Independent predictors of comprehensive knowledge of HIV in general population: findings from the Myanmar Demographic and Health Survey (2015-16)

REVIEWER 1
Adam K. Richards, Community Partners International, Berkeley, CA, USA 
REVIEWER
This brief report uses data from the 2015-2016 Myanmar DHS to explore factors related to “comprehensive HIV knowledge” in the general Myanmar population. The authors should make it clear that the crude estimates presented in Table 1 combine the sex-specific tables from the full Myanmar DHS report (table1 13.3.1 for women and 13.3.2 for men); the modest novelty of the manuscript as currently written lies in the results modelled in a multivariate analysis of these independent variables.

My sense is that the manuscript could be strengthened by making several modest changes as outlined below.


AUTHORS
Thank you for your time and kind effort providing comments. We agree with you that novelty of the manuscript lies in the multivariable analysis. As the result of reviewer input, we believe the manuscript has significantly improved. We have added the footnote in the Table 1 “The crude estimates presented here combine the sex-specific tables from the full Myanmar DHS report (table1 13.3.1 for women and 13.3.2 for men)”.

REVIEWER
Introduction
First paragraph: it is not at all clear that “the right knowledge and a positive attitude towards HIV… is a pre-requisite… [to meet the 90-90-90 goals”… at least when the definition of HIV knowledge is the one used by the DHS/in this manuscript. It is not unreasonable to propose that at least some of the knowledge domains (specifically, the ‘misperceptions’) may be less relevant when it comes to actual risk behaviors, care-seeking, adherence, stigma etc etc. [see below for additional discussion]Outcome
The authors should consider presenting results using knowledge of “HIV prevention measures” as an alternative, or perhaps secondary outcome.


AUTHORS
Thank you for your comment. We ran the model with the same method for “knowledge on HIV prevention measures”. We found that compared to those educated high school and above, those educated less than high school were 34% less likely [aPR: 0.66 (95% CI: 0.63, 0.69) : those with no formal education were 63% less likely [aPR: 0.37 (95% CI: 0.33, 0.41)] to have comprehensive knowledge. Other than education, we found no significant differences between categories across variables which is similar to the comprehensive knowledge result. Moreover, the percentage for the specific questions regarding HIV preventive measures and misperceptions were also already presented in the MDHS report. Therefore, we would like to keep current definition of outcome and we would like to compare it with the results of other studies conducted in different countries.

REVIEWER
I appreciate that the definition of ‘comprehensive HIV knowledge’ that is applied in other DHS surveys (and is comprised of four elements: one partner/condoms prevent HIV; PLWHA can be healthy; and 2 misconceptions NOT endorsed) though it is worth noting that most DHS surveys also present results for a subset comprosed of the first two of the four items in the comprehensive knowledge, often referred to as ‘HIV prevention measures.’ It would be useful to use this alternative definition, given the somewhat tenuous link between the misconceptions and HIV outcomes of commonly accepted value. Numerous studies demonstrate how individuals are capable of holding multiple beliefs about the etiology of disease while maintaining overall excellent adherence to risk behaviors, testing and treatment. For example, it is possible for TB patients in Bolivia to believe that traditional healers play a valuable role in the treatment of tuberculosis, but this belief does not impede their ability to seek out testing for TB and to strictly adhere to DOT (as long as structural barriers are addressed to enable access to care, payment for testing and treatment etc) – see An Ethnography of Nonadherence: Culture, Poverty, and Tuberculosis in Urban Bolivia. Jeremy A. Greene. Culture, Medicine and Psychiatry volume 28, pages 401–425(2004)¹. As the authors likely are aware, it is not uncommon in Myanmar for rural villagers to believe that malaria can be transmitted by eating papaya (or other fruits)… but this does not interfere with their excellent care seeking behavior and adherence to treatment – as evidenced by the impressive results of the large village malaria worker initiatives that have dramatically reduced transmission in endemic areas of Myanmar.
Stated more directly, there is a very real chance that it does not matter much whether or not someone believes HIV might be related to mosquitoes or food… as long as they also know that HIV is associated with real risk behaviors (unprotected sex; sharing needles etc), and that their practices align with lower risk behaviors.

I am not aware of studies demonstrating that misconceptions about HIV (specifically, mosquitoes and sharing food) meaningfully influence HIV risk behaviors, uptake of testing or treatment, HIV-related stigma or other meaningful outcome. If the authors are aware of such studies the might cite them as part of their choice to use the DHS definition of “comprehensive HIV knowledge” over alternative options that carry at least as much meaning – such as knowledge of HIV prevention messages.

I appreciate that knowledge related to PMTCT of HIV is not the specific emphasis of this brief report. Nonetheless, it is arguably at least as important an indicator of HIV related knowledge in the general population -- as the authors themselves imply in the introduction. (that is, HIV knowledge about the importance of limiting sex partners, condom use etc. is MUCH more valuable among key pops and perhaps migrants than it is among the general population; whereas every woman (and their male partners) who might get pregnant should be aware of HIV testing and treatment during pregnancy..

