Income inequalities and self-reported maternal health status: cross-sectional national survey
Dr S Petrou, National Perinatal Epidemiology Unit, University of Oxford, Old Road Campus, Old Road, Headington, Oxford OX3 7LF, UK. Email email@example.com
The objective of this study was to examine income inequalities in the self-reported health status of women during the postnatal period. We analysed data from a cohort of 18 523 birth mothers of children who participated in the Millennium Cohort Study. Data on income and self-reported health status were collected during face-to-face interviews conducted at 9 months postpartum. Total take-home household income from employment, government and other sources was estimated and equivalised to reflect household composition, while self-reported health status was converted into a dichotomous measure. Complex survey logistic regression models were used to explore the association between equivalised household income and fair or poor self-reported health status. Compared with mothers with equivalised household income in the first quintile (bottom 20%) of the income distribution, mothers in the third (OR 0.69; 95% CI 0.59–0.81), fourth (OR 0.43; 95% CI 0.38–0.50) and fifth (OR 0.32; 95% CI 0.27–0.37) quintiles had a decreased likelihood of reporting fair or poor health status (P < 0.001). However, following adjustment for other predictors of postnatal health status, only mothers with equivalised household income in the fifth quintile (top 20%) had a decreased likelihood of reporting fair or poor health status (OR 0.72; 95% CI 0.58–0.90; P= 0.004). We conclude that this study provides support for the existence of an income gradient for postnatal health status. Efforts to reduce income inequalities in adverse maternal health outcomes are likely to require macro and microeconomic initiatives.
Despite national1 and international2 initiatives to reduce socio-economic inequalities in adverse maternal outcomes, comparatively few studies have focused specifically on socio-economic differences in maternal health. Data from several countries at varying stages of economic development have showed an inverse association between socio-economic status and maternal mortality.3 However, the reasons for this association are not well understood. Discrepancies exist between women of varying socio-economic circumstances in their uptake of antenatal and delivery services, both in developed and in developing countries,4–7 but this is unlikely to be the sole explanation. Rather, it is far more plausible that absolute and relative poverty have wide-ranging effects on maternal mortality exerted through complex biopsychosocial pathways.
The association between socio-economic status and maternal morbidity has been far harder to assess, partly as a consequence of a paucity of adequate data sources. A limited number of observational studies conducted in developed countries have demonstrated that socio-economic disadvantage is associated with an increased risk of neural tube defect-affected pregnancy,8 backache,9 headache,9 migraine,10 anxiety10 and depression11 during the postnatal period. One study examined the association of individual household income and state income inequality with the self-reported health of a nationally representative random sample of American women with young children.12 The study demonstrated that both low individual income and high statewide income inequality are significantly associated with fair or poor self-reported health. However, a number of important socio-economic variables that are known to be associated with maternal health, including social class, cigarette smoking status and the employment status of the woman and her partner,3,6–11 were excluded from the authors’ multivariate analyses.
