Article Text
Abstract
Background Socio-economic influences over a lifetime impact on health and may contribute to poor physical functioning in old age.
Methods The authors examined the impact of both childhood and adulthood socio-economic factors on locomotor function at 63–86 years (measured with the get up and go timed walk and flamingo balance test) in the UK-based Boyd Orr (n=405) and Caerphilly (n=1196) prospective cohorts.
Results There was a marked reduction in walking speed and balance time with increasing age. Each year of age was associated with a 1.7% slower walk time and a 14% increased odds of poor balance. Participants who moved from a low socio-economic position in childhood to a high socio-economic position in adulthood had 3% slower walking times (95% CI −2% to 8%) than people with a high socio-economic position in both periods. Participants who moved from a high socio-economic position in childhood to a low adulthood socio-economic position had 5% slower walking times (95% CI −2% to 12%). Participants with a low socio-economic position in both periods had 10% slower walking times (95% CI 5% to 16%; p for trend <0.001). In Boyd Orr, low socio-economic position in childhood was associated with poor balance in old age (OR per worsening category=1.26; 95% CI 1.01 to 1.57; p=0.043), as was socio-economic position in adulthood (OR=1.71; 95% CI 1.20 to 2.45; p=0.003). Similar associations were not observed in Caerphilly.
Conclusion Accumulating socio-economic disadvantage from childhood to adulthood is associated with slower walking time in old age, with mixed results for balance ability.
- Aged
- gait
- physical performance
- social class
- socio-economic factors
- physical function
Statistics from Altmetric.com
Introduction
With increasing life expectancy, men in the UK aged 65 can expect to live a further 16.9 years and women a further 19.7 years.1 Ageing, however, is associated with impairments in many physiological systems,2 vulnerability to adverse outcomes (eg, acute illness, chronic disease, social vulnerability) and loss of functions important for daily living.3 4 Gait speed, rising from a chair and balance are simple, objective measures of physical performance that predict subsequent functional disability in older people.5–7 Socio-economic influences impact on health and mortality.8 9 Studies have found that low educational attainment is associated with the prevalence of disability in old age.10 11 Manual occupational social class and home and car ownership are associated with the risk of disability.12 Lower occupational social class has also been associated with poorer balance ability,13 walking speed14 and a physical performance battery.15 Few studies, however, have investigated associations of socio-economic circumstances in both childhood and adulthood with objective measures of physical performance in old age.
From a lifecourse perspective, different general models have been suggested to illustrate the possible influences of childhood and adult socio-economic circumstances on adult health16 17: one model proposes that adverse social circumstances may accumulate over a lifetime to increase the risk of poor health.18 Possible mechanisms linking early-life socio-economic influences with functional limitation in old age are illustrated in our research model (figure 1). Adverse childhood circumstances could be detrimental to final educational attainment, in turn influencing choices of health behaviours in adulthood, leading to chronic diseases that impair locomotor function. Other models postulate that there are critical periods in an individual's development during which there is increased vulnerability to external influences that may have lasting effects on later health.18 Like critical periods, sensitive periods are also times of rapid individual change, but exposures outside specific time windows have weaker long-term effects.18 We test the hypotheses that adverse socio-economic circumstances in childhood have long-term effects on physical function and are associated with slower walking speed and reduced balance times in old age. We test which of various lifecourse models (accumulation, critical or sensitive periods) of adverse socio-economic exposures at two time points (childhood and adulthood) in relation to physical performance are more appropriate19 for data from the Boyd Orr cohort and the Caerphilly Prospective Study (CaPS). We do not examine a separate social mobility model, as with social class measurements at only two time points, social mobility is imbedded in the accumulation, critical and sensitive period models, as others have suggested.20 We also test whether associations of social circumstances remain after taking account of health behaviours and diseases in adulthood.
Methods
Participants
The Boyd Orr study is a historical cohort based on the Carnegie (Boyd Orr) Survey of Diet and Health in Pre-War Britain, 1937–1939.21 22 In the original 1930s survey, 4999 children aged 0–19 years in 16 centres across the UK underwent physical measurements, the family completed a week long dietary account, and detailed assessments of socio-economic environment were made. In 2002, all 732 surviving study members who lived near clinics in Bristol, London, Wisbech, Aberdeen and Dundee, and had previously consented to clinical follow-up, were contacted; of these, 405 (55%) agreed to take part in a detailed clinical examination, including the get up and go23 and flamingo balance13 physical performance tests, when aged 63–83 years. Ethical approval was obtained from the Multicentre Research Ethics Committee for Scotland. All participants gave informed consent.
