Health and mortality strand abstracts


Health - Determinants of childhood health & mortality in developing countries: Monday 12 September 1:30pm 

Maternal schooling, child mortality, and pathways of influence: Evidence from a quasi-experiment in Ethiopia, Malawi, and Uganda
Liliana Andriano, Christiaan Monden, Department of Sociology, University of Oxford 

Since the 1980s the demographic literature has suggested that maternal schooling plays a key role in determining children’s survival chances in low and middle income countries. However, no studies have successfully distinguished between the causal and non-causal relationship between maternal education and child survival and to overcome the endogeneity problems inherent in this relationship. In order to identify the causal effect of maternal education on child survival we explore exogenous variation in maternal education induced by schooling reforms in the second half of the 1990s in Ethiopia, Malawi, and Uganda, which introduced Universal Primary Education. We use a fuzzy regression discontinuity design and the Demographic and Health Surveys data to test if increased schooling improved children’s survival chances across all three countries. We also seek to explore which are the pathways of influence explaining the effect of maternal education on child survival among socioeconomic status, attitudes towards modern health services, personal illness control, health knowledge, empowerment, and environment contamination and characteristics. Additionally, we examine if the intervening mechanisms are common across contexts or country-specific. 

Global deterministic patterns of socioeconomic determinants of neonatal mortality at the country level
Filippo Temporin, The London School of Economics and Political Science 

While determinants of infant and child mortality have received broad attention in literature, socioeconomic factors impacting neonatal mortality have only recently interested demographers and health researchers. The decrease in under-five mortality observed in developing countries in the last decades has not been followed by an equal drop in neonatal deaths, which has accounted for a gradually higher proportion of under-five mortality, from 37% in 1990 to 44% in 2013 globally (UNICEF, 2015), leading to a recent call for action about the hidden problem of neonatal mortality aimed at researcher and policy makers. This study aims at shedding light on the deterministic patterns of socioeconomic determinants of neonatal mortality, highlighting in particular differences in respect to post-neonatal deaths, by means of random coefficient regressions. By making use of DHSStatCompiler and World Bank data regarding 39 developing countries, this study assesses the influence of deprivation, income inequality, education, and other determinants on newborn’s mortality. Among the most significant findings, both poverty and economic inequality have a lower gradient of association with neonatal mortality compared to post-neonatal one. Moreover, their interaction term highlights how the effect of deprivation is significantly worsened in countries with greater income inequality. These findings have social policies implications in the fight for reducing neonatal mortality, and can partly explain the stagnation in neonatal mortality rates observed in the last decades. 

Birth order and sibling composition contributions to sex-differentials in child mortality
Ridhi Kashyap 1, 2, Kieron Barclay 3, 1 University of Oxford, 2 Max Planck Institute for Demographic Research, 3 London School of Economics and Political Science 

Excess female child mortality, linked to strong son preference and the sex-biased allocation of familial resources that these preferences give rise to, has been documented in the South Asia, most notably India and Bangladesh. This literature indicates that girls’ excess mortality risks in these settings are patterned by birth order and gender composition of the sibling group, with later-born girls with two or more sisters faring the worst. Younger cohorts in contexts with son preference however are increasingly born into smaller families and in families that may have practiced sex-selective abortion to facilitate desired family size and gender composition. Does falling fertility and the uptake of prenatal sex-selection diminish patterns of selective excess mortality for girl? We investigate how selective mortality for girls by birth order and sibling composition change across birth cohorts using a pooled dataset of over 7 million births from the Demographic and Health Survey data samples including 44 countries. We estimate post-neonatal mortality hazards featuring gender and birth year; gender, birth year and birth order; and gender, birth year and sibling composition interactions. By comparing contexts where son preference has been known to impact demographic outcomes (e.g India, Bangladesh, Pakistan, Azerbaijan, Armenia, Nigeria) and others where it has not (e.g sub-Saharan Africa, Latin America), we compare how mortality patterns differ across these settings. Preliminary results indicate that birth order effects continue to differ by gender in South Asia but less clearly in other parts of the world with higher birth order boys (especially 3+) showing lower mortality rates than earlier-born boys. Later birth-order girls, on the other hand, show higher mortality than earlier-born girls. 

