Journal of Public Health Advance Access originally published online on October 3, 2007
Journal of Public Health 2007 29(4):405-412; doi:10.1093/pubmed/fdm062
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
An analysis of the link between behavioural, biological and social risk factors and subsequent hospital admission in Scotland
P. Hanlon, Professor of Public Health1,
R. Lawder, Statistician2
A. Elders, Senior Statistician2
D. Clark, Senior Statistician2
D. Walsh, Public Health Programme Manager3
B. Whyte, Public Health Programme Manager3
M. Sutton, Professor in Health Economics4
1 University of Glasgow, Glasgow, UK
2 Information Services Division, National Services Scotland, Edinburgh, UK
3 Glasgow Centre for Population Health, Glasgow, UK
4 University of Aberdeen, Aberdeen, UK
Address correspondence to P. Hanlon, E-mail: phil.hanlon{at}clinmed.gla.ac.uk
Objective To determine the association between risk factors and hospital admission.
Methods The 1998 Scottish Health Survey was linked to the Scottish hospital admission database.
Findings Smoking was the most important behavioural risk factor (hazard ratio: 1.90, 95% CI: 1.59–2.27). Other behavioural risk factors yielded small but largely anticipated results. Hazard ratios for biological risks increased predictably but with some exceptions (blood pressure and total cholesterol). The top quintile for C-reactive protein showed almost double the risk of admission compared with the bottom quintile (hazard ratio: 1.93, 95% CI: 1.52–2.46). Elevated body mass index (BMI) increased the risk of serious admission (hazard ratio: 1.23, 95% CI: 1.03–1.47) and raised gamma-GT increased this risk by 20% (hazard ratio: 1.20, 95% CI: 1.04–1.38). Forced expiratory volume was the ‘biological’ factor with the largest risk (hazard ratio for lowest category: 1.82, 95% CI: 1.49–2.22). All the measures of social position showed variable effects on the risk of hospital admission. Large effects on risk were associated with self assessed health, longstanding illness and previous admission.
Conclusion The linkage of national surveys with a prospective hospitalization database will develop into an increasingly powerful tool.
Keywords: hospital admission, linked datasets, risk factors, Scottish Health Survey
 |
Introduction |
|---|
 TOP Introduction MethodsResultsDiscussionFundingAcknowledgmentReferences |
|---|
Epidemiology is the ‘study of the distribution and determinants of health-related states and events in populations’.1 However, within epidemiology, more is known about the aetiology of disease than is understood about some of the determinants of health service utilization.2 Yet, both are important.
We know that the factors that determine demand for health services are complex and interacting. They include the levels of disease in a population, the volume and nature of health service supply, the behaviour of key ‘gate keepers’, the expectations and help-seeking behaviours of the population, demographic factors, social capital and much else.3–6
Although these general insights are valuable, relatively little work has been done, using large national samples, on the interaction between established disease risk factors like obesity, raised blood pressure, elevated cholesterol and smoking on the pattern of hospital utilization. Still less is understood about newer risk factors like C-reactive protein and fibrinogen.7,8 Also, although the relationship between deprivation and high levels of health service demand is well established,9 the degree to which deprivation acts through known biological and behavioural risk factors is less well understood.
The purpose of this study is to address these areas of relative ignorance by taking advantage of a new resource created by the linkage of lifestyle and hospital utilization data across Scotland. It is only in the past 3 years that it has been possible, using probability matching techniques,10 to link Scottish Morbidity Records (SMRs) with the Scottish Health Survey (SHS). SHS respondents are asked at the end of their interview whether they would be willing to be re-contacted and to allow their records to be checked against NHS registers. The survey achieves a response rate of over 80% and 90% of respondents consent to record linkage.
The SMR system records details of all in-patient and day case admissions (but not accident and emergency attendances) to Scottish NHS hospitals. The record includes information on demographic factors (e.g. age, sex and address), diagnoses, clinical procedures and means of discharge. Using patient identifying information, acute hospitalization records (SMR1) are routinely linked to mental health hospital records (SMR4), cancer registrations (SOCRATES—formerly SMR6) and Registrar General death registrations, resulting in a linked database of all such patient records covering the period 1981 to the present day.