Data are also available from the Caerphilly Prospective Study (CaPS). The benefits of using two cohorts are to examine whether there is replication of results and to increase power where pooling the data are appropriate. CaPS recruited 2512 men aged 45–59 years between 1979 and 1983 from the town of Caerphilly, South Wales and the adjacent villages (http://www.epi.bris.ac.uk/caerphilly/caerphillyprospectivestudy.htm).24 A unique feature of the area that may have influenced the lifecourse exposures of the cohort is that Caerphilly prospered and expanded in the late 19th and early 20th century because of the success of the coal mining industry but subsequently declined from the mid-1980s with the closure of many coal mines across the UK. For the second examination (phase II 1984–1988), the original cohort was supplemented with all men of a similar age who had moved into the defined area, giving a final sample of 2398 men. Since then, the men have completed questionnaires and been examined on three further occasions: phases III (1989–1993), IV (1993–1996) and V (2002–2004). A total of 1196 men aged 66–86 years attended the phase V examination where the ‘get up and go’ and flamingo tests were undertaken. Ethical approval for the different phases of the study was given by the Ethics Committee of the Division of Medicine of the former South Glamorgan Area Health Authority and Gwent Research Ethics Committee.
Measurement of locomotor function
The get up and go test is a standardised objective measure of functional leg strength, power, mobility and balance that is strongly correlated with activities of daily living and integrates a number of basic mobility manoeuvres considered necessary for successful ageing and protection from later disability.23 The participant is observed and timed while they rise from a chair, walk 3 m, turn, walk back and sit down. The test was performed on each participant twice and the mean value used in the analysis to minimise measurement error and give a better estimate of the true underlying walking speed.
The flamingo test measures the ability to maintain postural stability in the upright position, a prerequisite for locomotor mobility.13 Balance disturbances are associated with future disability,25 are a frequent cause of hospital admission26 and predict mortality.27 The flamingo test was performed by timing how long subjects could lift one leg, while the other leg remained straight (eyes open). The position was held for as long as possible, for a maximum of 30 s. Participants used their preferred leg to stand on. The flamingo test was performed twice, unless the maximum score of 30 s was achieved in the first attempt. The best score was used in the analysis. The tests were measured using the same standardised protocol in both cohorts and the researchers trained by the same investigator (RMM).
Childhood socio-economic circumstances
In Boyd Orr, a detailed record was made of income and food expenditure for each family (both measured per head per week) at the original 1937–1939 survey, as described previously.28 Socio-economic position (SEP) in childhood was determined from the occupation of the male head of the household at the time of the original survey (four categories: I/II, professional and managerial; III, skilled; IV/V, partly skilled, unskilled, other, unclassifiable; unemployed), classified according to the Registrar General's Decennial Supplement for 1931. In CaPS, the men were asked at recruitment about their father's occupation (social class combined into five categories: I/II, professional and managerial; III skilled non-manual; III skilled manual; IV/V, partly skilled, unskilled, other, unclassifiable; unemployed).
Education
In Boyd Orr, participants were asked their age on leaving school and whether they had any part- or full-time higher education, in a follow-up questionnaire administered in 1997–1998. In CaPS, education was recorded at phase III, as the age at leaving school and the highest achieved qualification, dichotomised into whether the participant did or did not achieve higher education.
Adult SEP
In Boyd Orr, SEP in adulthood was based on the main employment of the subject (for men and unmarried women) or spouse (women) measured in 1997–1998 and classified using the 1966 Classification of Occupations in three categories (I/II, professional and managerial; III, skilled; IV/V, partly skilled, unskilled, other, unclassifiable). In CaPS, SEP in adulthood was based on participants' present or last job at phase II (1984–1988) and classified into four categories (I/II, professional and managerial; III skilled non-manual; III skilled manual; IV/V, partly skilled, unskilled, other, unclassifiable). In CaPS, information was also available on how much time the participant spent sitting, walking and lifting or carrying heavy things in their job.
