Sociological variety and the transmission efficiency of Mycobacterium tuberculosis: a secondary analysis of qualitative and quantitative data from 15 communities in Zambia

Objectives Selected Zambian communities formed part of a cluster randomised trial: the Zambia and South Africa TB and AIDS Reduction study (ZAMSTAR). There was wide variability in the prevalence of Mycobacterium tuberculosis infection and tuberculosis (TB) disease across these communities. We sought to clarify whether specific communities could have been more/less vulnerable to M. tuberculosis transmission as a result of sociological variety relevant to transmission efficiency. Design We conducted a mixed methods secondary analysis using existing data sets. First, we analysed qualitative data to categorise and synthesise patterns of socio-spatial engagement across communities. Second, we compared emergent sociological variables with a measure of transmission efficiency: the ratio of the annual risk of infection to TB prevalence. Setting ZAMSTAR communities in urban and peri-urban Zambia, spanning five provinces. Participants Fifteen communities, each served by a health facility offering TB treatment to a population of at least 25 000. TB notification rates were at least 400 per 100 000 per annum and HIV seroprevalence was estimated to be high. Results Crowding, movement, livelihoods and participation in recreational activity differed across communities. Based on 12 socio-spatial indicators, communities were qualitatively classified as more/less spatially crowded and as more/less socially ‘open’ to contact with others, with implications for the presumptive risk of M. tuberculosis transmission. For example, watching video shows in poorly ventilated structures posed a presumptive risk in more socially open communities, while outdoor farming and/or fishing were particularly widespread in communities with lower transmission measures. Conclusions A dual dynamic of ‘social permeability’ and crowding appeared relevant to disparities in M. tuberculosis transmission efficiency. To reduce transmission, certain socio-spatial aspects could be adjusted (eg, increasing ventilation on transport), while more structural aspects are less malleable (eg, reliance on public transport). We recommend integrating community level typologies with genome sequencing techniques to further explore the significance of ‘social permeability’. Trial registration number ISRCTN36729271.


