The age and sex dynamics of heterosexual HIV transmission in Zambia: an HPTN 071 (PopART) phylogenetic and modelling study

  1. Matthew Hall
  2. William Probert
  3. Lucie Abeler-Dörner
  4. Chris Wymant
  5. Francesco di Lauro
  6. Xiayoue Xi
  7. Rafael Sauter
  8. Tanya Golubchik
  9. David Bonsall
  10. Michael Pickles
  11. Anne Cori
  12. Justin Bwalya
  13. Sian Floyd
  14. Nomtha Bell-Mandla
  15. Kwame Shanaube
  16. Blia Yang
  17. Peter Bock
  18. Deborah Donnell
  19. Mary Kate Grabowski
  20. Deenan Pillay
  21. Oliver Ratmann
  22. Sarah Fidler
  23. Helen Ayles
  24. Richard Hayes
  25. Christophe Fraser
  26. the PANGEA-HIV consortium
  27. the HPTN 071 (PopART) study team

  • Curated by eLife

    eLife logo

    eLife Assessment

    This important study provides convincing evidence for our understanding of HIV transmission dynamics by age and sex in Zambia during the PopART trial. By combining phylogenetic and individual-based mathematical modelling (IBM), it adds depth to the epidemiological literature and may inform more strategic allocation of HIV prevention resources in sub-Saharan Africa. The authors employ two complementary and well-established methodologies (phylogenetics and IBM), and this dual approach is a notable strength.

Reviewed by

Read the full article

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

Log in to save this article

Abstract

While much progress has been made in reducing the incidence of HIV-1 infection in sub-Saharan Africa in recent years, bringing the epidemic to an end will require identification of the demographic groups that continue to contribute to transmission. Pathogen phylogenetics and individual-based mathematical models (IBMs) of transmission are approaches that enable researchers to explore such questions. Here, we used both methods to characterise the ages and sexes of the individuals involved in heterosexual transmission in the context of the HPTN 071 (PopART) trial in Zambia.

The two methods produced largely concordant results, strengthening confidence in both. A principal finding was that when the age gap in transmission (the difference of ages between the two individuals) was stratified by recipient age, the largest differences were for the youngest female recipients and the smallest for the youngest males. For women under 21 this stood at a male 9.87 years older (95% CI: 8.02 - 11.59) in the phylogenetics, compared to 6.93 (95% HDI: 6.56 - 7.32) in the IBM. As the age of female recipients increased, this gap decreased towards parity. Conversely, the under-21 male recipients saw the smallest gaps with the female older by 0.14 years (95% CI: −2.95 - 3.23) in the phylogenetics and 1.38 years (95% HDI: 0.98 - 1.68) in the IBM. As the age of male recipients decreased, this gap steadily increased. The consequence of this pattern is that transmission to new age cohorts first entering into sexual activity is driven predominantly by male-to-female transmission.

We also showed that targeting interventions at younger adults captures most of the benefit of population-wide approaches. We used the IBM to simulate the PopART universal testing and treatment intervention into the future, showing that effective treatment of under-35-year-olds would account for 94.3% (95% HDI: 65.8% - 126.6%) of the reduction in incidence by 2039 that would be achieved by treating the entire population, while effective treatment of under-35 men accounts for 60% (95% HDI: 23.2% - 92.1%).

Finally, we simulated a one-year cessation of ART treatment for the whole population, which resulted in an immediate increase in both incidence and the average age at transmission of both sources and recipients. The magnitude of this was 4.6 years (95% HDI: 2.17 - 6.24) for female recipients, 5.3 (95% HDI: 2.74 - 7.09) for male recipients, 5.24 (95% HDI: 2.78 - 6.97) for female sources, and 6.04 (95% HDI: 2.92 - 8.09) for male sources. These changes would be slow to reverse even after ART was restored.

These findings indicate that substantial reductions in HIV incidence can be achieved through intensified testing and treatment of individuals aged under 35, and in particular young men, a group that drives the infection of younger women and for whom engagement with care remains disproportionately low.

