An ecological study on the association between early childhood caries and intimate partner violence in 20 low- and middle-income countries: 2007-2017

Background: Women are the worst affected by intimate partner violence (IPV), and this impacts negatively on the health of the children they care for. This study aimed to determine the associations between IPV and the prevalence of early childhood caries (ECC) in 3-5-year-olds. Methods: This was an ecological study using IPV (physical, sexual, emotional) data extracted from the Demographic Health Survey of 20 low- and middle-income countries and ECC data for 3-5-year-olds of the same countries for the period 2007-2017. Linear regression analysis was used to assess the relationship between the percentage of 3-5-year-olds with ECC (outcome variable) and IPV indicators (physical, sexual, emotional). The model was adjusted for the country’s gross national income and the percentage of women with secondary or higher education. Partial eta squared (ηp 2), regression coefficients, confidence intervals and p-values were calculated. Results: Data on ECC in 3-5-year-olds and IPV were available for six low-income-countries, 10 lower-middle-income-countries and four upper-middle-income-countries. The most prevalent form of IPV was physical violence (10.09%). The Democratic Republic of Congo had the highest prevalence of physical violence (45.8%), sexual violence (25.4%), and ECC (80.0%). The strongest association was between the prevalence of ECC and emotional violence (ηp 2=0.01), followed by physical violence (ηp 2=0.005), and sexual violence (ηp 2=0.003). For every 1% higher prevalence of emotional violence, there was 0.28% higher prevalence of ECC, and for every 1% higher percentage of physical violence, there was 0.21% higher prevalence of ECC. On the contrary, for every 1% higher prevalence of sexual violence, there was 0.35% lower prevalence of ECC prevalence. Conclusions: Emotional and sexual violence where the two types of IPV associated with the prevalence of ECC. The associations were minor and the directions of their effects were difference. These findings need to be studied further.


Introduction
Intimate partner physical, emotional and sexual violence (IPV) is a recognized threat to the health and rights of women globally 1,2 . Global estimates in 2010 indicated that about 30% of women have a history of IPV with this being highest in central (65.6%) and western (41.8%) sub-Saharan Africa and South Asia (41.7%) 3 . IPV is defined as the abuse that takes place between a couple who are either married or living together and may or may not have a child 4 . IPV not only has an impact on maternal health and welfare 5 , It also affects the nutritional status of children of mothers affected by IPV: these children are more likely to be stunted 6-12 . In addition, children who are exposed to IPV have increased risk of behavioural, emotional, psychological and social problems including mood and anxiety disorders, post-traumatic stress disorder, substance abuse and school-related problems 13 . Abused women are also more likely to physically abuse their children [14][15][16] . Children of mothers who experience IPV are also less likely to be immunized; have higher rates of diarrhoeal disease; and/or are at greater risk of dying before the age of five 17,18 The negative impact on the mental health of women affected by IPV 19 results in poor nutritional care of children 20-25 causing an increased risk for stunting 6-10,12 , and poor or delayed medical care access by affected children 12,23-25 . In addition, women who experience IPV may resort to substance use as a way to cope with the emotional, physical and mental stress 26 . Substance use may also be a way for affected persons to maintain and/or establish intimate relations with others 27 . Substance use disorder is a risk factors for child abuse and poor management of the oral health of the child. It may affect a parent's ability to function as a caregiver and provide basic health care needs of the child 28 .
Though several studies have established the negative association between maternal mental health and increased risk for ECC 29-32 , there is little known about the association between IPV and ECC. Lorber et al. 33 studied children's exposure to domestic violence and the risk for ECC and found no significant relationship between them though a positive trend was observed. However, in households where women were aggressive towards men, the risk for ECC was higher due to high exposure to cariogenic diets 34 .
The social cognitive theory may explain the relationship between IPV and ECC. The theory recognizes that an individual's behavior, cognition, and environment reciprocally and dynamically influence one another 35 . Key social environmental and personal factors play roles in determining an individual's engagement in health-promoting and disease-preventing behaviors 36 . Selfefficacy is a central cognitive construct in the social cognitive theory that may predict and explain health behaviors and outcomes 37-39 . Thus, a person's belief or confidence in their ability to perform certain actions influences the decisions to perform these actions 40 , including tooth brushing 41 , which is a significant risk factor for poor oral hygiene and ECC 42 . IPV undermines the self-efficacy of affected individuals 43,44 who are mostly women 45 serving as caretakers for children 46 . IPV can therefore compromise the ability of women to care for children who are entirely dependent on them for daily oral hygiene and access to professional oral health care services.
The aim of this study was to determine the relationship between the country-level prevalence of IPV and the prevalence of ECC in children aged 3-5-year-olds in countries where data on IPV and ECC are available. We conducted a country-specific analysis given the evidence suggesting differences in the relationship between IPV and ECC 47 .

