We carried out a survey among adults aged 16 years and above who were able to read in English, had access to the internet, and had access to a device with which to complete the survey. We invited participation through Twitter, email, and word of mouth and administered the survey through the Qualtrics online platform. Individuals were encouraged not to participate unless they were able to do so in a private place where they would not be disturbed or their answers would not be read by another person. There was no upper age limit. Based on study group members' locations, eight countries were targeted for recruitment: the United Kingdom, Bangladesh, India, Zambia, Ethiopia, South Africa, Kenya, and Uganda. However, people living in any country were eligible to participate. The survey website presented prospective participants with details of support organisations for IPFV and mental health before the survey questions. In Kenya, trained data collectors verbally informed eligible community members about the survey and offered participants a mobile device, if they did not have one, on which to complete the survey. The full survey including accompanying information and item wordings is included in the Supplementary Material .

Ethical approval was provided by the King’s College London Psychiatry, Nursing and Midwifery research ethics subcommittee (reference: HR-19/20-19295).

We analysed data using Stata 17 (36). We reported the prevalence of CMDs, and partner and family relationship difficulties by all covariates, using counts and proportions. To describe the data (see Table 1 ), we dichotomised partner and family difficulties into few/some difficulties and moderate/severe difficulties—we handled these variables as continuous variables for the modelling described below. We dichotomised all items on pandemic-related change into no change/improved and worsened, and we dichotomised perceived social support into low and high, at the median, which was 10 (see Table 1 ).

As described under the Measurement section, the outcome for analysis was (dichotomous) CMD. Assumptions of odds ratios as measures of relative risk may be violated where the outcome is very common (e.g. greater than 10%). Given the high prevalence of CMDs in our data, we used robust (or modified) Poisson regression to model the outcome, as recommended by Zou (37). To evaluate the association of partner difficulties and family difficulties with CMD, we used robust Poisson regressions estimating the risk ratio (RR) for the association of a unit change in each exposure variable with CMD. After inspecting unadjusted estimates including only the exposures and CMD (model I), we included basic sociodemographic variables (age, gender, and continent of residence: model II). We then included socioeconomic and social support variables (employment status, educational attainment, persons per household room, and perceived social support score: model III) and then pandemic-related change (model IV). Finally, we included general difficulties for each exposure (model V): for partner relationship difficulties, we adjusted for non-partner relationship difficulties; for family relationship difficulties, we adjusted for non-family relationship difficulties; and for home safety, we adjusted for perceived safety outside the home environment. Lastly, to evaluate associations stratified by gender and continent of residence, we estimated models including a multiplicative interaction term for each of these variables. We produced post-estimation fitted estimates for the association in each group and tested for significant heterogeneity using likelihood ratio tests.

The highest prevalence of CMD was in Europe (46.15%), then Asia (40.16%), the Americas (33.33%), and Africa (22.17%, see Table 1 ). The prevalence of CMDs among those with low perceived social support was 40.56%, and among those with high social support, it was 18.88%. Among those reporting difficulties in non-partner or non-family relationships, CMD prevalence was 49.65%, and in those not reporting these difficulties, it was 23.71%.

In the unadjusted model, there was an association of greater partner difficulties with CMD (RR = 1.50, 95% CI: 1.30, 1.72, p < 0.001, Table 2 ). In comparison, the association of greater family difficulties with CMD was closer to null but remained statistically significant (RR = 1.22, 95% CI: 1.04, 1.42, p = 0.012). The association of family difficulties with CMD remained similar after all adjustments but was no longer statistically significant (RR = 1.18, 95% CI: 0.97, 1.44, p = 0.89). There was some attenuation of the association of partner difficulties with CMD after all adjustments to RR = 1.30 (95% CI: 1.06, 1.60, p = 0.01).

On stratifying by gender, in the final models, both partner and family difficulties were significantly associated with CMD in women [partner difficulties: RR = 1.37 (95% CI: 1.09, 1.72, p = 0.003); family difficulties: RR = 1.31 (95% CI: 1.05, 1.65, p = 0.023)]. However, in men, the estimates did not reach statistical significance and were closer to null. We found no evidence for significant heterogeneity in the association of either partner or family difficulties with CMD between men and women.

On stratifying by continent, there was statistical evidence for the association of partner difficulties with CMD in Europe [RR = 1.34 (95% CI: 1.01, 1.78, p = 0.033)] and Africa (RR = 1.55, 95% CI: 1.14, 2.10, p = 0.005), but not in Asia. The estimates for the association of partner difficulties with CMD were non-significant and very close to null for Asia (RR = 1.01, 95% CI: 0.67, 1.51) and statistically significant for Europe (RR = 1.34, 95% CI: 1.01, 1.78) and Africa (RR = 1.55, 95% CI: 1.14, 2.10), while estimates were not produced for the Americas owing to small numbers of respondents in this group. There was statistical evidence for interaction by continent in the association of partner difficulties with CMD (p = 0.007), but not of family difficulties with CMD (p = 0.7988).

Self-reported partner difficulties and family difficulties were associated with CMDs after accounting for possible confounders, including pandemic-related change in each exposure variable. Estimates for the association of partner difficulties and family difficulties, with CMD were stronger in women than in men, but we did not find statistical evidence for interaction by gender. While there was statistical evidence for interaction in the association of partner difficulties with CMD by continent, this was based on small numbers of respondents from the Americas, so this result should be treated with caution.

