Mapping the Burden of Hypertension in South Africa: A Comparative Analysis of the National 2012 SANHANES and the 2016 Demographic and Health Survey
, Ronel Sewpaul 5, Adlai Davids 5,6 and Sasiragha Priscilla Reddy 5,6Round 1
Reviewer 1 Report
I accept your explanation.
Author Response
We would like to thank you for the immense efforts, commitments and very constructive comments from the reviewers and editor aimed at improving the quality of the manuscript.Reviewer 2 Report
I've noticed no obvious changes compared to the previous version.
Author Response
Thank you. We appreciate your initial comments which have immensely helped to improve the manuscript. Indeed, we have made a couple of revisions since the last review. For example, we have further clarified the overarching aim of the study. The primary aim of this study is to investigate the geographic variation, at a province level, in the prevalence of hypertension in South Africa in both 2012 and 2016. Thus, the primary exposure variable of interest is province, and the covariates/explanatory variables included in the model are behavioural, socio-economic and demographic variables, which included race and gender. In reporting on its primary objectives, the results report significant differences at provincial level in 2016. Regarding the covariates; significant differences in hypertension were found by gender, race (Coloured vs Black African), urban/rural residence, BMI, waist circumference, blood cholesterol and age. Similar findings regarding racial differences have been shown in Peltzer & Phaswana-Mafuya (2013) and Berry et al (2017) (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3721893/, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5656122/). While South Africa is still racially and economically segregated, the country had experienced rapid urbanisation and income growth resulting in lifestyle, stress and dietary changes among all South Africans. The hypertension prevalence is now very high among all race groups, between 28-45% between race groups (see for example Berry et al (2017)). The regression models adjust for race, gender, wealth index and a variety of covariates, as well as for spatial autocorrelation of the data. In this way, they adjust for the confounding effects of racial and economic factors by province. The results reported therefore show the associations of all independent variables with hypertension, controlling for the behavioural, socio-economic and demographic variables. We have included some of this text in the manuscript to improve alignment of the objectives and results. Besides, by simultaneously controlling for both spatially correlated and uncorrelated random effects along with the key covariates (which is the main strength of the approach), we have been able to separate noise from ‘true’ effects and gained higher precision in parameter estimation.
Reviewer 3 Report
Generally, the main elements to include in the introductions are research significance (the reason why the authors did this research). In the introduction section, I suggest including information regarding the importance of your research and what is the difference between your research and the previous research.
On the title, you used “A comparative analysis”. Normally, comparative analysis aims to see the difference in the average of the independent variable between two or more groups. I did not find any explanation regarding how did you compare them and what kind of goodness of fit did you used.
Lines 137-161, Statistical Analysis: I would suggest completing the model in Equation (1). Did you use the fully Bayesian approach? In the Bayesian approach, the variance parameter sigma is considered as random variables and has to be supplemented with appropriate prior assumptions. What kind of prior, hyperprior did you use? As the statistical analysis used the Bayesian method, using a sensitivity analysis is crucial to examine the influence of the priors on the estimation of the posterior distribution. How did you define spatial random effect? Why you did not account for the unstructured spatial effect on your model in Equation (1)? This is not in line with what you stated in lines 294-295 that “we fit several models with and without the structured and random components”.
Lines 167-172: Why the paragraph is italicized?
Line 191: “… (Fig 3 and 4). It should be “…(Figs 3 and 4).
Line 197, Figure 3: “Blue to red correspond to low risk to high-risk provinces”. There is no red on the map. How did you explain “white” on the map? I would suggest revising these all maps (a-d). Please also revise all maps in Figure 4!
Lines 212-222: You only interpreted some variables. I would suggest including more explanations of other variables such as region of residence, heart attack, smoking status, and drinking status.
Table 1, under “Waist circumference (tertile), you define 1(lowest), 2, 3(highest). Please define 2!
Table 2: You included DIC on the table, but there is no further explanation. Please add some explanations!
Table 2: Why you did not include “ Wealth Index” in Table 2 as you did on Table 1”?
Author Response
We would like to thank you for the immense efforts, commitments and very constructive comments from the reviewers aimed at improving the quality of the manuscript.
