The English are healthier than the Americans: really? (2024)

Background: When comparing the health of two populations, it is not enough to compare the prevalence of chronic diseases. The objective of this study is therefore to propose a metric of health based on domains of functioning to determine whether the English are healthier than the Americans.

Methods: We analysed representative samples aged 50 to 80 years from the 2008 wave of the Health and Retirement Study (N = 10 349) for the US data, and wave 4 of the English Longitudinal Study of Ageing (N = 9405) for English counterpart data. We first calculated the age-standardized disease prevalence of diabetes, hypertension, all heart diseases, stroke, lung disease, cancer and obesity. Second, we developed a metric of health using Rasch analyses and the questions and measured tests common to both surveys addressing domains of human functioning. Finally, we used a linear additive model to test whether the differences in health were due to being English or American.

Results: The English have better health than the Americans when population health is assessed only by prevalence of selected chronic health conditions. The English health advantage disappears almost completely, however, when health is assessed with a metric that integrates information about functioning domains.

Conclusions: It is possible to construct a metric of health, based on data directly collected from individuals, in which health is operationalized as domains of functioning. Its application has the potential to tackle one of the most intractable problems in international research on health, namely the comparability of health across countries.

Key Messages

• Comparing the health of populations based on the prevalence of health conditions does not give the full story about health comparisons. This approach does not take the severity of health conditions into account and assumes that two people with a health condition such as diabetes always have the same level of health, or that someone with hypertension is equally unhealthy as someone with diabetes.

• It is possible to construct a metric of health based on data directly collected from individuals, in which health is operationalized as domains of functioning.

• The English have better health than the Americans when population health is assessed only by prevalence of selected chronic health conditions. The English health advantage disappears, however, when health is assessed with a metric that integrates information about functioning domains.

Data

We use data from the 2008 wave of the HRS for the US data, and wave 4 of ELSA (May 2008–July 2009), for English counterpart data. The HRS7 and ELSA8 are biannual, longitudinal and nationally representative surveys that focus on adults aged 50 and over. Both datasets are openly available after registration. HRS data are available from the corresponding website [http://hrsonline.isr.umich.edu/index.php?p=data] and ELSA data are available from the UK data service [http://discover.ukdataservice.ac.uk/].

To operationalize health state in terms of domains of human functioning, we identified questions and measured tests addressing those domains that are common to both surveys. 34 self-report questions were identified. They consisted of impairments in body and mental functions (‘Are you often troubled with pain?’ and ‘How much of the time during the past week did you feel depressed?’), and difficulties in activities of daily living (ADLs) and instrumental activities of daily living (IADLs) (‘Because of a health problem, do you have any difficulty with bathing or showering?’ and ‘…do you have any difficulty with managing your money – such as paying your bills and keeping track of expenses?’). The response options for these selected questions were coded or recoded so that higher values indicated worse health.

Six variables were selected from the measured tests used in both surveys. Grip strength was assessed with a hand dynamometer. Lung function was assessed with peak expiratory flow rate (PEFR). Balance was evaluated with three progressively more difficult stances: side-by-side, semi-tandem and tandem.9 These results were recoded into a polytomous variable with four response options (0 = ‘ability to perform tandem stand’, 1 = ‘ability to perform semi-tandem but not tandem stand’, 2 = ‘ability to perform side-by-side stand but not semi-tandem stand’ and 3 = ‘not able to perform side-by-side stand’). Cognitive functions were assessed by immediate and delayed recall of 10 common nouns and an orientation test consisting of reporting day, date, month and year. For grip strength, lung function, immediate recall and delayed recall the sample was then divided into three groups: low (<one standard deviation (SD) below the mean), medium (± one SD around the mean) and high (>one SD above the mean). The distributions of all measured tests in both populations were very similar so that the same thresholds, based on the mean of both surveys, were applied; for grip strength and lung function, the thresholds were defined separately for males and females.

Analysis strategy

Descriptive statistics were used to characterize the sample, taking the sampling weights into account.

Our strategy was, first, to replicate the analysis of Banks etal.1 with the more recent waves of data to ensure consistency of results when using the population between ages 50 and 80. We calculated the age-standardized disease prevalence of diabetes, hypertension, all heart diseases, stroke, lung disease, cancer and obesity. We selected these conditions because they were the ones selected by Banks etal. and because other conditions had been captured with different approaches in both surveys, making the comparison very difficult. Also following Banks etal., we selected from both surveys the demographic variables of age, sex and ethnicity and the socioeconomic (SES) variables of education and household income, and divided the SES variables into three groups: low, medium and high education and income. To ensure that the results are not affected by health-related features of minority populations in the two countries (Blacks and Hispanics in the USA, Blacks and Asian immigrants in England), analyses were restricted to White populations in the two countries. As a result, the sample sizes were 10 349 for HRS and 9405 for ELSA. We recognize that this limits the generalizability of the results and that when we talk about ‘the English’ and ‘the Americans’ we exclusively refer to the White populations in both countries.

