Article Text
PDF
Abstract
Background Alcohol use is a consistent correlate of intimate partner violence (IPV) in low-income and middle-income countries (LMICs). However, the magnitude of this association differs across studies, which may be due to contextual and methodological factors. This study aims to estimate and explore sources of heterogeneity in the association between alcohol use and IPV in 28 LMICs (n=109 700 couples).
Methods In nationally representative surveys, partnered women reported on IPV victimisation and male partner’s alcohol use. We estimated the relationship between alcohol use and IPV using logistic regression and full propensity score matching to account for confounding. Country-specific ORs were combined using a random-effects model. Country-level indicators of health and development were regressed on ORs to examine sources of variability in these estimates.
Results Partner alcohol use was associated with a 2.55-fold increase in the odds of past-year IPV victimisation (95% CI 2.27 to 2.86) with substantial variability between regions (I2=70.0%). Countries with a low (<50%) prevalence of past-year alcohol use among men displayed larger associations between alcohol use and IPV. Exploratory analyses revealed that colonisation history, religion, female literacy levels and substance use treatment availability may explain some of the remaining heterogeneity observed in the strength of the association between alcohol use and IPV across countries.
Conclusion Partner alcohol use is associated with increased odds of IPV victimisation in LMICs, but to varying degrees across countries. Prevalences of male alcohol use and cultural factors were related to heterogeneity in these estimates between countries.
- violence
- alcohol
- low-middle income country
- gender
- epidemiology
Data availability statement
Data may be obtained from a third party and are not publicly available. Data may be requested from the Demographic and Health Survey program (https://dhsprogram.com/data/).
Statistics from Altmetric.com
Introduction
Intimate partner violence (IPV), defined as behaviour within an intimate relationship that causes (or risks causing) physical, psychological or sexual harm, is a leading risk factor for poor health outcomes among women globally.1 The lifetime prevalence of physical/sexual IPV is estimated to be highest in Southeast Asia (37.7%), the Eastern Mediterranean (37.0%) and sub-Saharan Africa (36.6%).2 Alcohol use is believed to directly increase risk for IPV perpetration via the psychophysiological effects of alcohol,3 while also sharing common risk factors, such as inequitable gender norms, which vary across cultures and contexts. Most studies evaluating the relationship between alcohol use and IPV have been conducted in high-income countries (HICs).4 5 Emerging research from low-income and middle-income countries (LMICs) finds a similar association between alcohol and IPV,6–8 but the magnitude of this relationship is inconsistent across studies. This variability may be due to contextual factors that influence gender roles, drinking culture and their relationship.9 Existing literature suffers from methodological limitations including non-representative samples, poor alcohol and IPV measurement, and insufficient methods to account for potential biases in observational studies.
There are numerous potential confounders of the association between alcohol use and IPV among individuals in LMICs occurring at the individual level, relational level, community level and societal level.10–12 For example, gender norms may influence socially acceptable patterns of alcohol use and gender roles, while also directly impacting behaviours and intimate partner dynamics. Not accounting for these shared risk factors may distort the true association between alcohol use and IPV. By clarifying this association and accounting for shared risk factors, we may be better positioned to identify efficient opportunities for preventing IPV. Attempts to account for these sources of bias have incorporated confounders as covariates in multivariate analyses.13 While this may mitigate bias by adjusting for the relationship between the confounder and outcome, it does not account for differences in the distribution of the confounder by exposure. Covariate imbalances between partners who drink and those who do not drink alcohol may lead to biased effect estimates. Applying causal inference methods, such as propensity scores, provides a method to address these limitations and reduce bias in estimates of the alcohol–IPV association. This study aims to estimate the unbiased association between partner alcohol use and IPV victimisation in LMICs. Second, we will explore contextual factors that explain between-country heterogeneity in the alcohol–IPV relationship.