The SHS is a national survey that collects in-depth information covering a wide range of health and behavioural topics; socio-demographic information (social class, housing tenure, car ownership, state benefits, etc.) and physiological measurements (taken by nurses) for a large representative sample of the Scottish population.11 At the time this project was initiated, there had been two waves, the first in 1995 in which 7932 adults (aged 16–64) were interviewed and the second in 1998 in which 9047 adults (aged 16–74) were interviewed. The results from the third SHS, conducted in 2003, had not at that stage been released.
|
Methods |
|---|
|
TOPIntroductionMethodsResultsDiscussionFundingAcknowledgmentReferences |
|---|
Ethical approval was sought and obtained from the Multi-centre Research Ethics Committees and the Privacy Advisory Committee's approval for the linkage was also sought and granted.
The aim was to estimate the risks of hospital admission associated with a wide range of risk factors. All analysis was undertaken on the 1998 SHS respondents. This survey population was selected (in favour of the 1995 or combined SHS populations) because there was a wider age group (16–74 years), more variables of interest were included (e.g. C-reactive protein), better measures of risk factors like exercise were employed and it became possible to restrict the outcomes to the most costly admissions.
Table 1 shows the reduced list of variables from the SHS employed by this study. A total of 8305 respondents from the 1998 SHS were linked to the April 2006 version of the Scottish hospital admission database which contains details of all SMR01 general hospital admissions, SMR04 psychiatric admissions, death records from 1981 to the end of September 2005 and cancer registrations from 1981 to December 2003. In each case, the date of event is recorded.
|
We analysed the risk of hospital admission in a ‘time to first event’ framework implemented via a Cox proportional hazards model. We accounted for different exposure times across individuals using the full date of interview and the date of death for those that died without any admission.
We began by analysing the time to first hospital admission. However, some of these admissions are brief, low cost and increasingly being dealt with in community settings or in hospital outpatient clinics. We refined the analysis to focus on admissions above average cost, which we label ‘serious’. This was achieved using published costs for each Healthcare Resource Group (HRG) to create a ‘severity index’: calculated as the reference cost for each HRG divided by the average cost for all SMR1 admissions. A severity index of more than one indicated an above average cost and was defined as a potentially serious admission. It turned out that the value of 1 was assigned to acute myocardial infarction without complications because that was equal to the average cost of all SMR1 admissions. So, a serious/costly admission was at least as costly as an acute myocardial infarction.
The results using all hospital admissions were similar to, but less powerful than, those restricted to serious hospital admissions and we present these latter results in this paper.
To estimate the impact of emigration or other reasons for loss to follow up, the SHS datasets were linked to the then most current Scottish Community Health Index (CHI) (CHI is a general practice based population register in Scotland).12 Respondents who were known to have died or to be currently on the CHI comprised 96.2% of the 1998 survey respondents which reduced the subjects to 7876. Two Cox's proportional hazard models were run. First, analysis was carried out on ‘all respondents’ and then excluding emigrants. Both models had identical statistically significant risk factor categories and there was almost no difference between the two models in terms of hazard ratios. It was, therefore, decided that all analysis would be run excluding emigrants as this would result in a cleaner dataset.
Each of the 33 risk factors was modelled individually (in age and sex adjusted Cox's proportional hazards regression models) and we represent the differences in risk using hazard ratios.
|
Results |
|---|
|
TOPIntroductionMethodsResultsDiscussionFundingAcknowledgmentReferences |
|---|
Consent was granted for 8305 SHS responses including person-identifiable information to be made available. The linkage of the SHS data to the April 2006 version of linked SMR01 ‘catalogue’ [The SMR01 catalogue is a linked file that, as well as SMR01 records, includes SMR04 records, cancer registrations and death records)] successfully linked 75% of the survey records to an admission record between March 1981, when the linked catalogue began, and September 2005. The remainder were assumed not to have experienced a hospital admission. Only admissions that occurred after the date of the SHS (1998) were used in the analysis. During the 7.5 years of follow-up, 1715 individuals had a serious hospital admission. Tables 2
–5 show the results of the age and sex adjusted models for four different categories of risk factors.
