Biomeasure Collection in Social Surveys - Challenges and Opportunities 2 |
|
| Chair | Mr Matt Brown (Centre for Longitudinal Studies, UCL Institute of Education ) |
| Coordinator 1 | Dr Emily Gilbert (Centre for Longitudinal Studies, UCL Institute of Education) |
| Coordinator 2 | Ms Anne Conolly (Health and Biomedical team, NatCen Social Research) |
| Coordinator 3 | Dr Shaun Scholes (Research Department of Epidemiology and Public Health, University College London) |
The integration of physical measures and biomarkers in population surveys has become increasingly common in recent years. These measures are considered to be an important complement to self-reported health measures that are typically collected in large surveys. Yet, the value of physical measures and biomarkers and their associations with self-reported measures has not been fully demonstrated. This paper aims to address that gap by comparing associations between performance-based and self-reported measures of functioning using data from the Health and Retirement Study (HRS).
The HRS is an ongoing national panel study of people over age 50 in the U.S. that began in 1992. Since 2006, HRS has administered performance measures of grip strength, lung function, walking speed and balance (along with other physical measures and biomarkers) as part of the biennial in-person interview. These measures are administered by field interviewers and participation rates have been high--95% among those asked to do the measures and 80-85% of all eligible sample members. Because participants who do not complete the performance measures tend to be different than those who do, the interview also includes self-reported questions corresponding with each functional domain, which are asked of all respondents and can be used for imputation or weighting purposes.
In this paper we use data from the 2006-2012 waves to examine associations between the performance and self-reported measures and how they differ by sex and age. Using logistic regression, we also compare the extent to which the performance and self-reported measures are predictive of mortality, nursing home admission and hospitalization, controlling for demographic and health factors.
Preliminary results suggest strong, but far from perfect associations between the performance and corresponding self-reported measure, with correlations ranging between 0.22 for lung function and 0.37 for both walking and hand strength. The correlations tend to decline with age, being strongest for persons under age 65 and weakest for persons age 80+. Gender differences in the correlations are small.
For all four functional domains, the performance measures are much stronger predictors of mortality, and generally also of nursing home admission and hospitalization, than the self-reported measures. However, most of the self-reported measures remain statistically significant after controlling for the corresponding performance measure (plus demographic and health factors), suggesting that the self-reported measures capture some aspect of health or functioning that is not captured by the performance measures. We also find gender differences in the association between the performance measures and health outcomes—e.g., walking speed, grip strength and lung function are strong predictors of mortality for both men and women, but the effects are significantly larger for women than for men.
We argue that both performance and self-reported measures have distinct value in surveys and may contribute to our understanding of the physiological mechanisms of health changes in later life and how perceived health may mediate or moderate those changes.
Measures of physical function and physical performance are collected in many large-scale surveys which track health and the ageing process. Historically, there has been no standard set of equipment used to assess blood pressure, grip strength or lung function; different devices have been used across studies and have been changed between waves of longitudinal surveys due to obsolescence or technological improvement. CLOSER, the UK’s Cohort & Longitudinal Studies Enhancement Resource, funded a randomized, controlled trial with repeated measures to determine whether statistical adjustment is needed to compare between surveys or to estimate longitudinal trajectories. The experiment involved two models of sphygmomanometer for blood pressure (Omron 705 and Omron 907), two spirometers for lung function (Micromedical Plus and Easy-on NDD) and four dynamometers for grip strength (Nottingham Electronic, Jamar Hydraulic, Jamar Electronic and Smedley). In total, 118 healthy individuals were randomly and equally assigned to a sequence of measurements. Assessments were carried out by trained researchers in a consistent setting, following prescribed protocols.
For the three measures, we derived continuous and categorical outcomes as appropriate, tested for order effects and assessed differences of paired measurements using dependent t-tests, Bland and Altman plots and multi-level models to account for clustering of observations in measurement devices and respondents. We further use regression models to test whether differences between measurement devices are constant or vary with respondent characteristics. We draw conclusions about the need for adjustment for each of the three measures and comment on implications for future data collection.
This experiment focused on equipment used in UK surveys such as the British birth cohorts, Whitehall II and the English Longitudinal Study of Ageing, so has direct relevance for analysts using these datasets. However it is also has broader relevance to international studies such as the Health and Retirement Survey and the Survey of Health and Retirement in Europe and to broader initiatives such as CLOSER and the NIH Toolkit which seek, in different ways, to increase the comparability and utility of biosocial data.
(The authors of this paper are Lessof, Wong, Hardy and the Equipment Comparison Team whose names will be listed during the presentation. If this paper is accepted, if possible the conference information about authors should include mention of the Equipment Comparison Team.)
The Survey of Health, Ageing and Retirement in Europe (SHARE) is a multidisciplinary and cross-national panel database of micro data on health, socio-economic status and social and family networks of individuals aged 50 or over covering 27 European countries and Israel (www.share-project.org).