Article activity feed

  1. Version published to 10.7554/elife.107082.2 on eLife
  2. Version published to 10.7554/elife.107082 on eLife
  3. eLife

    eLife Assessment

    This important study provides convincing evidence for our understanding of HIV transmission dynamics by age and sex in Zambia during the PopART trial. By combining phylogenetic and individual-based mathematical modelling (IBM), it adds depth to the epidemiological literature and may inform more strategic allocation of HIV prevention resources in sub-Saharan Africa. The authors employ two complementary and well-established methodologies (phylogenetics and IBM), and this dual approach is a notable strength.

  4. eLife

    Reviewer #1 (Public review):

    Summary:

    This manuscript describes the results of phylogenetic and epidemiological modeling of the PopART community cohorts in Zambia.

    Comments on revised version:

    Thank you for the opportunity to re-review this interesting paper.

    This reviewer struggled to follow along with the author's response letter. It was challenging because responses were brief and did not list the specific changes made, leaving the reviewer to search for changes in the text. As best I could tell, there were no changes made that matched some of the highest-priority suggestions.

    Critique #1 - This reviewer did not find the presentation of confidence intervals in the Abstract and other sections, which were suggested. Please note the format that was suggested in the original critique from Reviewer 1.

    Critique #2 - regarding removal of unsubstantiated claims and use of a p-value to compare analysis to a null hypothesis - it seems the authors skipped over this critique and did not address it.

    Regarding bias: the authors answered a different question than the one asked. The reviewer asked what proportion of transmissions were sampled; the authors stated that only communities from which phylo data was acquired were modeled. Was sampling 100% in those communities? Please provide the percentage and provide analysis that shed light on how sampling bias could impact the analysis.

    Regarding "cherries" - the reviewer did not understand the author's response. The query was regarding what percent of the total number of phylogenetic pairs (denominator) were the 355 that had high confidence in directionality (numerator). The response could be expressed be a proportion.

    The expectation of ART reducing the age of sources of transmission seems unrealistic to this reviewer. People on ART are not always adherent and can still transmit during gaps in adherence. ART dramatically increases life expectancy with HIV, which would have the opposite effect.

  5. eLife

    Reviewer #2 (Public review):

    Summary:

    The authors analyzed PopART data to better characterize the age and sex specific transmission dynamics in Zambia with a goal of allocation of resources.

    Strengths:

    Important analysis to hone in on key driver of HIV transmission in Zambia, which hopefully can be used to tune prevention efforts to maximize effect while limiting required resources. Two analytic approaches used, and while the phylogenetic data was markedly more limited, it mirrored the simulated epidemic. The authors did a nice job reviewing the limitations of the data and the analyses and providing analyses to support their goals and hypothesis, and this work may have more impact now that resources in SSA for HIV prevention and treatment may become more scarce.

    Comments on revised version.

    The revised manuscript clarifies the impact and utility of this work and better allows the comparability of the two methods. Highlighting the differences (or lack thereof) between the undiagnosed and diagnosed population) simplifies the public health approach.

  6. eLife

    Author response:

    The following is the authors’ response to the original reviews.

    eLife Assessment

    This important study provides evidence for our understanding of HIV transmission dynamics by age and sex in Zambia during the PopART trial; by combining phylogenetic and individual-based mathematical modelling (IBM), it adds depth to the epidemiological literature and may inform more strategic allocation of HIV prevention resources in sub-Saharan Africa. The authors employ two complementary and well-established methodologies (phylogenetics and IBM), and this dual approach is a notable strength. However, the evidence supporting key conclusions is incomplete, with several claims insufficiently substantiated by the data presented. Improvements in data presentation (e.g., quantification of qualitative statements, statistical estimates, and clearer description of results) would substantially strengthen the paper.

    We thank the editor and reviewers for their positive comments. We have revised the manuscript in response to the points raised, as described below.

