Non-attendance at re-examination 20 years after screening in the British Regional Heart Study

Transcription

1 Journal of Public Health Medicine Vol. 24, No. 4, pp Printed in Great Britain Non-attendance at re-examination 20 years after screening in the British Regional Heart Study M. C. Thomas, M. Walker, L. T. Lennon, A. G. Thomson, F. C. Lampe, A. G. Shaper and P. H. Whincup Abstract Background Maintenance of high participation rates in longitudinal studies is critical to their validity because of the possibility of bias associated with non-participation, which may differ between studies. This paper examines factors associated with participation status over time, by comparing the characteristics of non-attenders and attenders at a 20 year follow-up examination in a cardiovascular cohort study with an initial participation rate of 78 per cent. Methods A baseline examination was carried out between 1978 and 1980 and subjects have been followed up through regular reviews of general practice records, postal questionnaires, and a clinical re-examination of survivors in Data obtained by questionnaire in 1996 (Q96) and at baseline examination have been used to compare the characteristics of these recent non-attenders and attenders who remain in the cohort for further follow-up. Results The non-attendance rate of available survivors at re-examination was 23 per cent (n 1313). Rates of nonattendance were highest in Scotland and lowest in Southern England. Non-attenders were older than attenders, more likely to smoke, and reported more disabling conditions and greater use of multiple medications at Q96, but recalled similar rates of diagnosed coronary heart disease. Marked differences in social factors were identified. Total mortality rates within 1 year of re-examination were over three times higher amongst non-attenders than attenders. Conclusions Non-attendance is related to health status, risk factor status and social circumstances and may affect some estimates of disease prevalence, but does not appear to have a major impact on the estimated prevalence of coronary heart disease. Information collected on non-attenders in longitudinal studies helps to estimate, adjust and minimize these effects. Keywords: non-attendance, follow-up, bias, methodology Introduction Non-participation at recruitment is a problem for every epidemiological study, as it is widely accepted that participants and non-participants may differ from each other in many ways. Smoking is consistently found to have a positive association with non-participation, 1 3 whereas other factors, such as manual social class, heavy drinking and lack of physical activity have been identified as characteristics of non-participants in some studies. 4 6 Self-assessed health status has been shown to influence participation, 7 9 but there is little additional information available on self-reported symptoms, reported doctor diagnosed conditions or medication. Most investigators report a consistency of certain characteristics in non-participants, but it has also been asserted that non-participants do not necessarily constitute a homogeneous group, and that generalizations about them should be made with caution. 1,4 Much of the information on non-participation or nonresponse is derived from postal questionnaires and is often based on a small sub-sample of the non-participating or nonresponding population. 5,6,10 Moreover, few studies have examined factors influencing continuing participation in long-term cohort studies. The aim of this investigation has been to compare the cardiovascular and other characteristics of non-attenders and attenders at the 20 year follow-up examination in the British Regional Heart Study, and in addition to examine the possible reasons for the lapse in their participation. It is also possible to compare the M. C. Thomas, Research Assistant M. Walker, Senior Lecturer in Epidemiology L. T. Lennon, Research Assistant A. G. Thomson, Computer Programmer F. C. Lampe, Lecturer in Statistics A. G. Shaper, Emeritus Professor of Clinical Epidemiology Department of Primary Care and Population Sciences, Royal Free and University College Medical School, Rowland Hill Street, London NW3 2PF. P. H. Whincup, Professor of Cardiovascular Epidemiology Department of Public Health Sciences, St. George s Hospital Medical School, London SW17 0RE. Address correspondence to Ms Mary Thomas. mary.thomas@pcps.ucl.ac.uk Faculty of Public Health Medicine 2002