GENERAL COMMENTS
OVERALL COMMENTS Thank you for the opportunity to review this manuscript. The data presented in the manuscript comprises a mixed-methods analysis to understand how social factors facilitate the spread of tuberculosis (TB) in 15 neighbourhoods in Zambia. The findings suggest that there are differences in how specific how socio-spatial factors influence TB transmission. Social factors were broadly categorised into those that impact spatial crowing and social permeability of neighbourhoods. Importantly, the findings suggest that some of these factors ('routine' as described by the authors) can be addressed via public health and information interventions. In contrast, others ('structural' as defined by the authors) might be more difficult to change over the short-medium term. There are three key strengths of the study: 1. It's critical that the contribution of social factors to the spread of TB is investigated with the same rigour as the biomedical and clinical aspects. While some of the results in the manuscript may seem intuitive (and some have been previously observed in other regions), the fact that interdisciplinary data-driven findings have been presented around how socio-spatial factors impact TB transmission is noteworthy. 2. The combination of quantitative (for TB transmission) and qualitative (for socio-spatial factors) is interesting, and as noted by the authors, the interdisciplinarity of the approach is unique.
3. The study uses data from and builds on a previous large-scale interventional study of TB (ZAMSTAR) and thus maximises its findings and usefulness. It's also good to see that the approach to Patient and Public Involvement (PPI) is described in the methods. PPI is not always published in manuscripts, leaving little opportunity to assess if and how it was conducted. Other manuscripts should follow this example, and a description of PPI work in publications should become the accepted standard in the field. While the study's limitations are clearly noted, some additional aspects as described below should be addressed to further strengthen the manuscript. MINOR REVISIONS INTRODUCTION 1. Either here or in the section of the discussion that mentions potential limitations -I think it would be important to note some of the caveats of using the ratio of Annual Risk of Infection (ARI) to TB prevalence to determine transmission efficiency. Others have published valid and relevant caveats that could influence the reliability of the measure, for example, https://bmcinfectdis.biomedcentral.com/articles/10. 1186/1471-2334-11-156). Two of the potential caveats that have been previously described: a) that the baseline prevalence of TB is measured in children and b) that the measure can't detect multiple exposures/infections are particularly important in higher-burden settings. While (ARI:TB Prevalence) is a well-known metric and therefore appropriate to use, it's important to mention the caveats.
2. The data the authors have used focus on Zambia. Given that the ZAMSTAR study was conducted in South Africa, it would be useful to know why data for South Africa was not available/included in this studythis could be important for the generalisability of the findings. METHODS 1. It would be useful to describe how neighbourhoods were selected for the study. The original CODA and ZAMSTAR studies are cited here, but it's not clear how many of the same neighbourhoods were included here and a brief description here would also mean the reader would not have to check the methods of previous papers to assess whether there is any potential bias in selection of communities. 2. While observations of 'social permeability' are really important and interesting, there could also be a time component to thisthe longer people spend either outside their own neighbourhood or that people from outside of the neighbourhood are present could influence TB transmission risk. Moreover, the nature of activities (e.g., whether in well-ventilated areas or not) could also play a role. Could the authors suggest a way to take these dynamics into account, either here or in the discussion? 3. The potential limitations of the broad-based survey (BBS) and its ability to detect all social and spatial engagements has been well-described. One thing that was noted or mentioned is how household contacts/engagements might interact with non-household engagements to influence TB transmission. Again, this could be mentioned here or in the discussion. 4. Page 9, line 10 -11, could the authors provide a reference for ranking the burden of infectiousness of men's involvement higher than woman's and adults higher than children. These ideas make sense, but a citation would increase validity. RESULTS 1. Could the authors note if some neighbourhoods that were studied more rural or urban, and if this might have contributed to some of the findings. 2. It would be interesting to note if there were differences in the availability of healthcare services between neighbourhoods and whether this could this contribute to the overall risk of TB transmission/ 3. Importantly, schools have been noted as potential space for TB transmission: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0039246. It would be important to mention whether the BBS noted if children travelled outside of neighbourhoods for schooling, and if so, whether this could this have influenced TB transmission risk. 4. For people who travelled outside of the neighbourhood or people who travelled into new neighbourhoods, did this always occur via public transport? Could the mode of transportation into or out of neighbourhoods interact with the frequency of travel to impact TB transmission risk? 5. Figure 1 is not clearly annotated in the manuscript and I couldn't see a figure legend. Moreover, given the relatively large amounts of data presented, it might make the figure easier to interpret if some of the data points were colour-coded according to them. For example, structural vs routine socio-spatial factors and/or spatial and permeability factors could be distinguished using different colours. DISCUSSION 1. An essential issue to that could be addressed in more detail in the discussion is the generalisability. The study uncovers interesting and relevant interactions between socio-spatial factors and provides relevant areas for follow up. Moreover, the idea of having contextualised rather than one-size-fits all public health interventions is a good one. However, how would the authors envisage generalising these findings to other regions (where social patterns and co-morbidities are likely to significantly vary)? Given the social uniqueness of different communities, would this kind of study need to be conducted in all high-burden areas to determine neighbourhood-specific interventions? It could be useful to have suggestions for how these kinds of studies could be achieved and contribute to regionappropriate contextualised public health interventions. 2. It could be interesting to discuss how whole-genome sequencing of M. tuberculosis might aid in confirming/ruling out sites of transmission as well as cases that were a result of permeability of the neighbourhoods. 3. Finally, social permeability and levels of spatial crowding likely interact to impact TB transmission risk. It could be interesting to discuss how the data could be analysed, taking these interactions into account.

