Cohort Profile: The Scottish Health Surveys Cohort: linkage of study participants to routinely collected records for mortality, hospital discharge, cancer and offspring birth characteristics in three nationwide studies

Background

Although life expectancy is increasing in Scotland,¹ the nation still has the highest rates of coronary heart disease (CHD) and selected malignancies in the UK and higher rates than most countries in Western Europe.^2,3 The Scottish Health Surveys (SHeSs)—conducted in 1995, 1998 and 2003––were established to provide detailed, contemporary health information on a large, representative sample of the Scottish population. By capturing a range of behavioural, biological, psychological and social characteristics, their purpose was to monitor health in order to assist in policy formulation and the development of new health initiatives across the whole of Scotland.

How did the study come about?

Commissioned by the former Scottish Executive Health Department, the cross-sectional SHeSs took place in 1995/1996 (hereafter termed ‘1995’),⁴ 1998/1999 (‘1998’)⁵ and 2003/2004 (‘2003’).⁶ Similar in content to the Health Surveys for England, the SHeSs are principally focused on cardiovascular disease (CVD) and related risk factors.⁷ All three have been conducted by the Joint Health Surveys Unit of the National Centre for Social Research and the Department of Epidemiology and Public Health, Royal Free and University College Medical School, London; in 2003, the MRC Social and Public Health Sciences Unit and the Scottish Centre for Social Research were also partners in the survey.

What does it cover?

Data were gathered in two stages: a face-to-face interview was followed by a nurse visit for the collection of biological material. Each survey consists of information on somatic and psychological health with dedicated modules on specific conditions and risk factors, such as asthma, dental health, physical activity, eating habits, smoking and drinking.⁸ Additionally, anthropometric and, for a subsample, biological measurements such as blood pressure and blood and saliva specimens have been taken (Table 1). Early and current socio-economic information is available, with a range of measures of socio-economic status at the time of the survey as well as a measure of parental occupational social class.

Table 1.

Baseline SHeS data

Health measures

CVD

Diabetes

Respiratory health

Accidents

Food poisoning

Self-assessed general health

Longstanding illness

Acute sickness

Psychiatric morbidity

Health-related quality of lifea

Dental health

Use of health services

Use of dental services

Health-related behaviours

Alcohol consumption

Cigarette smoking

Dietary characteristics

Physical activity

Use of prescribed drugs and supplementsa

Immunizationsa

Infant feedinga

Exposure to environmental tobacco smoke

Biological measurements

Anthropometrya

Respiratory function

Blood pressure

Blood analytesa

Urine measurementsa

Biochemical measurement of smoking

Electrocardiograma

Individual socio-demographic characteristics

Economic activity status

Occupational social class

Education

Ethnicity

Religiona

Parental social classa

Household characteristics

Composition

Relationships of householdersa

Tenure

Car ownership

Receipt of state benefits

Incomea

Economic status/occupation of household reference persona

Area deprivation

^a2003 only; ^bprescribed drugs, contraceptive pills, vitamin supplements and nicotine replacement therapy; ^cheight (length, demispan), weight, waist, hip and mid-upper arm circumference (1998 and 2003); ^dtotal cholesterol, high-density lipid cholesterol, C-reactive protein (1998 and 2003), γ-glutamyl transferase (1995 only), fibrinogen, glycated haemoglobin, blood lead (1995 only) (for adults) and ferritin, total and house dust mite-specific immunoglobulin E (1998 and 2003) and haemoglobin (for children 11–15 years); sodium, potassium and creatinine (for 2003 adults).

A prospective element of the surveys is provided by linkage to data on hospitalizations and mortality. The Information Services Division (ISD) of National Health Service (NHS) Scotland maintains a database of deaths and Scottish Morbidity Records (SMRs) of cancer registrations and discharges from NHS hospitals in Scotland, linked to each other for individual patients (Table 2). The latter hold information including the reason for the visit, length of stay and waiting time. Emergency, transferred and elective admissions are included in total hospital admission; audits have shown that these data are ∼90% accurate in identifying the correct diagnosis,⁹ and completeness of SMR data is ∼99%.¹⁰ During the original survey interview, participants were asked to consent to their (or their child's) name, address and date of birth being sent to the ISD for confidential linkage to their health records. Over 90% of SHeS participants consented at each survey. Since 2004, they have been followed up with regular mortality and hospital discharge data linkage from 1981 to December 2007, and cancer registrations (also from 1981) to December 2005 (Figure 1); ongoing linkage is planned for the surveys being conducted from 2008 to 2011.¹¹ Retrospective data from 1981 until conduct of survey interview provides information on hospital diagnoses of any pre-existing morbidity. Similar linkage had previously been carried out in a randomly selected sample of individuals admitted to Scottish general hospitals in the late 1960–1970s¹² and registered users of general practices in the mid-1980s.¹³

Table 2.

Data available according to each SMR scheme

Code	Record type
SMR00	Outpatient
SMR01a	General/acute inpatient/day case
SMR02	Maternity
SMR04a	Mental health inpatient/day case
SMR06a	Cancer register
SMR11	Neonatal discharge
SMR50	Geriatric (long stay)

^aAvailable in the minimum SHeS–SMR.