Economic Returns to Investment in Family Planning in India and China
Sabu S. Padmadas 1, Bohua Li 2, Min Qin 1, 2, Fiifi Amoako Johnson 1, 1 Department of Social Statistics & Demography, University of Southampton,  2 China Population and Development Research Centre, Beijing   

Since the early 1970s, family planning programs in China and India have largely focused on promoting long term reversible and irreversible methods of contraception. However, there is little population research on the economic impact of family planning in India and China. Two inter-related questions are addressed in this research: (i) does the use of post-partum long-term contraception increase the odds of economic opportunities for women in China and India? (ii) does women’s education mediate the relationship between use of long-term contraception and likelihood of post-maternity employment? We use cross-sectional population data from the 2006 China National Population and Family Planning Survey (NPFPS) and the 2005-06 Indian National Family Health Surveys (NFHS). Random intercept multinomial logistic regression models were fitted with women’s economic empowerment as outcome variable defined in terms initiating or returning to employment for cash income. Type of contraceptive method used after the last birth and parity were considered as primary predictors along with relevant sociodemographic, spatial and household variables. The findings show a positive relationship between post-partum long-term contraception and economic opportunities for women in India and China. The effect of sterilization is more accentuated and significant in India which lags behind China in terms of overall contraceptive use coverage and progress in female education. On the other hand, the investments and benefits of long-term contraception seem to accrue in both countries, across women’s reproductive life course, in not only limiting fertility and averting unwanted pregnancies but also boosting family income and reducing poverty. 

Health inequalities across the life course: early life circumstances & health: Tuesday 13 September 09:00am 

An exploration of educational outcomes for children with disabilities
Fiona M Cox, Alan Marshall, University of St Andrews

Evidence from cross-sectional and panel studies, qualitative research and reports drawn primarily from snapshot Government statistics indicate that children with disabilities face particular barriers to achieving success within and beyond education. However to date little or no longitudinal research has been published investigating the causal relationship between disability and education and employment outcomes. This lack has been noted by WHO in their World Report on Disability (2011) which repeatedly calls for more longitudinal research in order to “allow researchers and policy-makers to understand better the dynamics of disability.” Education is often key to future participation in the labour market; the DWP ‘Fulfilling Potential’ report goes so far as to say that “for those who are born with an impairment or a health condition, education and other early life experiences influence the whole of their life chances.” A lack of robust measures and classification issues make it difficult to accurately estimate the numbers of children with a disability, however DWP statistics suggest that around 7% of children in the UK are covered by the Equality Act (approx. 0.9M children). The inclusion of more detailed health questions in the 2011 Scottish Census, along with the Scottish Longitudinal Study’s linkage to education data from ScotXed has created a unique opportunity to conduct research in this area. This paper presents part of a wider project and investigates the influence of childhood disability – and other possible confounding factors such as type of disability, parental disability and socioeconomic status – on educational attainment. 

A complex relationship between ethnicity, socio-economic status and the risk of child obesity/overweight in the UK
Alice Goisis 1, Melissa Martinson 2, Wendy Sigle1 1, 1 London School of Economics and Political Science, 2 University of Washington 

A range of studies report a robust association between family socio-economic status and the prevalence of child obesity as children from poorer backgrounds are, on average, more likely to be obese than children from more advantaged families. However, some recent studies have suggested that the relationship between disadvantage and the prevalence of childhood obesity might be more complex than previously supposed. Studies in the U.S. have shown that for ethnic minority children the income gradient in child overweight/obesity is either non-existent or reversed, suggesting that we should be careful in assuming that higher socioeconomic status is protective (against obesity) for all groups of the population. In this paper, we aim to contribute to this emerging stream of research by analysing these issues in the U.K., where research on this topic has been rather limited so far but where rates of obesity are particularly high for children of ethnic minority parents. 

Qualitative evidence on the relationship between breastfeeding and postnatal depression
Maria Iacovou, Department of Sociology, University of Cambridge 

Postnatal depression is estimated to affect around one in 13 women; as well as posing risks to mothers’ current and future wellbeing, there is increasing evidence that postnatal depression has negative and enduring effects on children’s later cognitive and psychological development. Previous survey-based research has demonstrated a substantially elevated risk of postnatal depression among mothers who intended to breastfeed, but who in the event did not go on to breastfeed their babies (Borra, Iacovou and Sevilla, 2014). In this paper, I use extended interviews with mothers to investigate in depth some of the factors underlying the relationship between breastfeeding and postnatal depression, and to make an initial assessment of the possible efficacy of toolkit-based interventions by healthcare providers. A sample of 50 mothers has been selected, who (a) struggled to varying degrees with breastfeeding, and who (b) experienced a period of low mood (including, but not limited to, a formal diagnosis of postnatal depression) following the birth of their babies. To date, just under half of these mothers have been interviewed, with semi-structured interviews ranging from 25 minutes to over 2 hours in duration. An important finding, which was not anticipated in advance of the research, is that protracted and frightening labours, and in particular induced labour, figured strongly in mothers’ accounts; this finding may have relevance as a diagnostic tool for caregivers, and may also be important in the formulation of clinical protocols. 