Study design and data sources
This ecological study was based on IPV data extracted from the domestic violence module of the Demographic Health Survey (DHS), a nationally representative household survey that collects information on population-based indicators of health and nutrition in low-and middle-income countries (LMICs). The module includes questions on types and experiences of violence (physical, sexual and emotional) mostly perpetrated by the current or most recent husband/ partner 48 .
The present study also used data on ECC collected by El Tantawi et al. 49 . The data on the prevalence of ECC between 2007 and 2017 was extracted from the WHO and other Country Oral Health Profile databases as well as from published journals and government reports after conducting a search applying no language filter. Data on ECC was generated for 113 countries using this methodology. Early childhood caries ECC is identified in children < 72 months of age with one or more decayed (cavitated and non-cavitated), missing teeth due to decay or filled primary tooth surfaces 50 . For each country, databases were searched for data on ECC prevalence: (1) MEDLINE, (2) Web of Science (3) Scopus and (4) Google Scholar (1 st 100 search results) using a country's name combined with search terms. This was complemented by searching

Amendments from Version 2
In line with the expected revisions raised by the reviewer, we had changed all the term "relationship" changed to "association". In addition, we changed all references to 3-5-year-olds to 36-71-months olds. We also agree with the reviewer's suggestion that what was captured in the data were women who had experienced physical violence, sexual violence and emotional violence and not women who had experienced physical violence only, sexual violence only and emotional violence only. We have included this as a study limitation.
Any further responses from the reviewers can be found at the end of the article REVISED master and doctoral degrees theses about ECC, articles published in local journals and government reports. No language filter was applied.
Studies were included if they (1) were conducted on humans in epidemiologic, non-laboratory setting, (2) assessed ECC using the definition of the American Academy of Pediatric Dentistry or data was available to allow its estimation and (3) included healthy children. Excluded studies were (1) duplicates such as theses and resulting publications, (2) publications from which needed information could not be extracted (article with no full text or information not mentioned) and (3) studies published before 2007.
We extracted data from studies reporting prevalence figures from nationally representative samples if they were available. Otherwise, we used the retrieved data to calculate ECC prevalence for a country by dividing the total number of children affected by ECC in the study/s by the total number of children examined and multiplying by100.
The search strategy for Pubmed was: "Dental Caries" [Mesh] OR caries OR dental decay; Filter by age "infant: 1-23 months" and "preschool child"; Filter by date: from January No language filter was applied. The retrieved data were used to calculate the ECC prevalence for each country by dividing the number of children affected by ECC by the number of children examined and multiplying by 100. In the present study, we used the prevalence of ECC for 3-5-year-old children. Further details were reported in our previous paper 49 .

Intimate partner violence
We investigated the history of physical violence (the intentional use of physical force with the potential to cause injury or harm), sexual violence (any experience of unwanted or forced sexual activity and sexual coercion) and emotional violence (psychological abuse including humiliation, insults and intimidation) 51 . Self-reported lifetime exposure to IPV was separated into three categories: (i) physical violence only; (ii) sexual violence only; and (iii) emotional violence only. Details on the source of the IPV is available in Table 1.

Confounders
Income level was reported to be associated with ECC 49 and IPV 3 . We adjusted for country income level based on the 2015 gross national income (GNI) per capita calculated using the World Bank Atlas method 52 since it was the latest available for the study period. In addition, we also obtained information about the percentage of women between 15 and 49 years of age who completed secondary or higher education per country from the DHS data 53 .

