Review of 'Associations between the household environment and stunted child growthin rural India: a cross-sectional analysis'

General comments

This paper is potentially very interesting as it looks across the economic social and environmental conditions in children who have and have not suffered stunting in an Indian state. It thus examines three main pillars of sustainable development.

The data is complex if in part somewhat old (15 years) and more needs to be done to describe the study’s findings about the two populations of stunted and unstunted children and to set the background of these two groups of children in a wider context much of which currently appears too late in the paper to inform the reader in a timely way.

Response: We have now set the context earlier in the paper and highlighted why we are using the dataset from 2006. See our edits included below:

“India’s National Family Health Survey provides data from representative sample at national and state levels on a comprehensive list of domains that include health, nutrition, fertility, mortality and family planning with focus on women and children. The household interviews also provide critical estimates on different household characteristics. In recent years, NFHS has been the basis of claims on successes of National Rural Health Mission. (Wal, 2018)  The value of using NFHS data to improve nutritional outcomes through ICDS and NHRM has been recognised. (Lahariya & Khandekar, 2007) Conducted in 2005-06, NFHS-3 provides a profile of important baseline statistics on the association of different factors related to larger household environment and nutrition at the commencement of National Rural Health Mission, which was launched in April 2005 which is why we base our study around this database.”

The context setting should explain more fully how this study sits in relation to other studies on stunting in India (e.g. with respect to urban vs rural communities) so that it can be more clearly established in the reader’s mind what this study contributes. There has been a lot of work on stunting and spelling this out will help the reader appreciate the importance of the work an may help guide the discussion of the results which at present is rather too short on explaining the wider significance of the work and how it relates to the policy positions of the state concerned on such issues as open defecation.

Response: Thank you for your comment. In our introductory section we highlight the gaps in evidence for rural settings in India, the complex linkages with environmental factors and why we selected the state of Rajasthan. We discuss open defecation and implications in our discussion now and place it in the context of Clean India Mission in India.

It may be more productive to conduct other forms of statistical analysis than the ones chosen. Some calculation of relative risk rather than the odds-ratio approach may allow for a more rounded interpretation of the results. The issues involved here seem to be much more in the public health rather than the medical domain. The study is fundamentally examining the wider determinants of health where the Bradford Hill criteria can be helpful in organising data and analysis.  Much depends on the authors’ view of their data and whether they are viewing the study as being exploratory of a real world situation or one where the unstunted children are a de facto control group such that the study becomes one where the intervention is the unimproved economic, social and environmental circumstances of the child. The first approach may be preferable, in which case perhaps the whole data set could be analysed by some approach based on the principle components of the dataset or on some simple multiple regression that might in effect allow the relative importance of the large number of variables (some of which are parametric and some non-parametric) to be more clearly set out for a reader. Such approaches may have the advantage of allowing a re-examination of the data once main principle components have been established and consideration given to underlying mechanistic or process-based cause and effect pathways. The authors may think this approach to discursive but it may avoid corralling a wide ranging study into a narrow interpretation set determined by the odds-ratio format.

Response:  In our study, we focus on informing factors that determine risk. Even if we were to use risk ratios versus odds ratio the main model would be derived from the same model with the list of statistically significant predictors being the same.  In addition, we are not intending to compare outcomes for different population groups with a control group where some of the other modelling techniques such as risk ratios would be appropriate.

There would be room in a methods or approach section to describe more fully the reasoning behind the statistical approach selected. This might help the field in general. More discourse on analytical approaches is needed to avoid misunderstandings arising in operational or policy arena. This might include something about whether the study’s objective is to examine how various variables might be related or whether there is enough evidence to describe relative causal significance.

Response: We sincerely thank the reviewers for their insightful comments. The intent of this paper is to inform the predictors of a binary outcome, stunting, based on national level cross-sectional survey data. Logistic regression is a widely used technique for the same and similar analysis technique has been applied to same survey data.¹ Multiple regression is a technique that is used for continuous outcomes and cannot be applied to binary outcomes. However, given that odds ratio often produces wider confidence intervals for the point estimates, we have mentioned that as a limitation in the newly included statement of limitation. We have also elaborated the methods for variable selection.

Detailed comments on text (comments on Tables follows)

L27-29: Use of the word “unparalled” is difficult to understand in the sentence is thus the sentence if unclear.

Response: We agree with the reviewer and now have removed the word “unparalled”

L35-37: This text is unclear.

Response: We agree with the reviewer – we have now further clarified more clearly why the first 1000 days are vital for development.

L60-62: The reference to UNICEF should perhaps move up the text to L56 at least. Perhaps this is part of the contextual material that sets the scene?

Response: We agree with the reviewer. Changes included in text. Additional changes in text have also been made to make the flow more coherent in the context.

L70-71: A statement of robustness would be more appropriate in a medical journal in its current form it means that some of this material could be worked into to other sections of the paper and have greater impact.

Response: We agree and the statement of robustness has been removed.

L72: Are there any other factors linked to stunting that might be important that could not be studied in this work due to lack of access to relevant data? If so these could be stated in order to inform readers of the limitations of the study – perhaps describing the limitations, scope and focus of this kind of study is more important than a description or statement of robustness?

Response: We have now an extended section highlighting limitations of the study. See below:

“The first limitation is data availability as we could only include variables which were captured in the DHS-3 and therefore, potential mediators, moderators or even predictors could have been missed. For example, although improved water source is used as an indicator of higher probability of safe water the DHS-3 data did not include biological indicators of pathogenic contamination that might influence infection risk.

Secondly, the source data did not include information regarding any intervention and any intervention, which would have introduced either locally or nationally within 5 years period prior to the onset of study would have disproportional effect on nutritional status of the children.