Our results suggest that partner difficulties and family difficulties are associated with CMD and are not attributable to differences in individual sociodemographic characteristics, socioeconomic status, or perceived social support. Point estimates suggested that the relationship between family difficulties and CMDs may be somewhat weaker in men than women, although we did not find statistical evidence for interaction, and this requires further study. Across the whole sample, the relationship of partner difficulties and family difficulties with CMDs appeared to persist, irrespective of the experience of change in difficulties since the beginning of the pandemic. Based on our analysis, the association of partner difficulties and family difficulties with CMD is unlikely to be due to the non-specificity of the markers themselves. That is, it is unlikely that the relationship between partner difficulties and CMDs is explained by difficulties in wider relationships (family, friends, and neighbours), and it is unlikely that the relationship between family difficulties and CMDs is explained by difficulties in wider relationships (partner, friends, and neighbours), as we adjusted for these factors in our analyses. This adds confidence that our estimates reflect the impact of difficulties in each specific relationship domain, rather than non-partner or family-related factors, which may result in endorsing these items.

We found that partner difficulties and family difficulties were associated with CMDs across diverse geographical settings, including in LMICs, which largely enforced less restrictive lockdowns than HICs (38–41). This could indicate that the degree of restrictions at a policy level had a limited impact on the mental health impact of IPV in contrast to other factors such as economic insecurity or fears of infection, which may have varied less across countries, or that the effect of restrictions was partly explained by other factors, such as local variation in compliance with these restrictions (42). Our finding of a stronger association of partner difficulties with CMDs in Europe and Africa, compared to Asia, warrants further research in large representative samples.

Our results are consistent with the effect of partner difficulties and family difficulties on CMDs during the pandemic. Mechanisms could include prolonged anxiety and psychosocial stress (for example, about finances and livelihoods) and psychological mechanisms reflecting a sense of entrapment in abusive relationships. Stronger point estimates for the association of family difficulties with CMDs in women compared to men are in line with previous research suggesting that family demands have more damaging effects on women’s mental health, compared to men (43, 44), but again, we did not find statistical evidence for interaction, and this hypothesis should be tested in future studies.

The collection of research data on IPFV during the pandemic shifted in large part to remote methods (45). Following international ethical guidance, we employed self-reported partner difficulties, family difficulties, and perceived home safety as proxy markers for IPFV. Several studies have demonstrated a relationship between (directly measured) IPV and CMD during the pandemic, although many were conducted in HICs (4, 8, 26, 27, 46–49). There have been limited attempts to consider the impact of pre-pandemic IPV on CMDs during COVID-19. By accounting for pre-pandemic IPV models, we increase confidence that associations of partner difficulties and family difficulties with CMD are not attributable to pre-pandemic difficulties in relationships.

Although some studies reported increased family violence (FV) during the pandemic (50, 51), we present the first evidence (to our knowledge) that family relationship difficulties are associated with CMDs in a pandemic context. Previous research found that dissatisfaction with family relationships was strongly associated with psychological distress (scoring moderate or high on the 10-item Kessler Psychological Distress Scale) in New Zealand during the pandemic (42). Although this study also measured the experience of family violence, associations of family violence with distress were not presented.

A limited number of studies has examined the mental health impact of partner difficulties in the pandemic context. Although partner difficulties are associated with CMDs (19, 20, 52–54), evidence for this relationship during COVID-19 is limited, despite emerging evidence on the prevalence of IPV during this period (14, 15).

Our analysis of this cross-sectional survey cannot determine the temporal relationship of partner difficulties or family difficulties with CMDs. Our sample was not representative of all people in the countries sampled, and respondents with access to devices, the internet, and social media are likely to represent a source population with higher levels of educational attainment and socioeconomic status. Both factors influence the risk of IPFV and CMD, as well as the likelihood of adverse impacts of pandemic restrictions. Our findings may therefore under-estimate the prevalence of both IPFV and CMDs. Survey response was variable between continents surveyed—for example, a small number of respondents in the Americas rendered association estimates for partner difficulties in this region unstable. We measured CMDs using an instrument designed to define community prevalence, and our results may not generalise to the association of IPFV with clinically ascertained depression. While variation in the prevalence of CMDs, for example, a greater prevalence in Europe compared to Africa, ascertained in our study is consistent with other studies carried out during the pandemic (55), we cannot discount the possibility of variable under-ascertainment of CMDs across geographical settings, which would introduce bias.

Items measuring partner and family difficulties did not specify forms of IPFV (physical, sexual, and emotional) in detail to safeguard respondents. We were therefore unable to consider the correspondence of our exposure data with direct self-reported measures of IPFV, such as those collected using the Conflict Tactics Scales and other measures (56). We also did not measure coercive control, so we could not disaggregate the presence of coercive control within the group reporting partner difficulties or family difficulties. The relationship that we identified between proxies for IPFV and CMDs during the pandemic may have been confounded by the availability of mental health support and services, which we did not measure and which varies between geographical settings (19). In this paper, we employed partner and family relationship difficulties as indirect indicators of IPFV. Although meta-analytic evidence suggests that self-reported relationship satisfaction is strongly negatively correlated with partner violence (57), this should be considered when generalising our results from partner and family difficulties to IPFV.