    First of all, we would like to summarise what we have changed regarding the presentation of summary statistics throughout the text. We agree that many of the statements in the original submission tended towards being qualitative. This was the result of shying away from presenting two separate estimates, with different ways of quantifying uncertainty, in the text. The phylogenetics could be presented as mean and confidence interval, while the IBM would need some measure of centrality (mean or median) and the highest density interval for a summary statistic (e.g. the mean age gap) as it varies over the posterior. These are not directly comparable. We have now changed this to present both where appropriate, with cautionary note about the difference between the CIs and HDIs (lines 257-260).

    We also were somewhat arbitrary regarding where we chose to summarise the posterior in the IBM or look at the best-fitting single simulation, and where we presented the mean as opposed to the median. We have done a considerable overhaul of what is presented in this revision:

    (1) We always present the posterior summary unless the level of detail is such that summarising uncertainty over the posterior is not feasible (e.g. in figures 3, 4 and 5). In the latter case we still use the best-fitting IBM replicate.

    (2) In the main text we always present the mean. For the phylogenetics the summary statistics are mean and confidence interval. For the IBM this is the posterior mean, and 95% HDI, of the mean of a particular statistic as calculated in each of the 1000 IBM replicates. For example, each replicate will have its own distribution of male source ages which have a mean value. These means also vary over the posterior, and a mean of them is calculated, as well as the HDI interval to represent posterior uncertainty. This “mean of means” may be a slightly confusing piece of terminology at first glance, but it allows us to properly capture posterior uncertainty in a way we mostly avoided in the first submission.

    One result of 1) above is a change to figure 6. It is now summarised over the posterior, with the result that time trends that were previously not evident become clear. This changes our conclusions slightly (lines 529-537) but it should be noted that the magnitudes of the trends remain small.

    Public Reviews:

    Reviewer #1 (Public review):

    Summary:

    This manuscript describes the results of phylogenetic and epidemiological modeling of the PopART community cohorts in Zambia. The current manuscript draft is methodologically strong, but needs revision to strengthen the take-home messages. As written, there are many possible take-away conclusions. For example, the agreement between IBM and phylogenetic analysis is noteworthy and provides a methodological focus. The revealed age patterns of transmission could be a focus. The effects of the PopART intervention and the consequences of a 1-year disruption could be a focus. It is important, though, that any main messages summarized by the authors are substantiated by the evidence provided and do not extrapolate beyond the data that have been generated. I recommend that the authors think deeply about what the most important, well-supported messages are and reframe the discussion and abstract accordingly.

    We have rewritten the abstract, and also made changes to the discussion in order to centre our message around the contribution of particular of demographic groups to transmission, and how, with that contribution revealed, such groups can be selected for specialised interventions.

    Strengths/weaknesses by section:

    (1) ABSTRACT

    The Abstract summarizes qualitative findings nicely, but the authors should incorporate quantitative results for all of the qualitative findings statements.

    The abstract in the revision is extensively revised, and contains quantitative estimates throughout, from both methodologies where appropriate.

    The ending claim is not substantiated by the modeling scenarios that have been run: "targeted interventions for demographic groups such as under-35 men may be the key to finally ending HIV." It is straightforward to run this specific scenario in the model to determine whether or not this is true.

    Our modelling framework is not set up to model the “last mile” of HIV elimination, notably as it has no component for MSM or FSW transmission, and we do not feel that we could confidently present results regarding it. As a result, this statement has been greatly softened in the new abstract (lines 75-78).

    The authors should add confidence intervals to the quantitative metrics, such as the 93.8% and 62.1% incidence reduction.

    These have been added.

    (2) RESULTS

    The authors should check the Results section for any qualitative claims not substantiated by the analyses performed, and ensure the corresponding analyses are presented to support the claims.

    The Results and Methods describe the model's implementation of the PopART intervention differently. The Methods describes it as including VMMC, TB, and STI services, while the Results only mentions intensified HIV testing and linkage.