SHARE’s main aim is to provide data on individuals as they age in order to analyse the process of individual and population ageing in depth. One key area covered by SHARE is health. From the first wave (2004) on, SHARE combined self-reported health with objective health measurements in form of physical performance measurements. In its sixth wave (data collection in 2015), SHARE included the collection of blood samples in a major part of the participating countries. Blood was collected in form of Dried Blood Spot (DBS) samples. The technique of DBS collection and analyses offers great opportunities to large scale surveys like SHARE as DBS samples may be collected by trained lay interviewers and shipped using a standard mailing service. 12 SHARE countries participated in the DBS collection. Informed written consent given by the respondent was needed in all of them prior to the blood collection.
The entire data collection in SHARE, including the DBS part, is ex-ante harmonized. This includes the questionnaire development, the interviewer training, and for example the composition of the used DBS collection kit. Thus, all interviewers should face similar conditions when trying to gain informed written consent to the blood collection. But the observed consent rates vary remarkably over countries and some interviewers had more success than others. Obviously, consenting is a respondent’s decision and therefore may depend on respondent’s characteristics. But interviewer characteristics seem to play an important role, too, as well as other factors that vary from country to country. This work aims on developing strategies to identify factors that determined consent rates for the DBS collection in SHARE. Respondent, interviewer, and country level characteristics will be taken into account. The knowledge of such influencing factors may help to further improve the consent gaining process and its success.
Increasingly, population-based studies are incorporating measures of blood-based biomarkers as part of a larger effort to elucidate mechanisms linking social factors and health. Studying the interactions between social, physical, and biological factors on health and aging has become increasingly important as many countries experience a demographic shift toward an older population distribution. However, understanding the validity of different methods of biomarker collection in the field is essential. In addition, understanding the effect of different collection protocols on consent and potential nonresponse bias is also a crucial consideration when adding biological measurement.
Blood-based biomarkers can be collected in the field by a variety of methods including peripheral blood collection used for dried blood spot (DBS) assays and venous-blood collection by trained phlebotomists. The validity of DBS assessments used by population-based studies can be addressed by comparing assay values from DBS collected in the field with values from venous blood samples. Similarly, collection and shipping protocols for venous blood can be evaluated by comparing venous blood collected in the field, often in a respondent’s home, to blood collected and immediately processed in a laboratory setting.
In this paper we use DBS and venous blood collected from 10,000 respondents during the 2016 wave of the Health and Retirement Study (HRS) to compare results from 5 assays: total cholesterol, HDL cholesterol, HbA1c, CRP, and cystatin C. For each pairing of assays, we show descriptive statistics (e.g., mean, SD, range), the correlation, and the equation linking the two. We also provide Bland Altman plots for the venous blood and DBS comparisons to provide a method of comparing the differences between two assays across the range of values and to see how frequently outliers occur.
A larger set of assays are available to compare results from the venous blood collection in the HRS to the U.S. National Health and Nutrition Examination Survey (NHANES), another nationally representative survey of adults. The HRS collected VBS samples in respondent homes and shipped samples overnight to the processing lab. In contrast, NHANES collects specimens in a mobile lab unit, processing or freezing immediately. Taking advantage of the fact that these are both nationally-representative surveys, we can compare the distributions of results from these studies to examine field vs. laboratory collection.
The DBS approach generates values that are strongly related to venous blood levels of HbA1c, cystatin C, and C-reactive protein. Assessing lipid levels reliably with DBS appears to be more difficult. However, even when DBS values and values from venous blood are highly correlated, raw values are often very different, and using conventional cutoffs may be misleading. Venous blood result comparisons between HRS and NHANES are highly correlated showing the effects of field collection, delayed analysis (48 hrs) are negligible across a variety of assays.
In addition, we also compare consent rates for blood-based biomarker collection across demographic groups within the HRS. While consent rates for biological data collection are high, certain population subgroups are disproportionately less likely
This study investigates the creation of polygenic scores for population-based cohort studies. Mathematically, polygenic scores are a linear, sometimes weighted, combination of risk alleles that estimate the cumulative genetic risk or genetic susceptibility of an individual for a particular trait. While conceptually simple, there are numerous ways to estimate them, not all achieving the same end goals. In this paper, we systematically investigated the impact of four key decisions in the building of polygenic scores from published genome-wide association meta-analysis results: 1) which single nucleotide polymorphisms (SNP)s to include in the score, 2) whether to use genotyped or imputed data, 3) whether to account for linkage disequilibrium, and 4) which type of method best captures the correlation structure among SNPs (i.e. clumping vs. pruning). Using the Health and Retirement Study – a population-based longitudinal panel study of American over the age of 50 – we examined the variability and covariability in polygenic scores, as well as predictive ability, arising from different approaches to estimating polygenic scores for four human traits (height, body mass index, educational attainment, and depression).