AUTHORS
Thank you for the detailed comment. We realized the value of having HIV prevention measures as a secondary outcome. However, since our study’s focus was on comprehensive knowledge of HIV and short reports does not allow us to expand the manuscript (maximum 2 tables/figures and 2500 words); we would like to stick to our current outcome. We also found that misconceptions of HIV transmission have been associated with HIV testing in some DHS paper (mentioned below). In this connection, in order to meet the first 90 of UNAIDS 90-90-90 goal, we decided to keep comprehensive knowledge (preventive knowledge + misconceptions) as our outcome of interest.
https://www.dovepress.com/what-are-the-determinants-of-misconception-about-hiv-transmission-amon-peer-reviewed-fulltext-article-HIV
https://bmcinthealthhumrights.biomedcentral.com/articles/10.1186/s12914-016-0089-8#Tab3

REVIEWER
Exposures
The DHS includes questions related to media exposure (print, radio and television; and these answers are often combined into a composite indicator), and it would be valuable to include this in the model, as it might suggest possible means by which the low knowledge of HIV in Myanmar might be ameliorated. The authors should either include one or more of these variables, or explain their choice to omit them.


AUTHORS
Thank you for pointing out this. We included exposure to at least one of the mass media (television, newspaper, radio) at least once a week in the model and added the values in the table 1.

REVIEWER
Did the authors consider estimating knowledge among important subpopulations who can be classified using DHS data? Unfortunately, information on most high risk HIV behaviors is not available from the DHS, though it may be useful to document HIV knowledge among respondents who are: recently sexually active; reported an STI; engaged in payment for sex; or who lived in particular states/regions. For example, given the extremely high prevalence of HIV among PWID in Kachin and Shan (from the recent IBBS), as well as the relatively common practice of injection drug use in those Northern regions, it is not inconceivable that HIV will become a generalized epidemic in that area; and partners of PWID at are at especially high risk.

AUTHORS
Thank you for this interesting comment. Because short reports does not allow us to expand the manuscript (maximum 2 tables/figures and 2500 words), we could not cover all the aspects and we decided to estimate knowledge among general population in this paper. We have published a paper regarding HIV testing among STIs which knowledge variable was an independent variable. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7310811/

REVIEWER
The lack of information on other HIV risk factors / key populations in the DHS (eg condom use; illicit injection history; same-sex-practices; migration history etc) is a limitation worth noting in the manuscript.


AUTHORS
Thank you. We have stated “The study population might include some key affected population that could influence the true prevalence among general population” as a limitation in our manuscript. LINE 201 of revised manuscript with track changes

REVIEWER
The model among adolescents likely should take advantage of the DHS variables specific to this group (ever attended school; attended school in the past year etc).

AUTHORS
Thank you. We agree with your point. We have also analyzed separate models for adolescents and non-adolescents and found similar results as combined one except regional variation prominent among adolescents. In the adolescent model, compared to delta and lowland region, those residing in hilly region (aOR=0.51, 95%CI: 0.34,0.78) and plain region (aOR=0.67, 95%CI: 0.51, 0.88) had significant lower comprehensive knowledge. Considering this is a short report with limited tables and the need for a clear and simple message, we decided against presenting another model for adolescents in this manuscript. We hope this is fine.  

REVIEWER
Statistical analysis
I appreciate and agree with the choice to present prevalence rate ratios (and not odds ratios). In the setting of an outcome of modest prevalence, most would choose to use glm with a binomial distribution and a log link function, and not a robust poisson regression. See for example
Martin R Petersen and James A Deddens. BMC Med Res Methodol. A comparison of two methods for estimating prevalence ratios. 2008; 8: 9². 
The results are unlikely to be qualitatively different using either method, though the authors might explain their choice to the reader (or consider using glm with a binomial distribution)

AUTHORS
Thank you. We first tried a log binomial model. But because of lack of convergence, we proceeded with the modified Poisson regression to estimate the prevalence ratios. We have also mentioned in the manuscript. LINE 141 of manuscript with track changes

REVIEWER
Modelling choices
Stratification / separate models for adolescents
The authors appropriately note that HIV knowledge is especially low among adolescents, though it is less clear why this might be the case. A reasonable a priori hypothesis is that the factors associated with HIV knowledge among adolescents are quite different than risk factors among young/middle age adults. [Some authors would even submit a separate manuscript J]. The authors should explain why they chose to estimate a single model for both older adults as well as adolescents. If the authors first conducted stratified analyses and observed similar associations; or if they formally explored the possible presence of interactions between adolescent and older ages, then they should say so. If these were not done, then I strongly encourage the authors to explore possible effect modification among adolescents, and to comment on whether results appear similar

AUTHORS
Thank you for your comment. We tried the potential interaction across variables (including age category and other variables), included in the model and tested for model selection using AIC/BIC results. However, the models with interaction(s) was not significantly improved compared to the simple model. Therefore, we decided to present the simple model in our manuscript. We have added the information in the manuscript. LINE 148 of revised manuscript with track changes

REVIEWER
Interactions
Did the authors consider exploring effect modification by other variables in addition to adolescent vs adults? For example, are SES gradients of similar magnitude in urban and non-urban settings? A brief comment on whether interactions were considered for other/any variables would be useful.