SEP from childhood to adulthood
Using the occupation of each participant's father and their own SEP in adulthood, we dichotomised social class into ‘high’ and ‘low’ and determined which of four lifetime social class groups the participants fell into: high SEP in both childhood and adulthood (high to high); low in childhood and high in adulthood (low to high); high in childhood and low in adulthood (high to low); and low in both childhood and adulthood (low to low). We classified ‘high’ SEP as the top third of participants. In Boyd Orr in childhood, this was classes I and II; in adulthood social class I. In CaPS, ‘high’ SEP was considered to be classes I, II and III non-manual at both time points; the remaining groups were ‘low’ SEP.
Measures of health behaviours and status
Boyd Orr questionnaire data included general health, major diseases (cancer, stroke), the Rose angina questionnaire29 and health-related behaviours (smoking, alcohol and exercise). A diagnosis of diabetes was derived from a combination of self-reported doctor diagnosis and fasting glucose measurement. Cancer prevalence at the time of the measurement of locomotor function was based on either self-reported cancer or a notification of cancer registration received from the National Health Service Central Register. Body mass index (BMI) was measured at research clinics (weight (kg)/height (m)2) and divided into three groups (<25; 25 to <30; 30+ kg/m2 (obese)).
In CaPS, a detailed medical and lifestyle history was obtained at research clinics, including major diseases (cancer, stroke, angina and diabetes) and health-related behaviours (smoking, alcohol and exercise30), and BMI was obtained from measurements of height and weight.
Statistical analyses
All analyses were carried out using Stata version 11 (Stata, College Station, Texas).31 Walking times were natural log transformed because of a skewed distribution. Linear regression models were used to investigate associations of childhood and adulthood SEP, education, health behaviours, diseases and BMI with walking time, adjusting for age, sex and clinic location. We back-transformed the regression coefficients to obtain the percentage change in walk time per unit change in exposure (percentage change=100×(exp(β)–1)).32 Robust standard errors were calculated to account for clustering of siblings within families. Over a third of participants achieved the maximum balance of 30 s, so we dichotomised the flamingo test at the lowest 20% of performers, using a cut-point of <5 s (poor balance). Logistic regression models examined associations of socio-economic circumstances on the binary balance outcome, adjusting for age, sex and clinic location, giving the OR and 95% CI (95% CI) for poor balance. To assess whether it was suitable to pool the data from the two cohorts, we tested for interactions and applied a more liberal p value of <0.1, as the sample size may be too small to detect significant interactions at the conventional threshold. We treated age and age at leaving school as continuous variables, as there was no evidence of non-linear relationships (p values ranged from 0.24 to 0.69).
To test if either the accumulation or the critical period models best fit the data, we adapted an approach presented by Mishra, Nitsch and colleagues.19 We compared a series of nested models (representing either the accumulation or critical period models) with a fully saturated model that assumes that each of the possible trajectories in SEP between childhood and adulthood is associated with physical function (see appendix for further details). Large p values indicate that the goodness of fit of the nested model is as good as the saturated model and therefore that the hypothesis for the nested model of interest is supported by the data. We test for a sensitive period using the ‘lincom’ command in Stata,31 with small p values suggesting a difference between the time periods, providing support for a sensitive period model. In a sensitivity analysis, we classified the flamingo test into three approximately equal sized groups (0 to 6.9; 7 to 29.9; and 30 s) and performed ordinal logistic regression, to see if this method of classification detected any non-linear effects. The validity of the proportional odds assumption was checked using a Brant test.33 Performance tests often cannot be used for people who have difficulty undertaking the tests34 and non-participation can be higher in subjects who are cognitively impaired, had fallen in the previous year, use a walking aid, or have impaired activities of daily living.35 To address these concerns, we assessed the impact of missing data in CaPS by conducting a sensitivity analyses using multiple imputations by chained equations.36 37 The analysis assumes any systematic difference between the missing values, and the observed values can be explained by differences in observed data. We used the ‘ice’38 command in Stata to impute confounder and outcome missing data. Ten cycles of regression were carried out and 25 datasets imputed. In Boyd Orr, only one participant did not do the timed walk, and two participants did not attempt the balance test.
Results
The characteristics of the study members are shown in table 1.
