, it is similar to observations in models evaluating ‘antibiotic cycling’ in the intensive care unit environment. These models also suggest that random allocation minimises selective pressure over time.13 Among threshold strategies, time horizon was an extremely important consideration: maximising the prevalence threshold before a drug recommendation is changed most markedly reduced gonorrhoea incidence over time but would provide practical and ethical difficulties: could a physician or public health agency advocate treatment of an individual with a therapeutic agent that is likely to be ineffective? More nuanced threshold strategies that considered differential antimicrobial recommendations by the risk group were projected to provide the most durable reductions in NG incidence over time, by effectively building an antimicrobial resistance ‘firewall’ between the core group individuals who disseminate resistant strains and individuals outside the core group. However, such divergent recommendations are also likely to raise both practical and ethical concerns.

Finally, we evaluated the projected benefits that might be expected with a rapid, accurate, point of care test for identification of antimicrobial-resistant infection; such a test would be highly desirable and could provide long-term preservation of antimicrobial class effectiveness comparable to other highly favourable strategies, without raising the equity and ethical issues that would be inherent in risk-based strategies or more stringent thresholds for changing first-line therapeutic recommendations. However, such technologies are not yet available for clinical use, may be costly and may not prove appropriate for use in outreach or low-resource settings.

Like any model study, this one has limitations. Our model is simple, which provides transparency and makes it an imperfect representation of real-world complexity. Nonetheless, the model does represent several key attributes of how gonorrhoea spread is dependent on the existence of a high-risk core group and the effects of group-targeted treatment on the overall prevalence trend of NG. There is uncertainty in our parameter estimates, and sensitivity analyses showed that using different parameter values changes the time horizon for prevalence rebound, but the overall qualitative projections of the model were robust. Finally, our model tests a fairly limited constellation of strategies and does not consider the possibility that current gonorrhoea dynamics are driven, for example, by risk compensation due to decreasing fear of HIV,10 wider availability of high-risk sex partnership via the internet25 or other social dynamic changes. We feel this was appropriate, as our focus here was on trends in dissemination of resistant NG rather than crude changes in NG incidence. Future models should incorporate greater complexity and realism to better inform specific policy choices related to the treatment of gonorrhea.

Thus, perhaps the most important insight to emerge from our model is that NG has developed a challenging clinical endgame: NG can only be effectively controlled at current rates if the core group individuals are targeted, but targeting of core groups will disseminate resistant strains, undermining control efforts. Given the slow pace of development of novel antibiotic classes (and indeed, the reduced availability of such easy-to-administer oral preparations for NG treatment as oral cefixime in the USA26), the emergence of NG resistance at high prevalence highlights the need for non-drug strategies for control of this disease. Vaccine development for NG is challenging, given the ‘antigenic switching’ characteristic of NG.27 Nevertheless, condoms remain a very effective means of preventing NG,28 although they are commonly overlooked or taken for granted. As a reminder of condoms' effectiveness in preventing sexually transmitted infections in the context of HIV prevention, Dr Kent Sepkowitz wrote: “[w]ere a vaccine to be comparably effective…the F.D.A. would scramble to approve it.”29 In the absence of a technological solution to the current problem of rising gonorrhoea rates, these are words worth remembering.

Key messages

  • Gonorrhoea control strategies that focus on treatment of core groups represent a major recent public health success.

  • Increasing antimicrobial resistance in N gonorrhoeae threatens to erode this achievement.

  • A mathematical model suggests that, paradoxically, the focus on core groups needed for effective control of gonorrhoea accelerates the spread of antimicrobial resistance.

  • The durability of antibiotic effectiveness for gonorrhoea would be improved via strategies other than those based on ‘resistance thresholds’, currently in common use.


Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Files in this Data Supplement:


  • Competing interests None.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement All data used in the model are publicly accessible. Model code can be obtained from Dr. Fisman.