|
|
|
|
Behavioural risk factors (data are presented for males and females combined but analysis has been done by sex)
Heavy smokers had almost twice the risk of a serious admission (hazard ratio: 1.90, 95% CI: 1.59–2.27) (Table 2). Smoking is the behavioural risk factor associated with the largest single risk. Reaching the recommended level of physical activity was associated with decreased risk. For example, the risk of a serious admission for those reporting the recommended levels of activity was 21% lower (hazard ratio: 0.79, 95% CI: 0.69–0.91) than those reporting not reaching the recommended levels. Not reaching daily fruit and vegetable targets was associated with an elevated risk of 15% (hazard ratio: 1.15, 95% CI: 1.00–1.33). These results are as might have been expected but the results for alcohol consumption require more interpretation. High risk of admission for ex-drinkers may reflect their previous rather than current consumption and the fact that moderate drinkers were less at risk of admission than light drinkers may or may not reflect a protective effect of moderate alcohol consumption.
Biological risk factors
In general, for this group of risk factors, hazard ratios increased as risk increased. There were, however, some interesting exceptions (Table 3). Blood pressure, when elevated, is now routinely detected and treated. Therefore, being on treatment reflects a risk status; in broad terms, being treated carried approximately a 30% additional risk of admission but being hypertensive and untreated carried no additional risk. This probably reflects the fact that those measured as hypertensive during the survey who were not on treatment were either giving falsely elevated readings or were suffering from very mild hypertension. For total cholesterol, the moderately raised category showed less risk of serious admission (hazard ratio: 0.74, 95% CI: 0.60–0.91). However, those with a low (less desirable) level of HDL cholesterol showed 27% greater risk of hospital admission (hazard ratio: 1.27, 95% CI: 1.11–1.47). C-reactive protein and fibrinogen have both been established as risk factors for heart disease.7,8 The top quintile for C-reactive protein showed almost double the risk of a serious admission compared with those in the bottom quintile (hazard ratio: 1.93, 95% CI: 1.52–2.46), although the equivalent increase in risk for the top quintile in the distribution of fibrinogen was 73% (hazard ratio: 1.73, 95% CI: 1.33–2.25).
Obesity, reflected in an elevated BMI (obesity is defined here as a BMI > 30), was associated with a higher risk of serious admission (hazard ratio: 1.23, 95% CI: 1.03–1.47). The waist–hip ratio also predicted admission, e.g. the chance of a serious admission was elevated by 26% (hazard ratio: 1.26, 95% CI: 1.12–1.42) for those with raised, compared with ‘normal’, waist–hip ratio. The risks measured as a result of obesity were slightly higher for females than for males (not shown). Gamma-GT, which can be elevated through alcoholic liver damage, was associated with increased risk: for example, serious admissions were higher by 20% in those with an elevated level (hazard ratio: 1.20, 95% CI: 1.04–1.38) compared with those with a ‘normal’ level. Forced expiratory volume (FEV) is a strong predictor of a serious admission (hazard ratio: 1.82, 95% CI: 1.49–2.22) among those with the lowest recorded levels.
Social risk factors
Those with markers of poverty or lower social position were at greater risk of experiencing a hospital admission than those without (Table 4). For example, poorer educational achievement, being in lower social class, not owning a car, renting rather than owning your home or being unemployed were associated with a higher risk of serious admission. Typically, these markers were associated with increased risks of between 20% and 40%.
The level of neighbourhood deprivation had a measurable effect. The more deprived your neighbourhood, the greater the risk of experiencing a hospital admission. Being resident in the most deprived quintile, compared with the least deprived quintile, showed 63% increased risk for a serious admission (hazard ratio: 1.63, 95% CI: 1.35–1.96). However, measures of central heating, overcrowding, rurality, drive time to GP, straight-line distance to A&E and drive time to the nearest hospital had no significant association with the risk of admission.