    This is a slight misreading of the text. That paragraph in the Methods is describing the trial itself, not the modelling framework.

    A limitation of the model is that HIV disease progression is based on the ATHENA cohort in the Netherlands, which is a different HIV subtype (B) than the one in the research setting (C). The model should be configured using subtype C progression data, which have been published, or at least a sensitivity analysis should be conducted with respect to disease progression assumptions.

    The available literature does not suggest a significant difference in progression between subtypes B and C, and we have added text and citations to this effect (lines 699-701).

    In Table 2, the authors should consider adding a p-value to establish whether or not IBM and phylogenetics estimates are different.

    We have done this; the appropriate test was a posterior predictive check. See lines 261-263, 575-579 and 805-814.

    (3) DISCUSSION

    The literature review and comparison of study results to previously published phylogenetic studies is very nice. The authors could strengthen this by providing quantitative estimates with CIs for a more scientific comparison of the study results vs. prior studies, perhaps as a table or figure.

    We have expanded the discussion on this point (lines 504-527). We considered adding a table, but the existing literature that directly answers the questions we ask is quite limited and fragmentary. For example, Monod et al do not present a complete treatment of age gaps. The literature using regression analyses to identify predictors of HIV prevalence or incidence related to partner age is extensive, but those results are not directly comparable to ours.

    The authors state that due to "the narrow geographical catchment area... The results should not be automatically extrapolated to apply to other SSA settings." The authors should exercise this caution when comparing the results to studies in South Africa and elsewhere.

    We have made more explicit acknowledgements of these limitations (lines 598-600).

    There are many other limitations to the analysis, including some mentioned above, that are not acknowledged. The authors should think carefully about what the most important limitations are and acknowledge them honestly at the end of the Discussion section.

    The limitations paragraph has been revised (lines 598-605).

    Reviewer #2 (Public review):

    Summary:

    The authors analyzed PopART data to better characterize the age and sex-specific heterosexual HIV transmission dynamics in Zambia, with the goal of allocating resources.

    Strengths:

    Important analysis to hone in on the key driver of HIV transmission in Zambia, which hopefully can be used to tune prevention efforts to maximize effect while limiting required resources. Two analytic approaches were used, and while the phylogenetic data were markedly more limited, they mirrored the simulated epidemic. The authors did a nice job reviewing the limitations of the data and the analyses. The authors did a nice job of providing analyses to support their goals and hypothesis, and this work may have more impact now that resources in SSA for HIV prevention and treatment may become more scarce

    Weaknesses:

    To increase the impact and utility of this work, it would be helpful to parse the analysis just a bit further to estimate the roles of undiagnosed vs diagnosed and untreated subpopulations on this transmission. PopART is a multifaceted intervention, but the cost, effort, and approach to reengagement in care vs testing/treatment can be quite different.

    We have now provided stratified results by diagnosed and non-diagnosed status of the source, as well as an overall summary of the proportion of undiagnosed sources by age and sex. See lines 305-310, 539-547, and table 3.

    Recommendations for the authors:

    Reviewing Editor:

    We commend you for conducting a rigorous and comprehensive study titled "The age and sex dynamics of heterosexual HIV transmission in Zambia: an HPTN 071 (PopART) phylogenetic and modelling study" that significantly advances the understanding of HIV transmission dynamics in sub-Saharan Africa. The study utilizes an innovative dual-methodology approach integrating individual-based mathematical modelling (IBM) and pathogen phylogenetics to characterize heterosexual HIV transmission patterns by age and sex during the PopART trial in Zambia.

    This manuscript reports on HIV transmission dynamics in Zambia using data from the PopART study, combining individual-based modelling and phylogenetic analysis. The use of two independent methodologies enhances confidence in the consistency of the findings and enables robust cross-validation. The work addresses an important topic in HIV prevention, particularly in settings where resources may become more constrained, and offers insight into potential demographic targets for intervention.