REVIEWER
Dara, M World Health Organization Office at the European Union, Brussels, Belgium REVIEW RETURNED 10-Mar-2021

GENERAL COMMENTS
Congratulations with a great paper. I suggest the following points to improve your paper: -Please expand on the results of the prevalence survey and the ARI estimates, e.g. which method had been used for the survey and any other consideration relevant to that survey (changes with any previous survey ...)..
-Please consider adding in the background and in the discussion sessions more data or their absence of any racial differences or other determinants (e.g. nutritional status, level of income prevalence of HIV,) or any other determinants which may have had an impact on the results. Thanks and regards, Dr Masoud Dara

Reviewer Comment Author Response
Reviewer Your point that ARI in adults may be different to that measured in children is also well taken, and indeed one of the main arguments in our previous paper (now reference 13).
However, since we are interested in how the relative rather than absolute value of the ARI/prevalence ratio compares across communities, it only matters that the relationship between measured ARI with average infection risks does not vary. Our previous paper [13] does present data showing that reported contact patterns by age (which mediate the relationship between adult and child ARI) show remarkable consistency across communities, which helps defend our use of a metric based on ARI measured in children for our exploratory comparisons.
Additional limitations with this metric include the dependency of ARI on TST interpretation method, and the measurement uncertainty associated with both TB prevalence and ARI estimates.
We have tried to capture some of this by adding to the limitations paragraph in the Discussion as you suggest: "…in future work. 2. The data the authors have used focus on Zambia. Given that the ZAMSTAR study was conducted in South Africa, it would be useful to know why data for South Africa was not available/included in this studythis could be important for the generalisability of the findings.
Our analysis was heavily reliant on BBS Data from the ancillary study, CODA. The qualitative component of CODA was conducted in the Zambian ZAMSTAR communities only, primarily due to practical and financial constraints. Therefore, there is no equivalent set of BBS data available from the South African sites. We have made the following minor adjustment to the Introduction to flag that CODA was integral and only conducted in Zambia:

"We used a novel mixed methods interdisciplinary approach to conduct a secondary analysis of Zambian data sets, collected during a cluster randomised trial (CRT) -the Zambia and South Africa TB and AIDS Reduction Study (ZAMSTAR) -and as part of an ancillary study, named the Contact Observations of Daily Activities (CODA). [12] [13]"
Owing to the relational and qualitative foundation of the analysis, findings are necessarily context specific.
Having said this, the BBS method itself, which allows researchers to develop contextually informed indicators, is easily transferable and the dual dynamic of social permeability and crowding could be relevant to variations in transmission efficiency across other settings.
To convey this, we have adapted the final Discussion paragraph on limitations (new text underlined): "Findings are context specific and data were effectively cross-sectional. We focused on interactions within localised geographical boundaries, and did not capture interactions on the far side of extra-community travel. Furthermore, fieldwork was conducted at a particular time of year and observations took place during the daytime, which limits the generalisability of our observations accordinglynot only across but also within the study communities. We anticipate relevant socio-spatial features and their role in mediating TB transmission will vary between contexts. However, the dual dynamic of spatial closeness (crowding) and social openness (permeability) may be relevant in other settings, with further investigations valuable.
In addition, we revised our Summary of strengths and limitations to include: "Results are context specific, therefore further investigation in other settings would be valuable to assess the generalisability of key concepts." Reviewer 1 -Methods Also see minor addition to beginning of methods: "This secondary analysis used whole communities, rather than the individual, as a unit for analysis and..." 3. The potential limitations of the broad-based survey (BBS) and its ability to detect all social and spatial engagements has been well-described. One thing that was noted or mentioned is how household contacts/engagements might interact with non-household engagements to influence TB transmission. Again, this could be mentioned here or in the discussion.
We assume that this comment should have read "one thing that was not noted or mentioned is…", and we have responded accordingly.
The BBS method used to generate qualitative data is rapid and focused on communities at large, providing a broad-brush picture of the collective. It is a valid method for participant observation in public spaces. Less so within households, although day long household observations were conducted to understand who was staying at home and where residents went when they left. We argue that, for a valid qualitative understanding of household contacts and how they interact with the broader community, one would need deeper ethnographic work, directed at a selection of households. We opted for the community level focus due to transmission being more likely to take place in the wider community in this particular setting, although targeting tuberculosis-affected households for tuberculosis screening, HIV testing, and referral for treatment of tuberculosis or M. tuberculosis infection remains a priority.
While households are not the focus of this particular study, they are obviously part of the sociological topography of communities. BBS does capture this, both through the 'Housing Density' indicator and the indicator, 'Who Remains at Home'. Interestingly, households did not emerge in this BBS data as significant 'hotspots' for community transmission. We are aware of other settings where households have emerged as hotspots. This has been in cases where boundaries between the private home and wider community are particularly blurred, e.g. where private households start functioning more as public churches.
In addition, we do include where children are coming into contact with adults in the broader community and this does start to give an indication of where household contacts are engaging with others (e.g. video shows, bars and on transport). In sum, we feel that the design of our qualitative methods and the study setting, mean that the delineation of gender and age is more appropriate than that between household and community.
We addressed this point by adding the following text with new reference to our 'Qualitative Data' segment in Methods, rather than the discussion, as it pertains more to the framework of our study: "The qualitative BBS data indicated how public places and transport were used by different age groups and genders within each community. Our focus on the broader community is justified as we know that transmission is less likely to take place within households in this particular setting. [21] 4. Page 9, line 10 -11, could the authors provide a reference for ranking the burden of infectiousness of men's involvement higher than woman's and adults higher than children. These ideas make sense, but a citation would increase validity.
Thanks for this. We have modified this sentence ('in Mixed Methods Analysis').
Old version:

