Biomeasure Collection in Social Surveys - Challenges and Opportunities 3 |
|
| 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) |
BACKGROUND
It is well-established that in general the lower an individual’s socioeconomic position the worse their health: i.e. a social gradient in health that runs from the top to the bottom of the socioeconomic spectrum. Typically, estimates of this socioeconomic gradient have relied heavily on self-reported measures. Differences in self-reported measures may reflect differences in individuals’ expectations of their own health or differences in awareness levels of their adverse health condition; and yet both expectations and awareness are themselves socially patterned. Estimates of the socioeconomic gradient in health may therefore be underestimated using reported as opposed to objectively-measured data. Using data from the 2014 Health Survey for England (HSE), we compared the relationship between socioeconomic circumstances and health using objective and self-reported measures of the same condition.
METHODS
We focused on outcomes for which participants had contemporaneous objective and self-report data: hearing (n=5334); hypertension (n=4669); and diabetes (n=3894). Objectively-measured hearing loss was defined as impairment in the better hearing ear to the level of ≥35dB HL at a frequency of 3 kHz; self-reported hearing difficulty was defined as perceived difficulty with hearing, or report of current hearing-aid use. Hypertension was defined objectively as systolic / diastolic blood pressure ≥140/90mmHg; self-reported current hypertension was defined as reported hypertension or high blood pressure as an illness lasting or expected to last ≥12 months. Objectively-measured diabetes was defined using a value of glycated haemoglobin ≥6.5%; self-reported diabetes was defined as reported diabetes including hyperglycaemia as an illness lasting or expected to last ≥12 months. Equivalised income quintiles, and quintiles of the area-based Index of Multiple Deprivation (IMD), were used as measures of socioeconomic status (SES). Logistic regression on age-standardised data was used to examine the statistical significance of differences across SES groups adjusting for sex. An SES-by-sex interaction term was used to test if the socioeconomic gradient varied by sex.
RESULTS
Using income quintiles as the measure of SES, socioeconomic gradients in hearing loss were more apparent using data from the hearing test (P=0.022) as opposed to self-reported data (P=0.807); a similar pattern was found using the IMD measure (P=0.016 and P=0.096 respectively). Prevalence of self-reported hypertension was similar across income groups (P=0.807); objectively-measured hypertension varied significantly across income groups for women but not for men (P=0.042 for SES-by-sex interaction). The opposite pattern was found for diabetes: prevalence of self-reported diabetes including hyperglycaemia varied significantly across the income- (P=0.001) and IMD-quintiles (P=0.042); prevalence of objectively-measured diabetes based on the values of glycated haemoglobin was similar across the SES groups (P=0.137 and P=0.111 respectively).
CONCLUSIONS
Socioeconomic gradients were more apparent when using objective measurements for hearing (both sexes) and for hypertension (women only). Analyses are still ongoing and will include other health conditions.
Health status indicators can be obtained through surveys of probabilistic samples of population, using self-reported or objectively measured data. Self-reported data is the most common method used in survey research however its validity is often questioned due to report bias. Undiagnosed disease, misunderstanding of medical terms used in questionnaire are major sources of measurement error associated to self-report. Hypertension and hypercholesterolemia are chronic conditions known to have a long asymptomatic period adding extra complexity to use of self-reported measure.
This study aims to compare self-reported and examination-based hypertension and hypercholesterolemia prevalence in Portugal and to identify factors associated with measurement error in self-reports.
Portuguese National Health Examination Survey was conducted in 2015 on representative sample of 4911 adults aged 25-74 years. It combines blood pressure measurement, blood collection and interview.
To assess accuracy of self-reported hypertension and hypercholesterolemia sensitivity and specificity were calculated with examination-based results as the reference standard.
Logistic regression was used to estimate odds ratios (OR) of incorrect hypertension and hypercholesterolemia self-reports according to sex, age, education and general practitioner visit in the past year.
The examination-based hypertension and hypercholesterolemia prevalence was 36.0% [95%CI:34.3-37.7] and 63.3% [95%CI:61.2-65.4], respectively, while self-reported prevalence was 26.0% [95%CI:24.2-27.9] for hypertension and 25.0% [95%CI:23.1-27.0] for hypercholesterolemia.
Self-reports of hypertension and hypercholesterolemia showed high specificity (99% and 98.1%, respectively) but low sensitivity (70.6% and 38.6%, respectively).
Incorrect report of hypertension was associated to male gender (OR=2.1, [95%CI: 1.6-2.9]), groups aged 45-54 years (OR=3.7, [95%CI:1.9-7.1], 55-64 years (OR=2.4 [95%CI:1.2-5.0]) and 65-74 years (OR=2.6 [95%CI:1.5-4.4]), lack of general practitioner visit in the past year (OR=1.5 [95%CI:1.1-2.0]) and basic education (OR=2.2 [95%CI:1.5-3.1] e OR=1.9 [95%CI:1.3-2.7]), for 1 cycle of basic education and 2-3 cycle of basic education, respectively.
Individuals with secondary education (OR=1.4 [95%CI:1.2-1.6]), groups aged 35-44 years (OR=1.4 [95%CI:1.1-1.8]), and 45-54 years (OR=1.7 [95%CI:1.2-2.3]), and whose who have not visited general practitioner (OR=1.4 [95%CI:1.1-1.8]) were more likely misclassify themselves as not having hypercholesterolemia.