    However, several aspects of the manuscript limit its current impact. The main take-home messages are diffuse and not clearly presented. Some conclusions in the abstract and discussion appear to go beyond the scope of the presented data. For instance, the claim that targeting under-35 men may be key to ending HIV is not directly tested in the modelling scenarios and should be reframed or removed unless supported by new analyses. Furthermore, important quantitative details, such as confidence intervals, p-values, and precise age group estimates, are lacking in key sections (e.g., the Abstract and Results).

    The authors are encouraged to clearly identify and communicate their central findings, ensure all claims are fully supported by their analyses, and make the data more accessible to readers by adding detailed, quantitative summaries where needed.

    The following are our recommendations to the Authors:

    (1) Clarify Study Objectives and Central Messages

    Reframe the abstract and discussion to highlight a clear, well-supported set of main findings.

    Avoid overgeneralized or unsubstantiated claims, especially those not directly tested by your model (e.g., the effectiveness of targeting under-35 men).

    As stated above, we have revised this text accordingly.

    (2) Support Qualitative Claims with Quantitative Data

    Provide numerical results, including effect sizes and confidence intervals, wherever qualitative trends are mentioned.

    For example, restate: "The largest gaps for female recipients were among the youngest" as "... in the age group XX-YY with OR = Z.Z (95% CI: A.A-B. B)."

    As mentioned at the top of the review, we have overhauled the treatment of summary statistics extensively, and now give confidence or highest density intervals throughout the text.

    (3) Improve the Results Section

    Check that all claims are supported by the analyses, and ensure figure references are accurate.

    The statements that went beyond what was supported, notably about ending the epidemic by targeting young men, have been removed. The typo in table references has been fixed.

    Annotate Figure 6 with trendline coefficients and p-values where applicable.

    The takeaway message of figure 6 has now changed and we no longer see no trend, just a minor one.

    Revise Figure 4 for clarity or consider replacing it with a tabular format.

    We would prefer to keep the current figure 4, as we have not found any clearer way to illustrate the patterns, which are the consequence of the phenomenon observed in figure 5. We have put more explicit descriptive text in the discussion, linking the two figures (lines 470-476).

    (4) Address Potential Bias and Model Assumptions More Rigorously

    Explain sampling bias in IBM and phylogenetics (e.g., how the 355 high-confidence phylogenetic pairs were selected).

    The reviewer comment regarding the 355 pairs was based on a misapprehension; we used all the pairs we found using the phyloscanner pipeline. There are no sampling bias issues involved in the IBM as every individual in the simulations is considered. Appendix 2 includes some sensitivity analysis results if the procedure used to find the 355 is changed.

    Discuss how the use of subtype B disease progression data from the ATHENA cohort may impact results in a subtype C setting. A sensitivity analysis would strengthen this.

    Subtype B progression data was used in the absence of any appropriate data from subtype C, but the literature does not suggest any major difference between the two (lines 699-701).

    (5) Include More Detail on Undiagnosed Populations and ART Effects

    Estimate the roles of undiagnosed and untreated subpopulations in driving transmission.

    As mentioned above, this analysis has been added.

    Clarify mechanistically how ART might influence age gaps in transmission dynamics.

    This now is clarified in the introduction (lines 127-129).

    (6) General Improvements

    Provide p-values where comparisons are made (e.g., in Table 2).

    Use consistent terminology and definitions across Methods and Results.

    Add more discussion on limitations, especially regarding generalizability to other SSA settings.

    All of these have been inserted as previously mentioned.

    By addressing these points, the manuscript would present a more coherent narrative and a stronger, evidence-based contribution to the field. We appreciate you all for your fantastic effort and hope you will reflect the feedback in your final paper.

    Reviewer #1 (Recommendations for the authors):

    Thank you for the opportunity to review this interesting manuscript.

