A Novel Method to Estimate the Indirect Community Benefit of HIV Interventions Using a Microsimulation Model of HIV Disease

Abstract

Background:

Microsimulation models of human immunodeficiency virus (HIV) disease that simulate individual patients one at a time and assess clinical and economic outcomes of HIV interventions often provide key details regarding direct individual clinical benefits (“individual benefit”), but they may lack detail on transmissions, and thus may underestimate an intervention’s indirect benefits (“community benefit”). Dynamic transmission models can be used to simulate HIV transmissions, but they may do so at the expense of the clinical detail of microsimulations. We sought to develop, validate, and demonstrate a practical, novel method that can be integrated into existing HIV microsimulation models to capture this community benefit, integrating the effects of reduced transmission while keeping the clinical detail of microsimulations.

Methods:

We developed a new method to capture the community benefit of HIV interventions by estimating HIV transmissions from the primary cohort of interest. The method captures the benefit of averting infections within the cohort of interest by estimating a corresponding gradual decline in incidence within the cohort. For infections averted outside the cohort of interest, our method estimates transmissions averted based on reductions in HIV viral load within the cohort, and the benefit (life-years gained and cost savings) of averting those infections based on the time they were averted. To assess the validity of our method, we paired it with the Cost-effectiveness of Preventing AIDS Complications (CEPAC) Model – a validated and widely-published microsimulation model of HIV disease. We then compared the consistency of model-estimated outcomes against outcomes of a widely-validated dynamic compartmental transmission model of HIV disease, the HIV Optimization and Prevention Economics (HOPE) model, using the intraclass correlation coefficient (ICC) with a two-way mixed effects model. Replicating an analysis done with HOPE, validation endpoints were number of HIV transmissions averted by offering pre-exposure prophylaxis (PrEP) to men who have sex with men (MSM) and people who inject drugs (PWID) in the US at various uptake and efficacy levels. Finally, we demonstrated an application of our method in a different setting by evaluating the clinical and economic outcomes of a PrEP program for MSM in India, a country currently considering PrEP rollout for this high-risk group.

Results:

The new method paired with CEPAC demonstrated excellent consistency with the HOPE model (ICC=0.98 for MSM and 0.99 for PWID). With only the individual benefit of the intervention incorporated, a PrEP program for MSM in India averted 43,000 transmissions over a 5-year period and resulted in a lifetime incremental cost-effectiveness ratio (ICER) of US$2,300/year-of-life saved (YLS) compared to the status quo. After applying both the direct (individual) and indirect (community) benefits, PrEP averted 86,000 transmissions over the same period and resulted in an ICER of US$600/YLS.

Conclusions:

Our method enables HIV microsimulation models that evaluate clinical and economic outcomes of HIV interventions to estimate the community benefit of these interventions (in terms of survival gains and cost savings) efficiently and without sacrificing clinical detail. This method addresses an important methodological gap in health economics microsimulation modeling and allows decision scientists to make more accurate policy recommendations.

Graphical Abstract

INTRODUCTION

Microsimulation models of human immunodeficiency virus (HIV) disease progression and treatment are valuable tools that can help assess the clinical and economic outcomes of HIV interventions and determine cost-effectiveness among a set of competing strategies [1-3]. These models follow individual persons at risk of acquiring HIV and persons with HIV (PWH) using group- and country-specific transition probabilities of HIV disease acquisition, progression, and treatment. HIV microsimulations often model individuals one at a time and record statistics of interest for each person. Therefore, they are memory-efficient and can incorporate extensive individual-level clinical details such as current and past HIV viral load and immune function (CD4 count), current and prior treatments, opportunistic infections and chronic diseases associated with HIV, loss to follow-up and return to care, and other clinical details that may affect disease progression, morbidity, mortality, and cost. Hence, they can accurately estimate the direct “individual benefit” (e.g., survival gains and HIV cost savings) of different HIV testing, prevention, and treatment interventions.

However, interventions that treat HIV or prevent the transmission of HIV to people at risk of acquiring the virus not only benefit those receiving the intervention, but also benefit persons in the larger transmission network (e.g., sexual network) who are not reached by the program, since these partners are subsequently at lower risk of acquiring HIV due to fewer of their partners living with HIV (on average) [4]. In this case, the overall health benefits of an intervention are greater than the sum of individual benefits. Therefore, to fully capture the benefits of interventions that can prevent HIV transmissions, HIV microsimulation models could more accurately account for the direct individual benefit of the intervention to persons who receive it, as well as the broader indirect benefit (“community benefit”) of averting transmissions to all persons in the same transmission network, whether or not they receive the intervention. It is, however, challenging to estimate the community benefit without a detailed dynamic transmission model of the HIV epidemic in a given country; because individual patients are tracked one at a time in a typical microsimulation, these models traditionally lack detail in terms of how individuals transmit HIV to one another [5].

In this paper, we seek to develop, validate, and demonstrate a novel, practical method to estimate the community benefit of HIV interventions that help prevent transmission of HIV without a dynamic transmission model. Examples of such interventions include: (1) routine HIV testing, which can help detect newly-infected individuals and decrease the number of individuals who present with advanced HIV disease, thus reducing HIV transmissions [6]; (2) prevention interventions, such as HIV pre-exposure prophylaxis (PrEP) that can protect individuals from acquiring HIV [7]; and (3) treatment interventions, such as antiretroviral medications (ART), which can also decrease HIV transmissions by achieving virological suppression (undetectable viral load) among PWH and prevent onward transmission [8]. The model incorporates the community benefit into incremental cost effectiveness ratios (ICERs), defined as the change in cost over the change in life expectancy, a common metric used to inform health policy decision-making [9].

This paper contributes to the literature by providing a novel method that allows microsimulation models of HIV disease to fully capture the broader community benefit of HIV interventions, specifically the indirect gains (in terms of life-years gained and cost savings) of such interventions. By developing a novel extension to already existent microsimulation models of HIV, we enable them to incorporate the community benefits, allowing existing models to maintain advantaging from their clinical detail. In a case study, we assess the clinical and economic outcomes of a PrEP program targeted to HIV-uninfected men who have sex with men (MSM) in India. This case study was selected for illustrative purposes, as MSM’s sexual network includes non-MSM individuals, such as transgender women (TG), substantial data exists regarding incidence and transmission in this population, and PrEP rollout for MSM is currently under consideration by the National AIDS Control Organization (NACO) in India [10-12].

METHODS

Analytic Overview

We first describe the details of a new method to estimate transmissions averted and account for the indirect community benefit of HIV interventions. Then, by applying our approach within the Cost-Effectiveness of Preventing AIDS Complications (CEPAC) model, a validated and widely-cited microsimulation model of HIV disease, we simulate a PrEP strategy for HIV-uninfected high-risk MSM and people who inject drugs (PWID) in the US and estimate the number of transmissions averted with a 5-year PrEP program [1, 2]. We validate our method by comparing our transmission results against outcomes estimated by the HIV Optimization and Prevention Economics (HOPE) model – a dynamic transmission model of HIV disease maintained by the Centers for Disease Control and Prevention (CDC) – for the same analysis [5]. Then, to illustrate our method, we leverage the CEPAC model and this new method to estimate the number of HIV transmissions averted and the community benefit of offering PrEP to HIV-uninfected MSM in India. We assess how including the indirect community benefit can influence the estimated clinical outcomes, costs, and cost-effectiveness of PrEP in this analysis.