Figure 1.

Linkage of study members of the SHeSs to hospital discharge, cancer registry and mortality records

The current frequencies of SMR events and deaths in the Scottish Health Surveys Cohort are shown in Table 3. Additionally, matching of maternity (SMR02),¹⁴ and neonatal (SMR11)^14,15 discharge records (separate systems from those of the SMR01 and SMR04 hospital discharges) is also possible (Table 2), providing an intergenerational component.¹⁵

Table 3.

Frequencies of deaths and hospital discharges from CVD to December 2007, and cancer registrations to December 2005, by baseline survey year

	1995	1998	2003	Total	Percentage
Number at risk	7363	8305	10 470	26 138
CHD	313	370	167	850	3.3
Acute myocardial infarction	120	156	62	338	1.3
Cerebrovascular disease	130	182	74	386	1.5
Lung cancer	52	80	19	151	0.6
Bowel cancer	36	38	10	84	0.3
Prostate cancer	14	25	12	51	0.2
Breast cancer	52	70	16	138	0.5
All cancers combined	443	541	150	1134	4.3
All deaths	504	736	331	1571	6.0

Who is in the sample?

The surveys are based on a stratified, clustered random probability sample of individuals living in private households across the whole of mainland Scotland plus the larger inhabited islands, with one in three—over 300—postcode sectors (average population of 5000) in Scotland selected at each wave. Over time, the range of ages included in the surveys has widened. The survey in 1995 only included adults up to the age of 65 years; in 1998, children over 2 years of age and adults up to the age of 75 years were sampled, and, in 2003, the full age range was surveyed. The health surveys have been designed to provide data at both the regional as well as national level, allowing geographical as well as socio-economic comparisons.¹⁶ Weighting has been applied to take account of disproportionate sampling within health regions, differing probabilities of selection within households of different sizes and within multi-occupied addresses, and differential response. SHeS interview response has generally been high, varying from 81% in 1995 and 76% in 1998 to 60% in 2003. Overall, 91–93% of adult survey participants in 1995–2003 consented to their records being linked to NHS administrative data and parental consent was given for 92% of the participating children in 2003 (parental consent was not requested in 1998) (Figure 2), yielding a combined data-set of 26 138 individuals (Table 4). Nurse visits were made to around three-quarters of adults and children who consented to such a visit (Figure 2). Whereas blood samples were taken for only one in three of 11–15-year-old children, they were provided by most adults.

Figure 2.

Respondent composition by linkage, nurse visit and provision of blood specimen in the linked SHeS–SMRs data. ^aPercentages: percentage who consented to linkage of those in the original survey sample. ^bPercentages: percentage who had a nurse visit of those who consented to linkage. ^cPercentages: percentage who gave a blood sample of those who consented to linkage. ^dBased on a target sample of 1002 children aged 11–15 years

Table 4.

SHeS respondents providing consent to linkage according to sex and age

Survey year (% target population)	Survey participants	Consenting to follow-up	Percentage
1995 (81)
16–24	1022	957	94
25–34	2000	1881	94
35–44	1803	1677	93
45–54	1534	1411	92
55–64	1573	1437	91
Total	7932	7363	93
Total men	3524	3304	94
Total women	4408	4059	92
1998 (76)a
16–24	927	866	93
25–34	1738	1621	93
35–44	1836	1703	93
45–54	1590	1469	92
55–64	1492	1351	91
65–74	1464	1295	88
Total	9047	8305	92
Total men	3941	3664	93
Total women	5106	4641	91
2003 (67)
0–15	3324	3045	92
16–24	740	668	90
25–34	1055	956	91
35–44	1620	1486	92
45–54	1411	1300	92
55–64	1411	1299	92
65–74	1091	994	91
75+	820	722	88
Total	11 472	10 470	91
Total men	3610	3283	91
Total women	4538	4142	91
All surveys
Total men	11 075	10 251	93
Total women	14 052	12 842	91
Total adults	25 127	23 093	92
Total children	3324	3045	92
Total	28 451	26 138	92

^aConsent for linkage of 3892 surveyed children aged 2–15 years in 1998 was not sought.

What is attrition like?

Since the population of Scotland is relatively stable, with low emigration,¹⁷ follow-up is available on the vast majority of consenting SHeS respondents. The SHeS data have been linked to the Community Health Index¹⁸ as at January 2008, providing details on whether respondents have been registered with a Scottish general practice at the end of 2007. This allows identification of a small number (∼4 and 7% for 1995 and 1998 surveys, respectively¹⁹) of emigrants for whom follow-up morbidity records in the linked datasets may be incomplete.