    In the public review, I have recommended that the authors should incorporate quantitative results for all of the qualitative findings statements. As one example, I would recommend that "We found the largest gaps for female recipients were among the youngest of those recipients" is re-written as "The largest gaps for female recipients were in the age group XXX-YYY with OR=ZZZ (XXX-YYY)." such as odds ratios, and specific outcome definitions including ages. To give one more example: "immediate increase in the average age at transmission of both sources and recipients" could be rephrased as "increase in the average age at transmission by XXX (YYY-ZZZ) years for sources and XXX (YYY-ZZZ) for recipients over [TIME PERIOD]."

    We hope the revisions we have made to the statistical presentation are satisfactory as a response to this request.

    Again in the public review, I recommended checking the Results section for any qualitative claims not substantiated by the analyses performed, and ensuring the corresponding analyses are presented to support the claims. An example is: "Trends are minor or non-existent in the former two variables." - please annotate Figure 6 (assuming the authors meant to reference Figure 6 and not 7 here?) to show over what period trendlines were fit and provide the coefficient and CI. To support the stated claim even more strongly, a p-value might be apt with a null hypothesis of a slope of zero.

    Please check the numbering on all figure references in the text, as some appear to be misnumbered. E.g., where the text refers to Figure 7, I believe the authors meant to reference Figure 6.

    The change to how we handled the statistics has changed the message of figure 6 (which is now figure 7) and rendered this somewhat moot. We have checked that all figure and table references are now correct.

    Figure 3 is very nice, but if the axes were flipped on one panel, it would make them easier to compare, and then adding some statistics to assess whether the patterns are the same or different when a man vs woman is the source.

    We have flipped the axes here.

    Figure 4 was too complicated for me. I could not follow the Sankey flows because there is too much going on and overlapping. Consider revising to make it easier to digest... perhaps to table format?

    As mentioned above, we would prefer to keep this figure, but we have situated it better in the text.

    Reviewer #2 (Recommendations for the authors):

    A few points that would improve the clarity and the strength of the manuscript

    (1) There is a need to clarify more about how the IBM and phylogenetic data does not suffer from sampling bias. For e.g.,

    Line 205: What proportion of the transmissions modeled in the IBM from Zambia?

    All of them. We confined the analysis of the IBM to the Zambian communities from which phylogenetic data was acquired (lines 755-758).

    Line 217: What proportion of the phylogenetic pairs (cherries) suggesting transmission were the 355 that had high confidence in directionality. How do these pairs compare to the others

    There was no identification of “cherries” involved in picking these pairs; the phyloscanner procedure does not use that step. We confined our analysis solely to the pairs for which we did identify a direction of transmission; that is the 355. Appendix 2 includes a sensitivity analysis involving varying the parameters by which these were identified.

    (2) I appreciate the authors noting that MSM transmissions are unlikely to be playing a role in this cohort, as noted in previous work by the group. However, systematic undersampling of men is common in other study cohorts of HIV. While the MSM and heterosexual networks may be relatively distinct, undersampled men who are bridging the networks could impact the estimates. Can the authors use the time to diagnosis analysis (HIV phyloTSI) to estimate rates of undiagnosed men and women?

    We feel that this is beyond the scope of this work. The phylogenetics dataset in its totality could be used for this purpose (although it is probably highly biased towards undiagnosed individuals due to the considerable majority of samples coming from the healthcare facilities). However, we concentrate here solely on the subset involved in our probable transmission pairs, which is fairly small. Extending the scope to an exploration of the full dataset would seem like a separate study, which we do have plans to do.

    We have used the IBM for this question instead (lines 303-321), however, as MSM transmission was not modelled, it is also not ideal for answering this question. Ultimately we feel that the way these studies were implemented makes it an unsatisfactory tool for answering the MSM question, important as it is.

    (3) Expanding on the point above, in other settings, transmission to young men has been associated with partnerships with older men, and if these young men then transmitted to young women, would we see a similar effect as noted in these models (assuming the young men were less well sampled).

    Our previous work (Hall et al., 2024) suggested no excess of identified male-male pairs in the phylogenetics dataset which might suggest cryptic male-to-male transmission. The age disparities would be worth exploring had this been found, but is curtailed by the lack of it.

