The differential impacts of non-locally acquired infections and treatment interventions on heterosexual HIV transmission in Hong Kong

Introduction Heterosexual infections have contributed to a high proportion of the HIV burden in Asia and Eastern Europe. Human mobility and non-local infections are important features in some cities/countries. An understanding of the determinants of the sustained growth of the heterosexual HIV epidemics would enable the potential impacts of treatment-based interventions to be assessed. Methods We developed a compartmental model for heterosexual HIV transmissions, parameterized by clinical and surveillance data (1984–2014) in Hong Kong. HIV sequence data were included for examining genetic linkages and clustering pattern. We performed sensitivity analyses to evaluate effects of high-risk sexual partnership and proportions of non-locally acquired infections. Four hypothetical interventions (a) immediate treatment, (b) enhancement of retention in care, (c) HIV testing campaigns, and (d) test-and-treat strategy, were examined. Results Data of 2174 patients (723 female and 1451 male) diagnosed with HIV between 1984 and 2012 in Hong Kong were collected for model parameterization. Among 1229 sequences of non-MSM (men who have sex with men) patients, 70% were isolates and 17% were either dyads or triads. In base-case scenario, the total estimated number of new infections in 2012–2023 would be 672 for male and 452 for female. Following 100% retention in care intervention, the total proportion of averted new infections in 2012–2023 would be 7% for male and 10% for female. HIV testing campaign in 2012 and 2017 followed by 100% immediate treatment strategy would avert 5% and 9% of male and female new infections, respectively. In the epidemic model, an increase of high-risk sexual partnership from 6% to 9% would increase the epidemic growth (annual number of newly diagnosed and newly infected cases) by about 10%. If no non-locally acquired infection occurred as from 2012, the epidemic growth would slump. To control the heterosexual epidemic, periodic HIV testing at 5-year intervals with immediate treatment would avert 5–13% of annual new infections in 2013–2023. Conclusions Enhanced HIV testing with immediate treatment is most effective in controlling the heterosexual epidemic, the impacts of which might however be attenuated by any increase of non-locally acquired infection, assuming little variations of high risk partnership over time.


Introduction
Worldwide, sexual transmission accounts for a majority of prevailing HIV infections. In the past decade, HIV transmission in men who have sex with men (MSM) has escalated globally, an observation also seen in Asian countries. [1,2] In contrast, delineation of epidemiological patterns in heterosexual infection has so far received relatively less attention, [3][4][5][6] especially outside sub-Sahara Africa. In Asia and Eastern Europe, heterosexual infections have contributed significantly to the overall HIV viral burdens at population levels. [7] In 2017, for example, heterosexual transmission accounted for 68% of all reported new HIV diagnoses in Eastern Europe. [8] In the development of intervention programmes, key populations including MSM, people who inject drugs (PWID) and female sex workers (FSW) have often been the main focus of HIV prevention programme, while heterosexually infected persons are neglected. Despite much lower risk of HIV transmission, the large population size could mean a high number of HIV infections and the challenge of effective prevention program implementation. When transmission of HIV in the key populations could explain the emergence of concentrated HIV epidemics, it could not well explain the sustained transmissions of the virus in the general population.
Partner concurrency could be a driver of the generalized heterosexual HIV epidemics in Africa, though such proposition has continued to be hotly debated. [9][10][11] The predicted impacts of interventions targeting concurrency were however moderate. [9] The existence of small networks, impacted by concurrency, has been examined through stochastic modelling to understand the sustained transmission of the virus in the community. [12] To explain the sustained heterosexual transmission, non-locally acquired infections were suggested to be another key parameter. In Europe, migrants accounted for 41% of new HIV diagnoses in 2017. The declining trend of reported heterosexual cases was affected by the falling number of heterosexual migrant cases originating from generalized HIV epidemic countries. [8] In Australia, 35% of heterosexual HIV diagnoses in 2002-2011 were attributable to non-local exposure. [13] Some HIV epidemic modelling studies have also focussed the influence of migration on local epidemic growth, the impact of which varied by local context. [14][15][16] In Hong Kong, an international city in Asia, the heterosexual HIV epidemic grew rapidly in the 1980s-1990s but became stabilized from 2000 onwards. Heterosexual infections have accounted for some two-fifths of all reported cases despite the low overall HIV prevalence in the population, [17] a scenario shared by other low prevalence countries in Asia. Following expanding coverage of antiretroviral therapy (ART) for HIV-infected individuals in care, the impacts of treatment-as-prevention could play a role in shaping future epidemiology. To explain the transmission pattern of heterosexual HIV infections, a compartmental model was designed to assess the potential impacts of interventions centring on the promotion and enhancement of ART, and non-locally acquired infections.

