Imperial College London Estimating the impact of improvements in the HIV care cascade on HIV incidence among men who have sex with men in the US: mathematical modelling for HPTN 078 Good evening! I’m Kate Mitchell. I work for the HPTN modelling centre, and I’m based at Imperial College London. Today I’ll be presenting some of the modelling work we’ve done for HPTN 078, estimating the potential impact on HIV incidence of improving engagement with the HIV care cascade among HIV-infected MSM in the US, particularly focussing on one of the trial sites in Baltimore. Kate M Mitchell Imperial College London London, UK 25th July 2017

Trial sites: Atlanta GA, Baltimore MD, Birmingham AL, Boston MA The HPTN 078 trial Enhancing Recruitment, Linkage to Care and Treatment for HIV-Infected Men Who Have Sex with Men (MSM) in the United States Trial sites: Atlanta GA, Baltimore MD, Birmingham AL, Boston MA The full title of the HPTN 078 trial is Enhancing Recruitment, Linkage to Care and Treatment for HIV-Infected Men Who Have Sex with Men (MSM) in the United States.

Trial sites: Atlanta GA, Baltimore MD, Birmingham AL, Boston MA The HPTN 078 trial Enhancing Recruitment, Linkage to Care and Treatment for HIV-Infected Men Who Have Sex with Men (MSM) in the United States Trial sites: Atlanta GA, Baltimore MD, Birmingham AL, Boston MA The trial began enrolling participants in early 2016, at 4 sites: Atlanta, Georgia, Baltimore Maryland, Birmingham Alabama, Boston Massachusetts.

The HPTN 078 trial Enhancing Recruitment, Linkage to Care and Treatment for HIV-Infected Men Who Have Sex with Men (MSM) in the United States Intervention (n = 178) Case Manager Intervention Package There are 2 main components to the trial: first of all deep-chain RDS is being used to identify and recruit HIV-infected men who are not virally suppressed The recruited MSM are then randomized to either a case manager intervention package, designed to enhance linkage to care and treatment, or to standard of care The trial outcomes will be, for the RDS component, the number of HIV-positive MSM who are not virally suppressed who are identified and recruited, and for the case manager intervention, the % of men who are virally suppressed after 24 months. MSM Not virally suppressed Control (n = 178) SOC for Linkage and Treatment Deep-Chain Respondent Driven Sampling Individual Randomization

Mathematical modelling in HPTN 078 Before/during the trial - to estimate levels of viral suppression that must be reached to attain HIV incidence reduction targets After the trial - to estimate reduction in HIV incidence achieved by trial interventions In HPTN 078, modelling will be used to conduct two analyses: one before and during the trial, to estimate the level of viral suppression that needs to be reached in order to attain HIV incidence reduction targets. This gives us an idea of what the trial needs to achieve in order to get a big reduction in HIV incidence.

Mathematical modelling in HPTN 078 Before/during the trial - to estimate levels of viral suppression that must be reached to attain HIV incidence reduction targets After the trial - to estimate reduction in HIV incidence achieved by trial interventions Modelling will also be used after the trial is completed to estimate the likely reduction in HIV incidence that would be achieved in the wider MSM population at each site given the levels of viral suppression reached.

Mathematical modelling in HPTN 078 Before/during the trial - to estimate levels of viral suppression that must be reached to attain HIV incidence reduction targets After the trial - to estimate reduction in HIV incidence achieved by trial interventions Today I will be presenting results from the first analysis.

HIV prevalence: 30% in 2014 (NHBS*) HIV epidemic among MSM in Baltimore HIV prevalence: 30% in 2014 (NHBS*) Virally suppressed: 40% of diagnosed in 2015 (Maryland DHMH) PrEP use: 2.4% in 2014 (NHBS*) At the moment, we have results for one of the sites, Baltimore. I’ll begin with some background about the HIV epidemic among MSM in Baltimore. HIV prevalence is rather high - 43% in 2011, and 30% in 2014 in the CDC’s NHBS surveys (National HIV Behavioural Surveillance). *CDC National HIV Behavioral Surveillance

HIV prevalence: 30% in 2014 (NHBS*) HIV epidemic among MSM in Baltimore HIV prevalence: 30% in 2014 (NHBS*) Virally suppressed: 40% of diagnosed in 2015 (Maryland DHMH) PrEP use: 2.4% in 2014 (NHBS*) Estimates of the proportion of HIV positive MSM who are virally suppressed in Baltimore are low but increasing over time – latest estimate from 2015 was that 40% of those diagnosed were virally suppressed *CDC National HIV Behavioral Surveillance