    (4) Related to the point above, is there an estimate of the populations (age and sex) that are undiagnosed in the IBM model? Can this be teased out... is transmission from men to women more likely 2/2 lack of diagnosis... or lack of engagement in care?

    We have explored results by diagnostic status as it pertains to age and sex, but we feel that moving on to a more general exploration of the role of diagnosis and lack of engagement in care is again going beyond the scope of what is already a long paper.

    (5) I'm still not fully clear as to why ART might affect age gaps. Can this be explained in more detail?

    See lines 127-129.

  7. Version published to 10.7554/elife.107082.1 on eLife
  8. eLife

    eLife Assessment

    This important study provides evidence for our understanding of HIV transmission dynamics by age and sex in Zambia during the PopART trial; by combining phylogenetic and individual-based mathematical modelling (IBM), it adds depth to the epidemiological literature and may inform more strategic allocation of HIV prevention resources in sub-Saharan Africa. The authors employ two complementary and well-established methodologies (phylogenetics and IBM), and this dual approach is a notable strength. However, the evidence supporting key conclusions is incomplete, with several claims insufficiently substantiated by the data presented. Improvements in data presentation (e.g., quantification of qualitative statements, statistical estimates, and clearer description of results) would substantially strengthen the paper.

  9. eLife

    Reviewer #1 (Public review):

    Summary:

    This manuscript describes the results of phylogenetic and epidemiological modeling of the PopART community cohorts in Zambia. The current manuscript draft is methodologically strong, but needs revision to strengthen the take-home messages. As written, there are many possible take-away conclusions. For example, the agreement between IBM and phylogenetic analysis is noteworthy and provides a methodological focus. The revealed age patterns of transmission could be a focus. The effects of the PopART intervention and the consequences of a 1-year disruption could be a focus. It is important, though, that any main messages summarized by the authors are substantiated by the evidence provided and do not extrapolate beyond the data that have been generated. I recommend that the authors think deeply about what the most important, well-supported messages are and reframe the discussion and abstract accordingly.

    Strengths/weaknesses by section:

    (1) ABSTRACT

    The Abstract summarizes qualitative findings nicely, but the authors should incorporate quantitative results for all of the qualitative findings statements.

    The ending claim is not substantiated by the modeling scenarios that have been run: "targeted interventions for demographic groups such as under-35 men may be the key to finally ending HIV." It is straightforward to run this specific scenario in the model to determine whether or not this is true.

    The authors should add confidence intervals to the quantitative metrics, such as the 93.8% and 62.1% incidence reduction.

    (2) RESULTS

    The authors should check the Results section for any qualitative claims not substantiated by the analyses performed, and ensure the corresponding analyses are presented to support the claims.

    The Results and Methods describe the model's implementation of the PopART intervention differently. The Methods describes it as including VMMC, TB, and STI services, while the Results only mentions intensified HIV testing and linkage.

    A limitation of the model is that HIV disease progression is based on the ATHENA cohort in the Netherlands, which is a different HIV subtype (B) than the one in the research setting (C). The model should be configured using subtype C progression data, which have been published, or at least a sensitivity analysis should be conducted with respect to disease progression assumptions.

    In Table 2, the authors should consider adding a p-value to establish whether or not IBM and phylogenetics estimates are different.

    (3) DISCUSSION

    The literature review and comparison of study results to previously published phylogenetic studies is very nice. The authors could strengthen this by providing quantitative estimates with CIs for a more scientific comparison of the study results vs. prior studies, perhaps as a table or figure.

    The authors state that due to "the narrow geographical catchment area... The results should not be automatically extrapolated to apply to other SSA settings." The authors should exercise this caution when comparing the results to studies in South Africa and elsewhere.

    There are many other limitations to the analysis, including some mentioned above, that are not acknowledged. The authors should think carefully about what the most important limitations are and acknowledge them honestly at the end of the Discussion section.

