Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
KNOWLEDGE 4 HEALTH Innovative Analytical Techniques in Health Research: A showcase of Structural Equation Models Trutz Haase Jonathan Pratschke Knowledge 4 Health Conference, 25th May 2016, Royal Hospital Kilmainham HEALTHY IRELAND “Accelerating the take-up of new knowledge and innovating through advances in scientific knowledge is a key aspect of how we will achieve the four high-level goals. The consistent application of evidence of what works and what interventions positively impact on health behaviours, in a cost-effective way, is critical to setting policy and investing in prevention programmes. Excellent population health analysis capability is required for understanding and predicting threats to public health.” Department of Health, 2013, p. 28 A NEW DIRECTION The Healthy Ireland strategy entails a paradigm shift involving: a broader definition of health which embraces physical, mental and social well-being a “whole-of-society approach” to population health a concern with the wider determinants of health a stronger evidence-based approach to policy a focus on the social determinants of health as direct and indirect causes of illness an emphasis on reliable indicators of health, well-being and risk factors CHALLENGES FOR HEALTH RESEARCH This implies a number of difficult challenges for health researchers: a move from specialist sectoral analyses to more systemic research a shift away from random control trials and towards observational data on the population a drive towards more fully-specified explanatory models a greater emphasis on causal inference a concern with complex networks of direct and indirect influences a demand for more comprehensive composite indicators with high reliability ISSUES THAT NEED TO BE ADDRESSED IN HEALTH RESEARCH 1. Definition of composite indicators for health, well-being and other key concepts 2. Exploration of the multi-dimensional structure of health and wellbeing 3. Identification of the risk and protective factors that influence health and well-being 4. The importance of taking account of mediated effects when designing policy interventions 5. The monitoring of health outcomes and risk factors over time This presentation aims to demonstrate the contribution that Structural Equation Modelling can make to these tasks. 1. WHY DO WE NEED COMPOSITE INDICATORS? Ease of interpretation – compared with results from many individual variables, each with its own specificities In harmony with the Common Risk Factor approach central to Healthy Ireland Can be used to assess progress over time and to facilitate benchmarking and monitoring (powerful policy impacts) Use of broadly-defined concepts facilitates communication of research findings to the general public Example: Socio-economic Position as a “latent variable” COMPOSITE INDICATORS: EXAMPLE: TILDA WAVE 1 - MEASURING SEP Third-level Education Assets .52 / .53 Income .46 / .47 .70 / .70 Occupation .55 / .57 SocioEconomic Position Parameters shown are the standardised coefficients for the male and female sub-samples 2. MULTI-DIMENSIONAL CONCEPTS Many concepts in health research comprise distinct components (e.g. overall health comprises physical, cognitive, mental, socioemotional dimensions) The definition of these concepts should be driven by theory This means that we should start by assessing/testing the dimensionality of key concepts Each dimension may be measured indirectly using a set of criteria/variables, treated as “partial manifestations” Example: Dimensionality of HP Deprivation Index based on Social Class, Labour Market Deprivation, Demographic Decline MULTI-DIMENSIONALITY EXAMPLE: HP DEPRIVATION INDEX d1 Age Dependency Rate d2 Population Change d3 Primary Education only d4 Third Level Education d5 Persons per Room d6 Professional Classes d7 Semi- and Unskilled Classes d8 Lone Parents d9 Male Unemployment Rate d10 Female Unemployment Rate Demographic Growth Social Class Composition Labour Market Situation HP Deprivation Index 3. RISK AND PROTECTIVE FACTORS Healthy Ireland