A fuller understanding of the magnitude of socio-economic inequalities in adverse maternal health outcomes might inform government policies in this area and, in particular, help identify areas where government efforts on tackling health inequalities might be targeted. Thus, the purpose of this study was to examine income inequalities in the self-reported health status of women during the postnatal period. We focused on women with young children because, uniquely in several respects, the postnatal period offers health and social care professionals opportunities to intervene to influence lifestyle and behavioural factors. Furthermore, there is a recognition that early intervention in the most vulnerable mothers can prevent deleterious consequences in their subsequent health and wellbeing and foster the cognitive and social–emotional development of the child throughout childhood and, potentially, through adulthood.13
Data from the Millennium Cohort Study (MCS) formed the basis of the investigation.14 The MCS is the fourth of the UKs national longitudinal birth cohort studies that are following large samples of individuals through the course of their lives. The first MCS survey was carried out between September 2000 and November 2001. The sample design allowed for disproportionate representation of families living in areas of child poverty, in the smaller countries of the UK and in areas of England with high ethnic minority populations. Parents of the children were recruited from child benefit records after being given a chance to opt out by post, telephone or on the doorstep. The survey included 18 553 families, which after allowance for 246 sets of twins and 10 sets of triplets, amounted to 18 819 children in the cohort. A total of 18 553 face-to-face interviews were given, almost entirely by mothers, when the children were aged 9 months. The first survey recorded the circumstances of pregnancy and birth, as well as those of the all-important early months of life, and the social and economic background of the family into which the children have been born. Further details about the MCS, its origins, objectives, sampling procedures, content, methodology and response rates are reported elsewhere.14–16
As part of the face-to-face interviews conducted during the first survey, the child’s principal carer was asked a series of questions using computer-assisted techniques about their own state of health and its impact upon their life. This included a question that asked the principal caregiver to categorise their general state of health as excellent, good, fair or poor. Self-assessed health status has been shown to have high internal consistency, construct validity and reliability and is generally considered to be a good predictor of morbidity and mortality.17,18 For the purpose of this study, we converted the measure of self-assessed health status into a dichotomous variable by dividing the sample into those whose health status was reported as excellent or good and those whose health status was reported as fair or poor.
Principal caregivers were also asked a series of questions about their personal and household income from employment, government and other sources. They were then asked to locate total take-home household income from all these earnings and sources after tax and other deductions on one of two sets of 18 annual household income bands; one for couple households and one for lone parent households. The two sets of income bands were constructed to show approximately equivalent purchasing power when reading from one scale to another scale; couple household bands approximating at 1.6 times the equivalent band for a lone parent household.16 The midpoint for the selected income band was calculated and then equivalised for each principal caregiver using the formula recommended by Gravelle and Sutton19 to reflect household composition (equivalised income = household income/√[adults + 0.5 × children]).20 Equivalised household income was then categorised into quintiles using data from the entire cohort that ranked households from the poorest to the most affluent (with the poorest quintile representing the baseline in subsequent analyses).
All analyses were restricted to the 18 523 birth (biological) mothers of MCS children who were interviewed in their role as principal caregivers (representing 99.8% of all principal caregivers). Complex survey logistic regression models were used to explore the association between equivalised household income and fair or poor self-reported health status. The regression analyses incorporated sampling weights that reflected the probability of a household being selected within one of the MCSs primary sampling units (398 UK electoral wards). Univariate and multivariate methods were used to obtain unadjusted and adjusted odds ratios and their respective confidence intervals. The multivariate models adjusted for a number of factors considered a priori as potentially predictive of postnatal health status on the basis of a review of the existing empirical evidence in the broader literature.5,12,21,22 This included categorical variables for maternal age, occupational social class,23 marital status, educational status, partner’s educational status, ethnicity, employment status, partner’s employment status, smoking status, income inequality within region of residence and self-perceived neighbourhood deprivation. No automatic selection method was used to retain independent variables in the multivariate models. Exponential increase in the adjusted odds ratios for the ordered categorical variables was tested using the adjusted Wald test for linear trend on the logarithmic scale. Linear increase in the adjusted odds ratios was tested by regressing them onto a linear score variable with unit increase for each ordered category and weighting by the inverse of variance of the adjusted odds. Separate models were also constructed for women with and without self-identified longstanding illness, disability or infirmity following initial analyses which suggested that a longstanding limiting illness, disability or infirmity is highly predictive of fair or poor self-reported health status (unadjusted OR 9.93; 95% CI 8.70–11.33). We chose a significance level of 0.05 (two tailed). All analyses were performed using STATA 9SE (StataCorp LP, College Station, TX, USA).
Of the 18 523 birth mothers of MCS children who were interviewed in their role as principal caregivers, our analyses were based on 17 870 women (96.5%) who reported their self-assessed health status. There were no significant differences in the demographic, socio-economic and clinical characteristics of the women who did and did not report their self-assessed health status.