Health status
The SHS asks about self-assessed health, longstanding illnesses and receipt of incapacity benefit (Table 5). The General Health Questionnaire (GHQ 12) is also used to assess aspects of psychosocial illhealth. Those who perceived their own health as ‘very bad’ had an increased risk of a serious admission nearly five times that of those who answered ‘very good’ (hazard ratio: 4.91, 95% CI: 3.35–7.18). A score of 4 or more on the GHQ12 was associated with a near doubling of the risk of admission compared with those with a score of zero (hazard ratio: 1.92, 95% CI: 1.68–2.20). Reporting a longstanding illness was associated with an increased risk of admission. That risk increased if the longstanding illness was described as limiting daily activities but also with increased reporting of the number of longstanding illnesses. Being in receipt of incapacity benefit also showed increased risk (hazard ratio 2.24, 95% CI: 1.81–2.77).
Finally, the relationship between previous hospital admissions and the risk of subsequent hospital admissions was investigated. Not surprisingly, there was a strong relationship. For example, those with four or more recorded admissions in the previous 5 years had a 4-fold increase in risk of any hospital admission (hazard ratio: 4.22, 95% CI: 3.68–4.84) compared with those with none.
|
Discussion |
|---|
|
TOPIntroductionMethodsResultsDiscussionFundingAcknowledgmentReferences |
|---|
Main findings of the study
One of the main reasons for carrying out this study was to create a resource for the future. With each successive wave of the SHS and rapidly escalating numbers of person years of hospitalization to analyse, it will be possible to create a much more sophisticated understanding of the way in which risk factors affect hospital admission. This study represents the start of this process and establishes that this approach is practical and potentially useful.
The second important finding is that a large selection of established risk factors do associate with the risk of hospital admission. This is not surprising but the relative size of each association was less predictable. Before embarking on this study, a range of local experts were informally asked by the authors for their estimation of the size of effect and relative importance of established risk factors on hospital admission. Most answered that the data were simply not available and that they would find it hard to estimate the scale of the impact. The value of this paper, therefore, is that it demonstrates how record linkage can establish the absolute scale and relative importance of these relationships.
Of the behavioural risk factors, smoking had the largest effect. This almost certainly reflects the large number of diseases known to have a causative relationship with smoking. However, it is striking how modest was the effect of, for example, diet and physical activity.
Of the biological markers, FEV1 had the largest association with the risk of hospital admission, confirming a finding from a previous study on a much older cohort.4 C-reactive protein has attracted increasing attention in recent years because of its association with inflammation, chronic stress and the metabolic syndrome13, so it was interesting to note that higher levels were associated with increased risk of admission. However, it should also be noted that the results for high blood pressure and total cholesterol were less straightforward. Most individuals with high levels of blood pressure are now treated as are many with elevated cholesterol. Consequently, their measurement of risk at the time of SHS may have been reduced through treatment but their chances of requiring a hospital admission may be higher because of the underlying condition. Consequently, it is difficult to interpret these findings.
All the measures of social position had variable but unsurprising associations with the risk of hospital admission: none of these associations were large. It is likely that much of this association is mediated through higher levels of behavioural and biological risk but further analysis, as data are accumulated over time, will allow us to establish the degree of interaction between social, biological and behavioural risk factors.
The relatively small size of the increased risk of hospital admission associated with behavioural, social and biological factors may surprise some. In contrast, large effects on risk were associated with self-assessed health, longstanding illness and history of previous admission. It may seem obvious that indicators that reflect the existence of established disease are associated with the risk of subsequent admission but the size of these effects is worthy of note.
What is already known on this topic
It is well established that record linkage can be a powerful tool for health service research.14 A large literature has established each item in Table 1 as a risk factor for disease. A very few studies have partially established the nature and size of the link between risk factors and hospital utilization4–6 but not for general population cohorts. This is the first time that an analysis has been carried out for a national population cohort.
What this study adds
A large and expanding linked database has been used to establish the presence and scale of association between risk factors and hospital admission. With the passage of time, person years of follow-up will increase. The result will be an increasingly powerful tool for analysing factors that influence hospital utilization.