Methods
Participants
Data for this study come from the Demographic and Health Surveys (DHS), which are nationally representative cross-sectional surveys conducted in LMICs.14 In each country, 5000–30 000 households are selected using multistage sampling.15 The target population is women aged 15–49 years and children under the age of five. The majority of surveys include adult men of similar age. Interviewers were sex matched, fluent in the local language, typically had a secondary education and underwent intensive training and supervision.15
This analysis was restricted to women and their male partners (n=109 700 couples) to whom they were married/cohabiting at the time of the interview in 28 countries that included questions about IPV and alcohol in the most recent survey conducted between 2005 and 2016. The countries included in the analysis are listed in table 1.
Distribution of couples (n=109 700) by country
Measures
Questionnaires were translated, adapted and piloted in clusters not selected for the survey in each country.15 Interview items included in propensity score models covered demographic characteristics, socioeconomic status, reproductive health and gender norms/attitudes (see online supplementary materials). The domestic violence module is administered privately to ever-partnered females. Most (99.8%) eligible women consented to complete the domestic violence module and 1% of those that agreed to participate did not complete the interview because privacy was not possible.
Supplemental material
Partner alcohol use was measured by asking the female respondent whether her male partner drinks alcohol and, if so, how often he gets drunk (often, sometimes or never). IPV victimisation among female participants in the past year was measured using a modified version of the Conflict Tactics Scales (CTS2),16 which has been used to measure physical, psychological and sexual IPV in LMICs.17 The modified version of the CTS has shown good internal consistency (α≥0.79) and construct validity in HICs.16 We defined IPV as any physical, sexual or psychological violence perpetrated by a female’s current male partner in the past year.18 19
Patient and public involvement
We did not directly include patient and public involvement (PPI) in this analysis, but the DHS surveys were developed with PPI.
Statistical analysis
We estimated the distribution of sociodemographic variables for the weighted sample of females who completed the domestic violence module and their male partners. Variables included age, marital status, employment, residence, whether the household has electricity, number of children ever born, pregnancy history, and whether the female knew that her father ever beat her mother. We calculated the prevalence of partner alcohol use, frequency of drunkenness and IPV victimisation.
Propensity score matching
We evaluated the ability of five propensity score matching/weighting methods to balance potential confounders between male partners who drink and those who do not drink alcohol. We estimated the odds of partner alcohol use as a function of covariates including sociodemographics, reproductive health, gender norms/attitudes, family history of domestic violence and other health behaviours (see variables in online supplementary material). Participants were matched or weighted using nearest neighbour matching (1:1, 1:2 and 1:3), full matching and inverse probability of treatment weighting to examine the standardised mean differences in the distribution of observed covariates in matched/weighted samples using R.20 21 The distribution of observed covariates is expected to be similar across exposed and unexposed participants with similar propensity scores and thus the difference in the odds of IPV between women with partners that drink versus those that do not drink alcohol can be considered an unbiased estimate of the association assuming no unmeasured confounders.22 23 This method has been used previously with DHS data to estimate an ‘unbiased effect size’ describing the association between employment and IPV.24 Full matching displayed superior performance relative to other matching methods as indicated by the smallest standardised bias (figure 1).
Standardised bias by matching method note: outliers (variables) removed from graph to improve resolution of box plots.
Meta-analysis
The relationship between partner alcohol use and IPV victimisation was estimated using country-specific logistic regression models in the fully matched sample using Stata V.14.25 DHS sampling weights were combined with the frequency weights produced by full matching for the control group. Age of the male and female partner, family history of domestic violence and socioeconomic status were adjusted for in the outcome logistic regression models to produce doubly robust estimates of the alcohol–IPV association. These estimates were combined using random-effects models to calculate the overall and regional pooled estimates. We calculated I2 estimates of heterogeneity, which describe the variation in effect estimates across studies.26
Meta-regression
The outcome was the log OR of IPV as a function of partner alcohol use at the country level. Independent variables were derived from external sources (see online supplementary material 2) and included human development index, urbanicity, domestic violence legislation, gross domestic product per capita, female literacy, past-year alcohol use and heavy drinking among men, patterns of alcohol use, substance use treatment services, alcohol policy, recent armed conflict, religion and colonisation history. We used the adjusted R squared, ORs and CIs to evaluate the results given the limited power in this country-level meta-regression analysis.