Model structure
A deterministic compartmental model was constructed in R3.5.1, composing of a heterosexual male and a heterosexual female sub-model, with interactions between but not within the two. Bisexual and transgender sub-models were not separately constructed but subsumed under MSM, the epidemiology of whom was not modelled in this study. Each sub-model was structured by compartments of susceptible (sexually active adults aged 15-64 for female and aged 15-84 for male) and infected individuals progressing through the following disease stagesacute infection, chronic infection categorized by CD4 levels of �200/μL, 201-350/μL, 351-500/ μL, >500/μL, and AIDS-and along the cascade of HIV care (undiagnosed, diagnosed, treatment initiated and lost to follow-up) (S1 Appendix). The model was an open system with population inflow (net susceptible population growth and non-locally acquired infections) and outflow (death). The epidemic growth was simulated from 1983 to 2011, and projected from 2012 to 2023.

Data source and model parameters
Anonymous longitudinal clinical data in 1985-2014 of HIV-infected heterosexuals aged 18 or above at diagnosis were retrieved from an HIV cohort database built with data accessed retrospectively from all three HIV specialist clinics in the public service in Hong Kong. [18] In addition, HIV-1 genotype resistance testing sequences in 1994-2012 of HIV-infected non-MSM patients were accessed. Descriptive statistics were performed to describe the demographics, baseline clinical status and HIV treatment status of these individuals.
The collected data were used to estimate the value of most of the model parameters, including diagnosis rate, pre-treatment disease progression rate, treatment initiation rate, loss to follow-up rate, viral load suppression (SVL, �500 copies/mL) and rebound (>500/mL) rate (description on the derivation of parameter values from clinical data in S1 Appendix). Annual number of reported new diagnoses and proportion of suspected non-local HIV infection were retrieved from surveillance epidemiological datasets. Parameters from overseas clinical trials, such as transmission risk, were referenced from literature.
Approvals of the Joint Chinese University of Hong Kong-New Territories East Cluster Clinical Research Ethics Committee (CREC), Kowloon West Cluster Research Ethics Committee, and Kowloon East Cluster Research Ethics Committee were obtained. Data access approval was granted by Department of Health, Hong Kong Special Administrative Region Government in compliance with the Personal Data (Privacy) Ordinance. Informed consent was waived as the collected data were anonymised and had been collected retrospectively. This study was conducted in compliance with Declaration of Helsinki.
To describe the clustering pattern of cases, collected HIV-1 sequences (protease and partial reverse transcriptase of pol gene) were aligned by MUSCLE and analysed by neighbour-joining tree in MEGA6 to identify clusters with bootstrap value �90, 1000 times simulation. A total of 1229 sequences of patients were analysed (976 for heterosexual contacts, 18 for contaminated blood transfusion contacts, 204 for injection drug use, and 31 for undefined transmission mode). Among them, 142 sequences (12%) were genetically connected with 1 other sequence, forming 71 dyads, and another 60 sequences (5%) were connected with 2 other sequences in 20 small triads, 164 sequences (13%) were connected with at least 3 other sequences in 23 clusters (S1 Appendix, p.9), and the rest (70%) were isolated sequences. Different from HIV-infected MSM who formed a number of clusters, [19] most of the sequences were isolates, dyads or triads for HIV-infected heterosexuals. In this model, there were no further delineation of population in heterosexual male and female submodels. Instead, sexual partnership was ranked in our model. A susceptible individual was assumed to be at 1) high risk of HIV transmission if he/she and his/her sex partners were in concurrent partnership; 2) medium risk if one was in serial monogamy while the other in concurrent partnership; 3) low risk if both partners were in serial monogamy. We further assumed that the risk decreased by one level after HIV diagnosis, as a result of ART and/or reduction of behavioural risk. [20] Estimation of the force of infection is described in the S1 Appendix.
We fitted model predictions over time to the reported prevalent number of HIV-infected heterosexual male and female in 1984-2012 in Hong Kong under negative log likelihood in R (stats4 package) using male function. In this calibration process, [21] we simultaneously varied the duration of partnership for male and female at medium risk, duration and proportion of partnership at low risk within data ranges suggested in the previous studies to find the best fit value. Modelling results were validated with number of reported prevalent cases in 2013 and annual number of newly diagnosed cases. [22] We defined non-locally acquired infections as HIV transmissions occurring either outside Hong Kong, or from a non-local HIV positive individual within the city. It was categorized by gender and ethnicity (Chinese and non-Chinese) in model parameters. A Chinese person with non-locally acquired HIV infection was assumed to be physically present in Hong Kong during acute infection, on the premise that he/she was either infected during short travel overseas or had contracted the virus from non-local persons within the territory. Non-locally acquired HIV infection in a non-Chinese was however added to the compartment of patients diagnosed with CD4>500/μL, assuming that they were infected beyond the period of acute infection before entering Hong Kong. Annual surveillance reports [17] published by the Department of Health were consulted to derive parameter values.

