Non-selective distribution of infectious disease prevention may outperform risk-based targeting

The study examines how to optimally distribute primary prevention tools (such as vaccines or PrEP) when individual infection risk is hard to measure, particularly in stigmatized or marginalized populations like MSM. Conventional guidance favors targeting individuals at highest risk (e.g., those with many contacts who can drive superspreading). However, accurately estimating risk and contact structures is difficult and often infeasible for routine public health practice. The authors hypothesize that the optimality of risk-based targeting depends critically on the individual-level infection-preventing efficacy of the intervention. They propose that below a critical efficacy threshold, non-selective (random) distribution may outperform risk-based strategies, and aim to derive this threshold analytically and test its implications for PrEP distribution for HIV among MSM worldwide.

Existing guidelines and cost-effectiveness analyses often recommend offering PrEP to those at high risk of acquiring HIV. Prior network epidemiology work shows that protecting high-degree nodes (potential superspreaders) can reduce transmission and has inspired risk-based immunization strategies. However, practical implementation is hindered by poor accuracy of risk metrics, stigma, punitive laws, and limited availability of detailed contact network data. Extensions of network-based prioritization exist, but most require information that is rarely available in routine surveillance. There is also recognition that partial efficacy leads to breakthrough infections, especially among highly exposed individuals, complicating the assumption that targeting hubs is always optimal.

The authors develop a theoretical framework using heterogeneous mean-field (HMF) theory on annealed networks with degree distribution p(k), focusing on settings where contact dynamics are fast relative to disease progression. They model transmission with a Susceptible-Infectious-Susceptible (SIS) process where the per-contact transmission rate is λ and recovery rate is μ. Individual-level prevention efficacy e (ε) reduces the probability of infection upon exposure; e=1 implies complete protection. They derive a linear response function f(k) quantifying the change in endemic prevalence resulting from providing a small amount of prevention to individuals in degree class k starting from zero coverage. This response decomposes into direct protection (F_dr), which may peak at finite k when e<1 due to breakthrough risk increasing with exposure, and indirect protection (F_ind), which increases with k because protecting highly connected individuals reduces onward transmission risk to others. Combining these effects yields an optimal targeted degree k* and an analytically derived critical efficacy e_c that separates two regimes: for e≥e_c, prioritizing high-risk (high-degree) individuals is optimal; for e<e_c, non-selective distribution can outperform risk-based targeting. They analyze how e_c depends on baseline endemic prevalence (in the absence of prevention) and network heterogeneity (e.g., coefficient of variation of a negative binomial degree distribution), showing that higher prevalence and greater heterogeneity increase e_c. They further assess the invasion stage via the epidemic threshold, finding that early epidemic dynamics always favor high-risk targeting. Empirically, they parameterize MSM contact heterogeneity with negative binomial distributions informed by surveys and compute effective prevalence (i.e., fraction of individuals capable of transmitting HIV) using data on HIV prevalence, treatment coverage, and viral suppression for 58 countries and 24 cities. They test robustness to assortative mixing typical of MSM networks and perform sensitivity analyses to different PrEP efficacy values.

- Theoretical result: There exists a critical prevention efficacy e_c (analytically computable) that determines the optimal distribution strategy. Above e_c, prioritizing high-degree (high-risk) individuals is optimal; below e_c, non-selective distribution can yield larger reductions in community prevalence. - Direct vs indirect effects: With imperfect efficacy (e<1), direct protection F_dr can peak at a finite degree due to higher breakthrough risk among hubs, while indirect protection F_ind increases monotonically with degree. The combined effect yields a finite optimal targeted degree k* when e<e_c. - Dependence on context: e_c increases with baseline endemic prevalence and with network heterogeneity (e.g., higher coefficient of variation in degree), implying that the same intervention efficacy may call for different strategies across settings. - Application to HIV PrEP in MSM: Using effective prevalence estimates for 58 countries and 24 cities and a negative binomial degree model, setting PrEP efficacy to 60% placed 34 of 78 communities in the high-efficacy region and 44 in the low-efficacy region, with 4 in a transition zone. - Geographic patterns: Many European communities lie in the high-efficacy region (supporting risk-based targeting), while several in southern Africa fall into different regions despite geographic proximity (e.g., Namibia vs Botswana). - Sensitivity to efficacy: Lowering efficacy to 44% (trial estimate) would shift only 2 of 76 communities across regions; increasing to 86% (IPERGAY) would shift 5 of 76, indicating robustness of regional assignment to efficacy assumptions. - Assortative mixing: Incorporating empirically estimated assortativity produced minimal changes; only a few communities changed region, and large changes occurred only under unrealistically strong assortativity. - Implication across epidemic stages: While invasion-stage analysis supports risk-based targeting, in endemic contexts with lower efficacy, non-selective distribution can outperform, particularly in high-prevalence, low-treatment-coverage settings.

The findings support a context-dependent approach to distributing prevention: when individual-level efficacy is high relative to a community’s prevalence and contact heterogeneity, risk-based targeting of high-degree individuals remains optimal. However, in many real-world endemic settings with imperfect efficacy, especially where HIV prevalence is high and treatment coverage is low, non-selective distribution can reduce community-level prevalence more effectively and is logistically simpler. As treatment coverage expands and viral suppression increases, communities may transition toward the high-efficacy region where risk-based targeting regains optimality. Advances such as long-acting injectable PrEP (e.g., CAB-LA), with higher efficacy than oral PrEP, can accelerate movement into high-efficacy regimes, potentially shifting optimal strategies over time. Overall, the study reframes prevention allocation by highlighting the trade-off between protecting hubs (high indirect benefit) and their higher breakthrough risk when efficacy is modest, offering a principled, analytically grounded criterion to guide policy.

The paper introduces an analytical framework demonstrating that a critical efficacy threshold governs whether risk-based targeting or non-selective distribution best reduces community prevalence. Applying this to HIV PrEP among MSM shows that many communities, given current efficacy and epidemiological conditions, would benefit more from non-selective approaches. The work provides a practical, data-informed way to tailor prevention strategies to local prevalence, treatment coverage, and network heterogeneity. Future research should integrate richer behavioral dynamics, heterogeneous transmission probabilities, and side-effect considerations; extend to other pathogens (e.g., malaria) with appropriate transmission models; and refine community-specific estimates of contact structure to sharpen critical efficacy calculations.

- Modeling simplifications: The SIS-based, heterogeneous mean-field model is coarse-grained and does not capture detailed behavioral dynamics, heterogeneous per-act transmission probabilities, or potential risk compensation behaviors. - Scope of prevention effects: The analysis focuses on infection-preventing efficacy and does not address morbidity/mortality reduction benefits, which can drive different allocation priorities (e.g., COVID-19 vaccines). - Access and adherence: Factors that jointly affect access, adherence, and thus realized efficacy are not explicitly modeled; stigma and criminalization can reduce adherence and effective efficacy. - Side effects and ethical constraints: The framework does not incorporate scenarios where targeting only high-exposure individuals is necessary due to risk–benefit trade-offs tied to side effects. - Data and parameter uncertainty: Estimates of effective prevalence, contact heterogeneity, and assortativity carry uncertainty; some reported sample sizes vary across analyses, and high-resolution contact data are limited. - Generalizability: While invasion-stage results favor risk-based targeting, the main results pertain to endemic contexts; translation to other diseases requires model adaptation (e.g., vector-borne transmission, different mixing patterns).