RR:C19 Evidence Scale rating by reviewer:
Reliable. The main study claims are generally justified by its methods and data. The results and conclusions are likely to be similar to the hypothetical ideal study. There are some minor caveats or limitations, but they would/do not change the major claims of the study. The study provides sufficient strength of evidence on its own that its main claims should be considered actionable, with some room for future revision.
Overall the results and conclusions appear plausible, and substantiated by a solid model data-driven calibration approach – some of them (such as that more service interruption will increase incidence, whereas more sexual distancing decreases incidence) of course trivial. For HIV incidence, the results are in line with simulations with several independent models (https://www.thelancet.com/journals/lanhiv/article/PIIS2352-3018(20)30211-3/fulltext ).
We have 3 major comments concerning the overall findings:
1. The model parametrization is an elegant, indeed innovative approach, of translating comprehensive detailed partnership and network data from USA MSM into statistical estimates that then feed as priors into a Bayesian model calibration. What is missing to confirm the adequacy, is demonstration of the resulting model fit for ‘basic’ overall HIV and STI epidemiology. Please show, in some additional graphs (or tables), the model fit for:
· HIV prevalence, by age, (current and/or lifetime) partnership number, and by calendar year over the past say 10 years – similar to the calibration by ethnicity shown in Suppl. Figure 8, and for diagnosed fraction of HIV-infected MSM in Suppl. Figure 4, and other clinic-epi indicators in Suppl. Figures 6 and 7.
· NG and CT prevalence and incidence, also by age, by the same stratification – if there is such data for the Atlanta area or a similar or overall USA MSM population?
· HIV and/or STI incidence rates, if any reliable representative data (from cohort studies, or case records, if believed reasonably complete ) are available?
For example, Supplement section 12.1 states there were ‘target values’ for STI incidence, but I have not spotted the sources for those nor seen the values and the model fit. I recommend that the epidemic fit be shown in several graphs, in turn for HIV and one or more STIs, each showing data alongside model prediction, over time, by age, by partnership number etc.
2. One curious result is the contrast in impact on HIV versus on STI, of the two COVID-related tendencies, in the default scenario with 18 -months sexual distancing and service interruption. Why, if as long-term impact STI incidence keeps falling, this is not translated into a corresponding ongoing fall in HIV incidence? Would that not be expected, given the cofactor effect of STI on HIV incidence? Also in (https://www.thelancet.com/journals/lanhiv/article/PIIS2352-3018(20)30211-3/fulltext the net effect of service interruptions alongside sexual distancing, across the models in that study, was toward decreasing HIV incidence. I welcome more explanation on this, beyond the general and rather vague ‘higher transmissibility of … and potential for reinfection with STIs’, stated in the main manuscript on page 8.
3. The model calibration accounts for assortative mixing by age and ethnicity, reproducing ethnicity-specific HIV prevalence reassuringly (Supplemental Figure 8). In reality, there is likely additional assortativeness within age/ethnicity groups, by ego degree: for example, visitors of the same website or club may all have preference for one-off contacts, or for longer-term relations -- whereas here the model (implicitly) assumes proportionate mixing. This within-subgroup mixing is hard to measure and probably there were no calibration data available, but would it be possible to show, e.g. in a sensitivity analysis (or alternatively, at the least discuss), if and how additional assortativeness by ego degree (given ethnicity and age) would impact the results and conclusions?
INTRODUCTION: ‘EpiModel was developed by the authors for simulating…’:
I suggest to remove ‘by the authors’ to avoid confusion with the authors of the current study.
RESULTS: Figure 3, it seems odd that Sexual distancing alone, without service disruption, the green line, would not reduce HIV incidence the least at all (whereas it does, of course, for STI)? Please explain this.
Once more, commending the authors’ great attention to empirical detail. They also recommend further empirical studies. It would be useful to also describe what type of studies would be most called for. That might be a stimulus and guide for such empirical studies, for example as student projects.
3.1: Cross-type degree: The conceptual description is clear, but it remains unclear how to read the corresponding values in Supp. Table 1: Is the left or the right ‘Mean degree … for main, given current casual’ and left or right the vice versa?
3.1.9: ‘a fixed sexual role preference (exclusively insertive, exclusively receptive, versatile)’: I found this confusing at initial reading – and later see my supposition confirmed, that most men (54.4%) fall into the ‘versatile’ category, i.e. NOT have a ‘fixed preference’. Suggest to rephrase.
Page 43: .from empirical data (Table S1) from NHBS’: Table S1 does not give gonorrhea or chlamydia screening data; please rephrase and update.
In regression tables, please make sure to always specify the dimension of the covariates. For example, instead of ‘Duration’ please write ‘Duration (years)’.
GRAPHICS (main and Supplement):
Figure 2: Could you add the corresponding result for a Counterfactual of No sexual distancing & No service interruption?
X-axis units: Years instead of Weeks would be more readable (with the same adjustment in Results text).