HIV prevalence: 30% in 2014 (NHBS*) HIV epidemic among MSM in Baltimore HIV prevalence: 30% in 2014 (NHBS*) Virally suppressed: 40% of diagnosed in 2015 (Maryland DHMH) PrEP use: 2.4% in 2014 (NHBS*) PrEP use is currently very low – in the analysis I present today we have assumed that there is no PrEP use in this population *CDC National HIV Behavioral Surveillance

Research Questions How much does the level of viral suppression need to be increased by to reduce HIV incidence among Baltimore MSM by 10, 20, 30 or 50% after 2, 5 and 10 years? By how much could HIV incidence be reduced if US National HIV/AIDS strategy (NHAS) targets met by 2020: 90% diagnosed 90% of diagnosed retained in care 80% of diagnosed virally suppressed What is the likelihood of local HIV elimination among MSM in Baltimore in the next 20 years under current levels of care and if these targets are met? We set out to answer three questions in this analysis: [READ]

Research Questions How much does the level of viral suppression need to be increased by to reduce HIV incidence among Baltimore MSM by 10, 20, 30 or 50% after 2, 5 and 10 years? By how much could HIV incidence be reduced if US National HIV/AIDS strategy (NHAS) targets met by 2020: 90% diagnosed 90% of diagnosed retained in care 80% of diagnosed virally suppressed What is the likelihood of local HIV elimination among MSM in Baltimore in the next 20 years under current levels of care and if these targets are met? We set out to answer three questions in this analysis: [READ]

Research Questions How much does the level of viral suppression need to be increased by to reduce HIV incidence among Baltimore MSM by 10, 20, 30 or 50% after 2, 5 and 10 years? By how much could HIV incidence be reduced if US National HIV/AIDS strategy (NHAS) targets met by 2020: 90% diagnosed 90% of diagnosed retained in care 80% of diagnosed virally suppressed What is the likelihood of local HIV elimination among MSM in Baltimore in the next 20 years under current levels of care and if these targets are met? We set out to answer three questions in this analysis: [READ]

Deterministic compartmental model Sexual HIV transmission Mathematical model Deterministic compartmental model Sexual HIV transmission We developed a new mathematical model for this analysis. Our model is a deterministic, compartmental model of sexual HIV transmission among MSM in Baltimore.

Disease progression: CD4 decline stratified by viral load Model structure Disease progression: CD4 decline stratified by viral load Risk groups: age (<25, 25+) x race (black, white) Care cascade: HIV disease progression is modelled as the decline in CD4 count, stratified by viral load

Disease progression: CD4 decline stratified by viral load Model structure Disease progression: CD4 decline stratified by viral load Risk groups: age (<25, 25+) x race (black, white) Care cascade: Model risk structure: divide MSM into two race groups, and two age groups

Disease progression: CD4 decline stratified by viral load Model structure Disease progression: CD4 decline stratified by viral load Risk groups: age (<25, 25+) x race (black, white) Care cascade: On treatment Never testing, undiagnosed Testing, undiagnosed Diagnosed In care Achieving suppression Adherent Dropped out We’ve included a lot of detail about the care cascade in the model – including testing behaviour, diagnosis, linkage to care, ART initiation and dropout. The colouring highlights the relative infectiousness of those on ART and adherent to their treatment – I’ve shown them in pale pink, much less infectious than those in other states, shown in dark red. Non-adherent

Model parameters Domain Examples Data source Disease progression Initial CD4 and viral load distribution HIV-related mortality CD4 progression rates Published studies: cohorts in North America and Europe Infection probabilities Per-sex-act transmission probability Relative infectiousness different disease stages Published studies: meta analyses, study of Australian MSM Intervention efficacy Reduction in per-sex-act HIV transmission risk: condoms, ART Published studies: clinical trials, meta-analyses Sexual risk behaviour Number and type of partners Condom use Age and race of partners NHBS surveillance data, (eventually 078 trial) Baltimore Intervention behaviour HIV testing Linkage/dropout from HIV care ART initiation, adherence and dropout NHBS surveillance data, clinical cohorts, state health department data, (eventually 078 trial) Baltimore The model needs a large number of different inputs (also known as parameters). I’ve listed some examples here under different domains.