The Determinants of Breastfeeding Duration in the UK: A survival Analysis of the Early Postpartum Stage 
Chloe Harvey, University of Southampton   

Over three-quarters of mothers in the United Kingdom discontinue breastfeeding before six weeks postpartum, representing a critical period in which the promotion and support for optimal breastfeeding practices is crucial. In order to target support and intervention for breastfeeding mothers, it is necessary to determine and explore the predictors of breastfeeding duration and identify those most at risk of cessation. Therefore the overarching aim of this research was to investigate the determinants of breastfeeding duration, and in particular the predictors of early cessation within the first three months postpartum. Survival Analysis was conducted on a sample of 7,742 women, surveyed in the Infant Feeding Survey UK (2010), and Cox Proportional Hazards Model was the primary modelling technique. This study produces results which confirm that women who are younger, single, white and of a lower socio-economic status are more predisposed to the risk of cessation of breastfeeding, during the early post-partum period. Individual factors relating to the mother-infant dyad suggest that route of delivery is a direct determinant of breastfeeding cessation in the early postpartum stages, as women reporting a caesarean or assisted birth (with vacuum or forceps) experience a short-term, increased risk of discontinuation compared to mothers who had a natural birth. Predictors relating to the professional health care environment, suggest that not giving birth in a hospital or mid-wife led birth centre, actually contributes to a decreased risk of the termination of breastfeeding during the first three months postpartum. 

Health: Inequalities & local contextual factors. Tuesday 13 September 11:00am 

Unravelling Urban-Rural Health Disparities in England: Context or Composition 
Rebecca Allen, Hill Kulu,  Paul Williamson, University of Liverpool   

Background: Research presents significant health variations by residential contexts. Numerous studies report better health amongst rural populations in comparison to urban, whereas other research depicts the opposite. Upon investigation researchers are faced with a fundamental issue, that there is no universally accepted definition of what constitutes rural. As a result, academics suggest that any health variations reported could potentially be a data artefact. Aims: To investigate health variations in England by residential contexts and the causes of such differences. Further, to examine the sensitivity of results according to differing rural-urban classifications. Methods and Results: Using individual-level data from the 2001 UK census and applying various logistic regression models, utilising multiple rural-urban classifications, we demonstrate significant health variation by residential context. A clear gradient is apparent, with levels of LLTI increasing parallel to growing urbanisation. Unexpectedly however, London residents as a whole possess better than anticipated health, particularly outer London, with individuals holding health expectations similar to those in the most rural locations. Once we control for individual socio-demographic characteristics, variations between residential contexts significantly reduce; however, marked differences remain. The persistence indicates that although health variations are influenced by compositional factors, contextual issues play a significant role. Rural populations continue to have better health than that of urban, supporting the existence of a positive urban-rural health gradient, with the exception of the capital. Further, it is evident that health variations across the continuum are significantly influenced by the classification utilised. Thus, the way in which locations are defined is extremely important, and any investigation must justify objectively the classification practiced, as in this study. 

Neighbourhood context and allostatic load
N Crawford, Institute for Social and Economic Research, University of Essex, United Kingdom   

A number of studies suggest that residing in a deprived neighbourhood is associated with poorer health, independent of individual circumstance. Environment can, however, contribute to health inequalities in a myriad of ways. Interdisciplinary research at the social-biological interface is necessary to better understand the social and biological interactions that contribute to health inequalities. Set within the relative deprivation thesis, this paper models a curvilinear association between neighbourhood context, operationalized as an index of concentration at the extremes (ICE), and allostatic load (AL), a composite measure of health. ICE captures both concentrated advantage and disadvantage on a single scale, thereby recognising the importance of the proportional imbalance between affluence and poverty within a neighbourhood, as proposed by the relative deprivation thesis. This study utilises geocoded data collected at waves 2 and 3 of Understanding Society: the UK Household Longitudinal Survey. Understanding Society represents a unique opportunity to study the interplays between biology, place and individual circumstance in a nationally representative sample. Multilevel models tested the association between neighbourhood context and AL with cross-level interactions between individual level and neighbourhood resource introduced to model the relative deprivation hypothesis. AL was computed as a sum score across 9 biomarkers with a higher score indicative of poorer health. Results operated in the expected direction with poorer individuals residing in neighbourhoods with greater heterogeneity reporting increased AL scores. Findings support the relative deprivation thesis; residing in an area characterised by greater inequality is more detrimental for disadvantaged individuals than those residing in more homogenous areas. 