Terminology and Notation

See Table 1 for a description of terminology used throughout this manuscript.

Table 1.

Glossary of terminology

Term	Definition
Primary cohort	The cohort of interest being modeled in a microsimulation, which can include HIV-infected and/or HIV-uninfected individuals (e.g., MSM in a given country).
Susceptible cohort	The cohort to which the primary cohort transmits, which may or may not include part of the primary cohort. For example, if MSM are being modeled as the primary cohort, then other MSM are part of the primary cohort, whereas transgender women and female partners are not part of the primary cohort.
Viral load	A measure of HIV virus in the body. For a given individual, the higher the viral load, the higher the transmission rate [4]. It is known that HIV-infected persons with undetectable viral load (virally suppressed) do not transmit the disease [34, 35].
Status Quo strategy (SQ)	In the SQ strategy, we assume that age-stratified incidence in the population does not vary from model start, so as to mimic a stable epidemic.
Intervention strategy (INT)	In the INT strategy, a clinical intervention is implemented, such as a PrEP program, which causes age-stratified HIV incidence to decline over time due to both direct prevention at the individual level and the additional community benefit due to fewer HIV infections within the primary cohort (thus, fewer transmitters).
Inside transmission rate	Transmission rate from an HIV-infected member of the primary cohort to HIV-uninfected members of the same primary cohort.
Inside transmissions	Number of transmissions that occur between HIV-infected and HIV-uninfected members of the primary cohort over a given period (e.g., transmissions from HIV-infected MSM in the primary cohort to HIV-uninfected MSM in the same primary cohort, if MSM are being modeled as the primary cohort).
Inside-attributable incidence rate	The portion or component of the incidence rate due to transmissions from HIV-infected to HIV-uninfected members of the primary cohort (e.g., the component of MSM incidence attributable to HIV-infected members of the primary cohort, if MSM are being modeled as the primary cohort).
Outside transmissions	Number of transmissions from HIV-infected primary cohort members to those not in the primary cohort (e.g., transmissions from MSM to transgender women, if MSM are being modeled as the primary cohort).
Outside-attributable incidence rate	The portion of the incidence rate due to transmissions from those outside the primary cohort to those inside the primary cohort (e.g., transmissions from transgender women to MSM, if MSM are being modeled as the primary cohort, or from HIV-infected prevalent MSM to HIV-uninfected MSM if only HIV-uninfected MSM are modeled in the primary cohort).
Aggregate incidence	Inside- plus outside-attributable incidence (rate or probability).
Cumulative incidence	Cumulative incidence (rate or probability) throughout the duration of the intervention (e.g., from simulation time 0 to t), as opposed to a monthly incidence.
Individual benefit of the intervention (direct benefit)	The direct benefit of the intervention for individuals who receive it, in terms of life-years gained and HIV costs averted (costs saved).
Additional community benefit of the intervention (indirect benefit)	Benefit attributable to preventing transmissions from individuals who receive the intervention. Due to lower viral load in the transmission network, over time, compared to status quo (SQ), HIV incidence declines gradually, which results in additional life-years gained and HIV cost savings for individuals who receive the intervention (on top of the individual benefit of the intervention) and for those who do not receive the intervention but who benefit indirectly. “Additional” implies the benefit that does not overlap with the individual benefit.
Averted transmission	A transmission that occurred in the less effective strategy (e.g., the SQ strategy) but did not occur in the more effective strategy (e.g., the INT strategy) at a given point in time. Note that a transmission that is averted at a given time can still occur later.

HIV: human immunodeficiency virus; MSM: men who have sex with men; PrEP: pre-exposure prophylaxis

A complete list of variables and their definitions is provided in Table 2. For all variables, subscripts 0 and t denote simulation time 0 and t respectively, subscript 0:t represents a cumulative value from time 0 to t, the subscripts in/out denote attributable to inside/outside the primary cohort, and the subscript agg denotes aggregate (i.e., inside + outside).

Table 2.

Table of notation.

Notation	Description	Source
Rv	The calibrated country- and population-specific rate of transmission from HIV-infected members of the primary cohort in viral load stratum v ∈ V to any susceptible cohort of interest. V is the set of viral load strata (see Methods).	Derived
NHIV−	The number of HIV-uninfected persons in the primary cohort at risk of infection at index year (simulation time 0).	Literature
PSQ,0	Monthly incidence probability at time 0 in the primary cohort in the status quo strategy.	Literature
Pup	Proportion of the primary cohort members who participate in the intervention.	Literature
Peff	The relative incidence reduction of a prevention intervention for individuals who participate. For PrEP, this can be viewed as a combined adherence/efficacy parameter, and for ART and testing, this can be assumed to be 0. This parameter accounts for the instantaneous incidence reduction at the time the person initiates the prevention intervention.	Literature
PINT,0	Monthly incidence probability at time 0 in the primary cohort in the intervention strategy. This value should be lower than PSQ,0 if both Pup and Peff> 0.	Derived
ISQ,agg (IINT,agg)	Aggregate incidence rate in the status quo (intervention) strategy, including inside- and outside-attributable incidence.	Literature (Derived)
ISQ,in (IINT,in)	Incidence rate in the status quo (intervention) attributable to those inside the primary cohort (inside-attributable incidence).	Literature (Derived)
ISQ,out (IINT,out)	Incidence rate in the status quo (intervention) attributable to those not in the primary cohort (outside-attributable incidence)	Derived
t	Duration of the intervention (in months).	User-defined
PSQ,0:t (PINT,0:t)	Cumulative incidence probability within the primary cohort in the status quo strategy (intervention strategy).	Derived
ISQ,0:t, agg	Aggregate infections within the primary cohort in the status quo strategy from time 0 to t.	Derived
TSQ,0:t,in (TINT,0:t,in)	The number of inside transmissions in the status quo (intervention) from time 0 to t.	Model output
ISQ,0:t, in (IINT,0:t, in)	The number of inside-attributable infections in the status quo (intervention) from time 0 to t.	Model output
ISQ,0:t,out (IINT,0:t,out)	Cumulative infections within the primary cohort from time 0 to t in the status quo (intervention) strategy attributable to those outside the primary cohort.	Derived
IINT,0:t,agg	Aggregate infections within the primary cohort in the intervention strategy from time 0 to t.	Derived
(PSQ,0:t,out (PINT,0:t,out)	The cumulative probability of acquiring HIV between time 0 and t in the status quo (intervention) strategy due to transmissions attributable to outside the primary cohort.	Derived
${\overline{\mathit{P}}}_{\mathit{INT}}$	The average monthly incidence probability (from time 0 to t) within the primary cohort in the intervention strategy.	Derived
PINT,t	The monthly incidence probability at time t in the intervention strategy.	Derived
y	The year (since model start) in which a hypothetical transmission is averted due to the intervention. We assume y ∈ (1,2,…,Y}. Y is an arbitrary user-defined horizon and may correspond to either the duration of the intervention, the lifetime of the primary cohort, or another relevant horizon.	User-defined
TSQ,y,out (TINT,y,out)	Total transmissions to outside the primary cohort (“outside transmissions”) in the status quo (intervention) in year y.	Model output
TAy	Transmissions averted in year y due to the intervention.	Derived
LHIV+,y	Average life-years of a susceptible cohort that starts off with acute HIV infection (for a transmission in year y).	Model output
LHIV−,y	Average life-years of a susceptible cohort that starts off HIV-uninfected but is subject to an HIV incidence (for a transmission in year y).	Model output
CHIV+,y	Average HIV cost of a susceptible cohort that starts off with acute HIV infection (for a transmission in year y).	Model output
CHIV−,y	Average cost of a susceptible cohort that starts off HIV-uninfected but is subject to an HIV incidence (for a transmission in year y).	Model output
tilde (~)	Denotes that life years/costs have been further discounted based on the year the transmission was averted and thus have been brought back to time 0 of the primary cohort (index year). For instance, for a transmission that is averted in year y, L is the discounted average life-years in year y (automatically discounted in the model), whereas $\tilde{\mathit{L}}$ is the corresponding value further discounted and brought back to time 0 of the primary cohort (this is done manually outside the model via Eq.’s 6-10).
${\tilde{\mathit{L}}}_{\mathit{TA}}$	Total life-years gained via all the transmissions averted over Y years.	Derived
${\tilde{\mathit{C}}}_{\mathit{TA}}$	Total change in HIV-related costs due to all the transmissions averted over Y years. Note that this number is expected to be negative, as averting transmissions is likely to reduce costs (but does not have to for the method to work).	Derived
d	Model discount rate (often set to 3%, per convention [16]).	User-defined
$\Delta \tilde{\mathit{L}}$	The average life-years difference between the status quo and intervention for the primary cohort, discounted at d (often 3%).	Model output
$\Delta \tilde{\mathit{C}}$	The average cost difference between the status quo and intervention for the primary cohort, discounted at d (often 3%).	Model output
N	The size of the primary cohort.	Literature or model run size
ICER	Incremental cost-effectiveness ratio (change in cost divided by change in life expectancy), including the benefits of averting HIV transmissions.	Derived