emphasises health risk behaviours, which influence many different outcomes This Common Risk Factor approach needs to be built explicitly into the design of research projects This approach can be operationalised by analysing the effects of risk/protective factors on health and well-being within a SEM model If this is done using Structural Equation Modelling techniques, we can use latent variables and control for measurement error Example: Estimation of the effect of risk/protective factors and social context on health and well-being RISK AND PROTECTIVE FACTORS EXAMPLE: TILDA WAVE 1 Goodness of Fit (M/F): N: CFI: RMSEA: 3,740 / 4,423 .954 / .956 .024 / .026 SocioEconomic Position Age Abused Childhood Smoker Lives Alone Regular Drinker Intimate Relationship Social Participation Problem Drinker Physical Exercise Social Network Overall Health Unemployed R² = .33 / .46 male or female effect only 0.03 to 0.10 0.11 to 0.20 0.20 to 0.50 All effects significant at p < .05 Religiosity Personal Well-being R² = .55 / .60 4. MEDIATED EFFECTS Classical regression models allow us to estimate the net direct effect of a variable, without considering inter-relationships To make research more relevant to policy-making, we have to model how effects are generated (underlying mechanism) It is essential to start with a theoretical model and to use path diagrams to translate this into a statistical model Example: Analysis of GUI 9-year-old cohort confirms importance of mother’s well-being as mediator in relation to child well-being It also reveals how many contextual influences are also mediated, in line with the ecological model of child well-being MEDIATED EFFECTS EXAMPLE: GUI 9-YEAR OLDS – WAVE 1 Goodness of Fit: N: CFI: RMSEA: 4,881 .951 .023 Financial Difficulties - . 08 - . 10 Local Problem Scale SCG Well-being Non-Irish Ethnicity - . 10 Low Social Class R²=.04 - . 06 Local Services Scale Equivalised Household Income Decile Haase-Pratschke Deprivation Score - . 15 ESRI Basic Deprivation - . 11 PCG Well-being . 09 R²=.17 Health Status (Child) - . 10 - . 06 - . 11 . 41 . 04 - . 10 Life Events (Child) Gender (Child) . 08 - . 07 - . 04 . 07 Child Well-being R²=.31 All effects significant at p < .05 - . 28 Low Education (PCG) Health Status (PCG) - . 04 . 12 Age (PCG) 5. MONITORING HEALTH OUTCOMES AND RISK FACTORS OVER TIME Monitoring key health policies should be carried out in relation to the overall health and well-being of the population Structural Equation Modelling allows for… the specification of outcomes as latent concepts the investigation of their inter-relationships their change over time measurement of the effects of risk behaviours and socio-economic factors SEM provides the most powerful analytical framework to respond to the challenges posed by the Healthy Ireland “paradigm” Example: Longitudinal study of the determinants of health and well-being using data from Waves 1 and 2 of TILDA LATENT VARIABLES IN LONGITUDINAL RESEARCH EXAMPLE: TILDA WAVES 1 AND 2 Stability Factor R2 = 0.72 Well-being Wave 1 -0.62* Depression 1 -0.70* Loneliness 1 0.52* Well-being Wave 2 0.75* 0.84 Life Satis. 1 -0.63* Quality of Life 1 Depression: 20-item CESD score (Radloff, 1977) Loneliness: 5-item UCLA Loneliness Scale (Russell, 1996) Life Satisfaction: single item with a 7-point response scale Quality of Life: 19-item CASP scale (Hyde et al., 2003) Depression 2 -0.67* Loneliness 2 0.48* 0.81 Life Satis. 2 Quality of Life 2 RISK FACTORS IN A COMPLEX LONGITUDINAL MODEL EXAMPLE: TILDA WAVES 1 AND 2 Gender (M) Age Social Participation -0.04* Well-being Wave 1 -0.09* 0.75* 0.03* Well-being Wave 2 Intimate Relationships -0.09* 0.10* Cog. Function Wave 1 0.07* 0.02* Cog. Function Wave 2 0.89* Social Network -0.04* -0.03* Smokes 0.08* -0.05* Phys. Health Wave 1 Phys. Health Wave 2 0.79* -0.02* Drinking problem -0.03* Lives Alone 0.03* Physical Exercise 0.04* Social Class Comparative Fit Index (CFI): 0.96 Yuan-Bentler Corrected CFI: 0.96 Yuan-Bentler Corrected RMSEA: 0.036 (CI: 0.035, 0.037) For further information on our research: www.trutzhaase.eu