Integrating small area census data to investigate social change in Scotland 1981-2011
Daniel J Exeter 1, Jinfeng Zhao 1, Zhiqiang Feng 2, Paul Norman  3,  1 School of Population Health, The University of Auckland,2 School of Geosciences, University of Edinburgh,  3 School of Geography, University of Leeds 

Background: Between 1981 and 2011, Scotland’s population increased by approximately 5%, from 5.035 million to 5.26 million. The analysis of social change among smaller spatial units is more problematic, as they are prone to boundary changes over time. This is a considerable issue in social epidemiology where reliable population numerators and denominators are required to investigate disease patterns over time. In this study, we describe the development of a consistent small area geography to reliably analyse health and social change in Scotland between 1981 and 2011, without the need for population estimation techniques. Methods: Using Geographical Information Systems (GIS) we merged the 2001 and 2011 OA boundary files and used AddressPoint household location data to distinguish between erroneous ‘sliver’ polygons that arose due to differences in digitising from those actual boundary changes that occurred between 2001 and 2011. We assumed that any polygon containing two households or fewer were sliver polygons and removed. This resulted in a ‘Merged 2001 and 2011 OA’ boundary file, comprising 36,921 unique zones. We used a similar merging approach to integrate the ‘Merged 1981 ED and 1991 OAs’ with the Merged 2001 and 2011 OAs to create the final CATT2011 boundaries. Results: Initially the merging process created 8364 unique boundaries that enabled the small area analysis of census data from 1981, 1991, 2001 and 2011 reliably. However, there were 30 zones with more than 50,000 residents per zone. For these areas we relaxed the sliver polygon rule and considered areas with 5 or fewer households as slivers, reducing the data set to 8556, with a marginal effect on the mean and standard deviation. Many of the remaining zones with large populations were those in rural areas with many sliver polygons according to our criteria. Using ArcGIS Districting software, we visually modified these remaining areas with populations exceeding 20,000 by considering the boundaries of ‘Merged 1981 ED and 1991 OA’ and ‘Merged 2001 and 2011 OA’ and slivers where there are fewer number of address points. This resulted in 8611 CATTs in total, with populations ranging between 51 and 19,438, while the mean was approximately 615, depending on the census year of interest. Conclusions: This research provides a robust small area geographical boundary file to analyse social change. We are using these to examine widening health and social inequalities in Scotland between 1981 and 2011. 

The same but different: Social correlates of local area mortality in Australia
Jon Anson, Dept. of Social Work, Ben-Gurion University of the Negev, Israel

Over the past half century, demographers have spent considerable effort documenting the effects of resources on mortality, at various levels of aggregation, from the individual to the nation state. Nonetheless, the concept of social inequality, and in particular, the nature of the resources which give the rich their advantage, have been relatively unexplored. In the present analysis we consider mortality risks in local, small area populations in Australia, and we relate these risks, to three dimensions of capital (wealth generating resources) and of family structure. In this way, we propose to identify what, in particular, gives certain people an advantage over others, as well as to propose a metric for comparing Australian social inequality with that to be found in other countries. Our main findings are: 1. Net of local area social characteristics, remote areas of the country are disadvantaged 2. Indigenous communities have particularly high mortality, even after allowing for their very low levels of wealth and other resources 3. Of the three forms of capital identified by Bourdieu, cultural capital has the largest effect in terms of reducing mortality. 4. Stable family structures, in particular marriage as a universal institution, is no less important than the level of cultural capital in reducing mortality 5. Male mortality is consistently higher than female mortality, and the effects of social conditions are greater on male than on female mortality. 6. Despite the significant effects of social conditions, it is important to note the relatively high homogeneity of mortality patterns, pointing to the critical effects of state social and health services in reducing mortality for all sections of the population. We conclude with a short discussion of possible implications of these findings. 