"Expectations of the burden of infectious tuberculosis led to men's involvement being ranked higher than women's involvement, and adults' involvement higher than children's."
New version: "Previous work on these communities combining the TB prevalence results, ARI estimates and quantitative social contact data, suggested men are responsible for more infections than women.
[13] Children, especially age <5 years, typically have lower bacteriological load than adults and are likely to contribute less to transmission than adults of either sex." Reviewer 1 -Results

1.
Could the authors note if some neighbourhoods that were studied more rural or urban, and if this might have contributed to some of the findings.
The setting is mostly urban and there is no doubt that this contributed to our findings: BBS is a participatory and the observational qualitative method is by definition contextually sensitive. Three of the sixteen Zambian ZAMSTAR communities were classed as peri-urban with an urban centre (in a district town or on a major road), with certain parts spreading into more rural areas. We did not conduct BBS in the rural spread of these communities, so our findings are from urban and periurban settings. We have restructured our abstract, also in line with the Editor's comments, to add the following: "Setting: ZAMSTAR communities in urban and periurban Zambia, spanning five provinces." Please also see new text in Methods, comment 1 (supra), which includes the following: "…All communities were predominantly high density urban communities (three were peri-urban) located along the line of rail and/or major trading routes. Healthcare services were therefore comparable to the extent that they all provided TB diagnosis and treatment; had trained health care providers; and voluntary TB supporters. A degree of parity is implied. The maximum distances needed to travel to the diagnostic centre did vary across communities, but it was beyond the scope of this piece to assess the impact of any nuances in state healthcare services on the overall risk of TB transmission, beyond 'crowding' at facilities, as captured by the indicator "Crowding at Health Centres" (Table 1) Please also see new text in Methods, comment 1 (supra). Which includes the following text: Schools do play an important role within the communities and formed sites for TST testing for the ZAMSTAR trial. Therefore, in an indirect way, transmission in schools is accounted for in the ARI data: Infection in school children was used as marker of what was going on in the broader community.
This secondary analysis, however, was not investigating potential hotspots for transmission, but rather trying to classify, or type, sociological variation in meaningful way for transmission efficiency.
BBS findings did show that in four communities, a notable number of children were observed out of school.
Please see Table 1, 4 th to last row. We couldn't identify whether these children were not attending school, or simply had a different time structure to their school day and were unable to meaningfully investigate. There was one community where children were observed travelling to another community to attend secondary school, but again this didn't emerge as significant. As mentioned earlier, a limitation of the BBS is that it doesn't document extra-community activity. Our text has been revised to include this. Please see revised Summary text and Discussion, as highlighted in comment 2 for Methods (supra) and adjusted text to limitations paragraph in Discussion, with new reference: This is an interesting question. However, in almost all instances minibuses were the predominant mode of extra-community travel, except in one community where residents predominantly travelled by foot. Please see page 11, lines 6-15, of original submission. As such, we are not able to meaningfully investigate this. Figure 1 is not clearly annotated in the manuscript and I couldn't see a figure legend. Moreover, given the relatively large amounts of data presented, it might make the figure easier to interpret if some of the data points were colour-coded according to them. For example, structural vs routine socio-spatial factors and/or spatial and permeability factors could be distinguished using different colours.