Scenarios constructions
In base-case scenario, we simulated the HIV epidemic in 1983-2010, parameterized by longitudinal clinical data and literature. The epidemic growth for heterosexuals in the medium term (2012-2023) was then projected with treatment initiation criteria and diagnosis rate of 2011. In response to the recommended test-and-treat strategy for controlling the HIV epidemic, [23][24][25] four scenarios of treatment-based interventions were designed for introduction from 2012, while keeping other conditions unchanged. These scenarios were A) rate of retention in care before and on ART at 50% and 100% from 2012; B) immediate treatment for all diagnosed cases from 2012; C) HIV testing campaigns in 2012 and 2017 (assuming 100% diagnosis rate in the respective year); and D) test-and-treat-HIV testing campaign in 2012 and 2017 with immediate treatment from 2012.

Sensitivity analyses
To examine the impact of uncertainty of key parameters on model simulation results, sensitivity analyses were performed for the influences of varying the risk ranking ratios of sexual partnership, and proportion of non-locally acquired infections, with the starting point of 1) 1983 and 2) 2012 respectively. Five combinations of partnership types were simulated in the sensitivity analyses by varying the ratios of high:medium:low risk partnership: A) 6%:38%:56% (base-case); B) 3%:38%:59%; C) 3%:41%:56%; D) 9%:35%:56%; and E) 9%:38%:53%. Sensitivity analyses for zero, twice and thrice the annual number of non-locally acquired infections by gender were performed. Other model parameters were kept unchanged and no intervention were introduced in these analyses. Separately, sensitivity analyses for non-locally acquired infections were performed from 2012 with intervention scenarios.

Characteristics of heterosexually acquired HIV infections in Hong Kong
Between 1984 and 2012, 2174 (48%) of 4508 patients aged 18 or above at diagnosis in the cohort database had contracted HIV through heterosexual contacts. Of these, 1844 (85%) were diagnosed in or after 1997, when ART became regularly offered to patients under a CD4guided approach. The male-to-female ratio of these patients was 2:1. CRF01_AE was the commonest HIV subtype, accounting for 66% of patients whose virus had been genotyped. The patients' characteristics were compared by gender using univariate analyses (odds ratio) and Mann-Whitney U test (U-test). Females were more likely to be non-Chinese and were younger at diagnosis compared to males (33 year-old vs 40 year-old, p<0.001). (Table 1) Overall, males were at a poorer clinical state at diagnosis with a lower median CD4 count (116/μL vs 205/μL, p<0.001) and higher viral load (median log 10 4.9/mL vs log 10 4.6/mL, p<0.01). A higher proportion of male had progressed to AIDS at HIV diagnosis, or that the diagnosis was made within three months of AIDS (defined as late diagnosis in this study).
At the latest assessment, 78% of the patients have been or were on ART. There was no gender difference as regards treatment status and responses. The median interval from diagnosis to ART initiation, and from treatment initiation to SVL, was four months and three months respectively. Overall, some 78% (1469/1878) had achieved SVL.