ART: antiretroviral therapy; HIV: human immunodeficiency virus; PrEP: pre-exposure prophylaxis.

Transmission Structure

HIV transmission has been shown to be dependent on both HIV prevalence and the viral load of people living with HIV (PWH) [4, 13]. For members of the primary cohort (the cohort being modeled in a microsimulation), natural history data on HIV viral load dynamics can be used to determine how an HIV-infected person’s viral load changes as he or she links to care, becomes virally suppressed, or becomes lost-to-follow-up [14, 15]. We stratify viral load into strata v ∈ V [4]. We define R_v as the calibrated country- and population-specific rate of transmission from HIV-infected persons in viral load stratum v ∈ V to any susceptible cohort of interest (for more information on the calibration process and the derivation of R_v, see the Supplementary Appendix). As we track the individual viral load of PWH in our microsimulation, we determine how the number of HIV transmissions changes over time, as the intervention reduces incidence and increases viral suppression.

Transmission Inside the Primary Cohort

This section describes how our method reduces incidence within the primary cohort to reflect averted HIV transmissions within the primary cohort due to an intervention.

In the status quo strategy (SQ), we assume that age-stratified incidence in the primary cohort does not vary with time from model start (i.e., remains constant) to mimic a stable epidemic. However, in the intervention strategy (INT), if there is transmission inside the primary cohort (e.g., MSM to MSM), incidence will decline over time due to both the individual benefit (e.g., PrEP, a preventive therapy that can protect individuals taking it from acquiring infection) and the additional community benefit of the intervention (e.g., most individuals who receive PrEP do not become HIV-infected, thus do not transmit the disease to others in the transmission network).

We separate incidence attributable to primary cohort members (inside-attributable incidence) from incidence attributable to non-primary cohort members (outside-attributable incidence, see Table 1). This is necessary because the individual, immediate protection of an intervention such as PrEP equally affects both inside- and outside-attributable incidence, as an individual who takes PrEP is protected from acquiring HIV regardless of whether the transmitter is inside or outside the primary cohort. The community benefit, however, which exists because there are fewer individuals with HIV in the primary cohort (thus lower HIV viral load in the primary cohort), mainly affects inside-attributable incidence. In reality, the community benefit can slightly affect outside-attributable incidence through the subsequent chain of transmissions averted among persons outside the primary cohort. Here, we assume outside-attributable incidence is not affected by the community benefit for simplicity; we elaborate on this assumption in greater detail in the case study.

Following the steps presented below, we seek to estimate cumulative decline in HIV incidence during the intervention period (say time 0 to t); then, we integrate this community-wide incidence decline within the model.

Step 0: Define monthly incidence probability at simulation time 0 (index year) under the status quo strategy.

Monthly incidence probability at simulation time 0 in SQ, P_SQ,0, is obtained from the country and population-specific literature.

Step 1: Define monthly incidence probability at simulation time 0 (index year) under the intervention strategy.

Given monthly incidence probability at simulation time 0 in SQ, P_SQ,0, the intervention uptake, P_up, and the immediate, individual incidence reduction of INT, P_eff (e.g., immediate protection of PrEP, see Table 2), we can estimate the monthly incidence probability at time 0 in INT, P_INT,0, as follows:

$${\mathit{P}}_{\mathit{INT},0}=(1-{\mathit{P}}_{\mathit{up}}){\mathit{P}}_{\mathit{SQ},0}+{\mathit{P}}_{\mathit{up}}\left[1-(1-{\mathit{P}}_{\mathit{SQ},0}{)}^{(1-{\mathit{P}}_{\mathit{eff}})}\right].$$ (1)

See the Supplementary Appendix for the derivation and proof of Equation (1).

The following steps involve two simulation model runs: (A) a run for the status quo strategy with P_SQ,0 as the incidence input and inputs R_v for viral load stratum v ∈ V, defined as inside transmission rates (transmission from primary cohort members to other primary cohort members, see Table 1); and (B) a separate run for the intervention strategy with P_INT,0 as the incidence input and inputs R_v for viral load stratum v ∈ V, similarly defined as the inside transmission rates.

We define I_SQ,agg (I_INT,agg) as the aggregate incidence rate in the status quo (intervention) strategy, including inside- and outside-attributable incidence. We have I_SQ,agg = I_SQ,in + I_SQ,out and I_INT,agg = I_INT,in + I_INT,out. Figure 1A represents hypothetical status quo inside, outside, and aggregate incidences. Figure 1B depicts hypothetical intervention incidences. Note that the inside-attributable incidence per person-month increases at a slower rate in INT compared to that of SQ because fewer individuals acquire HIV in INT due to both the individual and the community benefits.

Figure 1. Illustration of Status Quo and Intervention incidence over time.

Panel A. Hypothetical incidence rates (infections per person-month) attributable to inside and outside the primary cohort in the status quo strategy (SQ). Panel B. Hypothetical incidence rates attributable to inside and outside the primary cohort in SQ and the intervention strategy (INT). We assume an HIV-uninfected cohort is being modeled; therefore, I_SQ,in at time 0 is 0, that is, at model start (simulation time 0) no HIV infection is caused by primary cohort members as all of them enter the model as HIV-uninfected. If the primary cohort included some HIV-infected individuals at time 0, I_SQ,in and I_INT,in would both have a positive y-intercept. Infections attributed to inside the primary cohort increase overtime (given that more individuals will become HIV-infected), while infections attributed to outside the primary cohort decrease (given that we assume aggregate incidence in the status quo is constant). We assume that both P_up and P_eff > 0 and that intervention uptake occurs at time 0, and thus, I_INT,agg < I_SQ,agg at time 0. If either P_up or P_eff = 0, I_INT,agg = I_SQ,agg at time 0 (same y-intercept). See Methods and Table 1 for notation.