Model parameters Domain Examples Data source Disease progression Initial CD4 and viral load distribution HIV-related mortality CD4 progression rates Published studies: cohorts in North America and Europe Infection probabilities Per-sex-act transmission probability Relative infectiousness different disease stages Published studies: meta analyses, study of Australian MSM Intervention efficacy Reduction in per-sex-act HIV transmission risk: condoms, ART Published studies: clinical trials, meta-analyses Sexual risk behaviour Number and type of partners Condom use Age and race of partners NHBS surveillance data, (eventually 078 trial) Baltimore Intervention behaviour HIV testing Linkage/dropout from HIV care ART initiation, adherence and dropout NHBS surveillance data, clinical cohorts, state health department data, (eventually 078 trial) Baltimore Inputs relating to disease progression, infection probabilities and intervention efficacy were drawn from the published literature, where possible from similar populations and settings.

Model parameters Domain Examples Data source Disease progression Initial CD4 and viral load distribution HIV-related mortality CD4 progression rates Published studies: cohorts in North America and Europe Infection probabilities Per-sex-act transmission probability Relative infectiousness different disease stages Published studies: meta analyses, study of Australian MSM Intervention efficacy Reduction in per-sex-act HIV transmission risk: condoms, ART Published studies: clinical trials, meta-analyses Sexual risk behaviour Number and type of partners Condom use Age and race of partners NHBS surveillance data, (eventually 078 trial) Baltimore Intervention behaviour HIV testing Linkage/dropout from HIV care ART initiation, adherence and dropout NHBS surveillance data, clinical cohorts, state health department data, (eventually 078 trial) Baltimore For sexual risk behaviour and intervention behaviour, its important to use local data where possible, and these have been taken from the CDC’s NHBS surveys and Maryland health department data for Baltimore. (N.B. data not yet available from the trial)

Model calibration outcomes Observed Baltimore data: HIV prevalence - by age and race (NHBS) MSM demography - age and race (NHBS, census) Care cascade - % retained in care, on ART, virally suppressed (NHBS, Maryland DHMH) Next, we calibrated the model to data. The calibration stage is really important, as it ensures that the model can reproduce observed data before we use it to make predictions about intervention impact. For this analysis, we wanted to be sure that the model could reproduce data on:

Model calibration outcomes Observed Baltimore data: HIV prevalence - by age and race (NHBS) MSM demography - age and race (NHBS, census) Care cascade - % retained in care, on ART, virally suppressed (NHBS, Maryland DHMH) HIV prevalence

Model calibration outcomes Observed Baltimore data: HIV prevalence - by age and race (NHBS) MSM demography - age and race (NHBS, census) Care cascade - % retained in care, on ART, virally suppressed (NHBS, Maryland DHMH) MSM demography

Model calibration outcomes Observed Baltimore data: HIV prevalence - by age and race (NHBS) MSM demography - age and race (NHBS, census) Care cascade - % retained in care, on ART, virally suppressed (NHBS, Maryland DHMH) The HIV care cascade, in particular the proportion of men in care, on treatment and virally suppressed

Defined plausible parameter ranges Model calibration Defined plausible parameter ranges Sampled parameter ranges randomly 4 million times With each parameter set, simulated HIV epidemic 1984 – 2015 Retained parameter sets for which model outputs agree with observed Baltimore data

Defined plausible parameter ranges Model calibration Defined plausible parameter ranges Sampled parameter ranges randomly 4 million times With each parameter set, simulated HIV epidemic 1984 – 2015 Retained parameter sets for which model outputs agree with observed Baltimore data

Defined plausible parameter ranges Model calibration Defined plausible parameter ranges Sampled parameter ranges randomly 4 million times With each parameter set, simulated HIV epidemic 1984 – 2015 Retained parameter sets for which model outputs agree with observed Baltimore data

Defined plausible parameter ranges Model calibration Defined plausible parameter ranges Sampled parameter ranges randomly 4 million times With each parameter set, simulated HIV epidemic 1984 – 2015 Retained parameter sets for which model outputs agree with observed Baltimore data

Model Fits: HIV prevalence by age and race These plots show that the model was able to reproduce age- and race- specific HIV prevalence trends well. We obtained 299 different ‘fits’ – unique combinations of input parameters which were able to reproduce all of the observed data [Dotted lines are 95% CI]