Second, item response theory (IRT) was used to construct a metric of health with all self-report questions and measured tests addressing the selected domains in each survey. For each survey separately, we evaluated the assumptions of IRT, namely unidimensionality, local independency and monotonicity. We then combined the data of the surveys after collapsing response options with very low frequencies and then, by using the Polytomous Rasch Model, we created a single health scale.10^,11 Then we tested for differential item functioning (DIF) for survey, gender and age groups (≤64 and >64) using iterative hybrid ordinal logistic regression with change in McFadden’s pseudo R-squared measure (>0.02) as DIF criterion.12^,13 Questions and measured tests showing DIF were calibrated separately for each of the two groups showing DIF. After DIF correction, we calculated a final Rasch model. Based on the calibrations of the included questions and measured tests, a summary score of the health state of each of the individuals in the sample was calculated. We transformed the resulting scores into more meaningful values14 ranging from worst health in the sample (value 0) to best health(value 100).

Third, to test whether the health of the English was better than that of the Americans, we calculated a linear additive model15 controlling for socio-demographic and SES variables taking the sampling weights into account. Age was modelled as a non-parametric effect using P-splines. ELSA was used as the reference population, male for gender and low income and low education for the SES variables.

All analyses were performed with R version 2.15.1.16

Table 1.

Sample characteristics of HRS and ELSA populations, including response frequencies of the general health question

Table 2.

Self-reported health conditions and health state variables, by education and income, in the USA and England

Table 3.

Health state variables included in the single health scale and their threshold parameters (Thr 1–3) for the final Rasch model

Table 4.

Regression coefficients, standard errors (SE), the 90% confidence interval and p-values resulting from the linear additive model using the health score resulting from the constructed health metric as independent variable. Age was modelled as non-parametric effect using P-splines

Strengths and weaknesses in relation to other studies

The novelty of our approach consists first in the use of IRT to calibrate in a single metric the information from two independent health surveys. The creation of the single metric made possible a cross-population comparison. Second, we used information directly collected from individuals by means of questions and measured tests. Third, we operationalized health as a continuous variable based on domains of functioning and not as a dichotomous variable (healthy/unhealthy).

IRT methods have been previously used for analyses of general population surveys. WHO constructed a metric of disability, using the data of the World Health Survey19 with similar methods to ours, for its World Report on Disability.20 Thereafter, Hosseinpoor etal.21^,22 used the score derived from the WHO metric of disability to investigate differences between men and women in the context of socio-demographic factors, and Chatterji etal.23 used the same score to compare two populations, China and India. All these studies, however, used data of a single survey implemented in different countries. In our case, we combined the data of two independent surveys. The challenge we faced was data harmonization, so that the surveys could be analysed together. It was a time consuming exercise to identify those questions and measured tests that are common between surveys and thereafter to recode if necessary the response options of questions and to harmonize the data collection approaches of the measured tests. This could be the reason why Chan etal.24 used only a small subset of questions when comparing the HRS and ELSA populations. They used, as we did, an IRT model, but their results and ours are not comparable because they compared both populations with a small number of questions from a limited age group (≥65). Future health comparisons like ours will be facilitated by recent initiatives to use the same data collection approaches in different countries. One example, that hopefully will be followed by other initiatives, is the effort to standardize surveys on ageing across the world.25

To capture health we use information directly obtained from individuals by means of questions and measured tests about domains of functioning. This approach contrasts with indirect approaches for comparing the health of populations, such as health gaps and health expectancy that rely on existing population data, e.g. mortality and morbidity statistics. We opted for a direct approach because we wanted to propose a methodology based on which the information from health surveys could be utilized for comparing health not only at the population level but also at the subgroup or individual level. When health differences are found, we can specify the extent of those differences. For example, based on our metric, people between 50 and 80 with diabetes in the USA are in slightly worse health (x¯= 49.99; SD 14.6) than the English counterparts (x¯= 51.32; SD = 15.0) (analyses are not shown but can be obtained from the authors).

Our intention was to capture health from the perspective of the intrinsic capacity of the person, without taking into consideration whether the environment had a positive or negative influence on that capacity. Measured tests clearly reflect intrinsic attributes of the individual, but so do questions that require the respondent to focus on the condition of her or his body, for example the question in ELSA: ‘Because of a physical or health problem, do you have any difficulty getting up from a chair after sitting for long periods?’. We carefully selected this kind of question for our investigation and disregarded those in which the exclusive focus on the internal capacity was not as clear.