Get up and go test
Each year of age was associated with a 1.7% slower walk time in old age in both Boyd Orr and CaPS (95% CI 1% to 2%; p<0.001 in both cohorts) (figure 2A). More deprived SEP in childhood was associated with a 3% (95% CIs: 1% to 5%; p=0.002) slower walk time per category of worsening SEP in CaPS; the estimate in Boyd Orr was a 2% slower walk time (95% CIs: 0% to 4%; p=0.12) per category of worsening SEP (table 2). For both cohorts, increased educational attainment and duration were associated with 2–4% faster walk times per extra year at school. Slower walk times were observed for those in more deprived adult SEP categories (4–5% slower per SEP category). Smoking, the lowest exercise group and greater BMI were all associated with slower walk times. The diseases with the largest effects on walk time were history of stroke and angina.
We combined the educational exposures for Boyd Orr and CaPS (p for interaction between cohorts and age left school=0.11 and for higher education=0.10), in order to maximise power when adjusting for variables shown in figure 1. For the pooled unadjusted model, each extra year at school was associated with a 2% faster walk time (95%: CI −3% to −1%; p<0.001), and achieving higher education was associated with a 5% faster walk time (95% CI −7% to −2%; p=0.001). The association between education and walk speed remained the same after adjusting for health behaviours and disease status. After adjusting for adult SEP, associations were attenuated by 50%: a 1% slower walk time was observed per extra year at school (95%: CI −2% to 0%; p=0.032), and achieving higher education was associated with a 3% faster walk time (95% CI −6% to 0%; p=0.055).
Flamingo test
Each year of age was associated with a 14% increased odds of poor balance in Boyd Orr (OR 1.14; 95% CI 1.06 to 1.21; p<0.001) and a 15% increased odds of poor balance in CaPS (OR 1.15; 95% CI 1.11 to 1.19; p<0.001) (figure 2B).
Those from more deprived SEP classes, fewer school years and lower family income and food expenditure in childhood had the shortest balance times in Boyd Orr (table 3). For CaPS, there was weak evidence that lower childhood SEP, but not fewer years at school, was associated with the poor balance. Poorer adult SEP was associated with poor balance in Boyd Orr but not CaPS. We were surprised by this and speculated that Caerphilly men, who come from a mining community, might have been protected from deterioration in balance in old age because of the physical nature of their work. Stratifying the analysis by levels of occupational physical activity showed that for men with low occupational physical activity (those who spent most of their time sitting while at work and little of their time walking or lifting or carrying heavy things) gives a 43% increased odds of poor balance among the IV/V/other category compared with social class III manual (OR 1.43; 95% CI 0.84 to 2.43). Conversely, for men with a high occupational physical activity, the OR of poor balance among the IV/V/other category compared with social class III manual was 0.63 (95%: CI 0.28 to 1.38). However, a test of statistical interaction between SEP in adulthood and work conditions on balance ability was consistent with chance variation (p=0.21). Poor balance was associated with smoking in adulthood, less exercise and increased BMI (table 3). In both cohorts, the disease with the highest odds for poor balance was diabetes.
Lifecourse models and SEP
There was no evidence that associations of lifetime SEP with walk time differed between cohorts (p interaction=0.77) so data were pooled. Participants who moved from low SEP in childhood to high SEP in adulthood had walking times that were 3% slower compared with people who had a high SEP in both periods (table 4). Participants who moved from a high SEP in childhood to a low adulthood SEP had walking times that were 5% slower compared with people with high a SEP in both periods. People who had a low SEP in both childhood and adulthood had walking times that were 10% slower (p<0.001). Adjusting for health behaviours and comorbidities attenuated the associations slightly, but the association still remained (p<0.001). There was evidence that the accumulation model provides the best fit to the data, as there was no difference from the saturated model (p=0.81). There was no evidence of a critical period (Childhood p<0.001; Adulthood p=0.03) and no evidence of a difference between the two time points, rejecting the sensitive period model (see appendix for a full description of the tests).
There was evidence that associations of lifetime SEP with balance ability differed between cohorts (p=0.04), so balance data were not pooled. In Boyd Orr, participants who had a low SEP in childhood but high SEP in adulthood had 61% increased odds of poor balance compared with those who had high SEP in both periods (table 4). Participants who moved from high to low SEP between childhood and adulthood had a 163% increased odds of poor balance and participants who had low SEP in both periods had a 292% increased odds of poor balance when compared with those with high SEP in both childhood and adulthood (p<0.001). There was evidence that the accumulation model provides a good fit to the data, as there was no difference from the saturated model (p=0.61; see appendix). However, there was also evidence that the critical period model in adulthood was also a good fit (p=0.40), but no evidence to support a critical period in childhood (p=0.014) or a sensitive period model. Similar graded associations were not observed in CaPS.