The unadjusted and adjusted results of the complex survey logistic regression modelling for the entire analysed sample are given in Table 1. Univariate analyses showed that, when compared with mothers with equivalised household income in the first quintile of the income distribution, mothers with equivalised household income in the third (OR 0.69; 95% CI 0.59–0.81), fourth (OR 0.43; 95% CI 0.38–0.50) and fifth (OR 0.32; 95% CI 0.27–0.37) quintiles had a decreased likelihood of reporting fair or poor health status (P < 0.001). However, following adjustment for other independent predictors of postnatal health status, only mothers with equivalised household income in the fifth quintile (top 20%) of the income distribution had a decreased likelihood of reporting fair or poor health status (OR 0.72; 95% CI 0.58–0.90; P= 0.004). The multivariate analyses also showed that being aged 36 years or older (P < 0.001), being of semi-routine or routine social class (P= 0.013), being of Indian (P < 0.001), Pakistani (P < 0.001), Black Caribbean (P < 0.001) or other (P= 0.002) ethnic origin, past but not current employment status (P < 0.001), having an economically inactive partner (P= 0.035), being a smoker (P < 0.001) and residing in a neighbourhood of medium (P= 0.001) or high (P < 0.001) deprivation was associated with an increased likelihood of reporting fair or poor health status. The adjusted Wald test showed a significant exponential decrease in the likelihood of reporting fair or poor health status with increasing levels of equivalised household income and educational status of the mother. It also showed a significant exponential increase in the likelihood of reporting fair or poor health status with decreasing levels of occupational social class (managerial or professional social class defined as the referent) and with increasing levels of age and regional Gini coefficient. A significant linear trend of increasing likelihood of reporting fair or poor health status was found for increasing maternal smoking and neighbourhood deprivation categories. Broadly, similar results were showed by the models constructed for women with and without self-identified longstanding illness, disability or infirmity (data available upon request).
Table 1. Results of complex survey logistic regression exploring the association between equivalised household income and fair or poor self-reported health status, all birth mothers (n= 17 870)
|Equivalised household income (quintile)||Second||3351 (24.59)||1.02 (0.89–1.18)||0.746||1.15 (0.99–1.35)||0.068|
|Third||3213 (18.11)||0.69 (0.59–0.81)||<0.001||1.06 (0.88–1.29)||0.535|
|Fourth||3069 (12.48)||0.43 (0.38–0.50)||<0.001||0.83 (0.68–1.02)||0.071|
|Fifth||3502 (9.11)||0.32 (0.27–0.37)||<0.001||0.72 (0.58–0.90)||0.004|
|Not informed||1519 (19.03)||0.67 (0.56–0.81)||<0.001||0.93 (0.76–1.14)||0.482|
|First (bottom 20%)***||3216 (23.41)||1.00***|| ||1.00***|| |
|Age (years)||14–18||545 (21.47)||1.76 (1.37–2.26)||<0.001||0.91 (0.70–1.20)||0.522|
|19–24||4610 (20.20)||1.44 (1.29–1.62)||<0.001||0.90 (0.80–1.02)||0.104|
|≥36||2660 (18.08)||1.14 (1.00–1.29)||0.048||1.30 (1.14–1.49)||<0.001|
|25–35***||10 055 (16.12)||1.00***|| ||1.00***|| |
|Occupational social class****||Intermediate||3004 (23.65)||1.19 (1.02–1.38)||0.025||0.92 (0.78–1.08)||0.321|
|Small employer/self-employed||628 (13.06)||1.23 (0.92–1.63)||0.156||0.95 (0.71–1.28)||0.734|
|Lower supervisory/technical||973 (22.10)||2.03 (1.64–2.50)||<0.001||1.23 (0.99–1.53)||0.061|
|Semi-routine/routine||6590 (21.70)||2.27 (1.99–2.59)||<0.001||1.22 (1.04–1.43)||0.013|
|Not informed||2057 (24.55)||2.61 (2.23–3.05)||<0.001||0.98 (0.62–1.56)||0.931|