Base-case model and its projection to 2023
The

Scenarios of interventions and parameter changes from 2012
In the simulation, full retention in care per se (Scenario A1) resulted in lower epidemic growth than implementing immediate treatment alone (Scenario B) (Fig 2, Table 3A and 3B). As shown in the sensitivity analyses, the effect of interventions was tested by changing the annual number of non-locally acquired infections. In Scenario D, the estimated annual number of new male diagnoses would increase from 53 (-14% above base-case) in 2014 to 96 (46% above base-case) in 2023 when the number of non-locally acquired infections doubled (Fig 3). Similarly, the male and female new infections above base-case would become 40 (69%) and 22 (53%) more, respectively, if the number of non-locally acquired infections doubled.
Without interventions, when no non-locally acquired infection was added to the model from 2012, the annual number of newly diagnosed and prevalent cases dropped linearly (S1 Appendix, p.11). The number of new diagnoses in 2023 would be 89 less (73% lower) than base-case, while the number of diagnosed prevalent cases would be 698 less (22% lower). Instead of a linear decline in annual new diagnoses, the number of new infections would be 37

PLOS ONE
Heterosexual HIV transmission  and 16 less for male (68% lower) and female (46%) than base-case respectively in 2013, and this proportion of reduction would almost remain unchanged afterwards.

Results of sensitivity analyses by parameter changes from 1983
From the sensitivity analyses with 1983 as the starting year, a higher proportion of high risk partnership (Scenario D and E in S1 Appendix, p.12) would drive the epidemic growth. Specifically, a 50% increase of high risk partnership (say, from 6% to 9%) generated about 10% increase in the annual number of newly diagnosed, newly infected and diagnosed prevalence cases. The resultant annual increase of new infections would be at least 5% higher in female compared to males since 2003. The impact of changing the proportional distribution of risk ranking of sexual partnership since 2012 on the epidemic growth (S1 Appendix, p.13) could be inferred from the changes of parameter values from 1983 in the sensitivity analyses.
In the absence of non-locally acquired infections from 1983, there would have been zero epidemic growth (results not shown). When the annual number of non-locally acquired infections doubled, the number of newly infected, newly diagnosed and diagnosed prevalent cases would likewise double. The epidemic growth differed when the proportion of male or female non-locally acquired infection was altered separately (S1 Appendix, p.14). By setting non-locally acquired male infections to zero, less than 400 newly diagnosed and newly infected male resulted in 1983-2023 (at least 70% lower than base-case) while there would be 896 newly diagnosed (39% lower) and 513 newly infected female (58% lower). Conversely, setting nonlocal female infections to zero would simulate 473 (61% lower) and 585 (42% lower) new female diagnoses and infections than base-case, while male epidemic growth became just 19% and 23% lower, respectively. Similarly, an increase of non-locally acquired male infections resulted in substantial growth of both the male and female epidemics, while an increase of non-locally acquired female infections impacted the female epidemic but caused minimal changes to the male epidemic. Table 3. Changes in heterosexually acquired infection following interventions introduced in 2012, as compared to base-case by 2023. A1-100% retention in care; A2-50% retention in care; B-immediate treatment for all diagnosed cases; C-HIV testing campaigns in 2012 and 2017; D-campaign in 2012 and 2017 followed by 100% immediate treatment.  [26][27][28] Our results further highlighted the differential influence of gender, suggesting that HIV-infected male was a major driver of the HIV epidemic in female, but not vice versa. In our simulation, the effect of high risk sex partnership on the epidemic growth was moderate, compared to results in network-based studies published elsewhere. [12,29] To control on-going heterosexual transmissions, new interventions targeting male appears to be crucial.