In response to the question/comments, we are pleased to say that we have effected all necessary changes and corrections, and these could easily be traced in one of the attached documents named ‘Revised Manuscript’.
Below is a point-by-point basis response to each of the questions/comments/points in the ‘Response to reviewers comments’ as detailed below.
Comments and Suggestions for Authors
Generally, the main elements to include in the introductions are research significance (the reason why the authors did this research). In the introduction section, I suggest including information regarding the importance of your research and what is the difference between your research and the previous research.
Response: Thank you: We have further clarified in the relevant sections of the manuscript the importance and novelty of the manuscript. For example, we have added a paragraph in both the introduction to say that “The primary aim of this study is to investigate the geographic variation, at a province level, in the prevalence of hypertension in South Africa in both 2012 and 2016. The novelty in the study is its application of a statistical approach that controls for the latent effects of geographical location, allows for spatially autocorrelated responses that arise in cluster stratified survey designs, and simultaneously accounts for linear and non-linear covariates (behavioural, socio-economic and demographic) and the data hierarchy. The Bayesian geo-additive modelling approach is novel and favourable over traditionally applied frequentist approaches. This is, to our knowledge, the first application of these methods to recent South African data on blood pressure. The manuscript presents the provincial variance in hypertension after adjustment for the above variations. The results also found that, after adjustment for all these statistical variations and covariates, the known risk factors for hypertension such as BMI, waist circumference, and blood cholesterol were associated with hypertension. Therefore the findings are similar to other studies, but the methods used are more robust, allowing for a straightforward quantification of uncertainty in parameter estimation with potentially higher precision.”
On the title, you used “A comparative analysis”. Normally, comparative analysis aims to see the difference in the average of the independent variable between two or more groups. I did not find any explanation regarding how did you compare them and what kind of goodness of fit did you used.
Response: Thank you. It is correct that the purpose of comparative studies is to see how the response varies over different contexts. Here, we have compared how the prevalence of hypertension varied geographically in South Africa while comparing evidence from two separate household surveys conducted in 2012 (2012 SANHANES) and 2016 (2016 SDHS). In addition, the advanced statistical approach utilised here controlled for other key socio-economic covariates and health indicators (e.g., BMI), thus, we were able to see how the effects of each of these covariates on the prevalence of hypertension in South Africa has changed over time (4-year period) across the datasets. This would serve a means of evaluating the potential impacts of concerted intervention efforts and several years of investments aimed at improving quality of life in South Africa.
Lines 137-161, Statistical Analysis: I would suggest completing the model in Equation (1). Did you use the fully Bayesian approach?
Response: Thank you. Yes, we used a fully Bayesian approach as samples of the parameter values were drawn from their joint posterior distribution using Markov chain Monte Carlo (MCMC) techniques, in particular, the iteratively weighted least squares (IWLS) updating scheme.
In the Bayesian approach, the variance parameter sigma is considered as random variables and has to be supplemented with appropriate prior assumptions. What kind of prior, hyperprior did you use?
Response: Thank you. Yes, the key advantage of Bayesian statistics is the ability to exploit expert knowledge (priors). For this study, we used inverse Gamma priors for the variance parameters , that is, , where and are shape and scale parameters, respectively.
As the statistical analysis used the Bayesian method, using a sensitivity analysis is crucial to examine the influence of the priors on the estimation of the posterior distribution. How did you define spatial random effect?
Response: Thank you. Although the default choices for the shape and scale hyperparameter values are , we varied these choices through sensitivity analysis to see how the various choices agreed with the model assumptions. However, in the end, the default choices were retained as they did not show any worse fit than the other choices. The spatial random effect was defined in terms of the total random effect which comprises the spatially structured (correlated) random effect and the spatially unstructured (uncorrelated) rand om effect.
Why you did not account for the unstructured spatial effect on your model in Equation (1)? This is not in line with what you stated in lines 294-295 that “we fit several models with and without the structured and random components”.
Response: Thank you. Following from the preceding response, both structured and unstructured spatial random effects were modelled separately to accurately capture both spatially dependent (due to shared characteristics among neighbouring spatial units) and spatially independent (due to spatial heterogeneity) effects. Specifically, the structured spatial random effect was modelled using Markov Random Fields (MRF) so that neighbouring locations with fewer observations can ‘borrow strength’ from neighbours with larger samples for parameter estimation. While, the unstructured spatial component was modelled as iid zero-mean Gaussian, that is, . Note that the output maps are based on the total spatial effect . We have further updated the relevant sections of the manuscript to make this clearer.