% of HIV-infected MSM virally suppressed Analysis – meeting incidence reduction targets HIV incidence % of HIV-infected MSM virally suppressed Having calibrated the model to the data, I then used the model to estimate the impact of different interventions. Here I illustrate how I did this analysis. For each parameter set fitting the data, I ran the projection forwards in time for 10 years (starting at the beginning of 2016), assuming levels of testing, linkage, ART initiation and retention stay at the level they were at at the end of 2015 – shown here for one of our selected parameter sets. This is our base case scenario

% of HIV-infected MSM virally suppressed Analysis – meeting incidence reduction targets HIV incidence % of HIV-infected MSM virally suppressed I then looked at 1,600 different combinations of higher rates of HIV testing, linkage, ART initiation and retention, assumed that these higher rates started at the beginning of 2016, and looked for combinations that enabled the model to meet the targets. The grey lines are from all the different enhanced intervention scenarios I looked at. In blue is one where the incidence after 5 years is 20% lower than the incidence in the base case, which is one of our targets – so I record for this the level of viral suppression that has been reached after 5 years (blue line) and the absolute increase in percentage of HIV-infected MSM who are virally suppressed after 5 years, compared to the initial level of viral suppression at the start of 2016.

Target: reduce HIV incidence by 10, 20, 30, 50% Increase in viral suppression needed to reach incidence reduction targets Target: reduce HIV incidence by 10, 20, 30, 50% This plot shows the increase in viral suppression required to reach different incidence reduction targets, from 10 to 50%. As you’d probably expect, the higher the target, the greater the increase in viral suppression needed to meet that target over a given time period.

Target: reduce HIV incidence by 10, 20, 30, 50% Increase in viral suppression needed to reach incidence reduction targets Target: reduce HIV incidence by 10, 20, 30, 50% 15 For example, after 5 years you can see that, to reach our 20% incidence reduction target, viral suppression needs to 15 percentage points higher than the initial level, reaching a final level of 55% of MSM suppressed

Target: reduce HIV incidence by 10, 20, 30, 50% Increase in viral suppression needed to reach incidence reduction targets Target: reduce HIV incidence by 10, 20, 30, 50% 32 15 To reach our 50% incidence reduction target after 5 years, viral suppression needs to increase by 32 percentage points above the initial level, reaching a final level of 72% of MSM suppressed

Target: reduce HIV incidence by 10, 20, 30, 50% Increase in viral suppression needed to reach incidence reduction targets Target: reduce HIV incidence by 10, 20, 30, 50% Note that it wasn’t possible to reach the 50% target after 2 years, even with very intense interventions achieving at least 60% virally suppressed after 2 years.

Impact of meeting NHAS targets If NHAS targets all met in 2020: 90% diagnosed – 90% of diagnosed in care - 80% of diagnosed virally suppressed Here is the predicted impact of meeting the US targets for diagnosis, retention in care and viral suppression in 2020 –

Impact of meeting NHAS targets If NHAS targets all met in 2020: 90% diagnosed – 90% of diagnosed in care - 80% of diagnosed virally suppressed Increase in viral suppression in 2020 33pp we predict that if all of these targets are met for Baltimore MSM, the level of viral suppression will be increased by 33 percentage points above 2016 levels, giving 73% of MSM virally suppressed in 2020

Impact of meeting NHAS targets If NHAS targets all met in 2020: 90% diagnosed – 90% of diagnosed in care - 80% of diagnosed virally suppressed Increase in viral suppression in 2020 Incidence reduction in 2020 33pp 53% , and this will lead to a reduction in HIV incidence of 53% in 2020 compared with the base case scenario,

Impact of meeting NHAS targets If NHAS targets all met in 2020: 90% diagnosed – 90% of diagnosed in care - 80% of diagnosed virally suppressed Increase in viral suppression in 2020 Incidence reduction in 2020 33pp 53% which is in fact very similar to reaching our 50% incidence reduction target after 5 years. Meeting NHAS targets: very similar effect to reaching 50% incidence reduction target after 5 yrs