3(a) Male
As ART could reduce heterosexual HIV transmission, [20] treatment-based interventions are potentially useful. The importance of linkage to care and retention in care has been highlighted in a study in South Africa [25] and the United States [30]. Our modelling results suggested that enhancing retention in care to 100% or launching HIV testing campaigns with immediate treatment could reduce new infections by 5-7% in male and 9-10% among female by 2023. We simulated the impact of setting up periodic HIV testing campaign (at 5-year intervals), which is theoretically less costly than an annual campaign. If a 100% diagnosis rate is achieved, very low epidemic growth could result. The benefit would be even bigger if periodic testing is combined with immediate treatment. Whereas epidemic control have been concluded in other studies, [23,31] a test-and-treat strategy in Hong Kong is not anticipated to achieve such significant outcome. The discrepancy might be due to the current high treatment coverage and high proportion of SVL in Hong Kong, leaving limited rooms of improvement. [18] In addition, a high proportion of non-locally acquired infection among the heterosexual male could attenuate the impact of any forms of test-and-treat interventions.
We acknowledge that there're limitations in our study. The compartmental design might have underestimated the impact of concurrency. Nonetheless, our projected curves were similar to the epidemic curves derived from reported numbers. On the contrary, network structured models might overestimate the impact of concurrency by ignoring the potentially significant reduction of transmission risk after diagnosis and treatment initiation. [20,32] Other limitations of our study subsuming bisexual population to MSM. Bisexual individuals accounted for only 4% of diagnosed HIV cases, [17] and less than 5% of MSM [unpublished data]. In places like neighbouring Mainland China where some 26% of MSM were bisexual, [33] their inclusion as a separate sub-population would be crucial. We had likewise not considered FSW and their clients separately in the compartment models because of the extremely low (<0.2%, http://www.chp.gov.hk/files/pdf/g220.pdf) HIV prevalence among FSW. Limited by data availability, phylogenetic analysis was only based on the partial polymerase (pol) gene that included protease and reverse transcriptase regions.
As regards parametrization, our study has included non-locally acquired infection as a key parameter. Our approach of not separating overseas infections and those from non-local people in Hong Kong was constrained by data availability. The transmission hazard for high risk sexual partnership was estimated by the formula of random mixing, so as to minimize model uncertainty. This might have underestimated the transmission risk as defined in other studies. [12,29] We only stratified non-locally acquired individuals by ethnicity, but not for locally acquired individuals, assuming that there was no significant difference of infection risk by ethnicity in local population. This may however not be true, and adjustment is needed in settings with significant difference. Furthermore, model estimation might vary by the different units of hazard estimation. For example, the transmission hazard could be expressed in per-act basis instead of unit of partnership in this study, the results of which are not comparable. The assignment for partnership duration was arbitrary, due to the lack of relevant local data for the definitions. Nonetheless, model estimation on the annual number of new diagnoses and prevalent cases were close to the reported numbers.
In conclusion, enhancing retention in care, launching periodic HIV testing campaign, and offering immediate treatment for all diagnosed persons would be most beneficial for controlling the heterosexual spread of HIV in Hong Kong. Such approach would also be most feasible for introduction within the existing healthcare system. To monitor the epidemic growth in conjunction with the implementation of new interventions, it would be desirable to include indicators reflecting the cascade of care. On the other hand, in view of the potentially high influence of non-locally acquired infections on the heterosexual epidemic, more related information would need to be collected from newly diagnosed cases. To improve our understanding of sexual partnership (concurrent and serial monogamy) and partner exchange rate among heterosexuals, tailored behavioural study regularly conducted locally would be most useful. Finally, while our models were designed for Hong Kong, we believe that the methods and results are applicable for areas with similar socio-ethnic background, HIV epidemic condition and HIV care and surveillance system, such as some Mainland Chinese cities and Taiwan.  Table. (XLSX)

Supporting information
The opinions and assertions contained herein are private views of the authors and do not necessarily reflect those of the Centre for Health Protection, Hong Kong Special Administrative Region Government Department of Health, or other affiliating institutions.