Sensitivity analyses
We performed ordinal logistic regression on the flamingo test outcome divided into three approximately equal-sized groups (no evidence against the proportional odds assumption). The association of balance ability with social class did not differ from the logistic regression results on the binary outcome. All estimates were similar in the multiple-imputation analyses compared with the complete case analysis, indicating that non-participation in performance tests was unlikely to have biased the results.
Discussion
There was a rapid reduction in walking speed and balance time with increasing age. Poor childhood socio-economic circumstances were associated with slower walking times in old age. The association between education and walk speed remained unchanged after adjusting for health behaviours and diseases and was partly attenuated after adjusting for adult SEP. We found evidence that greater accumulation of socio-economic disadvantage throughout the lifecourse was associated with slower walking times in old age, with participants experiencing low SEP in both childhood and adulthood having the slowest walk times. There were mixed results for the balance outcome: there was weak evidence of an association of childhood SEP with balance in CaPS but no evidence for level of education or adult SEP; conversely we found evidence that a greater accumulation of socio-economic disadvantage was associated with poor balance in Boyd Orr. There was also evidence to suggest that a critical period in adulthood could be operating for balance ability in Boyd Orr. However, our subjects were aged over 70 years on average, and the timing of SEP measurements was not equally spaced. The indicator of SEP in adulthood was measured as the main occupation in Boyd Orr, which would represent more years of exposure over the lifetime than childhood SEP. Adult smokers, those with lower exercise levels and those with an increased BMI had slower walk times and poorer balance in both cohorts. The disease with the largest effect estimate on walking time was stroke, while diabetes had the biggest impact on balance time. Walking speed and balance disturbances have been shown to be associated with risk of mortality in other studies27 39; these data indicate that these important measures of physical function are associated with adverse socio-economic circumstances and are, therefore, potentially modifiable outcomes.
In other studies, gait speed and physical performance scores were lower among individuals reporting a lower education level,40 41 and the occupation of a person's father was associated with reduced physical performance at 53 years,42 even after adjustment for the participant's own adult social class. The latter relationship was attenuated when behavioural risk factors were included in the model but was not changed by comorbidity. Poorer socio-economic conditions in adulthood contributed to poorer balance and chair rise times in middle age.13 These results are consistent with Boyd Orr, but there are some discrepancies with the CaPS cohort for associations of adult social class with balance. Repeating the analysis excluding men with high levels of occupational physical activity provided results consistent with Boyd Orr and previous research. Functional balance was better for those whose work also placed importance on balance abilities,43 suggesting that performance in functional balance tests is related to the balance demands of a person's work.
The main strength of the study is the prospective data collection, from childhood in Boyd Orr and from mid-life in CaPS. The cohorts provide multiple measures of socio-economic circumstances—for example, family income and expenditure on food in childhood, information on health behaviour risk factors, BMI and comorbidities. The studies have objective measures of physical performance, which are known to be associated with risk of progression to disability6 and death5 in older people.
Limitations of this study include low statistical power for identifying associations in the individual cohorts, and the total sample size for both studies may be too small to pick up conventionally significant interactions when pooling data. Loss to follow-up, a common problem for longitudinal studies, may have contributed to lack of power, but it seems unlikely to have generated any associations observed via selection bias. We did not always observe consistent findings between Boyd Orr and CaPS. This could be for a number of reasons, including: (1) Boyd Orr contains men and women, but CaPS is men only; (2) exposures were measured differently in the different cohorts—for example in Boyd Orr, occupational social class was based on the main employment in adulthood, but in CaPS it was based on the current occupation at phase II; (3) the timing of some variables differed—for example, in Boyd Orr, exercise information was recorded at the 2002 clinic, but in CaPS it was measured up to 20 years previously in Phase II. The larger amount of missing data in CaPS at the research clinics could have had an effect, but we were reassured by the results of the multiple imputation analyses. Reverse causality is an unlikely explanation for observed associations with childhood measures, as in Boyd Orr the measures of locomotor performance were taken 65 years after the childhood examination. However, physical impairments during life could influence employment in adulthood, making reverse causality possible for adult social class associations. Childhood social class was measured by recall in CaPS (but still with prospective ascertainment of outcomes) which could have more measurement error than adult SEP. Such a measurement error could dilute associations with physical function.