When we operationalize health as a continuous variable based on domains of functioning, we implicitly reveal our understanding of health as a unidimensional construct to which different human functioning domains contribute. In this investigation we took into consideration only a limited number of those domains, namely those that were in common in the two surveys. This was based on the practical consideration of data harmonization. Nevertheless, as the estimates for the position of persons and items in the health continuum (Figure 1) reveal, we successfully covered all health levels of the sample. To our knowledge this is the first time that measured tests have been combined with questions in a single metric using IRT methodology. This seems to be a good decision, since measured tests proved to be especially useful as they increased measurement precision in the lower margins of the person distribution.

The validity of our metric of health can also be derived from several results of this study. First of all, all of the well-known gradients of health—age, education and income levels—are captured by our metric of health. Second, the well-documented but variable differential in health between men and women is also captured by the metric.26^,27 Third, although it is well known that one has to be careful when interpreting the results of single self-rated health questions,28 the results reported in Table 1 are consistent with the results of our health metric that shows little difference in health between the two populations. When we group together the responses to the self-rated health question ‘excellent’ and ‘very good’, there is also little difference between English and Americans.

Although we rely on a unidimensional conceptualization of health, we do appreciate that, as commonly understood and confirmed in the literature on health status measurement,29^,30 health can also be treated as a multidimensional construct. We also appreciate that health can be conceptualized exclusively as the absence of disease, as Banks etal. and many others do. It is important, however, to be aware of the conceptualization used in each investigation because it will guide methodological decisions and the interpretation of results. Our conceptualization of health guided our decision to use a unidimensional approach despite the fact that cognition, emotion, sleep, vision and hearing were also loading high in other group factors in the bifactor analyses. Our decision was then confirmed in the sensitivity analyses.

It could be seen as a limitation of the study that we did not question whether health state is a linear function of the dummy variables income, education, gender and country. Should this not have been the case, the predicted values could have gone beyond the health state range 0 to 100. We decided to assume a linear function in order to facilitate the interpretation of the results and for two further reasons that can be inferred from Figure 1. First, there are very few extreme cases and, second, that the distribution of the health metric is close to normal. Thus, we can assume that the predicted values will fall within the range between 0 and 100 and that the linear model was the most appropriate.

Meaning of the study: possible explanations and implications for clinicians and policy makers

Based on these considerations, we think that the interpretation of the results of Banks etal. is not that English are healthier than Americans but that, in that section of the population that is comparable between both countries and that exclude minority populations such as Black, Hispanic and Asian immigrants, the prevalence of health conditions is higher in America than in England. As we have seen in this investigation, however, in the same section of the population English and Americans do not differ in health when assessed with a metric based on domains of functioning and when controlling for age, gender, income and education. To validate our results, we calculated the amount of variance explained by the model (using adjusted R²) including and excluding the survey. In both cases the variance explained was 17.7%, which confirms the tendency of no difference between both countries.

The most intuitive explanation of why there is little health difference, even though Americans have a higher prevalence of chronic health conditions, is that English are indeed doing worse than Americans. Langa etal.31 have found, using data from HRS and ELSA, that US adults scored better than English adults on the 24-point cognitive scale they created.To see if this holds in other domains, we compared the populations with respect to the percentage of persons having problems in specific functioning domains. The results, which are not presented in this investigation but can be obtained from the authors, show that the percentage of English having problems in memory, energy, sleep, depression or sadness, dizziness and pain is higher than corresponding percentages in America. The percentages are lower in favour of the English only in hearing, seeing and mobility. Unfortunately, the percentage in all domains cannot be compared because HRS used filters for a relatively high number of questions and the data from the whole population in each domain are not available. Nevertheless, the results for this small number of domains already supports our explanation. This explanation raises important questions for health policy, such as: ‘Is England doing a good job in prevention, but does not sufficiently take care of persons who already have health conditions?’. Further studies could shed light on this by investigating health care utilization and quality of care.

1. Banks J, Marmot M, Oldfield Z, Smith JP.Disease and disadvantage in the United States and in England. JAMA2006;295:2037–45. [PubMed] [Google Scholar]

2. Martinson ML, Teitler JO, Reichman NE.Health across the life span in the United States and England. Am J Epidemiol2011;173:858–65. [PMC free article] [PubMed] [Google Scholar]

3. Woolf SH, Aron L. (eds). National Research Council and Institute of Medicine. U.S. Health in International Perspective: Shorter Lives, Poorer Health .Washington, DC: National Academies Press, 2013. [PubMed] [Google Scholar]

4. Murray CJL, Ezzati M, Flaxman AD, et al.GBD 2010: design, definitions, and metrics. Lancet2012;380:2063–66. [PubMed] [Google Scholar]

5. Salomon JA, Mathers CD, Chatterji S, Sadana R, Üstün TB, Murray CJL.Quantifying individual levels of health: definitions, concepts and measurement issues. In: Murray CJL, Evans DB. (eds). Health Systems Performance Assessment: Debates, Methods and Empiricism. Geneva: World Health Organization, 2003. [Google Scholar]