Statistical analysis
The datasets (ECC and IPV indicators) were matched by country. Analysis was restricted to only women selected and successfully interviewed in the DHS domestic module and those who were formerly or currently in union or living with a man at the time of survey. Statistics, including means and standard deviations (SD), were calculated for IPV indicators, ECC, GNI and percentage of women completing secondary or higher education. Multiple linear regression analysis was used to assess the relationship between the percentage of 3-5-yearold children with ECC (outcome variable) and the three IPV indicators adjusted for GNI and percentage of women with secondary or higher education. Regression coefficients, 95% confidence intervals (CIs), p values and partial eta squared (η p 2 ) as the measure of effect size were calculated using SPSS version 22.0 (IBM Corp., Armonk, N.Y., USA). Significance level was set at 5%.

Results
A total of 190 publications on ECC from 88 countries (45.6% of the UN member states) were available 54 . Of these, data on ECC in 3-5-year-olds and IPV were available for 20 LMICs. These 20 countries included six low-income countries (30%), 10 lowermiddle-income countries (50%) and four upper-middle-income countries (20%). In view of the small number of countries with available data, analysis was conducted for all 20 countries collectively rather than categorizing by income level. In total 68,031 children <36 months old and 154,452 children 36-71 months old were included in the study. Table 2 lists the GNI, percentage of women with secondary or higher education, the percentage of 3-5-year-old children with ECC, and the proportion of women in the country who reported (i) physical violence only; (ii) sexual violence only; and (iii) emotional violence only. The prevalence (SD) of ECC in 3-5-year-olds in these 20 countries was 62.20% (18.66) and the mean (SD) GNI was 6219.47 (4083.99) US$ whereas the mean (SD) percentage of women with secondary or higher education was 57.15% (23.47). The most reported form of IPV was physical violence (25.46%, 10.09), followed by emotional violence (22.19%, 9.90), and lastly sexual violence (7.87%, 6.18).
The Democratic Republic of Congo (DRC), a war-torn country, had the highest prevalence of physical violence (45.8%). ECC prevalence in this country was also high (80%). Sexual violence was highest in DRC, followed by Uganda (25.4%). The highest prevalence of emotional violence (41.7%) was in Uganda while the prevalence of ECC in Uganda was 41%.  Table 3 highlights the association between the percentage of 3-5-year-old children with ECC and the percentage of women with various types of IPV in the 20 LMICs after adjusting for GNI and the percentage of women with secondary and higher education. The associations between ECC and each of the three categories of IPV were non-statistically significant. The strongest association was between ECC prevalence and exposure to emotional violence (ηp2= 0.01), followed by