Sensitivity analyses
To assess the sensitivity of our results to bias from an unmeasured confounder, we employed Rosenbaum’s primal model, which conservatively assumes that an unobserved confounder is collinear with IPV, but its relationship to partner alcohol use is unknown.27 We varied the magnitude of this association between 1.1 and 2.0. A value of 1.25–1.30 is considered robust to unobserved confounding.28 We assessed whether the findings were sensitive to our definition of IPV by comparing the meta-analytic results to a parallel analysis using another common IPV definition (ie, physical/sexual).
Results
Of the 109 700 couples included in the sample, most were married (76.88%), living in rural areas (66.10%) and had electricity (56.31%). On average, female participants were 5.84 years younger than the male partners. Females reported ever giving birth to an average of 3.24 children and 11.03% were currently pregnant. One-quarter of women reported that her father ever beat her mother (Table 2).
Demographic characteristics of the sample (n=109 700 couples)
Twenty-seven per cent of females reported experiencing past-year IPV (95% CI 22.72 to 32.53; table 3). The most common form was physical (20.09%), followed by psychological (16.81%) and sexual IPV (6.44%). Estimates of past-year IPV ranged from 10.23% in Azerbaijan to 54.24% in Afghanistan. The majority of women reported that their partner does not drink alcohol (66.90%; 95% CI 60.00 to 73.14). Among those that did report partner alcohol use, frequent drunkenness was common (66.41%). The prevalence of alcohol use was lowest in Afghanistan (0.39%) and highest in Cambodia (85.28%).
Prevalence of partner alcohol use and IPV victimisation among n=109 700 adult female DHS participants
Relationship between partner alcohol use and IPV victimisation
Partner alcohol use is associated with a 2.55-fold increase in the odds of IPV victimisation in LMICs (95% CI 2.27 to 2.86). There was substantial variability in these estimates across countries (I2=70.0%). The region-specific ORs in order of ascending magnitude were 1.68 in Latin America and the Caribbean (95% CI 1.47 o 1.92; I2=0.0%), 2.12 in South Asia (95% CI 1.91 to 2.34; I2=0.0%), 2.69 in sub-Saharan Africa (95% CI 2.35 to 3.07; I2=56.7%), 3.28 in East Asia and the Pacific (95% CI 1.39 to 7.75; I2=79.6%), and 3.70 in Europe and Central Asia (95% CI 2.92 to 4.69; I2=0.0%)(figure 2). The results of sensitivity analyses revealed that the observed association between partner alcohol use and IPV would persist even if an unmeasured confounder existed that differed in the odds of partner alcohol use between two matched participants by 55%, which suggests that the findings are robust and insensitive to hidden bias. These results were also insensitive to the outcome definition as indicated by similar results when applying an alternative definition of IPV (Physical/Sexual IPV: OR 2.56, 95% CI 2.29 to 2.86).
Country-specific and pooled estimates of the association between partner alcohol use and IPV. IPV, intimate partner violence.
Contextual factors related to the association between partner alcohol use and IPV
We identified country-level characteristics that explained variability in the magnitude of the alcohol–IPV association. Countries with a past-year prevalence of male alcohol use less than or equal to 50% displayed larger associations between partner alcohol use and IPV relative to countries with a high prevalence (>50%). Relative to countries where the primary religion (>2/3 of the population) was Muslim, primary Christian and mixed (Christian and Muslim) countries displayed a 35%–38% smaller association between partner alcohol use and IPV. The past-year prevalence of alcohol use among males and primary religion explained 21.5% and 20.3% of the variability observed in estimates of the alcohol–IPV association, respectively (table 4).