Health:  Health inequalities in a national/regional context. Tuesday 13 September 1:30pm 

Cross-national comparisons of adolescents' psycho-social health: The role of measurement invariance 
Nichola Shackleton 1, 2, John Jerrim 3, 1 COMPASS Research Centre, University of Auckland, 2 Department of Statistics, University of Auckland,, 3  UCL Institute of Education, University College London   

We aimed to use the European School Survey Project on Alcohol and Other Drugs (ESPAD 2007) to investigate cross-country comparisons of socioeconomic inequalities in adolescent psychosocial health (depressive mood and self-esteem). Of particular importance for this project was establishing measurement invariance (measurement equivalence) for the scales used to measure psychosocial health. Measurement invariance is a precursor to any group or time/condition comparison. It tells the researcher that any changes in the estimated means across groups are the result of real differences, and not because the measure operates in a different way across different groups. For this paper we focus upon testing the well-known and widely use measure of self-esteem, the Rosenberg self-esteem scale. The “psychosocial health” module was an optional aspect of the survey. 12 of 33 countries included the Rosenberg self-esteem scale. We found: 1) There was no evidence that factor loadings were invariant across the twelve countries Δχ2(108)=4476.28. This is the weakest form of measurement invariance. 2) That a two factor solution resulted in substantially improved model fit than a one factor solution Δχ2(1)=31130.21. 3) That treating individual items as equally weighted, by summing items together or taking mean responses across items, results in considerably worse model fit than allowing items to be weighted differentially based on the correlation between the factor and the item Δχ2(9)=2801.70. 

Mortality patterns and levels of natives and immigrants in Greece in 2011
Georgia Verropoulou, Cleon Tsimbos, University of Piraeus, Department of Statistics & Insurance Science 

The present study addresses for the first time in Greece the issue of levels of mortality among natives and immigrants, using vital registration data on deaths for the period 2010-2012 and population data derived from the 2011 census, by citizenship. The analysis is based on life tables regarding the two most numerous communities, Albanians and Bulgarians, and on standardised mortality ratios regarding smaller groups. In total, estimates for 26 ethnic groups are presented. The findings indicate that only Albanian males have a slight mortality advantage compared to natives; all other groups seem to experience higher mortality. Nevertheless, in several cases immigrants have significantly lower mortality compared to their countries of origin; in that sense, the research provides some support for the ‘selectivity of migrants’ hypothesis. Moreover, there is great heterogeneity between migrant groups and this should be taken into account when formulating public health policies. 

Regional mortality variations of the older population in Greece: 2010-1012
Cleon Tsimbos 1, Stamatis Kalogirou 2, Georgia Verropoulou 1, 1 University of Piraeus, Greece, 2 Harokopeio University, Greece   

Scope: The aim of the study is to explore recent variations of mortality in Greece among people aged 65 years old or higher at local authority level. Data The study uses vital statistics on deaths registered during 2010-2012 and population counts based on the 2011 population census. Data on deaths and population are cross-classified by sex, five-year age groups and place of usual residence. The vital statistics used in the analysis are unpublished and have been provided to the authors by the Hellenic Statistical Authority upon special request. Methods: We calculate age-standardized mortality rates separately for males and females aged 65+ employing the direct standardization technique; the age distribution of the total population of Greece (both sexes combined) is used as standard population. Standard errors of the standardized mortality rates are computed assuming the Poisson distribution. The analysis includes 290 (out of 325) local authorities exhibiting at least 20 annual deaths of persons aged 65+ and relative standard errors less than 25%. We employ mapping techniques to depict spatial patterns of mortality as well as sex differentials. We also examine correlations between mortality and environmental factors to identify potential links. Results: The spatial patterns are complex and suggest a significant variation in standardized mortality rates. High mortality rates appear in urban areas and in most of the municipalities located in Central and Northern Greece while low rates appear in affluent suburbs, islands and mountainous municipalities. 

Who gets killed when disaster strikes: Exploring demographic differential vulnerability and the role of human development in reducing mortality from natural hazards 
Erich Striessnig 1, 2, Raya Muttarak, 1, 2, 1 Wittgenstein Centre for Demography and Global Human Capital (IIASA, VID/ÖAW, WU), 2 Vienna Institute of Demography    

This paper evaluates the heterogeneous impact of weather events on populations along the dimensions of age, sex, and development. The analysis is based on previously untapped cause of death data by age and sex over the period 1988 – 2014 obtained from the WHO mortality database based on civil registration records of 133 countries/territories. Using these data, we first analyze patterns of heat-, cold- and other weather-related mortality in a multivariate framework. Preliminary results from logistic regressions suggest that mortality rates from meteorological events tend to be higher for men. Differential by sex is larger for adults than for young children or the elderly. Furthermore, the sex differential in mortality becomes smaller with improvements in human development. Using projections of future human development in the three dimensions of income, life expectancy, and education, as generated for the most recent set of Shared Socioeconomic Pathways, in order to assess the impact of future climate change we plan to project weather-induced mortality in different world regions. The paper will also discuss issues of data quality and data harmonization for the study of the differential demographic impact of natural disasters: one of the goals is to stimulate an interdisciplinary debate in this area. 