Factors
Regression

Discussion
This is the first study attempting to assess the association between IPV and ECC using national data available from multiple countries around the world. This data suggests that in countries with a higher proportion of women facing physical and emotional IPV, the prevalence of ECC is also higher. However, the prevalence of ECC was lower in countries with a higher proportion of women facing sexual IPV. The relationships between the forms of IPV and ECC examined in this study were not statistically significant likely due to the relatively few numbers of countries with available information from the DHS. Also, the effect sizes observed were small though this is not unexpected given the multifactorial nature of ECC, and that IPV may be partly explained by other social determinants of health such as income and education (for which we controlled for in our analysis). The study finding, however, highlights an area for future research as there are indications that the global incidence of IPV has been on the increase 55 ; and the association between both IPV and ECC seems to represent co-existing issues rather a causal relationship. Factors such as culture, lifestyle and socio-economic status may be etiological or predisposing factors for ECC and IPV. Although we controlled for income and education, other related underlying factors that are not measured may be able to explain some of the observed variance among countries in the prevalence of both health problems (IPV and ECC).
The study provides new information to strengthen existing evidence on how social context is associated with disease risks. Globally, the prevalence of ECC is high -it is the 10 th most prevalent childhood illness 56 . Most of the efforts to address this problem have largely been limited to interventions aiming at altering biological and behavioral risk factors. Little attention has been paid to addressing the social determinants of health, significantly impact the behavioral and biological risks.
The study results suggest that the children 3-5-years-old resident in LMICs may be at increased risk for ECC when the care-giver is exposed to IPV though the effect size may be small. The stronger link between emotional violence and ECC -stronger than physical and sexual violence -may be related to post-traumatic stress disorder 57 , suicidal ideation and low self-esteem 58 that results from the emotional violence. Emotional abuse often occurs prior to or concurrently with physical or sexual abuse and increases the trauma of physical and sexual abuse 59 . The risk for mental health consequences of emotional violence is higher than it is for physical and sexual violence 58 . Children of mothers with mental health disorders are more likely to have ECC than children of mothers without these mental health disorders 60 .
The non-significant associations between ECC and the different types of IPV and the small effect size observed may imply that the role of IPV as a structural determinant of the risk for ECC may be small for the children at this age. However, with our understanding that when children 3-5-years-old are exposed to IPV as an adverse childhood event, the risk of caries in adolescence and adulthood is significantly higher 61 . What this study has done is to indicate that there are signals of risk for caries when pre-schoolers are exposed to IPV. Existing studies and reviews of the IPV and oral health literature reveal that the link between IPV and ECC is quite plausible 62 . Integrating early childhood promotion and ECC prevention messages into programs that take place in women's shelters may be the first step to reduce the immediate possible risk of ECC and possible future risk for caries by offering practical methods to women to care for their children's oral health. This proposed management strategy also reiterates the importance of integrated maternal and child health care as a cost-effective strategy to manage diseases using the common risk approach 63 . There is currently very little evidence on approaches that promote the integration of maternal and child general healthcare with their -and especially the child's-oral health. As evidence emerges, it becomes needful to develop programs addressing maternal IPV and ECC in pre-school children as part of a general health screening process, especially in countries where the prevalence of IPV is high.
The inverse though non-significant relationship between sexual violence and the prevalence of ECC is difficult to explain. Studies suggest that some mothers who had experienced sexual abuse demonstrate parental reflective functioning, an essential capacity for providing sensitive care for children under routine parenting circumstances. This is also associated with the caregiver developing resilience 64,65 . This evidence of increased child protection by mothers who have a history of sexual violence may also be reflected in the improved oral healthcare the children receive. There is however, research that demonstrated negative psychosocial impact of sexual violence with negative implications for the oral health of children 66 while another found no association between women's experience of violence and ECC 67 . This finding needs to be studied further.
One of the strengths of the present study is the use of population-level data derived from the DHS, thus ensuring data quality 68 . We used the data from 20 LMICs to provide insight into a phenomenon that if further explored, may shape policy and practice around ECC management 69 . Our study, however, is limited by the low number of included countries thereby challenging the generalizability of findings. The low sample size was addressed by using partial eta squared, a measure of effect size, to compare the association of ECC with various forms of IPV and to decide on the importance of each variable, rather than using the p values to determine the significance of the associations. Ensuring the availability of IPV data in other countries with different income levels may help develop a better understanding of the nature of the association between ECC and IPV.
Another limitation is that IPV data is based on self-reporting with the possibility of under-reporting due to social desirability bias 70 . However, this is the standard method used to obtain information on IPV 71 . We also only extracted data on IPV and women though there is evidence suggestive that both males and females may be victims of IPV 72,73 ; and IPV affecting men may be associated with the risk of ECC 26 . Also, this is a crosssectional study. We are unable to determine a causal relationship between IPV and ECC and the direction of the relationship between the two variables though it seems less likely that ECC will cause IPV. In addition, the study is ecological with potential for ecological fallacy resulting from linking different independent data sets 74,75 . The 20 countries included in this study were not randomly selected to represent the 135 LMICs 76 .
Rather they were a sample of LMICs with complete data for the variables included in this study. This limits the generalizability of the study findings to all LMICs. However, the report can be generalized to countries with similar epidemiological profiles to those represented in this study. Like all ecological studies, our study suggests the plausibility of an association between ECC and IPV although the effect size in the present case is small. Future studies are needed to look into possible associations at the individual level and assess the presence of potential mediating factors.

Conclusions
Despite these limitations, the study provides new insight into the possible association between ECC and IPV: physical, sexual and emotional violence were associated with ECC prevalence though the associations were not significant, and the effect size of the associations were small. Women's exposure to emotional violence had stronger associations with the prevalence of ECC than physical and sexual violence.

Underlying data
The IPV data used in this study are available from the Demographic and Health Survey (DHS) website. Access to the dataset requires registration and is granted only for legitimate research purposes. A guide for how to apply for dataset access is available at: https://dhsprogram.com/data/Access-Instructions. cfm. Please find the urls to the specific datasets in Table 1.