Lines 167-172: Why the paragraph is italicized?
Response: Thank you. That was typo and is now corrected.
Line 191: “… (Fig 3 and 4). It should be “…(Figs 3 and 4).
Response: Thank you. The missing ‘s’ added.
Line 197, Figure 3: “Blue to red correspond to low risk to high-risk provinces”. There is no red on the map. How did you explain “white” on the map? I would suggest revising these all maps (a-d). Please also revise all maps in Figure 4!
Response: Thank you. The Figure legend has now been revised and made a lot clearer to avoid confusions.
Lines 212-222: You only interpreted some variables. I would suggest including more explanations of other variables such as region of residence, heart attack, smoking status, and drinking status.
Response: Thank you. Figure 5 and Figure 6 present only the posterior estimates of the smooth functions of the continuous covariates. Indeed, we also included other key covariates you mentioned. Please see Table 1 and Table 2.
Table 1, under “Waist circumference (tertile), you define 1(lowest), 2, 3(highest). Please define 2!
Response: Thank you. We have added ‘middle’ for 2.
Table 2: You included DIC on the table, but there is no further explanation. Please add some explanations!
Response: Thank you. Now included in lines 189 to 190 of the manuscript.
Table 2: Why you did not include “ Wealth Index” in Table 2 as you did on Table 1”?
Response: Thank you. While Table 1 contains the descriptive analysis results, Table 2 contains results from the Bayesian hierarchical geo-additive models with models choices based on the DIC.
Reviewer 4 Report
Dear authors The article is supporting and funding by a government so it is clear to me this study was important enough to be considered but I have some concerns that invite you to consider them as below: 1. The conclusion part seems too weak. A reader can feel like there wasn't anything important in the main body of the study! 2. Why this study has done for 2016 and a comparison between 2012 and 2016? 3. The introduction part could be enough if you had a literature review section. Right now the article doesn't have enough background.ÂAuthor Response
Thank you for the constructive comments aimed at improving the quality of the manuscript. We have effected all necessary changes and corrections in the attached revised manuscript.
We have provided a point-by-point response to each of the questions/comments/points in the ‘Response to reviewers comments’ as detailed below.
Comments and Suggestions for Authors
Dear authors The article is supporting and funding by a government so it is clear to me this study was important enough to be considered but I have some concerns that invite you to consider them as below:
Response: Thank you.
1. The conclusion part seems too weak. A reader can feel like there wasn't anything important in the main body of the study!
Response. Thank you. More texts and stronger points have now been added.
2. Why this study has done for 2016 and a comparison between 2012 and 2016?
Response: Thank you. Here, we have compared how the prevalence of hypertension varied geographically in South Africa while comparing evidence from two separate household surveys conducted in 2012 (2012 SANHANES) and 2016 (2016 SDHS). In addition, the advanced statistical approach utilised here controlled for other key socio-economic covariates and health indicators (e.g., BMI), thus, we were able to see how the effects of each of these covariates on the prevalence of hypertension in South Africa has changed over time (4-year period) across the datasets. This would serve a means of evaluating the potential impacts of concerted intervention efforts and several years of investments aimed at improving quality of life in South Africa.
3. The introduction part could be enough if you had a literature review section. Right now the article doesn't have enough background.
Response: Thank you. However, given that the purpose and novelty of the study is specifically on the advanced statistical methods used, we feel that the 15 studies already cited in the introductory section along with the 12 others would suffice.
Round 2
Reviewer 2 Report
Despite the average originality of the study, I agree with the explanations of the authors, I guess more analysis with the available data is difficult to perform.
Reviewer 3 Report
The authors have carefully revised the manuscript.
There were some typos under Figure 4 (Legend) lines 210-212
Please change it!
"Figure 3" with "Figure "4"
"Figure 3a and Figure 3b" with "Figure 4a and Figure 4b
"Figure 3c and Figure 3d" with " Figure 4c and Figure 4d"