What has it found? Key findings and key publications

As the linkage for the most recent (2003) survey only took place in 2007, publications using data from the combined surveys are few and currently largely limited to a series of conference presentations. Earlier work, based on linkage of the 1995 and 1998 survey data, found smoking, forced expiratory volume, C-reactive protein, fibrinogen and blood pressure at survey baseline to be significant predictors of morbidity and mortality.^19,20 A separate analysis investigating factors associated with mortality and CHD events found that not being married, being physically inactive, being underweight and heavy smoking were significantly associated with higher risk of all-cause mortality; CHD risk increased with the number of cigarettes smoked per day, and decreased with increasing alcohol consumption, with the lowest risk in the most physically active.²¹ Selection bias assessment comparing the characteristics of the study participants with those in the general population of Scotland, showed that men in the study had lower CHD mortality than the general population. However, CHD incidence was the same, and although women in the study had a higher rate of hospitalization for CHD, there was no difference in female mortality.¹⁷ An investigation of the relationship between self-reported general health and subsequent all-cause mortality in Scotland, accounting for a history of CHD and geographical clustering found self-reported general health to be associated with all-cause mortality.²²

Linkage of 1995 and 1998 SHeS data to the SMR04 system of psychiatric admissions and death records has allowed investigation of the association between psychological distress and first psychiatric hospital admission, with 0.9% of the survey population experiencing such admissions.²³ Accounting for geographical clustering, multi-level survival analysis showed a highly significant increasing trend in the risk of first psychiatric admission with increasing psychological distress, which persisted following adjustment for a range of demographic, socio-economic and lifestyle risk factors. In addition, not being married, being in receipt of benefits, being a current smoker, being unemployed and self-assessing general health as poorer than ‘very good’ were also independently associated with psychiatric admission.²³

By linking women with data on CVD and its risk factors in the 1995 and 1998 SHeSs to their offspring's birth characteristics in the SMR02 system, a transgenerational analysis examined birth weight of offspring in relation to maternal CVD and its risk factors.¹⁵ Lower offspring birth weight was found to be independently associated with higher incidence of psychiatric morbidity and lower incidence of diabetes but not CVD in the mother.

The full potential of the data from the SHeSs being linked to Scottish hospitalization and mortality as a research resource has yet to be realized²⁴ and there are a number of projects currently underway. These include investigation of psychological distress as a predictor of future CVD and mortality;²⁵ examination of the relationship between dietary characteristics and CVD;²⁶ and several studies on self-reported alcohol consumption and alcohol-related deaths and hospital admissions.¹⁸ Other work aims to uncover the individual and combined contribution of different lifestyle factors to the excess burden of ill-health, and model CVD risk in relation to different treatment protocols.

The extensive data available from this study offer further scope for future work, such as examining: the association between socio-economic status and CVD separately in men and women, while controlling for preventable risk factors (smoking, physical inactivity and raised blood pressure); links between binge drinking and CVD; assessing the possibility of differential relationships between health behaviours and disease by socio-economic status; the relative effects of area and individual socio-economic status on health endpoints; and the role of emerging biological risk factors for CVD.

What are the main strengths and weaknesses?

The linkage of pooled data from these three large surveys with follow-up for hospitalizations and mortality has generated a large prospective cohort study that facilitates the examination of the role of a range of social, psychological, lifestyle and biological factors in the development of a range of important chronic diseases in a representative sample of the Scottish nation.^19,20,27 With an additional survey underway and others planned, this resource will increase in size with continued follow-up through record linkage. As such, it is distinct from the Health Survey for England where linkage to morbidity records is not currently possible.

The surveys are representative of individuals living in private households and thus exclude those living in communal establishments, such as residential care and prisons or those in the armed forces. There are potential sources of bias including that arising from response to the original interview and agreement to linkage of records.²⁴ The assumption that individuals are alive, if there is no trace of a death record, leads to the possible misclassification of those who have emigrated and subsequently died or had an event. However, low emigration levels mean such an occurrence will be infrequent and the additional linkage to the Community Health Index allows identification of the small number of potential emigrants with incomplete morbidity records. Retrospective data on pre-existing hospital-diagnosed morbidity prior to the conduct of survey interview will be incomplete for survey participants immigrating to Scotland after 1981. With only up to 13 years of follow-up since the baseline surveys, and with the age restrictions that were applied to the early surveys, the number of outcomes is currently modest. This does, however, represent a total of almost 200 000 person-years of follow-up and the power of the study will of course increase as the cohort matures.

Can I get hold of the data? Where can I find out more?

For each of the three surveys, data on consenting respondents are available in two distinct formats: the ‘minimum’ datasets and the ‘full’ datasets. The minimum datasets contain a set of summary variables derived from the linked SMR data (Table 2) [e.g. causes of death, incidence of acute myocardial infarction, stroke, cancer and psychiatric admissions (SMR04),²⁸ with corresponding dates] along with the complete health survey record. The full datasets contain fields from individual, anonymized patient SMR records. The minimum datasets are freely available to the wider research community by request from Catherine.storey@isd.csa.scot.nhs.uk or David.clark@isd.csa.scot.nhs.uk at ISD. Those who wish to access the full data files may request permission from the Privacy Advisory Committee at the ISD by contacting pac@isd.csa.scot.nhs.uk.¹¹

Funding

MRC and National Centre for Social Research (WBS Code U.1300.00.001.00013.01); Wellcome Trust fellowship grant (WBS Code U.1300.00.006.00012.01). The MRC SPHSU is jointly funded by the MRC and the Chief Scientist Office of the Scottish Government Health Directorates.

Conflict of interest: None declared.