Step 2: Define cumulative outside-attributable number of HIV infections under the status quo strategy.

Given the monthly incidence probability at time 0 in SQ, P_SQ,0, the number of HIV-uninfected persons in the primary cohort at time 0 (index year), N_HIV−, and the number of inside transmissions in the SQ from time 0 to t obtained from simulation Run (A), T_SQ,0:t,in, we can estimate cumulative infections within the primary cohort from time 0 to t in the SQ attributable to those outside the primary cohort, I_{SQ,0:t, out}, as follows:

$${\mathit{I}}_{\mathit{SQ},0:\mathit{t},\mathit{out}}=\left[1-(1-{\mathit{P}}_{\mathit{SQ},0}{)}^{\mathit{t}}\right]\ast {\mathit{N}}_{\mathit{HIV}-}-{\mathit{T}}_{\mathit{SQ},0:\mathit{t},\mathit{in}}.$$ (2)

See the Supplementary Appendix for the derivation and proof of Equation (2).

Step 3: Define cumulative outside-attributable number of HIV infections under the intervention strategy.

Cumulative infections within the primary cohort from time 0 to t under the INT attributable to those outside the primary cohort, I_INT,0:t,out, can be derived as follows:

$${\mathit{I}}_{\mathit{INT},0:\mathit{t},\mathit{out}}=\left((1-{\mathit{P}}_{\mathit{up}}){\mathit{P}}_{\mathit{SQ},0:\mathit{t},\mathit{out}}+{\mathit{P}}_{\mathit{up}}\left[1-(1-{\mathit{P}}_{\mathit{SQ},0:\mathit{t},\mathit{out}}{)}^{(1-{\mathit{P}}_{\mathit{eff}})}\right]\right)\ast {\mathit{N}}_{\mathit{HIV}-},$$ (3)

in which, P_SQ,0:t,out is the cumulative probability of acquiring HIV between time 0 and t in SQ from non-primary cohort members, which is estimated as ^I_{SQ,0:t, out}/N_HIV−. See the Supplementary Appendix for the derivation and proof of Equation (3).

Step 4: Define cumulative aggregate number of HIV infections under the intervention strategy.

Aggregate infections within the primary cohort in the INT from time 0 to t, I_{INT,0:t, agg}, can be estimated as the sum of the inside-attributable and outside-attributable infections, via:

$${\mathit{I}}_{\mathit{INT},0:\mathit{t},\mathit{agg}}=\left((1-{\mathit{P}}_{\mathit{up}}){\mathit{P}}_{\mathit{SQ},0:\mathit{t},\mathit{out}}+{\mathit{P}}_{\mathit{up}}\left[1-(1-{\mathit{P}}_{\mathit{SQ},0:\mathit{t},\mathit{out}}{)}^{(1-{\mathit{P}}_{\mathit{eff}})}\right]\right){\mathit{N}}_{\mathit{HIV}-}+{\mathit{T}}_{\mathit{INT},0:\mathit{t},\mathit{in}}$$ (4)

in which T_INT,0:t,in is obtained from simulation Run (B). See the Supplementary Appendix for the derivation and proof of Equation (4).

To simulate the primary cohort over time under the intervention strategy, we need to estimate the monthly incidence probability from time 0 to t in INT. To do so, we need to determine the monthly incidence probability at simulation time t. Note that incidence probability at time 0 is already calculated via Equation (1).

Step 5: Define monthly incidence probability at simulation time t under the intervention strategy.

Given aggregate infections in INT from time 0 to t from Equation (4), one can estimate the cumulative incidence probability in INT, P_INT,0:t, as ^I_{INT,0:t, agg}/N_HIV−. The average monthly incidence probability within the primary cohort in INT during the same time interval, ${\overline{\mathit{P}}}_{\mathit{INT}}$, is then obtained via:

$${\overline{\mathit{P}}}_{\mathit{INT}}=1-(1-{\mathit{P}}_{\mathit{INT},0:\mathit{t}}{)}^{\frac{1}{t}}.$$ (5)

Given that ${\overline{\mathit{P}}}_{\mathit{INT}}$ represents the average between P_INT,0 and P_INT,t (incidence probability in INT at times 0 and t, respectively), we estimate that ${\mathit{P}}_{\mathit{INT},\mathit{t}}=2{\overline{\mathit{P}}}_{\mathit{INT}}-{\mathit{P}}_{\mathit{INT},0}$.

As the community incidence reduction is gradual because viral load decreases gradually over time, we model the incidence decline from P_INT,0 to P_INT,t over t months via an exponential decay function.

Finally, we are able to estimate life-years and costs under the status quo and intervention strategies while reflecting the difference in age-dependent incidence between the strategies due to the individual and community benefit of intervention. We run the microsimulation model two more times. In Run (A′) for the status quo strategy, P_SQ,0 remains the incidence input, but the transmission rate R_v for viral load stratum v ∈ V is set to the aggregate (total) transmission rate (not just the inside transmission rate as was the case in Runs A and B). Similarly, in Run (B′) for the intervention strategy, P_INT,0 remains the incidence input, but the community incidence reduction of INT, as estimated above, is incorporated into the model (as an exponential decay function), and R_v is set to the aggregate transmission rate. These two runs yield life-years, HIV costs, and number of new infections and transmissions for members of the primary cohort, accounting for both the individual and the community benefit of the intervention. However, if the intervention also results in averting HIV transmissions outside the primary cohort, further adjustment to the ICER is needed to account for life-years gained and HIV cost savings for persons outside the primary cohort (see the next section).

Transmission Outside the Primary Cohort

This section describes the method for estimating the benefits of averted transmissions to individuals outside the primary cohort, and for incorporating these gains in life-years and cost savings into ICERs for a given intervention.

The benefits of an averted transmission are the life-years gained and cost savings associated with a given transmission not occurring at the time it would have occurred. This is equivalent to subtracting the mean discounted life-years and costs of a susceptible cohort that begins the model as HIV-uninfected from those of a susceptible cohort that begins the model with a new (acute) HIV infection, starting from year y, the year at which the transmission would have occurred. It is worth noting that: (1) we group transmissions averted by year (rather than month) for simplicity; (2) a transmission can be averted in year y > 0, so life-years and costs must be discounted accordingly (i.e., must be brought back to simulation time 0); (3) a person whose transmission is averted in year y may still contract HIV later; (4) the age at which a susceptible cohort member’s transmission is averted is likely to depend on y; thus, cohort characteristics in model runs should be adjusted accordingly and separate model runs should be conducted for y ∈ {1,2,…,Y} (e.g., the female partners of MSM should age as their partner ages, so higher y should correspond to a higher initial age of the susceptible cohort).