2 286 JOURNAL OF PUBLIC HEALTH MEDICINE characteristics of non-attenders with non-participants in other studies and with baseline non-participants in this study. Methods In , 9932 men aged years were invited to attend a health screening in one selected general practice in each of 24 British towns, to investigate reasons for geographical variation in the incidence of cardiovascular disease. 11 Those who participated became the British Regional Heart Study cohort, comprising 7735 men (78 per cent response rate). Over the following two decades, follow-up of subjects has been maintained through biennial reviews of general practitioner (GP) records and three self-completed postal questionnaires, in (Q5, 5 years after screening), 1992 (Q92, years after screening) and 1996 (Q96, years after screening). 12 Survivors were invited to re-attend for a 20 year follow-up examination in (20 years after screening and 2 4 years after completion of Q96) at which the same measurements as made at baseline were carried out. Before the mailed invitation to re-examination, a total of 148 subjects were excluded (see the Figure for details). Thirty-one further subjects were known not to have received their invitation. The remainder of the cohort (n 5516) was then divided into two groups, non-migrants and migrants. Non-migrant subjects (n 4922) were those still registered with the general practice from which they were initially recruited, and also subjects who had changed their GP but still resided within 10 miles of their original screening site. Migrants (n 594), were those no longer resident within 10 miles of their original practice. Migrants were invited to attend an appointment either at their original town or at one of the other 23 screening sites scheduled for a 2 week visit from the screening team during the 2 year period of activity. They were informed that their travel expenses to and from the elected site would be reimbursed. Invitations were mailed out 4 weeks before the first day of screening in each town, and reminder invitations were sent out 2 weeks later to those failing to respond. The invitation pack included an appointment card offering a time, date and site of appointment, with the option to accept, to choose an alternative date and time if the one allocated was inconvenient, or to decline attendance. It was not possible to enquire from subjects the reasons for non-participation, because of objections raised by Local Research Ethics Committees. A standard letter outlined what participation would entail, and was supplemented with a general information sheet about the study. A map illustrating how to reach the screening site and a reply-paid envelope for the reply were also included. Characteristics of subjects who attended and subjects who did not attend a 20 year follow-up examination were compared using the most recent questionnaire data available on both groups (Q96), as well as baseline physical measurements taken at initial screening on all study participants (attenders and nonattenders). Participation and non-participation relate to recruitment, attendance and non-attendance relate to re-examination, and response and non-response relate to postal questionnaire. For categorical variables, logistic regression and 2 tests were used to adjust for age. For continuous variables, linear modelling was used to adjust for age; t-tests were used for statistical testing. Results Attendance rates The overall attendance rate of available survivors, both migrants and non-migrants (n 5516), at re-examination ( ) was 77 per cent (n 4252). The 23 per cent non-attenders (n 1264) included those who refused (n 658, 11.9 per cent), those who did not reply (n 579, 9.4 per cent), and those who agreed to attend but did not do so (n 87, 1.6 per cent). Of those attending the re-examination, 94 per cent (n 4016) had responded to Q96, whereas only 69 per cent (n 911) of non-attenders had responded to Q96. Non-migrants were more likely to attend re-examination (Figure), but similar percentages of migrants and non-migrants responded to the questionnaire (Q96). Geographical variation Four main regions in Great Britain were defined: 13 Scotland, Northern England, The Midlands and Wales, and Southern England. Subjects from Scotland (Ayr, Dunfermline, Falkirk) had the lowest response rate at re-examination with 69 per cent. Response rates of subjects from the Midlands and Wales (Mansfield, Newcastle-under-Lyme, Shrewsbury and Merthyr Tydfil), and Northern England (Carlisle, Hartlepool, Darlington, Harrogate, Burnley, Dewsbury, Southport, Scunthorpe, Grimsby and Wigan) averaged 76 per cent at re-examination, whereas the highest response rates at re-examination were achieved in Southern England 79 per cent (Exeter, Guildford, Maidstone, Gloucester, Bedford, Ipswich, Lowestoft). A similar ranking of response rates had been observed in the baseline survey (75 per cent, 78 per cent and 79 per cent), in Scotland, Midlands Wales Northern England, and Southern England, respectively. After accounting for smoking, social factors and health status (by logistic regression) the regional variation in attendance was attenuated but remained significant. Characteristics of attenders and non-attenders Age Non-attenders had a mean age 69.9 years compared with a mean age of 68.8 years for attenders. This is a reversal from initial screening in , when non-participants were on average 1 year younger than participants: 48.9 years versus 50 years, respectively. 14 Characteristics at baseline assessment Using data collected at initial screening, subjects who failed to attend re-examination 20 years later were on average shorter

3 NON-ATTENDANCE AT RE-EXAMINATION 287 Figure Follow-up of subjects from screening ( ) to re-examination in ONS, Office for National Statistics; RTS, returned to sender. and lighter, but had similar mean body mass index (BMI) to those who attended, and similar mean levels of total cholesterol (Table 1). Non-attenders had a higher mean baseline blood pressure, and a lower mean forced expiratory volume in 1 second (FEV1) than those who attended. Based on longest held occupation recorded at baseline, when 57 per cent of the sample were manual workers, the proportion of manual workers at re-examination was 64 per cent among the non-attenders and 51 per cent among attenders. Characteristics at Q96 The most recent information available on non-attenders comes from a mailed questionnaire in 1996 (Q96), 2 4 years before re-examination. Health status Slightly higher rates of a recalled doctor diagnosis of heart attack, heart failure, angina, high blood pressure and diabetes were reported amongst non-attenders. Markedly higher rates of stroke, peripheral vascular disease and bronchitis were each reported more frequently amongst non-attenders. There were no differences found in the recall of other conditions (including cancer), apart from a higher reporting of prostate problems Table 1 Baseline assessments (means) in attenders and non-attenders at re-examination Attenders Non-attenders (n 4252) (n 1313) p value Height (cm) *** Weight (kg) ** BMI (kg/m 2 ) NS SBP (mm Hg) *** DBP ** Lung function (FEV1) *** Cholesterol (mmol/l) NS Manual occupation (%) *** Smoking status (%) *** All values are age adjusted. NS, non-significant. SBP, systolic blood pressure; DBP, diastolic blood pressure. **p ***p amongst attenders (16.6 per cent as opposed to 13.8 per cent for non-attenders (p 0.004)). However, self-assessed poor or fair health status was more common among non-attenders. Lifestyle The most striking finding was that non-attenders were twice as likely as attenders to be current smokers in 1996 (Table 2). How-