Mohajer Abbass Hameed
Murdoch Children's Research Institute, Melbourne, Vic, Australia This study utilises a cross-sectional methodology to explore the association between IPV and ECC using national data (Demographic and Health Survey) available from multiple countries. Overall, whilst the findings are statistically not significant, the authors, on some occasions (e.g. in discussion) appear to over-emphasise the findings and may appear to indicate that there is an association. Nevertheless, the key message that the findings do not support the hypothesis for a significant association, is sufficiently articulated.
In the method section, the authors indicate that IPV data was extracted from the domestic violence module of the Demographic Health Survey and authors have provided relevant data websites. However, in terms of prevalence of ECC, the authors appear to have done a literature search and identified articles/publications; yet it is not clear as to whether this was a systematic search of the selected databases, and how screening of records were done. Or was this a rapid search of the literature to get the relevant articles, if so, how many articles were included/excluded? Furthermore, when IPV data is extracted from DHS and ECC from published papers, how were these linked together? In other words, IPV data is extracted from one source (DHS) and ECC data extracted from another source (literature search); then how were these merged or analysed together? It appears that there is some missing sentence/information in the method section to clarify this.
Throughout the manuscript, there are some typos and grammatical errors. For example punctuation needs to be corrected in page 4, "IPV not only has an impact on maternal health and welfare 5 , It also affects the nutritional status of children of mothers affected by IPV: these children are more likely to be stunted 6-12 ." Lastly, the title of the manuscript can be re-worded to reflect the study investigation. Currently, the title indicates "association between early childhood caries and intimate partner violence", I wonder if this should be reversed to "association between IPV and ECC".

Is the work clearly and accurately presented and does it cite the current literature?
Yes

Is the study design appropriate and is the work technically sound? Partly
Are sufficient details of methods and analysis provided to allow replication by others? Partly

If applicable, is the statistical analysis and its interpretation appropriate? Yes
Are all the source data underlying the results available to ensure full reproducibility? Yes

If applicable, is the statistical analysis and its interpretation appropriate? Yes
Are all the source data underlying the results available to ensure full reproducibility? Yes Are the conclusions drawn adequately supported by the results? Yes I still have a problem with including children below 72 months old. That means the included children were under 6 yrs (full 6 years) not five and what about the criteria of age above 3 years? If the authors only restricted the search to the children under 72 months that means that children younger than 3yrs are also included. Also in the result section, it has been mentioned 68031 children were below 36 months. That is in contradiction with the method section and aim that mentioned children 3-5 years are studied.

○
Please clarify the variables for the experience of IPV, the authors mentioned that they have 3 variables for IPV: "physical violence only" "sexual violence only" and "emotional violence only" does that mean that if the women experienced both physical and sexual violence for example they are not included? Violence is hardly a single act and usually, the women experience multiple forms of violence. Perhaps the authors meant the experience of "any physical violence" etc. rather than "physical violence only" ○ It is great that the authors used the weight for their statistical analysis but that is not enough when it comes to the p value and CI. They need to adjust for cluster sampling design by using the strata and cluster variables in the complex sample analysis. The current analysis is actually prone to type one error and the current p values after adjustment for the sampling design might be even bigger.

If applicable, is the statistical analysis and its interpretation appropriate? Yes
Are all the source data underlying the results available to ensure full reproducibility? Yes

Are the conclusions drawn adequately supported by the results? Yes
Author Response 04 Jan 2022