From these runs, for a hypothetical transmission averted in year y, we obtain average life-years (costs) of a susceptible cohort that starts off with acute HIV infection, L_HIV+,y (C_HIV+,y), and those of a cohort that starts off HIV-uninfected, but is subject to HIV incidence, L_HIV−,y, (C_HIV−,y). We discount these values (see point 2 above) to bring them back to simulation time 0 using Eq.’s (6)-(9), in which d is the yearly discount rate (often 3%) [16].

$${\tilde{\mathit{L}}}_{\mathit{HIV}+,\mathit{y}}=\frac{({\mathit{L}}_{\mathit{HIV}+,\mathit{y}})}{(1+\mathit{d}{)}^{\mathit{y}}}$$ (6)

$${\tilde{\mathit{C}}}_{\mathit{HIV}+,\mathit{y}}=\frac{({\mathit{C}}_{\mathit{HIV}+,\mathit{y}})}{(1+\mathit{d}{)}^{\mathit{y}}}$$ (7)

$${\tilde{\mathit{L}}}_{\mathit{HIV}-,\mathit{y}}=\frac{({\mathit{L}}_{\mathit{HIV}-,\mathit{y}})}{(1+\mathit{d}{)}^{\mathit{y}}}$$ (8)

$${\tilde{\mathit{C}}}_{\mathit{HIV}-,\mathit{y}}=\frac{({\mathit{C}}_{\mathit{HIV}-,\mathit{y}})}{(1+\mathit{d}{)}^{\mathit{y}}}$$ (9)

Given these values, we can estimate the life-years gained $({\tilde{\mathit{L}}}_{\mathit{HIV}-,\mathit{y}}-{\tilde{\mathit{L}}}_{\mathit{HIV}+,\mathit{y}})$ and HIV cost savings $({\tilde{\mathit{C}}}_{\mathit{HIV}-,\mathit{y}}-{\tilde{\mathit{C}}}_{\mathit{HIV}+,\mathit{y}})$ for a hypothetical transmission averted in year y, discounted back to simulation time 0.

Finally, we estimate how many transmissions are averted in each year of the simulation. To do so, we conduct two model runs: (A″) a run for the status quo strategy with P_SQ,0 as the incidence input and R_v for viral load stratum v ∈ V defined according to transmission outside the primary cohort (outside transmission rate); and (B″) a run for the intervention strategy with P_INT,0 as the incidence input and the same values of R_v. The difference in total number of yearly transmissions (model outputs) between these two runs yields the number of transmissions averted in each year y to persons outside the primary cohort, i.e., TA_y = T_SQ,y,out - T_INT,y,out (see Table 2 for complete definitions of notation).

Thus, total life-years gained (costs saved) via all the transmissions averted over Y years, ${\tilde{\mathit{L}}}_{\mathit{TA}}({\tilde{\mathit{C}}}_{\mathit{TA}})$, can be derived via Equations (10) and (11).

$${\tilde{\mathit{L}}}_{\mathit{TA}}=\sum _{\mathit{y}=1}^{\mathit{Y}}({\mathit{TA}}_{\mathit{y}}\times ({\tilde{\mathit{L}}}_{\mathit{HIV}-,\mathit{y}}-{\tilde{\mathit{L}}}_{\mathit{HIV}+,\mathit{y}}))$$ (10)

$${\tilde{\mathit{C}}}_{\mathit{TA}}=\sum _{\mathit{y}=1}^{\mathit{Y}}({\mathit{TA}}_{\mathit{y}}\times ({\tilde{\mathit{C}}}_{\mathit{HIV}-,\mathit{y}}-{\tilde{\mathit{C}}}_{\mathit{HIV}+,\mathit{y}}))$$ (11)

$$\mathit{ICER}=\frac{(\Delta \tilde{\mathit{C}}\times \mathit{N})+{\tilde{\mathit{C}}}_{\mathit{TA}}}{(\Delta \tilde{\mathit{L}}\times \mathit{N})+{\tilde{\mathit{L}}}_{\mathit{TA}}}$$ (12)

in which $\Delta \tilde{\mathit{L}}(\Delta \tilde{\mathit{C}})$ represents the average life-years (cost) difference between the SQ and INT for the primary cohort, discounted at d within the simulation model, and N is the size of the primary cohort. If the method described in the previous section is implemented, $\Delta \tilde{\mathit{C}}$ and $\Delta \tilde{\mathit{L}}$ will account for the benefit of averting transmissions within the primary cohort; ${\tilde{\mathit{C}}}_{\mathit{TA}}$ and ${\tilde{\mathit{L}}}_{\mathit{TA}}$ further adjust the numerator and denominator of the ICER to also account for the benefit of averting transmissions outside the primary cohort.

Model Validation: Comparison to HOPE Model in the US

We validated our model against the HOPE model, a dynamic transmission model of HIV disease, by simulating a five-year PrEP intervention (between 2016-2020) for MSM and PWID in the United States, equivalent to the intervention modeled in Khurana et al. [5]. In the modeled intervention, PrEP was administered according to CDC guidelines and included HIV testing every 3 months [5]. We compared infections predicted by CEPAC in both the status quo (meant to represent 2016 trends of risk behavior, HIV treatment, and linkage to care, without PrEP) and the PrEP strategy over 5 years with those predicted by the HOPE model. We used Khurana’s definitions of high-risk groups and estimated PrEP eligibility in our model validation, along with HOPE’S predicted efficacy (73% for MSM and 49% for PWID) and uptake of PrEP (40% for MSM and 10% for PWID) [5, 17, 18]. Additional key inputs used in the validation are listed in Table 3. For input parameters that were included in CEPAC but were not provided in the Kharana’s analysis, we used US-specific values from a similar US-based CEPAC analysis [19]. Similar to in Kharana et al., we varied PrEP efficacy and coverage for MSM and PWID in sensitivity analyses.

Table 3.

Key inputs parameters used to validate the CEPAC model with the community benefit incorporated against the HOPE model.

	Value
Parameter	MSM	PWID	Source
Number of PrEP-eligible individuals in 2016	2,206,379	386,209	[5, 36-38]
HIV prevalence, %	11.0	15.0	[39, 40]
HIV incidence, rate /100 PYs	0.6	0.9	Derived from [5]
PrEP efficacy, % (range)	73.0 (44.0-92.0)	49.0 (9.6-70.0)	[5, 17, 18]
PrEP uptake, % (range)	40.0 (20.0-60.0)	10.0 (5.0-15.0)	[5]
Linkage to care, %	73.4		[41]
Weighted average transmission rate within the primary cohort (rate/100 PYs)	3.3	7.1	Derived from incidence and prevalence
Transmission risk ratio by HIV RNA level (copies/mL)			[4]
>100,000	4.4
10,001-100,000	3.9
3,001-10,000	2.0
501-3,000	1.0 (reference)
≤500	0.1
Transmission risk ratio while in acute phase (vs. chronic)	5.3		[42]
Viral suppression on first-line ART (DTG) at 48 weeks, %	87.0		[14, 43]

CEPAC: Cost-Effectiveness of Preventing AIDS Complications model; HOPE: HIV Optimization and Prevention Economics model; DTG: dolutegravir; HIV: human immunodeficiency virus; mL: milliliter; MSM: men who have sex with men; PrEP: pre-exposure prophylaxis; PWID: people who inject drugs; PY: person-year.