AUTHORS
Thank you for your comment. We tried the potential interaction across variables (including age category and other variables), included in the model and tested for model selection using AIC/BIC results. However, the models with interaction(s) was not significantly improved compared to the simple model. Therefore, we decided to present the simple first model in our manuscript. We have added the information in the analysis. LINE 148 of revised manuscript with track changes

REVIEWER
Supplementary analyses
As above, as currently presented the unique novel contribution made by this manuscript is a multivariable/adjusted model. Most associations remain relatively similar in the adjusted model, and the simple analysis sets a somewhat low bar for publication in a peer reviewed manuscript, even as a Brief Report. It would be valuable to augment the manuscript with additional analysis that would require fairly little effort. Here are two straightforward suggestions:

Predicted probabilities
For example, the authors could present absolute predicted probabilities of comprehensive knowledge (or knowledge of HIV prevention practices) for different individuals who do and do not possess factors associated with HIV knowledge: for example, the authors might report the probabilities of HIV knowledge for residents of Myanmar who are:

1. highest wealth quintile; high school or higher education; urban; professional
    
2. poorest wealth quintile; no education; nonurban; agricultural worker

The authors might calculate predicted probabilities for two groups of adolescents; as well as two groups of adults (4 calculations total. Margins facilitates calculation of the difference between predicted probabilities, the standard error of that difference, confidence intervals etc though it may not be necessary to report that, due to the very large magnitude of the difference apparent on presentation of the point estimates.

What is the potential value of calculated predicted probabilities? First, it provides an intuitive presentation of adjusted estimates on an absolute scale, which complements the information presented on a relative scale (aPRR). Second, it helps to place the findings in context and more directly address several key questions the current manuscript touches on only indirectly. One such question is whether comprehensive knowledge is likely or possible, even when the most favorable conditions exist (ie could any combination of factors produce a predicted probability close to the UNGASS goal of 95% among adolescents?). In addition, presentation of predicted probabilities would create an even more stark contrast for disadvantages groups who comprise a large proportion of the population. This, in turn, might further highlight the important conclusion that efforts to improve HIV knowledge per se (via media, education etc) are unlikely to address the low overall HIV knowledge; and that improving HIV knowledge will require major development and social change. [see also the comment below related to school-based HIV education efforts –calculating predicted probabilities among adolescents who are advantaged in every way, including attendance at school, might provide a rough sense of the magnitude of the impact necessary to achieve targets set by UNGASS].

Predicted probabilities are easily accomplished in STATA using the ‘margins’ postestimation command (immediately after running the full model regression).


AUTHORS
Thank you for this insightful comment. We have included the predicted probabilities across categories in Table 2.


REVIEWER
Summarize SES inequities using concentration index or similar health equity metric
The SES gradients (wealth and educational attainment) are very large, independent of eachother, and not explained by observed variables in the model. Methods exist to summarize inequities in health outcomes using tools developed by others, such as the concentration index developed by the World Bank. User-written command concindex is available in STATA that facilitates calculation of the concentration index in the setting of complex survey such as the DHS (using the svy: prefix). I suggest that the authors report two relative CIs for HIV knowledge, using household wealth and educational attainment as the two respective ranking variables.
 
AUTHORS
Thank you for this interesting comment. We ran the analysis for concentration index/concentration curve and found that “The HIV comprehensive knowledge was concentrated among richer people, more educated people and who hold better job title”. However, short reports does not allow us to expand the manuscript (maximum 2 tables/figures and 2500 words); we would like to stick as it is. We hope this is fine.

REVIEWER
Discussion
The major finding was already known: that “comprehensive knowledge of HIV,” as defined by the DHS is low in Myanmar; and that the distribution of knowledge demonstrates stark patterning according to axes of power and advantage, such as wealth, education, occupation, urban residence etc. It likely is worth stating that adjusting for the factors included in the multivariable model had little influence on what was documented already (in the DHS report tables).

AUTHORS
We agree. We believe that reporting where the poor knowledge of HIV was concentrated can be beneficial for a resource constraint country like Myanmar. We hope this is fine.
 