5.
Thanks for this suggestion. We have improved the annotation of Figure 1 (and similar supplementary files) by using colour distinctions between the indicators of crowding, indoor activity and social permeability.
In fact, we did provide legends/captions for the figures, but these did not appear in the reviewer version alongside the figures.
The captions are: 1. An essential issue that could be addressed in more detail in the discussion is the generalisability. The study uncovers interesting and relevant interactions between socio-spatial factors and provides relevant areas for follow up. Moreover, the idea of having contextualised rather than one-size-fits all public health interventions is a good one. However, how would the authors envisage generalising these findings to other regions (where social patterns and comorbidities are likely to significantly vary)? Given the social uniqueness of different communities, would this kind of study need to be conducted in all high-burden areas to determine neighbourhood-specific interventions? It could be useful Owing to the exploratory and hypothesis generating nature of the work, generalisability is limited. Also see related response to Introduction, pt 2 (supra). Theoretically, the BBS method is designed to be implemented in more than one neighbourhood/community where there are tangible and cogent socio-geographical boundaries and where similar features can be compared. Practically, a particular research interest and logistical limitations will shape the number of communities researched and for what purpose BBS can take place. It is a method that lends itself to systematic comparison of similarities and differences across communities and countries and thereby, using a framework of meta-indicators, lends itself to generalisability around a core research question. We would therefore argue that the pattern of socio-spatial factors could be relevant to other urban places and BBS would be an appropriate rapid method to assess key features for the purposes of comparison. 3. Finally, social permeability and levels of spatial crowding likely interact to impact TB transmission risk. It could be interesting to discuss how the data could be analysed, taking these interactions into account For a more in-depth analysis of how social permeability and spatial crowding interact, we would advocate on a more limited scope, focusing on one community, which, for us, would require more data collection. The best way that we could come up with in our study, given the breadth across communities and nature of the data, was the ability to follow a particular community (number) through the graphics of FIGURE 1, which provides a sense, albeit compact and complex, of how permeability and crowding interact within a particular community. For example, if one takes community 14 into consideration, we see that although non-residents are coming into this community, crowding in public space is less than in most other communities and transport is not overloaded and poorly ventilated. For community 4, with higher transmission efficiency, non-residents are also coming in, but public spaces and transport were gauged as more crowded.
We have adjusted 'Patterns with respect to transmission efficiency' in Results to highlight how Figure 1 should be used in this way, in order to analyse interaction: "By tracing an individual community (number) through Figure 1, one is able to consider the interaction of the different socio-spatial variables identified. Where transmission efficiency was highest…" We think our changes above and minor tweaks in paragraph 2 of the Discussion create more clarity on how our qualitative work embraces, rather than reduces, this complexity. We hope this is sufficiently improved.
Reviewer Two -General Comments 1. Please expand on the results of the prevalence survey and the ARI estimates, e.g. which method had been used for the survey and any other consideration relevant to that survey (changes with any previous survey ...) Thanks for this comment. As we noted, the methods for the historic TB prevalence and TST surveys whose data we are using have been described in references [14][15][16]. However, we have added some relevant methodological and background results to save readers referring to these papers. 2. Please consider adding in the background and in the discussion sessions more data or their absence of any racial differences or other determinants (e.g. nutritional status, level of income prevalence of HIV,) or any other determinants which may have had an impact on the results.
Thank you for requesting. Reviewer 1, in a similar vein, also felt more explanation of selection criteria and study setting was needed. We have reformatted our Abstract to flag important parameters of the setting: "Setting: ZAMSTAR communities in urban and periurban Zambia, spanning five provinces.
Participants: Fifteen communities, each served by a health facility offering tuberculosis treatment to a population of at least 25,000. TB notification rates were at least 400 per 100,000 per annum and estimates of HIV seroprevalence high.
We have also included the following text at the end of the introductory paragraph to the Methods section: