Is 2 a “High Number of Partners”? Modeling, Data, and the Power of Concurrency

To the Editor

Go and Blower¹ claim 3 necessary conditions for concurrency to drive heterosexual HIV epidemics: (1) “there should be many concurrent partnerships in the population;” (2) “the number of concurrent partners should be fairly high for the average individual;” (3) “the duration over which the partnerships overlap should be fairly long.” Also, because “no heterosexual community in which these conditions are met has been identified,” they conclude that concurrency is irrelevant to African HIV epidemiology. They provide no citations for their conditions, with good reason: the first 2 are untrue. Condition 1 depends on the definition of “many,” but modeling shows that only a relatively small fraction of the population need concurrent partners to generate sizeable heterosexual epidemics. Condition 2 is entirely incorrect.

For example, Goodreau et al.² considered behavioral data from 18- to 30-year-old Zimbabweans, the ages where incidence is concentrated; 11% of men and 5% of women had concurrent partners. Mean momentary degree (MMD; i.e., number of ongoing partnerships at any point) was 0.66. Using per-stage transmission probabilities,³ they showed that this would generate 9% HIV prevalence. If sexual contacts and concurrency were slightly underreported (MMD increased to 0.7, concurrency by 2 percentage points), prevalence rose to 14%. The population is on a threshold, and small underreporting in concurrency (a socially undesirable behavior, for women especially) could generate a realistic epidemic. Among those with concurrent partners, 79% had the minimum possible (2); MMD for those with concurrent partners was 2.28. When modeling the same MMD and durations but without concurrency, the epidemic disappeared. Concurrent partnerships do not need to be very numerous to drive an epidemic.⁴

As explained elsewhere,⁵ the compartmental models predominating in HIV epidemiology for a quarter century cannot explicitly include concurrency. To generate heterosexual epidemics, they typically include mean lifetime partner counts across the entire population near or above 100,^6–14 and/or that approximately 7% to 26% of the population have lifetime partnerships numbering many hundreds to thousands, ^{7–9,11–13,15} although no population-based survey contains numbers resembling this. Some authors¹⁶ including Vardavas and Blower¹⁷ justify this by explicitly omitting behavioral data and assuming partnership counts to match observed HIV prevalence. Others have highlighted the behavioral implausibility required by compartmental models¹⁸ and shown that realistic serial monogamy parameters cannot generate observed African HIV epidemics.^19,20 However, Go and Blower accuse network modelers of being “unconstrained by empirical data” and suggest that we simply continue telling all Africans to avoid having many sex partners. This message may be useful to some—for example, women with 5 or more recent partners. However, for those with only 2, but which overlap temporally—and such women are more common and, as network models show, contribute more to HIV spread than common sense suggests—that message will continue to fall on deaf ears.

I implore more researchers to take the effort to understand modeling and closely analyze different models’ assumptions. No model is perfect, but in comparison with compartmental models that cannot include all consequences of relational overlap and, instead, assume unrealistic partnership counts—making policy recommendations driven by those assumptions—I am confident that network-based concurrency models will fare well in the evaluation of who is unconstrained by empirical data.