We have attempted to estimate the direct effect of education on walking time, after taking account of adult socio-economic position, health behaviours and diseased in adulthood. This analysis depends on the assumption that there are no unmeasured confounding factors on (1) education and walking time, and (2) the intermediate variables (adult SEP, health behaviours, diseases) and walking time.44 However, these assumptions may be violated, leading to a biased estimate if there is unmeasured confounding.
The findings may not be applicable to contemporary children, if experiencing poor childhood circumstances in the 1930s differs in effects compared with deprivation experienced today. However, socio-economic inequalities still exist for contemporary children, and socio-economic differences persist into later life.17 The results from this study have implications for potential public health interventions aimed at modifying possible mechanisms underpinning the adverse impacts of disadvantage throughout life. This work offers some explanation for data showing widening health inequalities in later life,45 and it also suggests that in order to reduce health inequalities in the current older population, health promotion advice should target older people of lower socio-economic status. Estimates of percentage change in lifetime SEP and walking speed may appear small, but they are clinically relevant as slower walking speed is strongly associated with cardiovascular mortality46—for example, people in the lowest third of walking speed had a threefold increased risk of cardiovascular death. Various mechanisms could link early-life exposures and accumulating socio-economic disadvantage with poor physical function in old age, and require further investigation. For example, growth and nutrition in early life may reduce the chances of attaining peak potential physical capacity in adulthood (eg, via the insulin-like growth factor system), thereby affecting ability and decline in physical performance in old age.
We conclude that there is strong evidence that accumulating socio-economic disadvantage from childhood to adulthood is associated with a slower walking speed, even after taking disease risk factors and comorbidities into account. There is mixed evidence for associations of socio-economic disadvantage with reduced balance ability in old age.
What is already known on this subject
Ageing is associated with loss of physical functions important for daily living. Gait speed, rising from a chair and balance are simple, objective measures of physical performance that predict subsequent functional disability in older people. Socio-economic influences impact on health and have been shown to be associated with locomotor limitation in later life.
What this study adds
We found that accumulating socio-economic disadvantage from childhood to adulthood is associated with a slower walking speed in old age, after taking disease risk factors and comorbidity into account. There is mixed evidence for the association between lifetime socio-economic disadvantage and reduced balance ability in old age.
Policy implication
Policy makers need to consider reducing both childhood and adult social adversity to maintain a healthy older population and this may have long-term influences for future generations.
Acknowledgments
We are very grateful to all cohort members who participated in the studies. The Caerphilly Prospective Study was undertaken by the former MRC Epidemiology Unit (South Wales), and the Department of Social Medicine, University of Bristol, acts as the data custodian. We thank P Morgan, director of the Rowett Research Institute, for the use of the archive of the Boyd Orr cohort and, in particular W Duncan, honorary archivist of the Rowett. The Boyd Orr follow-up was established by G Davey-Smith and S Frankel. We thank G Mishra for her help with the modelling.
References
Footnotes
Funding The current analysis was funded by a Research into Ageing PhD studentship to KB (reference: 302). The Caerphilly Prospective Study was undertaken by the former MRC Epidemiology Unit (South Wales), and the Department of Social Medicine, University of Bristol, acts as the data custodian. The Boyd Orr cohort has received funding from the Medical Research Council, the World Cancer Research Fund, Research into Ageing, United Kingdom Survivors, the Economic and Social Research Council, the Wellcome Trust, and the British Heart Foundation. The Boyd Orr DNA bank was established with a grant from the Wellcome Trust (GR068468MA) and the follow-up clinics in 2002 were funded within a Wellcome Research Training Fellowship in Clinical Epidemiology to RMM (GR063779FR). RMM and YBS are both members of the HALCyon collaborative research group, part of the UK New Dynamics of Ageing cross-council research programme. DG is a NIHR senior investigator.
Competing interests None.
Ethics approval Ethics approval was provided by the Multicentre Research Ethics Committee for Scotland (Boyd Orr) and Ethics Committee of the Division of Medicine of the former South Glamorgan Area Health Authority and Gwent REC (CaPS).
Provenance and peer review Not commissioned; externally peer reviewed.