Thirdly, the DHS-3 allows one selected answer in each category. Yet, households often have multiple sources of drinking water, sanitation and cooking fuels and the DHS-3 did not collect information related to consumption frequency and quality of drinking water. Additionally, children who are schooled, work and/or use public toilets may be exposed to other environmental pathogenic risks of stunting outside of the home. If so, there is greater cause for concern since our results may underestimate the true associations of environmental determinants and anthropometry. Of knowledge, the DHS-4 has included more open-ended questions (e.g. ‘how do you clean water’), which will allow for a comprehensive analysis of household environmental practices on childhood stunting subsequently.”

L85-92: Clarification of the wording here would again help the importance of the work come across more clearly.

Response: A complete restructuring of the text along with inclusion of new references have been done to provide more clarity and improve clarity on how various variable were selected and how they relate to our context.

L101-105: The rationale for choosing the factors included may need to be set out just a little more. For are they a standard set or are they what were available in this location?

Response: A complete restructuring of the text along with inclusion of new references have been done to provide more clarity and improve clarity on how various variable were selected and how they relate to our context.

L109: Is a comment needed about the age of the data – especially if there is a mis-match between age of data and any comparison made to more recent policy positions.

Response– We have added section 3.2.2 in the paper where we explain the relevance of age as an variable, challenges in data collection for age and also why this data is important, with appropriate references.

L135: Check the words about sub-categories here.

Response– As this is a secondary data analysis, we have included wording consistent with the dataset to clarify the sub-categories.

Line 151: Is a reference needed here?

Response– Reference has been included.

Line 153: There seems to something of a jump here in the wording on risks to older children compared to younger ones. The domestic situation must be hugely complex in the social circumstances being dealt with. This piece of text is one of several that suggests the authors main interests are considering risks and how these might be reduced rather than examining interventions per se (see general comments).

Response– We agree with the reviewer and for the same reason the article advances the need of formative research to understand the complex multilevel factors and their interaction. To make this more explicit, we have added this in the conclusion section as well.

Line 157: Determining age by cultural memory is an interesting approach doubtless used previously. As age classes were then determined the approach seems acceptable although some of the children may have been close to class boundaries. Can the authors comment on how many children were near class boundaries or the impact of using an approach using the absolute ages discovered. Maybe this data is not available.

Response– This point has been added as a limitation to our study in the limitation section.

L162-164: More needed here perhaps on the reasons why this view of other factors is the one to go with in this study (may link back to L101-105). A little ore explanation is needed as to how this importance of age vs other factors was determined.

Response– The criteria for confounding is now addressed and that eventually addresses this comment for other variables in Section 3.2.3.

L169: unclear sentence

Response– The sentence has been simplified.

L172: ditto

Response– same as above.

L174: The overall dataset is complex with many variables or even sub-variables so the wish to simplify it must be almost irresistible but the impact of dichotomising needs to be explained more clearly. It would be a shame if data richness was lost or the interpretive power of the analysis lessened.

Response– This has been explicitly stated in the statement of limitation. (L659-661)

L195-196: It is not absolutely clear why it is that age appears as the only confounding variable. Age is certainly a strong influence on the outcome of stunting but the text needs some addition to explain why it is given a confounding status when other variables are treated more as explanatory or even perhaps “causal” or at least correlative ones.

Response– We agree with the reviewer that the explanation needed more clarity. We have now explicitly stated the reason and referenced the criteria in Section 3.2.3.

L198: It seems as if there are relatively small numbers in some of the improved categories. Are these so small that it affects the approach to statistical analysis? The problem is not uncommon. Many environmental studies suffer from this kind of issue.

Response– This is now added as a limitation. However, in Logistic Regression rare event effect is mostly on the intercept. As we are not advancing any inference on the base effect without the effect of predictors, we have only included that as a limitation and did not attempt more rigorous machine learning techniques to check and prevent it.

Comments on Tables

Both Tables need a clear explanatory legend given their complexity. The Journal format is such that this is possible.

Response: We have discussed key variables in the introductory sections of the paper.

Should any detailed data on length and height be provided so that readers can see whether there was a clear separation between the stunted and non-stunted groups?

Response– We have considered stunting as the outcome and not the underlying anthropometric measurement and the variable has been defined as per the standard definition of the World Health Organization. Analysis of height is beyond the scope of this research as we are not considering linear growth as a variable, but only stunting to determine the predictors of the same.

In the last line of Table 1 there is an arithmetical error, hopefully not found in the statistical analysis.

Response– We apologise for the arithmetical error. The percentage figure in the last line has been corrected.

In describing the data in Table 2 has sufficient been said about the impacts of age on the outcomes for very young and older children – this could have a major impact on policies and interventions. This might be especially important for cases where water-borne health factors are involved and the pathways by which the health factors could be managed.

Response– The age-dependent effects through unadjusted and adjusted odds ratio has now been discussed in our paper.

The data as expressed in the Tables raises some issues about how this kind of data can best be explored/analysed.  For example when considering land ownership it appears from a quick look at the data as shown in the table that there are more stunted children in situations where land is not owned and rather less when land is owned. Does this kind of clear message emerge in the text clearly enough (provided it is supported by an appropriate analysis)? I think the numbers involved are No land ownership: 116/224 stunted vs 108/224 non-stunted whereas where land is owned the ratios are the opposite way round: 416/970 vs 554/970 and of course this is only one of the factors – what would happen if these results were broken down by age class? Would the effect be stronger still in some age classes?

Response– We have now clarified the role of land ownership and links with stunting as ascertained from the tables.