  10. eLife

    Reviewer #2 (Public review):

    Summary:

    The authors analyzed PopART data to better characterize the age and sex-specific heterosexual HIV transmission dynamics in Zambia, with the goal of allocating resources.

    Strengths:

    Important analysis to hone in on the key driver of HIV transmission in Zambia, which hopefully can be used to tune prevention efforts to maximize effect while limiting required resources. Two analytic approaches were used, and while the phylogenetic data were markedly more limited, they mirrored the simulated epidemic. The authors did a nice job reviewing the limitations of the data and the analyses. The authors did a nice job of providing analyses to support their goals and hypothesis, and this work may have more impact now that resources in SSA for HIV prevention and treatment may become more scarce

    Weaknesses:

    To increase the impact and utility of this work, it would be helpful to parse the analysis just a bit further to estimate the roles of undiagnosed vs diagnosed and untreated subpopulations on this transmission. PopART is a multifaceted intervention, but the cost, effort, and approach to reengagement in care vs testing/treatment can be quite different.

  11. eLife

    Author response:

    We thank the editors and reviewers for their positive and constructive comments. The three most substantial points raised by the public review are the following:

    No explicit modelling of targeting of young men as a course to ending HIV.

    We did not intend to imply that the epidemic could be ended by this alone, or even that targeting young men was the optimum strategy if resources were available for more general preventative interventions. The “last mile” for HIV will be a very complex scenario in which key populations will start to play an outsize role, and our modelling framework was not developed to consider it. As a result, we would not have confidence in modelling the decline of the viral population to zero. We shall be qualifying the existing language in the paper in order to make this clear.

    Subtype-specific disease progression data.

    The criticism is that our modelling of disease progression was based on subtype B, while the HIV viral population in Zambia is overwhelmingly subtype C. Sensitivity to subtype has not been looked at in detail in this analysis as the literature suggests that the rate of CD4 decline does not differ between subtypes B and C.

    While some studies have shown differences in CD4 cell decline between subtypes, they have generally highlighted that subtype D progresses faster than other subtypes. Little evidence has been published on the differences between subtype B and C, and studies that do include both subtypes concluded that there was no significant difference in rates of CD4 decline between subtypes.

    No significant difference between rate of CD4 progression by subtype is evidenced in the following publications:
    - Klein et al. (2014) (N=9772)
    - Bouman et al. (2023) (although no subtype B)
    - Easterbrook et al. (2010) (N=861)

    While some studies have illustrated that "progression changes with HIV subtype", an interrogation of the underlying data highlights that subtype B is not included, e.g.
    - Kanki et al. (1999) looked at A versus "non-A subtype" but included no subtype B data.
    - Vasan et al. (2006) claims differences in rate of CD4 decline by subtype when compared to subtype D but includes no subtype B data.
    - Baeten et al. (2007) claims subtype D has faster progression that subtype A but includes no subtype B data.
    - Kiwanuka et al. (2008) claims differences in rate of CD4 decline but includes no subtype B data.
    - Amornkul et al. (2013) has no subtype B data.

    Furthermore, to explain why we used subtype B data to parameterise the model: usually, statistical analyses of CD4 count progression do not report parameters in a form that can be directly imported into models. Analysing summary statistics to include in models results in under-specified models of disease progression in simulations. For this reason we use the estimates from Cori et al. (2015); where the statistical analysis was specifically tailored to generate modelling parameters. The trade-off is therefore to use subtype C data with model misspecification, or subtype B data without; neither choice is perfect, and we chose the subtype B correctly specified estimates.

    The role of undiagnosed versus diagnosed and untreated subpopulations.

    We will add an additional analysis us to compare age differences in sources and recipients according to the diagnostic status of the source.

    The rest of the comments in the public review ask for improvements in data presentation (including some additional statistical analyses) and to make sure qualitative claims are fully justified. We are happy to oblige with these, and will make our thinking clear on all points in the full response.

  12. Version published to 10.1101/2025.05.02.25326873 on medRxiv