Health:  Health inequalities: Mechanisms & trends. Wednesday 14 September 09:00am 

Self-reported general health and Body Mass Index: a U-shaped relationship?
Paul Norman 1, Lorna Fraser 2, 1 School of Geography, University of Leeds, 2 Department of Health Sciences, University of York   

A sizeable literature demonstrates an increased likelihood of self-reported poor health with increasing Body Mass Index (BMI). A U-shaped relationship between BMI and health is evident in research but it is not believed that the health of people who are underweight gets the attention it deserves. The relationship between BMI categories and poor general health including those persons who are underweight has been investigated. Data were obtained from the Health Survey for England pooled for 2006-08 to provide a sample of 26,596 adults aged 16 to 74 in a series of binary logistic regression models to determine the likelihood of people reporting their general health as being poor. A clear U-shaped relationship between BMI and health has been found, controlling for individual level demographic, socio-economic and health-related behaviour variables and area deprivation. The literature largely focuses on the relationship between obesity and health and this is understandable given that many more persons are obese than are underweight. A person who is underweight is about as likely to report poor general health as a person who is obese. This work is published and will be extended prior to the BSPS conference using the HSE long-run time-series dataset to see whether the U-shape is evident over time. 

Educational attainment and biomarkers of health: Evidence using genetic markers
Xuejie Ding, Nicola Barban, Department of Sociology, University of Oxford   

Education is an established correlate of health status in older adulthood, but the estimates seldom represent a causal inference. The goal of this paper is to conduct instrumental variable (IV) analyses of the association between education and allostatic load measured health (i.e., BMI, Hba1c, systolic blood pressure, diastolic blood pressure, cholesterol ratio, C-reactive protein, and Cystatin C), using genetic variants as instruments as well as quarter of birth. Our IV analyses (n=8870) in the Health and Retirement Study cohort (2008) used two sets of instruments: (1) a polygenic allele score constructed from 35 single-nucleotide polymorphisms (SNPs) and (2) quarter of birth. Using the polygenic allele score as an IV, there was a substantial reduction in undesirable health risk per year of schooling. Analyses evaluating the plausibility of the IV assumptions indicated estimates derived from analyses rely on both instruments provide the best estimates of the causal effect of education. Marital status may serve as an important mechanism in this causal link as an additional year of schooling increases 8% of the chance of being married. We conclude that IV analyses suggest education is protective against risk of negative biomarker health outcomes in older adulthood. 

Socioeconomic differentials in health expectancy with and without multimorbidity using electronic health records
Madhavi Bajekal 1,Mei Sum Chan 1, Melvyn Jones 1, Ardo van den Hout 1, Dr Mar Pujades 2, 1 University College London, 2 University of Leeds

Background: The socioeconomic gradient in life expectancy is well established, but little is known about the impact of variations in disease patterns and trajectories underlying lifespan inequalities. Multimorbidity is the presence of two or more chronic diseases in an individual, and is the norm at ages over 75. For our analysis, we selected 30 major chronic diseases. We aim to quantify the impact of differences in the age of onset and disease progression on survival: do disadvantaged groups acquire more, or more lethal combinations of, diseases; or do they simply become ill at younger ages? Data: Our CPRD (Clinical Practice Research Datalink) cohort consists of 1.3 million English patients aged 45+, with linked primary care, hospital and ONS death records, followed up from 2001 to 2009. Methods: We use a multi-state Markov transition model to summarise differences in health expectancies, and the ages of disease onset, disease progression and death. No recoveries from these chronic diseases are assumed: i.e. the model is progressive. We model transition rates between 5 health states – no disease, 1, 2, 3+ diseases and death, for each deprivation and sex sub-group. Life expectancies and years spent healthy and with 1, 2, 3+ diseases are then derived from these transition rates, for any age from 45 onwards. Results and conclusions: We will discuss the key operational choices and methodological challenges. Both total life expectancy and years spent in each health state are likely to vary significantly by socioeconomic circumstances, age and gender. 