REVIEWER
 If the authors believe that the more strict definition of HIV knowledge (that requires respondents to correctly identify mosquitoes and food as NOT influencing risk of HIV) is in fact important, that it would be fair to say that an overall prevalence of 20.4% is extremely (abysmally?) low. If, however, they believe that in fact it is not crucial that the general population possesses the HIV knowledge specific to these questions, then it may be reasonable to describe this percentage in less stark terms: perhaps simply ‘low’ would suffice; I would avoid saying ‘relatively’ low, as this begs the question ‘relative to what’? the ideal? Other countries?

 AUTHORS
Thank you. We have removed “relatively” in the manuscript. . LINE 206 of revised manuscript with track changes

REVIEWER
Absolute prevalence of comprehensive knowledge should be reported for each country compared to Myanmar (eg Nigeria, DRC and Uganda).

AUTHORS
Thank you. We revised it. LINE 209 of manuscript with track changes

REVIEWER
The adolescent figures can be placed in the context of the UNGASS goal to achieve comprehensive knowledge among 95% of adolescents. 

AUTHORS
Thank you. We revised it. LINE 207 of manuscript with track changes

REVIEWER
The SES gradients (wealth and educational attainment) are striking, independent, and are not explained by observed variables included in this model. This likely deserves greater emphasis.
Why single out “the poorest two quintiles” for having “poor comprehensive knowledge”… when only 41% possessed comprehensive knowledge even the highest wealth quintile? [hint: predictive margins may be useful here]

AUTHORS

Thank you for this comment. We have added predicted probabilities as Table 2.

REVIEWER
As above, it would be valuable to report a secondary outcome of the two HIV prevention methods, which results in a slightly higher prevalence 54% among women and 62% among men (p205 in the DHS). It would be valuable to know whether or not the associations appear similar when this more common outcome is used in place of the stricter “comprehensive knowledge” outcome is used (the PRRs will likely be somewhat smaller in magnitude if the denominator/baseline value is larger, though this can be taken into account in your interpretation).
The manuscript might contrast HIV Knoweldge in the general population with that reported among key populations (ie from the respective IBBS among PWID and FSW).

AUTHORS
Thank you. We have included comprehensive knowledge among key population in Myanmar according to IBBS survey in the discussion. LINE 202 of manuscript with track changes

REVIEWER
Given the concentration of HIV risk among key populations – that in Myanmar include migrants, as noted in the more recent strategic plans of MOHS/NAP – the authors could more explicitly state the value of reporting HIV knowledge among the general public. One way to do this is to highlight the emphasis of NAP on PMTCT, and make explicit that the women in the present analysis are women of reproductive age.

AUTHORS
Thank you for this valuable comment. We included “Furthermore, low proportion of comprehensive knowledge of HIV/AIDS among reproductive aged women in general population is alarming for effective prevention of mother-to-child transmission of HIV.” LINE 214 of manuscript with track changes

REVIEWER
If a large number of children and adolescents do not attend school (attendance ratios are 83% for primary and only 60% among adolescents) then it is not clear how school-based HIV education would address the gap in HIV knowledge among this vulnerable group.

AUTHORS
Thank you. We revised the sentence as “School health education and peer intervention programsshould be strengthened to improve comprehensive knowledge among late adolescent.” LINE 225 of manuscript with track changes

REVIEWER
Addressing the massive SES gradients in HIV knowledge is an admirable goal, though it’s not at all clear what a “targeted” approach might look like, or whether it makes much sense given the low knowledge overall, and in every subgroup presented – though calculating the predictive margins might help to highlight whether it would be reasonable to omit certain subgroups from a ‘targeted’ campaign.
 
AUTHORS
Thank you for this interesting comment. We have included predicted probabilities as Table 2.


REVIEWER

Minor points
Introduction
The findings from Uganda belong in the discussion (along with DRC and Nigeria).

AUTHORS
Thank you. We have revised it in the manuscript. LINE 209 of manuscript with track changes

REVIEWER
Methods: the paragraph that begins “Comprehensive knowledge was considered…” should read “present” [drop as ‘yes’]

AUTHORS
Thank you. We have revised it in the manuscript. LINE 127 of manuscript with track changes

REVIEWER
Results: second paragraph -- repeating “17,622 respondents” is redundant
 
AUTHORS
Thank you. We have revised it.

REVIEWER
Figure 1 – I am not sure what is added by presenting the sampled clusters on a map, in the particular manuscript. Consider mapping the prevalence of the primary outcome of HIV knowledge. Since the crude prevalence of comprehensive knowledge by state/region is already provided in the DHS report (albeit separately for men and women, and in tabulated format); then the authors might present predicted probabilities ‘adjusted’ (standardized) for age, sex and nonurban residence…

AUTHORS
 Thank you for your comment. We excluded the map figure and included the predicted probabilities by SES after adjusted model is presented in Table 2.