Limitations of the study
The fundamentals of this study are very sound. The SHS is large, well validated and rigorously conducted.11 The SMR covers all admissions to NHS facilities and is maintained with high standards of quality control10 and the record linkage methodology employed is tried and tested.10 The main limitation of this study is the concentration on univariate associations. Plans are in place to extend to multivariate analysis but, nonetheless, the size and pattern of univariate association are of sufficient interest to merit detailed reporting in the first instance. Further research will also investigate the association between risk factors and admissions resulting from specific diagnostic categories, once sufficient records and time have accumulated to perform this work.
Future applications
In time, many applications may be found for this tool. These include scenario modelling of changes in risk factor, providing further evidence to motivate behavioural change and to support anticipatory care intervention programmes.
|
Funding |
|---|
|
TOPIntroductionMethodsResultsDiscussionFundingAcknowledgmentReferences |
|---|
NHS Health Scotland provided financial support and ISD Scotland provided staff members and substantial logistic support.
|
Acknowledgment |
|---|
|
TOPIntroductionMethodsResultsDiscussionFundingAcknowledgmentReferences |
|---|
The project team wish to thank: NHS Health Scotland for contribution of staff members; ISD Scotland for staff members and substantial logistic support; The Chief Scientist Office, and in particular Dr Peter Craig, for enabling access to the source dataset and for continuing financial support which allows ongoing linkage of this important resource.
|
References |
|---|
|
TOPIntroductionMethodsResultsDiscussionFundingAcknowledgmentReferences |
|---|
- Last JM. A Dictionary of Epidemiology (2005) 4th edition. London: IEA.
- Capewell S. The continuing rise in emergency admissions. BMJ (1996) 312:991–2.
[Free Full Text] - Kendrick S. The pattern of increase in emergency hospital admissions in Scotland. Health Bull (Edinb) (1996) 54(2):169–83.[Medline]
- Hanlon P, Walsh D, Whyte B, Scott SN, Lightbody P, Gilhooly MLM. Hospital use by an ageing cohort: an investigation into the association between biological, behavioural and social risk markers and subsequent hospital utilisation. J Public Health Med (1998) 20(4):467–76.
[Abstract/Free Full Text] - Hanlon P, Walsh D, Whyte B, Scott S N, Lightbody P, Gilhooly MLM. The link between major risk factors and important categories of admission in an ageing cohort. J Public Health Med (2000) 22(1):81–9.
[Abstract/Free Full Text] - Hanlon P, Walsh D, Whyte B, Scott SN, Hole D, Lightbody P, Gilhooly MLM. Influence of biological, behavioural, health service and social risk factors on the trend towards more frequent hospital admissions in an ageing cohort. Health Bull (Edinb) (2000) 58(4):342–52.[Medline]
- Bhatia M, Rothwell P. A systematic comparison of the quality and volume of published data available on novel risk factors for stroke versus coronary heart disease. Cerebrovasc Dis (2005) 20(3):180–6.[Web of Science][Medline]
- Lowe GD. Fibrin D-dimer and cardiovascular risk. Semin Vasc Med (2005) 5(4):387–98.[CrossRef][Medline]
- Acheson D. Independent Inquiry into Inequalities in Health Report. (1998) The Stationary Office.
- Kendrick SW, Clarke JA. The Scottish record linkage system. Health Bull (Edinb) (1993) 51:72–9.[Medline]
- The Scottish Health survey. (2003) http://www.scotland.gov.uk/Publications/2005/11/25145024/50251.
- Lawder R, Elders A, Clark D, et al. Using the linked Scottish health survey to predict hospitalisation & death. An analysis of the link between behavioural, biological and social risk factors and subsequent hospital admission and death in Scotland. (2007) Edinburgh: Information Services NHS NSS.
- Bonora E. The metabolic syndrome and cardiovascular disease. Ann Med (2006) 38(1):64–80.[CrossRef][Web of Science][Medline]
- Newcombe HB. Handbook of Record Linkage (1988) Oxford: Oxford University Press.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||