Country-level contributors to observed heterogeneity in the association between alcohol use and IPV
Countries with higher female literacy and fewer substance use treatment beds displayed smaller associations between partner alcohol use and IPV. Female literacy and substance use treatment beds each accounted for over 30% of observed variability. Post-Soviet countries displayed larger associations between alcohol use and IPV relative to former British, Spanish and Portuguese colonies. We explored whether the observed associations between the log OR and substance use treatment beds, female literacy, religion and colonisation history were confounded by past-year alcohol use among men. Colonisation history, substance use treatment beds and female literacy appeared to be independent of past-year alcohol prevalence as evidenced by comparable coefficients in the unadjusted and adjusted models. When adjusting for religion, the effect of male alcohol consumption was attenuated.
Discussion
We found that 27% of partnered women in LMICs have experienced past-year IPV, which is comparable to other multinational surveys.29 We estimated a 2.55-fold increase in the odds of IPV as a function of partner alcohol use after accounting for a broad range of potential confounders. This doubly robust estimate suggests a strong, independent association between alcohol use and IPV that is consistent with previous research reporting adjusted ORs ranging from 1.5 to 4.0 (8, 9, 13). Direct comparisons should be made cautiously given differences in recall period, measurement and sampling between studies. We found substantial heterogeneity in country-level estimates of the association between partner alcohol use and IPV. LMICs in Europe and Central Asia displayed the largest association, which was almost twice as large as the pooled OR estimated for Latin American and Caribbean countries.
An exploratory analysis of contextual factors that explain this heterogeneity found that countries with a low prevalence of past-year alcohol use among men and those that were predominantly Muslim displayed larger associations between alcohol use and IPV suggesting that countries with a less normative drinking culture may yield greater alcohol-related consequences. Female literacy, an indicator of female agency and empowerment, was associated with an attenuated association between alcohol use and IPV and may provide preliminary support for the protective role of female empowerment in the context of IPV risk. The increased number of substance use treatment beds available in a given country may be a proxy for demand for services or less prioritisation of prevention and early intervention during periods when alcohol-related consequences such as IPV may be more effectively prevented.
Colonisation history contributed to the observed heterogeneity in the alcohol–IPV association. Previous research has identified an association between colonisation and alcohol use for coping with loss of culture and resulting feelings of anger, avoidance and distress.30 Colonisation may threaten traditional gender and familial roles while introducing patriarchal models of intimate partner dynamics and gender norms, each of which may reinforce violence and its relationship to alcohol.31 Countries previously part of the Soviet Union displayed a larger alcohol–IPV association relative to those that had been colonised by Britain, Portugal and Spain even when adjusting for the prevalence of alcohol use among men, which may be explained by drinking patterns32 or recency of independence.33 Unexplained heterogeneity remains after accounting for alcohol use, religion, services, literacy and colonisation. Other aspects of culture, context and methodological features, such as measurement, are likely to influence the observed alcohol–IPV relationship. In the current study, all surveys administered the same alcohol and IPV assessment tools. Future meta-analyses integrating data from studies that administered varied assessment tools should explore whether this source of variation yields different estimates of the association between alcohol and IPV.
Interpretation of these results must be made in light of the following limitations. First, measurement of alcohol use and past-year IPV were based on female partner reporting. Partner alcohol use, which did not have a specific recall period, may be subject to recall bias if women that experience IPV are differentially likely to report their partner’s alcohol use relative to women without a history of IPV, perhaps due to differences in the perceived impact of their partner’s alcohol use on their well-being. Any partner alcohol use does not distinguish between low-risk and high-risk patterns of drinking. While use of another measure, such as the frequency of drunkenness, may be a stronger indicator of high-risk drinking it is more prone to measurement error and differences in interpretation.34 Given these limitations, we selected any alcohol use as the exposure, which we expect would produce a more conservative estimate of the alcohol–IPV association.