We evaluated the consistency between the two models in terms of estimating the number of infections averted due to intervention via the intraclass correlation coefficient (ICC) measure using a two-way mixed effects model [20, 21]. The intraclass correlation is used when different scenarios are evaluated by multiple models to assess the reliability of models using either consistency (when systematic differences between the models are irrelevant) or absolute agreement (when systematic differences between the models are relevant) measure. Note that the two models (CEPAC and HOPE) have clear systematic differences as they use different structures and underlying assumptions. Given that our goal is to validate the new method to capture indirect community benefit (and not to validate CEPAC against HOPE), we evaluated the consistency (rather than the absolute agreement or other measures of error) between number of infections averted estimated by our method paired with CEPAC against that of HOPE (i.e., we did not focus on the systematic differences between the two models).

Case Study: HIV Pre-exposure Prophylaxis (PrEP) for MSM in India

We demonstrated proof of concept using a case study of offering PrEP to MSM in India. We modeled both the status quo – meant to represent current rates of HIV incidence and linkage to care – as well as a PrEP program made available to the estimated 3.1 million MSM in India [22]. We projected HIV transmissions, lifetime infection risk, life expectancy, costs of the program, and ICERs.

All individuals were HIV-uninfected at model start, and based on PrEP acceptability studies in India, we assumed 55.7% chose to participate in the program [23]. PrEP had an effectiveness of 65%, a combination of expected adherence and PrEP efficacy [7, 24, 25], and the program was assumed to continue until age 35. In accordance with WHO guidelines, PrEP was paired with HIV testing every 3 months [26].

Figure 2 depicts the MSM transmission network we considered. We accounted for HIV transmissions from incident MSM (those who acquire HIV after model start) to HIV-uninfected MSM (inside transmissions), as well as transmission from incident MSM to TG and female partners (outside transmissions), who are other key members of the MSM transmission network [11]. In Figure 2, arrow 1 represents inside-attributable incidence (or inside transmissions), arrows 2 and 3 represent outside transmissions, and arrows 4-6 represent outside-attributable incidence. We assume that MSM can be infected with HIV by other MSM or TG; the vast majority of MSM report having no casual or commercial female partners [11]. Some MSM in India are either married or in exclusive partnerships with women, thus we assume MSM can transmit to female partners [11, 27]. Finally, due to substantial sexual mixing between MSM and TG, with a much larger number of MSM than TG, we assume that TG infections are caused by MSM [11, 27].

Figure 2. Illustrative diagram of transmission network for MSM in India, including transmission within and outside the primary cohort.

This figure outlines how transmission was modeled in the case study for MSM in India. In Panel A, the dotted black boxes outline which groups are modeled as the primary cohort (i.e., HIV-uninfected MSM and incident MSM, or those who become infected during the model simulation) and which groups contribute to transmission but are not being modeled directly in the microsimulation and are therefore outside the primary cohort (i.e., HIV-infected prevalent MSM, HIV-infected prevalent TG, HIV-uninfected TG, incident TG, and female partners of MSM). The dotted orange arrows describe an individual’s possibility to transition from one group to another (such as HIV-uninfected MSM becoming HIV-infected). The blue arrows describe how individuals transmit to one another (for example, incident MSM can transmit to HIV-uninfected MSM in the primary cohort, or TG and female partners outside the primary cohort). Panel B depicts that the size of the incident MSM cohort, which gradually grows as more MSM become infected over the simulation horizon, increases at a slower rate in the PrEP strategy compared to the status quo. Arrow 1 represents inside-attributable infections; arrows 4, 5, and 6 represent outside-attributable infections; and arrows 2 and 3 represent transmissions to persons outside the primary cohort.

MSM: men who have sex with men; PrEP: pre-exposure prophylaxis; TG: transgender women.

We assumed outside-attributable incidence was not affected by the community benefit of PrEP, given that 87% of MSM have no TG partners [11], thus, for them, the outside-attributable incidence only corresponds to transmissions from prevalent MSM (Figure 2, arrow 4) whose viral load is unaffected by PrEP rollout. Moreover, for the remaining MSM (13% of the total MSM) who have at least one TG partner, the majority of outside incidence is attributable to prevalent MSM and prevalent TG (Figure 2, arrows 4 and 5), groups whose viral load is unaffected by PrEP rollout.

We ran the model both with and without the time-dependent incidence reduction due to the additional community benefit of PrEP (reduced HIV viral load among MSM leading to reduced transmission in addition to the individual protection of PrEP). For additional model inputs, see Table 5.

Table 5.

Base case model input parameters for an analysis of PrEP for HIV-uninfected MSM in India.

Parameter	Value		Reference
Characteristics of primary cohort
Age, years, mean (SD)	29.4 (5.7)		[44]
Estimated size of MSM population, millions	3.1		[22]
HIV prevalence among MSM, %	4.3		[45]
Estimated size of primary cohort (HIV-uninfected MSM), millions	3.0		Calculated
HIV incidence, infections/100 PY (IQR)	0.9 (0.4-1.2)		[10]
Linkage to care, %	87.5		[45]
Characteristics of sexual network members	Women	TG
Age, years, mean (SD)	24.1 (5.7)	29.4 (5.7)	[44, 46]
Linkage to care, %	80.0	91.5	[45, 47]
Intervention parameters
PrEP uptake, % willing to participate	55.7		[23]
PrEP effectiveness, % incidence reduction	65.0		[24, 25]
Transmission dynamics
Transmission inside the primary cohort, rate/100 PY, off-PrEP	17.6		[10]
Transmission outside the primary cohort, rate/100 PY	6.2 (to TG); 0.6 (to women)		[48-50]
Acute, off-ART transmission risk ratio	5.3		[42]
Clinical characteristics post HIV infection
Acute CD4 count, cells/μl, mean (SD)	553 (230)		[33]
First-line overall viral suppression at 48 weeks among patients prescribed ART, %	82.1		[51]
Costs*
PrEP associated costs
Drug cost ($/month)	5.55		[52]
HIV test cost ($/test)*	4.39		[33]
Clinic visit cost ($/visit)	5.99		[53]
Creatinine testing cost ($/test)	3.00		Assumption
Antiretroviral therapy costs			[52]
1st-line ART, $/month	9.10
2nd-line ART, PI-based regimen, $/month	22.74
Monitoring and care costs
HIV viral load test ($/test)	22.08		[53]
CD4 count test ($/test)	3.69		[53]
Routine care cost (conditional on CD4 count, $/month)	7.48-25.83		[54]

ART: antiretroviral therapy; CD4: cluster of differentiation 4; HIV: human immunodeficiency virus; IQR: interquartile range; MSM: men who have sex with men; PI: protease inhibitor; PrEP: pre-exposure prophylaxis; PY: person-years; SD: standard deviation; TG: transgender women.

^*All costs are reported in 2018 USD.

To further evaluate the impact of the community benefit, we performed sensitivity analysis comparing the number of HIV transmissions averted in our case study under different PrEP uptake levels with and without including the community benefit. To assess the potential impact of not including the community benefit on policy conclusions, we conducted sensitivity analysis on the net monetary benefit (NMB), defined as the financial benefit based on the willingness-to-pay (WTP) minus the cost of the program, of status quo and PrEP (with and without the community benefits). We defined the WTP as $1,940/year-of-life saved (YLS), India’s per capita GDP in 2018 [28].