Likelihood of HIV elimination Likelihood of HIV elimination (HIV incidence <0.1% per year) among MSM in Baltimore in the next 20 years: under current levels of care: 0 when incidence reduction targets are met: Incidence reduction target: 10% 20% 30% 50% 2 years 3.5% - 5 years 0.02% 10 years when NHAS targets are met: Finally we were interested in knowing how likely HIV elimination would be both in our base case scenario and when meeting these different targets. We limited the time horizon to the next 20 years, and defined HIV elimination as HIV incidence going below 1 new infection per 1000 susceptible MSM per year, which is a threshold that’s commonly used with deterministic models of HIV transmission. In our base case scenario, HIV elimination was never observed. target: 90% of infected diagnosed 90% of diagnosed in care 80% of diagnosed virally suppressed All 3 targets 4.64% 0.40% 0.42%

Likelihood of HIV elimination Likelihood of HIV elimination (HIV incidence <0.1% per year) among MSM in Baltimore in the next 20 years: under current levels of care: 0 when incidence reduction targets are met: Incidence reduction target: 10% 20% 30% 50% 2 years 3.5% - 5 years 0.02% 10 years when NHAS targets are met: When our incidence reduction targets were met, elimination was only rarely observed within 20 years – in at most 4% of runs target: 90% of infected diagnosed 90% of diagnosed in care 80% of diagnosed virally suppressed All 3 targets 4.64% 0.40% 0.42%

Likelihood of HIV elimination Likelihood of HIV elimination (HIV incidence <0.1% per year) among MSM in Baltimore in the next 20 years: under current levels of care: 0 when incidence reduction targets are met: Incidence reduction target: 10% 20% 30% 50% 2 years 3.5% - 5 years 0.02% 10 years when NHAS targets are met: Similarly, when the NHAS targets were met, elimination was only predicted in at most 5% of runs. So, HIV elimination within the next 20 years is very unlikely to occur, even if national targets are met. target: 90% of infected diagnosed 90% of diagnosed in care 80% of diagnosed virally suppressed All 3 targets 4.64% 0.40% 0.42%

Discussion Large increases in viral suppression are needed to significantly reduce HIV incidence among Baltimore MSM Achieving NHAS targets by 2020 projected to reduce HIV incidence by ~50% Local elimination of HIV unlikely to occur in the next 20 years, even if NHAS targets met Future modelling should also consider the impact of increased PrEP coverage

Discussion Large increases in viral suppression are needed to significantly reduce HIV incidence among Baltimore MSM Achieving NHAS targets by 2020 projected to reduce HIV incidence by ~50% Local elimination of HIV unlikely to occur in the next 20 years, even if NHAS targets met Future modelling should also consider the impact of increased PrEP coverage

Discussion Large increases in viral suppression are needed to significantly reduce HIV incidence among Baltimore MSM Achieving NHAS targets by 2020 projected to reduce HIV incidence by ~50% Local elimination of HIV unlikely to occur in the next 20 years, even if NHAS targets met Future modelling should also consider the impact of increased PrEP coverage

Discussion Large increases in viral suppression are needed to significantly reduce HIV incidence among Baltimore MSM Achieving NHAS targets by 2020 projected to reduce HIV incidence by ~50% Local elimination of HIV unlikely to occur in the next 20 years, even if NHAS targets met Future modelling should also consider the impact of increased PrEP coverage

These results suggest ambitious targets for HPTN 078 Implications for HPTN These results suggest ambitious targets for HPTN 078 Powered to detect an 18pp difference in viral suppression after 2 years → ~29% reduction incidence 50% incidence reduction requires ~30pp difference Results suggest that other interventions may be needed in parallel with expanded treatment to reduce HIV incidence rapidly among US MSM

These results suggest ambitious targets for HPTN 078 Implications for HPTN These results suggest ambitious targets for HPTN 078 Powered to detect an 18pp difference in viral suppression after 2 years → ~29% reduction incidence 50% incidence reduction requires ~30pp difference Results suggest that other interventions may be needed in parallel with expanded treatment to reduce HIV incidence rapidly among US MSM

HPTN Modelling Centre: Marie-Claude Boily, Dobromir Dimitrov The HPTN Modelling Centre is funded through the HPTN Statistical and Data Management Centre (UM1 AI068617) HPTN Modelling Centre: Marie-Claude Boily, Dobromir Dimitrov NHBS data: Gabriela Paz-Bailey, Brooke Hoots, Danielle German, Colin Flynn HPTN 078: Chris Beyrer, Robert Remien, Theresa Gamble, Protocol and site teams