Compositional changes in educational groups as an explanation for widening health differences 
Cecilia Potente, Christiaan Monden, University of Oxford    

According to many studies, educational disparities in health have increased in the United States during recent decades. However, two interrelated factors have not been fully taken into account in the analyses so far: the role of educational expansion in shaping educational differences in health and consequently the compositional change within various educational groups. First, the goal of this analysis is to understand how the relationship between disadvantageous conditions in childhood and education changes over time. Is the mechanism of selection into various educational categories changing after the educational expansion? In addition, this association is examined also over age for selected cohorts in order to understand the extent of mortality selection. Secondly, we study changes in the relationship between education and health after the educational expansion. We use height in order to understand whether compositional differences within educational groups has a role in explaining differences in health. In addition, we analyze the association between height and health over age. Data are from 39 pooled National Health Interview Surveys (NHIS) containing information about health of US citizens from 1976 to 2014. First, using logit models we examine whether there is a difference in height composition of various educational groups and how this difference changes over time. Second, we analyze whether the association between health and height in various educational groups varies over time. We hypothesize that individuals with shorter height and lower educational achievement have augmented cardiovascular disease risk, since education is expected to be a mediating factor for disadvantaged childhood circumstances. 

Behavioral factors drive geographic mortality inequality in Great Britain
Laura A. Kelly, University of Pennsylvania 

Health behaviours influence all-cause and preventable mortality. The common approach to estimating mortality attributable to health behaviors utilizes population attributable fractions, which are often biased due to their reliance on accurate estimation of risk factor distribution in the sample population, analytic sensitivity to relative risk estimates, and an assumption of the causal pathway between exposure and mortality outcome. This study uses an improved, innovative population approach to indirectly estimate the effect of health behaviors on geographic variation of preventable mortality. The contribution of health behaviors is quantified using exploratory factor analysis on cause-specific mortality information for males and females aged 15-74 from 1981-2009 for 39 temporally-consistent English and Scottish geographies. To the author’s knowledge, factor analysis has never been applied to cause-specific, subnational vital statistics in Britain nor has the data necessary to conduct this analysis been previously constructed. The majority (89%) of observed geographic variation in mortality is driven by causes of death linked to health behaviors, and similar two-factor models explain the majority of mortality variation for both sexes. These factors are structured by causes of death linked to tobacco-use and diet (Factor 1) and rurality and substance abuse (Factor 2). The geographic patterning of each factor reinforces a North-South disparity in behavioral-driven mortality for both sexes. Particularly for Factor 2, injury (unintentional and suicide) and substance-abuse mortality is starkly concentrated in Scottish areas. Understanding the behavioral drivers of preventable mortality is vital to quantifying health inequality, identifying potential areas for policy interventions, and regulating national population health. 

Health: Health inequalities across the life course: later life. Wednesday 14 September 11:30am   

Intergenerational Influences on Physical Activity among Older Adults in Europe Albert Sabater, Elspeth Graham, Alan Marshall, Department of Geography and Sustainable Development & ESRC Centre for Population Change, University of St Andrews   

It is widely acknowledged that tendencies for becoming more or less active as one ages occur within a developmental context spanning an individual’s life. However, previous empirical studies have failed to adopt an intergenerational approach to the understanding of physical activity levels of the ageing population despite the recognition that members of one age group can exert an influence on those from another age group. The purpose of this study is to evaluate the unexplored association between adult children’s education and the physical activity levels of their older parents. Using propensity score matching and data from waves 1 and 5 of the Survey of Health, Ageing and Retirement in Europe (N=10,465), our study compares physical activity levels among healthy individuals aged 50 and over with highly-educated adult children to matched controls, stratified by gender. By using this approach, we take into account selection characteristics that predict physical activity in later life. We identify children’s education as a “treatment” and examine whether older people with similar background characteristics are more likely to engage in moderate or vigorous exercise at least once a week if they have highly-educated children. We test different compound variables for children’s education and duration of the treatment, including a measure of “Cumulative Educational Exposure”. The results suggest that having highly-educated children is good for physical activity levels among older adults, particularly for women. These findings highlight the potential of using an intergenerational approach to influence physical activity levels in later life. 