|Managerial/professional***||4618 (11.00)||1.00***|| ||1.00***|| |
|Marital status||Single||6082 (20.90)||1.56 (1.40–1.73)||<0.001||1.04 (0.92–1.17)||0.530|
|Separated/divorced/widowed||1258 (22.34)||1.76 (1.47–2.09)||<0.001||1.13 (0.94–1.36)||0.200|
|Ever married***||10 530 (15.18)||1.00***|| ||1.00***|| |
|Educational status||Higher/further||4271 (10.58)||0.54 (0.47–0.62)||<0.001||0.86 (0.74–1.01)||0.059|
|A/As/S level or equivalent||1670 (14.37)||0.77 (0.64–0.91)||0.003||1.01 (0.85–1.21)||0.865|
|None of the above||3511 (25.98)||1.65 (1.48–1.83)||<0.001||1.09 (0.97–1.22)||0.155|
|O-level/GCE or equivalent***||8418 (18.37)||1.00***|| ||1.00***|| |
|Partner’s educational status||Higher/further||3543 (10.44)||0.56 (0.48–0.65)||<0.001||0.86 (0.74–1.01)||0.064|
|A/As/S level or equivalent||956 (12.97)||0.67 (0.53–0.85)||0.001||1.02 (0.86–1.22)||0.778|
|None of the above||7823 (21.85)||1.40 (1.26–1.55)||<0.001||1.08 (0.96–1.21)||0.177|
|O-level/GCE or equivalent***||5548 (17.07)||1.00***|| ||1.00***|| |
|Ethnicity||Mixed||187 (23.53)||1.36 (0.94–1.98)||0.105||1.16 (0.82–1.64)||0.408|
|Indian||443 (20.54)||1.49 (1.07–2.07)||0.019||1.92 (1.33–2.76)||<0.001|
|Pakistani||842 (25.42)||1.94 (1.56–2.40)||<0.001||1.81 (1.40–2.33)||<0.001|
|Bangladeshi||344 (22.09)||1.51 (1.05–2.16)||0.025||1.40 (0.98–2.01)||0.066|
|Black Caribbean||262 (30.15)||2.23 (1.65–3.02)||<0.001||1.97 (1.36–2.85)||<0.001|
|Black African||372 (20.16)||1.54 (1.12–2.12)||0.008||1.44 (0.93–2.22)||0.099|
|Other||384 (23.70)||1.63 (1.22–2.19)||0.001||1.68 (1.21–2.33)||0.002|
|White***||15 036 (16.49)||1.00***|| ||1.00***|| |
|Employment status||On leave||444 (18.24)||1.54 (1.13–2.09)||0.006||1.25 (0.90–1.72)||0.177|
|Past paid job||7620 (21.44)||1.83 (1.66–2.02)||<0.001||1.28 (1.14–1.43)||<0.001|
|No paid job ever||1862 (25.08)||2.46 (2.12–2.86)||<0.001||1.40 (0.86–2.27)||0.174|
|Economically active***||7944 (12.02)||1.00***|| ||1.00***|| |
|Partner working||No||5194 (24.28)||1.98 (1.80–2.18)||<0.001||1.15 (.01–1.32)||0.035|
|Yes***||12 676 (14.90)||1.00***|| ||1.00***|| |
|Current smoking status (cigarettes per day)||1–10||3304 (20.85)||1.65 (1.46–1.85)||<0.001||1.34 (1.18–1.51)||<0.001|
|11–19||1045 (27.85)||2.55 (2.14–3.04)||<0.001||1.90 (1.60–2.25)||<0.001|
|≥20||1010 (32.48)||3.04 (2.57–3.60)||<0.001||2.10 (1.74–2.53)||<0.001|
|Nonsmoker***||12 511 (14.72)||1.00***|| ||1.00***|| |
|Gini coefficient by region (tertiles)*****||High||6504 (18.19)||1.09 (0.92–1.28)||0.318||0.95 (0.83–1.09)||0.496|
|Medium||5818 (17.65)||1.00 (0.85–1.19)||0.957||0.92 (0.79–1.06)||0.228|
|Low***||5548 (16.94)||1.00***|| ||1.00***|| |
|Neighbourhood deprivation score******||Medium (3–4)||4786 (20.58)||1.51 (1.35–1.68)||<0.001||1.22 (1.09–1.36)||0.001|
|High (5–6)||1426 (31.00)||3.07 (2.65–3.55)||<0.001||2.11 (1.82–2.46)||<0.001|
|Low (0–2)***||11 658 (14.78)||1.00***|| ||1.00***|| |
Discussion and conclusions
The main finding of our analyses of this cross-sectional national survey of mothers of young children was that high individual income, as indicated by equivalised household income in the highest quintile of the income distribution, was associated with a reduced likelihood of reporting fair or poor health status. When compared with mothers in the poorest 20% of households, mothers residing in the richest 20% of households were 28%, on average, less likely to report fair or poor health status. Although mothers residing in households in the second, third and fourth quintiles of the income distribution were not significantly less likely to report fair or poor health status than those in the first quintile, the trend analysis for adjusted odds ratios showed a significant exponential decrease across successively increasing quintiles.