Morenike Oluwatoyin Folayan, Obafemi Awolowo University, Ile-Ife, Nigeria
However, I still think that the interpretation of the results and regression analysis is misleading and the fact that the p values are very large indicates that the findings might be by chance. So, although the authors mentioned that the findings are "insignificant" still the whole result section, discussion, and also abstract are written in a form that points to the existence of association rather than non-existence. I understand that the sample size is small but with the current sample size and p values, it is not possible to draw such conclusions. Further, the Cis were not adjusted for the cluster sampling design which is also a major issue to consider. Response: We understand and acknowledge the Reviewer's concern. Our study is ecological as indicated at the beginning of Methods. This means the units of observation and analysis are countries not persons. Because of this, our sample size is 20. This is very much under-powered to show significance. Statistical significance, however, is a function of sample size. Please check the following table. We replicated our data (n=20) several times using the same values so that the sample would reach a much greater number (n= 1440) and re-ran the same analysis. In the re-ran analysis with a greater sample size of 1440, the regression coefficients (B) and partial eta squared are exactly the same as that of our sample of 20 but the confidence intervals changed and the p values are all significant. Our study used an ecological design where the countries were the units of observation. If the same datasets are used and the analysis is re-run at person level, the sample size would be 154,452 instead of 20 and the significance would be even greater.
Because of this, we include the effect size (partial eta squared) and considered it safe to mention the presence of associations rather than their absence. We made sure that we clarified all over the paper than the association was not significant and included the data so that our results can be replicated. We contend that there are always associations and the analysis can show whether they are significant or not (based on sample size among other things) and can also show their strength (for which we calculated the partial eta squared).
We feel our reporting and discussions of the associations and effect size were very appropriate for this form of data analysis. We have however gone ahead to revise the manuscript to focus on the non-significant association observed It is great that the authors used the weight for their statistical analysis but that is not enough when it comes to the p value and CI. They need to adjust for cluster sampling design by using the strata and cluster variables in the complex sample analysis. The current analysis is actually prone to type one error and the current p values after adjustment for the sampling design might be even bigger.
Response: Because we used ecological design, we calculated means for all values at country level. If the same datasets are used so that the person is the unit of analysis, persons would be clustered in countries and multilevel analysis would be needed to account for the clustering. Please check our response above clarifying that in our design where the unit of analysis is the country not the person, the values used are means of country values not persons and therefore there is no clustering and no multilevel analysis is needed.
○ Some minor details: Please change the term "relationship" to "association" throughout the manuscript. The study is cross-sectional, and causality cannot be established.
Response: All the terms "relationship" changed to "association"  Response: We have edited this to 36 -71 months. Also, the data of children below 36 months was not included in the analysis. This was an error and we have edited this. Thanks for picking this up. ○ ○ Please clarify the variables for the experience of IPV, the authors mentioned that they have 3 variables for IPV: "physical violence only" "sexual violence only" and "emotional violence only" does that mean that if the women experienced both physical and sexual violence for example they are not included? Violence is hardly a single act and usually, the women experience multiple forms of violence. Perhaps the authors meant the experience of "any physical violence" etc. rather than "physical violence only" Response: The DHS data used series of items/questions to capture the form of violence experienced -physical, sexual, emotional. Women who answered positive to a particular item/question are categorized to have experienced the form of violence. While it is possible to generate a variable for women who experienced multiple forms of violence, this study however presents only an experience of each form of violence. I do not think that the authors can come up with such an outcome interpretation that for every 1% increase in exposure to certain violence how much the prevalence of ECC increases as the p values are far from significant and i.e. the reported outcome is not associated to the exposure. Even if there is a trend there.

Response:
In agreement with the Reviewer's comment and to clarify our meaning and avoid misleading readers, we modified the results to reflect that the association was not significant. We based our interpretation on statistical convention ( https://www.wiley.com/WileyCDA/Section/id-831952.html AND https://home.csulb.edu/~msaintg/ppa696/696stsig.htm) that tests of significance have p values that show whether associations are significant or not. Therefore, associations can be significant or non-significant. Also, significance is related to sample size and for the same variables and associations, increasing the sample size may change a non-significant association into a significant association. We thus based our interpretation of the analysis on the direction and magnitude of the association and now explicitly indicated that it was non-significant.
○ Discussion: I think the whole discussion is written in a sense that "there is an association between exposure to IPV and ECC" which is definitely not correct. Since there was no association found in the analysis, I encourage the authors to rewrite the discussion around that. In this format I think the discussion is misleading as it focuses on the reasoning why maternal experience of IPV is associated with ECC. Which is an explanation of nonsignificant findings. Response: Thanks for highlighting this limitation with the discussion. We have revised the discussion to highlights the non-significant relationship between IPV and ECC, and the implications of the findings. We hope the reviewer will find the revision a lot more representative of the study findings and suitable Competing Interests: The authors have no competing interest