6. Andrich D.Rasch Models for Measurement. Newbury Park, CA: Sage, 1988. [Google Scholar]

9. Crimmins E, Guyer H, Langa K, Ofstedal MB, Wallace R, Weir D.Documentation of Physical Measures, Anthropometrics and Blood Pressure in the Health and Retirement Study. Ann Arbor, MI: Survey Research Center, University of Michigan, 2008. [Google Scholar]

10. Masters GN.A Rasch model for partial credit scoring. Psychometrika1982;47:149–74. [Google Scholar]

11. Andrich D.Application of a psychometric rating model to ordered categories which are scored with successive integers. Appl Psychol Measurement1978;2:581–94. [Google Scholar]

12. Crane PK, Gibbons LE, Jolley L, van Belle G.Differential item functioning analysis with ordinal logistic regression techniques. DIFdetect and difwithpar. Med Care2006;44(Suppl 3):S115–23. [PubMed] [Google Scholar]

13. Choi SW, Gibbons LE, Crane PK.Lordif: An R package for detecting differential item functioning using iterative hybrid ordinal logistic regression/item response theory and Monte Carlo simulations. J Stat Soft2011;39:1. [PMC free article] [PubMed] [Google Scholar]

14. Andrich D, Sheridan B, Luo G.RUMM2020: Rasch Unidimensional Models for Measurement .Perth, WA: RUMM Laboratory, 2002. [Google Scholar]

15. Wood SN.Generalized Additive Models: An Introduction With R .Boca Raton, FL: Chapman Hall/CRC, 2006. [Google Scholar]

16. R Development Core TeamR: A Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing, 2013. [Google Scholar]

18. Reise SP, Morizot J, Hays RD.The role of the bifactor model in resolving dimensionality issues in health outcomes measures. Qual Life Res2007;16(Suppl 1):19–31. [PubMed] [Google Scholar]

20. World Health Organization. Technical Appendix C: Design and Implementation of the World Health Survey. Geneva: World Health Organization, 2011. [Google Scholar]

21. Hosseinpoor AR, Stewart Williams J, Amin A, et al.Social determinants of self-reported health in women and men: understanding the role of gender in population health. PLoS One2012;7:e34799. [PMC free article] [PubMed] [Google Scholar]

22. Hosseinpoor AR, Williams JS, Jann B, et al.Social determinants of sex differences in disability among older adults: a multi-country decomposition analysis using the World Health Survey. Int J Equity Health2012;11:52. [PMC free article] [PubMed] [Google Scholar]

23. Chatterji S, Kowal P, Mathers C, et al.The health of aging populations in China and India. Health Aff (Millwood)2008;27:1052–63. [PMC free article] [PubMed] [Google Scholar]

24. Chan KS, Kasper JD, Brandt J, Pezzin LE.Measurement equivalence in ADL and IADL difficulty across international surveys of aging: findings from the HRS, SHARE, and ELSA. J Gerontol B Psychol Sci Soc Sci2012;67:121–32. [PMC free article] [PubMed] [Google Scholar]

26. Verbrugge LM.A health profile of older women with comparisons to older men. Res Aging1984;6:291–322. [PubMed] [Google Scholar]

27. Gorman BK, Read JG.Gender disparities in adult health: an examination of three measures of morbidity. J Health Soc Behav2006;47:95–110. [PubMed] [Google Scholar]

28. Salomon JA, Nordhagen S, Oza S, Murray CJL.Are Americans feeling less healthy? The puzzle of trends in self-rated health. Am J Epidemiol2009;170:343–51. [PMC free article] [PubMed] [Google Scholar]

29. McHorney CA, Ware JE, Jr, Raczek AE.The MOS 36-Item Short-Form Health Survey (SF-36): II. Psychometric and clinical tests of validity in measuring physical and mental health constructs. Med Care1993;31:247–63. [PubMed] [Google Scholar]

30. Prieto L, Alonso J, Lamarca R.Classical Test Theory versus Rasch analysis for quality of life questionnaire reduction. Health QualLife Outcomes2003;1:27. [PMC free article] [PubMed] [Google Scholar]

31. Langa KM, Llewellyn DJ, Lang IA, et al.Cognitive health among older adults in the United States and in England. BMC Geriatr2009;9:23. [PMC free article] [PubMed] [Google Scholar]

32. Murray CJL, Frenk J.Health metrics and evaluation: strengthening the science. Lancet2008;371:1191–99. [PubMed] [Google Scholar]

33. Meijer E, Kapteyn A, Andreyeva T.Internationally comparable health indices. Health Econ2011;20:600–19. [PMC free article] [PubMed] [Google Scholar]