Because of the strong relation between income and some of the covariates included in the multivariate models, it is possible that the effect of income on postnatal health status has been over-adjusted leading to some underestimation of the strength of the relation. For example, partial collinearity was identified between income and maternal age (Pearson’s correlation coefficient 0.16), which is likely to explain the change in direction of age-related effects between the univariate and multivariate models. Weaker collinearity was also identified between income and educational status and between income and employment status. To account for collinearity between covariates, we conducted additional multivariate logistic regression models that omitted covariates closely related to income. The gradient for the logarithms of the adjusted odds of reporting fair or poor health status increased in significance (data available upon request) suggesting that the estimated effects of income that we present can be interpreted as probable lower bounds of its total effect on postnatal health status.
Our study was based on data from a large, nationally representative sample of mothers of young children across all four countries of the UK. The data set also contained a rich set of demographic, socio-economic and contextual covariates that allowed us to test alternative model specifications for the relationship between income and postnatal health status. There are two broad caveats to the study results, however, that should be borne in mind by readers. First, postnatal health status was measured at 9 months postpartum and, consequently, might have under-estimated the impact of income-related maternal physical and emotional health problems, such as postnatal depression, that have been shown to dissipate after 6 months postpartum.9,24,25 Second, in common with cross-sectional studies of this type, we are unable to say with any degree of certainty that the observed associations are causal or due to reverse causality, perhaps through ill health resulting in some women deciding not to return to work following childbirth, which might in turn lead to lower household income. Further research we are planning will involve analysing panel data from successive sweeps of the MCS with the view to investigating the complex causal pathways between income and adverse maternal and child health outcomes.
In conclusion, this study provides support for the existence of an income gradient for postnatal health status. As the present government intends, work towards reducing income inequalities in adverse maternal health outcomes must be addressed with urgency. This should include targeting low-income mothers with specific interventions, aimed at health promotion and protection, which are known to be cost-effective. In addition, however, the reduction of maternal health inequalities will almost certainly require broader macroeconomic initiatives, such as tax and benefit changes, aimed at reducing relative poverty.
We would like to thank the participants of the MCS who provided data for this study.
The MCS is funded by the Economic and Social Research Council and by a consortium of government departments led by the Office for National Statistics. The National Perinatal Epidemiology Unit, University of Oxford, is funded by the Department of Health in England. Dr Petrou is supported by a UK Medical Research Council Senior Non-Clinical Research Fellowship. The views contained in this paper are held by the authors and not necessarily by the funding bodies.
Ethical approval was not required for this secondary analysis of anonymised, publicly available data.