Inequalities in the time between stopping work and death: ONS Longitudinal Study
Emily T Murray, Nicola Shelton, Department of Epidemiology and Public Health, University College London, London 

Increasing the State Pension Age of the United Kingdom based on average life expectancy could be depriving certain sub-groups of the population of retirement years, due to inequalities in disability onset and mortality. Data sources – Office for National Statistics Longitudinal Study (LS); all members resident in England and Wales, aged 40-69 at the 2001 census and died by December 2013. Methods - We calculated the number of months between stopping work and death. If the LS member had stopped work in 2001 or 2011 they were asked the year they had stopped working. If working at 2001 or 2011 census, and died in the following interval, the average amount of time they could have not worked before death was taken. Linear regression was used to examine mean differences in “not working” time by age category, sex, employment status, social class, ethnicity, tenure, health status, industry grouping and region. Results – Of the 18,205 LS members who had died May 2001 - December 2013, the mean number of months between stopping work and death was 152.4 (SD 1.4). LS members with higher mean months of “not working” were more likely in 2001 to be female, of older age, not working (both sick and retired), II and IV social class, owner occupiers and renters, had worse health, resided in the North and Wales and worked especially in mining, manufacturing and construction industries. Potential Application – Pension eligibility may be more equitable if based on years worked alone, rather than in combination with a minimum age. 

The importance of material hardship and psychosocial factors for onset of rheumatoid arthritis
Cathie Hammond, University of Essex.   

Background: Socioeconomic inequalities in rheumatoid arthritis (RA) indicate that adversity matters for its onset, but little is known about which types of adversity matter. I investigate: Whether adversities predict RA onset; How adversities at different life stages combine to predict RA onset; Whether adversities associated primarily with material and psychosocial pathways independently predict RA onset. Data and methods: The English Longitudinal Study of Ageing is representative of community-living adults aged 50+. Retrospective life-history information was used to measure adversities and smoking history. Multiple imputation was used to impute missing values. Cox regression analyses were used to model RA onset, adjusting for birth cohort and gender. RA onset from age 45 was regressed on life-course adversity, number of life stages with adversity, adversity during each life stage with mutual adjustment, and material and psychosocial adversities with mutual adjustment and adjustment for smoking. Sensitivity analyses excluded respondents with pain, depression, and poor cognitive function. Results: Of 6,663 respondents, 208 developed RA from age 45. RA onset was predicted by each additional adversity (maximum 8) hazard ratio HR=1.086, 95% confidence interval 95%ci=1.004-1.174; each additional life stage with adversity (maximum three) HR=1.155, 95%ci=0.937-1.424; youth adversity (maximum three) after adjustment for childhood and early-adulthood adversities HR=1.253, 95%ci=1.052-1.492; adversities associated with material (maximum two) and psychosocial (maximum three) pathways after adjustment for smoking, HR=1.34, 95%ci=0.894-2.018, HR=1.28, 95%ci=1.043-1.577, respectively. Sensitivity analyses provide similar findings. Conclusion: These findings require replication, but indicate the importance of adversity in the development of RA and raise questions about the mechanisms involved. 

Multi-morbidity, functional status and hypertension: a study of older adults in South Africa
Philippa Waterhouse 1, Nele Van Der Wielen 2, Pamela Banda 3, 1 Faculty of Health and Social Care, Open University, 2 Department of Social Statistics, University of Southampton, 3 School of Public Health and Social Sciences, University of the Witwatersrand   

The prevalence of non-communicable diseases (NCDs) is increasing worldwide, leading to new and different pressures on the health system, especially in low and middle income countries. Furthermore, there is a rise in the numbers of individuals who suffer from multiple NCDs at the same time, referred to as multi-morbidity. Research into multi-morbidity has been concentrated within higher income contexts while the phenomenon in low and middle income countries remains under-researched. Taking the case of South Africa, the aims of this paper are to firstly explore the association between multi-morbidity and functional status among older adults. Secondly, the paper will consider the impact of diagnosis and treatment of hypertension on this relationship. This paper uses data from Wave 1 of the South African Study on Global Ageing and Adult Health. Multi-morbidity was defined as the co-existence of any two or more chronic diseases. All subjects completed the WHO Disability Assessment Schedule which was used to compute a measure of functional status. A four category variable on an individuals’ hypertension status was creating using information on their recorded blood pressure as measured at the time of the interview, and whether they self-reported being diagnosed with hypertension. Linear regression was used to investigate the associations between multi-morbidity and functional status. Three nested models were considered. The first model only contained the multi-morbidity measure as an independent variable, in the second model individual’s characteristics, such as age, was entered as controls, and in the last model the hypertension variable.