Second, all data were cross-sectional, which precludes our ability to make causal inferences or confirm the temporal precedence of the independent variables in our models. Other studies have examined this association longitudinally and found that alcohol use is associated with incident male-to-female IPV.35 We are unable to rule out the possibility of IPV leading to alcohol use, yet there is an absence of longitudinal evidence supporting this relationship. Future experimental intervention research testing the effect of alcohol interventions on IPV outcomes in LMICs may strengthen our understanding of the causal nature of the alcohol–IPV relationship.36 The goal of this analysis was not to estimate a causal relationship, but rather to estimate an unbiased measure of the association between alcohol use and IPV. It is possible that our findings may be biased due to unmeasured confounding, but our sensitivity analyses suggest that the association between alcohol use and the unmeasured confounder would need to be strong and independent of other covariates included in the propensity score models to invalidate our inference. Third, these data may not be representative of all LMICs. Fourth, we conducted a complete case analysis; however, less than 5% of the sample were missing data on variables included in the propensity score or outcome regression models thus reducing concern about bias resulting from non-random missing data patterns.37 Interpretation of female literacy and substance use treatment findings must acknowledge that data were missing on approximately one in three of countries and may not generalise to all included countries.
Despite these limitations, this study has several strengths. This is the first study to use propensity score methods to account for bias in estimation of the association between partner alcohol use and IPV. Second, this study used nationally representative data from 28 LMICs. Third, this study explored potential contextual factors that may modify the magnitude of this relationship.
In conclusion, this study strengthens evidence supporting the independent association between alcohol use and IPV by accounting for potential sources of selection and information bias and extending these findings to 28 LMICs representing diverse world regions, cultures and contexts. These findings reinforce the need to consider alcohol use as a central correlate of IPV in epidemiological research and a possible target for IPV prevention. The treatment and prevention of alcohol misuse has been recommended as an IPV prevention strategy, yet this guidance is rarely implemented practice in LMICs.38 There is emerging experimental evidence suggesting that brief alcohol interventions may reduce some forms of IPV.2 39 We also found that the alcohol-IPV association varied by country-level factors related to the historical context, female empowerment and alcohol norms. These findings suggest the potential for macrolevel, structural interventions to attenuate the association between alcohol use and IPV. When considered together these findings highlight the need for evaluations of multilevel intervention strategies to weaken the association between alcohol and IPV. Informed by ecological theories, these interventions should explore the combination of structural components to address the societal and community-level conditions that enable alcohol-related IPV and individual-based and couple-based components to reduce unhealthy alcohol use and IPV in LMICs.40
What is already known on the subject
Alcohol use is consistently associated with intimate partner violence; however, there is large variation in the prevalence of alcohol use, partner violence and their association across countries, cultures and contexts.
Existing studies on this topic possess several limitations related to design, sampling and measurement. In this study, we integrate several epidemiological survey, propensity score and meta-regression methods to overcome some of these limitations and examine sources of heterogeneity in the relationship between alcohol use and partner violence in 28 low- and middle-income countries.
What this study adds
This study found that partner alcohol use remained a significant correlate of intimate partner violence after using methods to account for selection and information bias.
We quantified the large variation that exists in the association between alcohol use and partner violence across countries and world regions.
Factors related to drinking norms, historical and sociocultural context explained some of this heterogeneity suggesting that the association between alcohol use and partner violence is not static and extends beyond the individual level. Further examination of these contextual factors and their role in modifying the relationship between alcohol use and partner violence may provide insight into novel mechanisms underlying this association as well as the potential benefit of interventions to reduce these public health challenges.
Data availability statement
Data may be obtained from a third party and are not publicly available. Data may be requested from the Demographic and Health Survey program (https://dhsprogram.com/data/).
Ethics statements
Ethics approval
The ICF International Institutional Review Board (IRB) and a local IRB from each participating country approved all DHS procedures. The ICF International IRB adheres to protection of human subjects regulations provided by the US Department of Health and Human Services (45 CFR 46).
References
Footnotes
Funding This work was supported by the National Institute on Drug Abuse (T32DA007292) and the National Institute of Mental Health (T32MH096724).
Competing interests None declared.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.