Results at the end of the run
| Prevalence | Incidence rate per 100 |
|---|
| Group | with Group 1 | with Group 2 |
|---|
An interactive two-group model of contact rates, mixing, and transmission, from the EpiModel ecosystem
The population is split into two groups. Each group has its own contact rate, and the mixing statistic Q sets how many of those contacts stay within the group. The model asks how these two features shape the incidence and prevalence of infection in each group, both when the groups start out equal and when one group starts with a higher burden.
It is a deterministic susceptible-infected (SI) model in an open population, originally built to study HIV disparities among men who have sex with men, but the dynamics apply to any infection spread through contact.
Set the parameters on the left and press Run Model. The plots show prevalence and incidence over time for each group, and the results table reports the values at the end of the run along with the mixing matrix implied by your settings.
Before each run, take a minute to predict what will happen to each group. The guided exercises below walk through the key comparisons. Copy link saves the current settings in a shareable URL.
| Prevalence | Incidence rate per 100 |
|---|
| Group | with Group 1 | with Group 2 |
|---|
Each exercise starts from the default settings (use Reset defaults). Predict first, then run.
With the defaults, everything about the two groups is the same, so both head to the same equilibrium: 70% prevalence and an incidence rate of 7.00 per 100 person-time. Now raise Q toward 1. Nothing changes, because when the groups behave identically it does not matter whom they mix with.
Set the Group 2 contact rate to 6 with Q = 0. Group 2's burden rises, and so does Group 1's, because under proportional mixing many of Group 1's contacts are with the more active and more often infected Group 2. Now raise Q to 0.9, then 1. Assortative mixing shields Group 1 and widens the gap between the groups.
Reset, then give Group 2 a higher initial prevalence (for example, 0.20) and set Q to 1. The groups start apart but converge to the same equilibrium. Assortative mixing alone, however strong, cannot maintain a disparity that arose from some earlier cause; incidence starts to converge right away, and prevalence follows.
Reset, then set the transmission probability from Group 2 to 0.10, as if infected members of Group 2 were less often virally suppressed. With Q = 0 the two groups still end up with equal prevalence, because each group's contacts are spread across both groups in the same way. Raise Q and a sustained disparity appears, with Group 2 carrying the higher burden. Changing the initial prevalence values now changes the path but not the destination.
Set one group's initial prevalence to 0. As long as Q is below 1, infection eventually reaches that group and the same equilibrium results. Only when Q = 1 (complete separation) does the uninfected group stay uninfected. Real groups that share a community are rarely completely separated.
The ability of a factor to sustain a disparity is tied to its ability to generate one. Assortative mixing by itself generates no disparity between groups that otherwise behave alike, so it cannot sustain one either. Combined with a real difference between groups, such as contact rates, transmission, or partnership timing, assortative mixing keeps the groups' epidemics apart and makes the disparity larger and more durable.
Everyone belongs to Group 1 or Group 2 and is either susceptible (S) or infected (I), with no recovery. People enter the population as susceptibles at the birth rate, split between the groups in proportion to each group's current size, and leave at a death rate specific to their group and infection status. Time is in arbitrary units; all rates are per unit of time.
A member of group i makes ci contacts per unit time. The share made with group j is gij, set from Q:
p1 = c1N1 / (c1N1 + c2N2), g11 = p1 + Q(1 − p1), g12 = 1 − g11
g21 = g12 c1N1 / (c2N2), g22 = 1 − g21
The second line makes contacts balance: the contacts Group 1 makes with Group 2 equal those Group 2 makes with Group 1. With two groups, Q = g11 + g22 − 1, the same Q statistic computed from an observed mixing matrix. Q can fall below 0 (disassortative mixing) only as far as keeps every share between 0 and 1.
The force of infection on susceptibles in group i, with ρj the per-contact transmission probability from an infected member of group j:
λ1 = c1 [ g11 ρ1 I1/N1 + g12 ρ2 I2/N2 ] λ2 = c2 [ g22 ρ2 I2/N2 + g21 ρ1 I1/N1 ]
dSi/dt = b Ni − λi Si − μSi Si dIi/dt = λi Si − μIi Ii
The incidence rate shown is λi × 100: new infections per 100 person-time among those at risk. The equations are solved with fourth-order Runge-Kutta at a step of 0.1.
Existing differences in HIV prevalence by race are often used to explain differences in incidence: with strong assortative mixing and higher prevalence among Black men who have sex with men (MSM), an HIV-negative Black MSM is more likely than a White MSM to have partners with HIV. That logic explains short-term differences in incidence well, and it is tempting to extend it to long-term disparities. Modeling theory says otherwise: for a pre-existing disparity to persist, something besides assortative mixing must differ between the groups (Simon and Jacquez 1992; Morris et al. 2009). This app lets you see that for yourself.