Self-reports underestimate prevalence of hypertension and hypercholesterolemia. The accuracy of self-reported hypertension was higher than compared to hypercholesterolemia. Age, education and general practitioner visit were identified as important factors affecting self-reports’ accuracy for both health conditions.
Adding objective measurements to self-reported questionnaire improve data accuracy and allow better understanding of socioeconomic inequalities in health.
Background. As prevalence rates of overweight and obesity have risen, there has been an increasing focus on maternal prepregnancy overweight and obesity, and excessive gestational weight gain, in relation both to the effects on the women themselves and their children. Prepregnancy weights are required for determining obesity status on entering pregnancy and gestational weight gain, but actual prepregnancy measurements are rarely available. Recalled weights and early pregnancy weights are often used but no studies have compared these measures in women from the general population who had actual measurements of prepregnancy weight taken shortly before conception.
Methods: The Southampton Women’s Survey recruited 12,583 women aged 20-34 years from the general population when they were not pregnant. At recruitment, research nurses measured the women’s weights using portable digital scales (Seca, Germany), which were regularly calibrated. Those who subsequently became pregnant were followed-up and, at an early pregnancy visit (around 11 weeks’ gestation), the women were again weighed on digital scales, calibrated using the same procedure as the scales used in the initial prepregnancy interviews. The women were also asked “How much did you weigh 3-4 months ago, i.e. before you became pregnant?”
Gestational age at birth was determined using a detailed algorithm combining last menstrual period date and early ultrasound data from which estimated date of conception was derived. The analysis was based on 198 women whose initial interview took place during the three months before the estimated date of conception, as this is conventionally considered to be the pregnancy-planning period.
Bland-Altman plots were used to compare the two ‘proxy’ measurements, (1) recalled prepregnancy weight obtained during early pregnancy and (2) measured weight in early pregnancy, with the ‘gold standard’ prepregnancy weight measure taken at the recruitment interview during the three months before conception. Linear regression was used to examine the trend in differences in relation to the mean of the measurements in order to assess whether the differences varied by prepregnancy weight.
Results: Mean (SD) recalled weight was 1.65 (3.03) kg lighter than measured prepregnancy weight, while measured weight in early pregnancy was 0.88 (2.34) kg greater. The limits of agreement for recalled weights were -7.59 to 4.29 kg, which were wider than those for the early pregnancy measured weights: -3.71 to 5.47 kg. The trend analyses revealed a negative trend for the recalled weights (regression coefficient -0.054 kg/kg (95%CI:-0.083 to -0.025, P<0.001), indicating that heavier woman are more likely to underestimate their recalled weight. No similar trend was seen for the early pregnancy measured weight.
Conclusions: These findings suggest that measured weight in early pregnancy is a better assessment of prepregnancy weight than recalled weight. We conclude that, where available, early pregnancy measures can usefully replace recalled weight in clinical practice and research. First trimester weight gain estimates need to be interpreted cautiously in the absence of actual measurements of prepregnancy weight.
People are inaccurate in self-reports of height and weight, as data from the United States National Health and Nutrition Examination Survey, for example, have shown (e.g., Wen and Kowaleski-Jones 2012). But how accurate are they when categorizing themselves and people they know using a general weight classification, that is, when labeling individuals as underweight, normal weight, overweight, or obese? In-depth, egocentric social network interviews with a small sample of U.S. adults (N=62) yielded some answers to this question.
Respondents were asked their height and were weighed in person by the researcher to later compute a BMI. They were also asked to place both themselves and 30 people comprising their social networks into one of four weight categories: underweight, normal weight, overweight, or obese. Respondents were not informed of the fact that these categories correspond to BMI ranges, nor to what BMI range each weight category corresponds. Respondents were further asked to provide assessments of body size for themselves and two alters from their network using Stunkard et al.’s (1983) standard figural stimuli. I acknowledge issues surrounding the utility of BMI as a measure of fatness, but leave it aside in this analysis.
When comparing BMI with self-assessed weight category, nearly half of all respondents incorrectly categorized themselves, and over 90% of those who miscategorized themselves did so by underestimating their weight categories. Respondents also miscategorized portions of their social networks in two ways: Respondent BMIs showed an inverse relationship with the portion of their social network that respondents said were overweight or obese; and, respondents assigned a greater range of the figural stimuli to each weight category than has been established as the norm (Bulik et al. 2001). The reasons for these inaccuracies are unclear, though some possible reasons that emerged in discussion with respondents include respondents not wishing to call themselves or people in their social networks overweight or obese due to stigma around these labels, and people with higher BMIs judging others’ weight status relative to their own body size.
On its face, a request for respondents to place themselves and people they know into weight categories might seem a suitable replacement for asking them for a more precise measure of body size or weight, such as BMI. In fact, some work has claimed that a different, broad four-category grouping is valid with some respondents (e.g., Leahey et al. 2015). However, my study shows that data of this nature can be quite problematic.
There are specific implications for studies of social networks and obesity. Christakis and Fowler’s (2007) well-known work, and others following it (e.g., Hrusckha et al. 2011), have shown that heavier respondents (egos) have heavier alters in their social networks. When alters are available to be measured, this finding bears out. When relying on ego to provide information about alters’ weight status, however, this correlation no longer appears to hold, rendering social network approaches less effective to understand the spread of obesity.