RESULTS

Validation Against HOPE

Table 6 outlines validation results for both MSM and PWID. Analogous to HOPE, we report total infections for CEPAC during the 5-year period as well as the number of infections averted. Base case and sensitivity analysis results both show that there are small systematic differences between CEPAC and HOPE; however, our method paired with CEPAC and the HOPE model are consistent in estimating number of infections averted with PrEP. The intraclass correlation coefficient (ICC) measure was 0.99 (95%CI: 0.87-1.00) for MSM and 0.98 (95%CI: 0.79-1.00) for PWID. An ICC measure >0.90 is often regarded as excellent consistency between the models [29].

Table 6.

Comparison of HOPE and CEPAC model transmission output.

	MSM		PWID
	HOPE	CEPAC	HOPE	CEPAC
Status Quo
Total infections	102,410	103,917	16,676	15,880
PrEP (base case)	73% efficacy; 40% uptake		49% efficacy; 10% uptake
Total infections	76,562	81,756	15,849	14,704
Infections averted (%)	25,848 (25.2%)	22,161 (21.3%)	827 (5.0%)	1,176 (7.4%)
Lower PrEP efficacy	44% efficacy		9.6% efficacy
Total infections	85,847	89,544	16,462	15,379
Infections averted (%)	16,563 (16.2%)	14,373 (13.8%)	214 (1.3%)	501 (3.2%)
Higher PrEP efficacy	92% efficacy		70% efficacy
Total infections	70,539	76,622	15,516	14,338
Infections averted (%)	31,871 (31.1%)	27,295 (26.3%)	1,160 (7.0%)	1,542 (9.7%)
Lower PrEP uptake	20% uptake		5% uptake
Total infections	89,179	92,298	16,262	15,260
Infections averted (%)	13,231 (12.9%)	11,619 (11.2%)	414 (2.5%)	620 (3.9%)
Higher PrEP uptake	60% uptake		15% uptake
Total infections	64,461	71,452	15,442	14,143
Infections averted (%)	37,949 (37.1%)	32,465 (31.2%)	1,234 (7.4%)	1,737 (10.9%)

CEPAC: Cost-effectiveness of preventing AIDS complications microsimulation model; HOPE: HIV Optimization and Prevention Economics dynamic transmission model; MSM: men who have sex with men; PrEP: pre-exposure prophylaxis; PWID: people who inject drugs.

Case Study in India

PrEP with only the individual benefit reduced lifetime infection risk from 12.3% with status quo to 10.2% and averted 42,840 transmissions within the primary cohort (Table 7). PrEP with both the individual and community benefits resulted in a much larger combined effect, reducing MSM lifetime infection risk further to 6.1% and averting 68,500 transmissions among MSM and 17,760 transmissions to TG and female partners.

Table 7.

Base case clinical and economic model outcomes of a PrEP strategy for MSM in India using the new method paired with the CEPAC microsimulation.

Strategy	Lifetime infection risk (%)	Transmissions averted through age 35, inside the primary cohort*	Transmissions averted through age 35, outside the primary cohort*		LYs per person, primary cohort, undiscounted	LYs per person, primary cohort, discounted	Cost per person, primary cohort discounted ($)	ICER, lifetime ($/YLS)
Women TG
Status quo	12.3%	-	-	-	39.66	22.09	320	-
PrEP with only individual benefit	10.2%	42,840	-	-	39.96	22.21	590	2,300†
PrEP with both individual and community benefit	6.1%	68,500	1,460	16,300	40.32	22.34	510	600†

ICER: incremental cost-effectiveness ratio; LY: life-years; YLS: years of life saved; MSM: men who have sex with men; PrEP: pre-exposure prophylaxis; TG: transgender women.

^*We simulated 10 million persons to obtain stable results but scaled transmission outputs to the true size of primary cohort (HIV-uninfected MSM in India), which was 2,967,000.

^†Numbers are compared to status quo. The overall incremental cost-effectiveness ratio (ICER) was calculated as the total increase in cost for primary cohort members plus the total cost savings of transmissions averted to outside the primary cohort, all divided by the total gain in life expectancy for primary cohort members plus the total gain in life expectancy for averted transmissions outside the primary cohort. For example, $600/YLS was calculated as $\frac{(\$510-\$320)\ast 2,967,000-(\$2,820\ast (1,460+16,300))}{(22.34-22.09)\ast 2,967,000+(4.80\ast (1,460+16,300))}$, where 2,967,000 is the size of primary cohort (number of HIV-uninfected MSM), $2,820 is the average cost-savings per averted transmission outside the primary cohort (to TG/women), and 4.80 is the average life-years gained per averted transmission outside the primary cohort.

PrEP with only the individual benefit included increased per-person lifetime discounted expenditures by $270 and increased discounted life expectancy by 0.12 years, resulting in an ICER of $2,300/year of life saved (YLS), compared to the status quo. PrEP with both the individual and community benefits included increased per-person lifetime discounted expenditures by $190 and increased discounted life expectancy by 0.25 years, resulting in a combined ICER of $600/year-of-life saved (YLS), compared to the status quo, including the life-years and cost benefit of averting outside the primary cohort transmissions.

When varying uptake in sensitivity analysis, the community benefit consistently increased transmissions averted by ~60-110% compared to the individual benefit only (Figure 3). With the individual and community benefits incorporated, the NMB of the PrEP strategy was consistently higher than the NMB of Status Quo and increased with higher values of uptake, reaching $42,900 at 55.7% uptake. However, with only the individual benefit incorporated, the NMB of PrEP was consistently near that of Status Quo (Figure 4).

Figure 3. HIV transmissions averted with PrEP when varying uptake in sensitivity analysis.

This figure shows total transmissions averted through age 35 (including transmissions from MSM to other MSM, TG, and women) when PrEP uptake was varied in sensitivity analysis between 10% and 90%. The solid blue line indicates PrEP with both the individual and community benefits incorporated, while the dashed yellow line indicates PrEP with only the individual benefit. The base case PrEP uptake value was 55.7%. PrEP with only the individual benefit averted 42,840 transmissions while PrEP with both the individual and community benefits averted 68,500 transmissions among MSM, TG, and women.

MSM: men who have sex with men; PrEP: pre-exposure prophylaxis; TG: transgender women.

Figure 4. Net monetary benefit of PrEP when varying PrEP uptake in sensitivity analysis.

This figure shows the net monetary benefit (NMB) of the status quo, PrEP with the individual benefit alone, and PrEP with the individual and community benefits when PrEP uptake was varied between 10% and 90%. NMB is calculated as average per-person life-years times the WTP ($1,940/YLS) minus the cost of the program. The solid blue line indicates PrEP with both the individual and community benefits incorporated, while the dashed yellow line indicates PrEP with only the individual benefit. The dotted black line shows the status quo. The base case PrEP uptake value was 55.7%.

LY: life year; PrEP: pre-exposure prophylaxis; WTP: willingness-to-pay.

DISCUSSION

With the goal of ending the HIV epidemic globally, transmission within sexual and drug-using networks is increasingly a key point of concern [30]. When making policy recommendations, specifically informed by cost-effectiveness analysis, decision scientists should seek to predict potential benefits not only for the target population for the intervention, but also for the populations impacted due to the chain effect of preventing transmission from the target population to others. This paper sought to expand the ability of microsimulation models that simulate patients one at a time by accounting for the broader community benefits of interventions, allowing these models to better capture an intervention’s indirect effects while still benefiting from the extensive clinical detail of these microsimulations. Several HIV interventions provide such indirect community benefit and can be more precisely evaluated using our method. For example, antiretroviral medications (ART) and interventions to promote adherence to ART can help lower viral load and achieve virological suppression, thus decrease HIV transmissions from HIV-infected individuals. HIV pre-exposure prophylaxis (PrEP) can protect HIV-uninfected individuals from acquiring HIV, and behavioral education can help decrease high-risk behavior, both of which result in preventing onward HIV transmissions.

Our method, paired with the CEPAC model, predicts transmission outcomes consistent with those of HOPE, a widely-validated dynamic transmission model of HIV disease, despite systematic differences between the two models [1, 5]. For example, CEPAC is a microsimulation model that simulates individuals infected with or susceptible to HIV one by one (thus, it does not explicitly model contact between individuals) and does not stratify risk groups by race or circumcision status. However, compared to the HOPE model, CEPAC incorporates substantially greater detail with regard to HIV disease progression, treatment, transmission by viral-load strata, and opportunistic diseases and other chronic diseases associated with HIV. Therefore, we would not expect transmissions to match precisely. However, our validation results showed excellent consistency between the two models in terms of estimating infections averted with PrEP.

Our case study for India demonstrated that the life-years gained and cost savings due to the community benefit may have an impact on policy decisions and thus should not be ignored. The lifetime risk of HIV infection was 10.2% for PrEP with only the individual benefit and 6.1% for PrEP with both the individual and the indirect community benefits. PrEP with only the individual benefit prevented 42,840 HIV transmissions whereas PrEP with both the individual and community benefits prevented a total of 86,260 HIV transmissions from MSM. The per-person lifetime discounted costs and life years were $590 and 22.21 years for PrEP with only the individual benefit. Considering both benefits, the per-person lifetime discounted costs and life years for PrEP were $510 and 22.34 years. With both the individual and community benefits incorporated, the ICER of a PrEP program for MSM in India was $600/YLS, which is deemed “cost-effective” according to recommended willingness-to-pay thresholds [28, 31]. However, the same PrEP program without the community benefit had an ICER of $2,300/YLS. Moreover, in sensitivity analysis, we demonstrated that the community benefit increased transmissions averted by ~60-110% and resulted in substantial changes to the NMB. Notably, the Status Quo strategy dominated PrEP (had a higher NMB) with only the individual benefit included, indicating that PrEP would not be deemed cost-effective without the indirect community benefit. However, when both the individual and community benefits were included, PrEP was the cost-effective strategy at all uptake levels (had higher NMB values than the status quo). Therefore, including the broader community benefit of HIV interventions can change the policy conclusion of an analysis.

Our method is designed to be paired with microsimulation models of HIV diseases to addresses a common limitation of these models, namely that they do not fully capture the indirect community benefit of HIV interventions and thus may underestimate the life-years gained and cost savings of such interventions. Many of these models (including CEPAC) are well-established and validated, though the common limitation described above can sometimes be important depending on the intervention being evaluated and the goals of the analysis. Per the recommendation of the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) and the Society for Medical Decision Making (SMDM) Modeling Good Research Practices Task Force, the indirect community benefits of an intervention in a health economic microsimulation model may be excluded if the model determines that the intervention is already cost-effective without them. In that case, including such broader benefits will make the policy conclusion stronger. On the other hand, if the intervention under consideration is deemed not cost-effective when only the direct individual benefits are accounted for (as was the case in our PrEP case study), it is necessary to incorporate the indirect community benefits and re-evaluate the cost-effectiveness of the intervention [32]. In both cases, including the indirect community benefits makes the clinical and economic results more accurate and the policy conclusions more reliable [32].

Our method makes several assumptions and has some limitations. First, as is commonly done in HIV microsimulation models, we consider the primary cohort to be a static group (i.e., no new members enter the cohort after simulation time 0) [2, 3, 33]. Second, uptake of a given intervention occurs at model start and remains constant. Third, in simulation Run B, we use a constant incidence probability, P_INT,0, to estimate inside transmissions between time 0 and t in the intervention strategy (T_INT,0:t,in). In reality, the incidence in the intervention strategy declines gradually over time, and thus, cumulative inside transmissions in the intervention strategy (T_INT,0:t,in) are slightly overestimated (or transmissions averted are underestimated), making the method slightly conservative in estimating the intervention’s benefits. Fourth, while we did account for the benefit of transmissions averted from the primary cohort to those outside the cohort (e.g., MSM to TG in the case study), we did not account for the subsequent chain of transmissions averted among persons outside the primary cohort (e.g., TG), which can then affect (decrease) the outside-attributable incidence rate for primary cohort members (e.g., incident TG to MSM incidence, arrow 6 in Figure 2) in the long run. This may also result in a slight underestimation of the benefits of the intervention. Fifth, we could not validate our method paired with CEPAC directly with real data because doing so would require a counterfactual analysis, that is, a comparison between HIV transmissions that occurred in actuality with an intervention and HIV transmissions that would have happened in the absence of that intervention (i.e., HIV transmissions averted). Finally, our case study of PrEP for MSM was meant to provide a proof of concept for the new method. A comprehensive analysis would be valuable to assess the full clinical and economic outcomes of PrEP in various settings.

In conclusion, our method to estimate HIV transmissions averted and the community benefit of HIV interventions can improve microsimulation models which aim to assess clinical and economic outcomes of HIV interventions and lead to policy conclusions. Including the indirect community benefit increases the accuracy of estimates regarding the benefits (life-years gained and cost savings) of an intervention and leads to improved policy conclusions. Furthermore, using a case study of offering PrEP to HIV-uninfected MSM in India, we demonstrated that not only does PrEP improve outcomes for individuals who take it, but also that its indirect community benefit substantially improves outcomes for many additional people in the same transmission network.

Table 4.

The effect of the individual and community benefits of PrEP on inside and outside incidence for the case of a PrEP program for MSM in India.

	Individual Benefit of PrEP (direct benefit)	Additional Community Benefit of PrEP (indirect benefit)
Inside Incidence	For PrEP uptakers, inside incidence drops immediately according to efficacy and adherence. Those who do not take PrEP do not receive any direct benefit.	For both PrEP uptakers and nonuptakers, due to lower viral load in the primary cohort over time compared to the status quo (fewer number of incident MSM, see Figure 2B), inside transmission, and thus inside incidence, declines gradually. This decline is in addition to the immediate drop for PrEP uptakers.
Outside Incidence	For PrEP uptakers, outside incidence drops immediately according to efficacy and adherence. Those who do not take PrEP do not receive any direct benefit.	For both PrEP uptakers and nonuptakers, we assume outside incidence remains unaffected given that the vast majority of outside incidence is attributable to prevalent MSM and TG (already HIV-infected at model start), for whom viral load is unaffected by PrEP roll out (i.e., arrows 4 and 5 are unaffected by PrEP in Figure 2).

MSM: men who have sex with men; PrEP: pre-exposure prophylaxis; TG: transgender women.

Highlights.

• Microsimulation models of HIV are valuable tools for evaluating HIV interventions

• These models may not fully capture indirect community benefit of HIV interventions

• We develop a novel method to address this shortcoming of HIV microsimulation models

• Proof of concept is provided using a case study of offering PrEP to MSM in India

• Including the community benefit of PrEP leads to improved policy conclusions