4 288 JOURNAL OF PUBLIC HEALTH MEDICINE Table 2 Characteristics of attenders and non-attenders to re-examination reported in Q96 Attenders Non-attenders Odds Confidence Significance Factor (n 4016) a (n 911) b ratio intervals level Recall of doctor diagnosis Heart attack (0.93,1.51) NS Angina (0.91,1.38) NS Hypertension (0.94,1.29) NS Heart failure (0.94,2.82) NS PVD (1.3,2.35) *** Stroke (1.3,2.38) *** Bronchitis (1.04,1.54) * Diabetes (0.76,1.4) NS Reported symptoms Self-assessed health status (poor/fair) (1.57,2.12) *** Depression (0.95,1.59) NS Fell last year (1.26,2.0) *** Disability, limiting activity (1.3,1.77) *** Exertional chest pain (0.8,1.26) NS Intermittent claudication (1.01,1.8) * Respiratory symptoms (1.18,1.64) *** Reported medication No medication (0.85,1.14) NS Four or more medications regularly (1.21,1.83) *** Vitamins or minerals (0.59,0.82) *** Socio-economic and lifestyle factors % % Married (0.49,0.7) *** Living alone (1.34,2.02) *** House owner (0.33,0.46) *** At school (0.59,0.81) *** Retired (0.66,0.97) * Car access (0.38,0.54) *** Had a holiday recently (0.44,0.59) *** Current smoker (1.95,2.79) *** Alcohol abstainers (1.12,1.65) ** More than 6 drinks daily (1.02,1.89) * All age-adjusted ratios. NS, non-significant. PVD, peripheral vascular disease. a Number of re-examination attenders who completed Q96. b Number of re-examination non-attenders who completed Q96. *p ** p *** p ever, the percentage of regular smokers in this age group (55 74 years) is 20 per cent lower in both attenders and non-attenders than at baseline screening, when 32 per cent and 47 per cent respectively were current smokers (see Table 1). There was a higher proportion of both heavy drinkers (drinking six or more drinks daily) and of abstainers amongst non-attenders (see Table 2). Significantly more attenders were married, home-owners, had access to a car, and were educated past the age of 16 (Table 2). Mortality rates Total mortality rates within 1 year of re-examination were significantly higher among subjects who did not attend an appointment (6.2 per cent; n 81), than those who did (1.7 per cent; n 74), even though exclusion of severely ill men (n 67) from the invitation lists had been made by the GP before invitation. The majority of deaths amongst attenders and non-attenders were non-cardiovascular disease related (54 per cent and 59 per cent, respectively). Discussion The high response rate achieved in this study, despite the age of the cohort, is probably due to the sense of commitment and loyalty on the part of the subjects involved. Our findings show that attendance was somewhat lower amongst migrants (those subjects living more than 10 miles away from the town where they were originally screened). They were, however, just as willing as non-migrants to complete a postal questionnaire in It has been shown possible and well worth while to maintain follow-up on this mobile group. The 1996 questionnaire was the main source of data drawn upon for this comparative analysis of attenders and non-attenders. A higher percentage of attenders completed this postal

5 NON-ATTENDANCE AT RE-EXAMINATION 289 questionnaire compared with non-attenders and, given the contrasting characteristics of the two groups, this difference in response may lead to an underestimation of health differences and socio-economic factors between the groups. Data from the baseline survey are also consistent with these conclusions. This study has found that characteristics of attenders and non-attenders differ considerably. Compared with attenders, a non-attender was more likely to be a smoker, a manual worker, shorter, thinner, older, living alone, have a physical disability, be on more medication, be on a lower income, have less education, and have a shorter life expectancy. It has also emerged that many of the characteristics found in other studies to be associated with non-participation reflect those displayed here by non-attenders. Town response rates When considered by region (Scotland, Northern England, Wales and the Midlands, and Southern England) overall response rates proved to be similar for baseline screening and re-examination, worsening slightly in the areas that had poorer response rates initially. Poorer response rates in Scotland and the north could be attributed to the less healthy populations in those areas, as illustrated by the morbidity and mortality data collected after a 7 year follow-up period. A less healthy population with restricted mobility may be unable to attend an examination, or may be reluctant to attend because of frequent routine hospital and surgery visits. The low response rates in Dewsbury (64 per cent at re-examination compared with 79 per cent at screening) and Maidstone (68 per cent at re-examination compared with 72 per cent at screening) may relate to the additional measurements carried out in 1996 in these two towns. In this additional study the response rates were 83 per cent for both towns. Some men in these two towns may therefore have felt reluctance for further examination so soon afterwards. Age effects Although some younger subjects might still be constrained by work, this appears to be overridden by the ill-health at the other end of the age range, resulting in the average age of nonattenders being a year older than attenders. Data collected at re-examination may therefore reflect a somewhat healthier population than the national profile of this age group. The trend for younger subjects not to participate at recruitment could be associated with work obligations, possible loss of wages, or other commitments associated with younger families. At the reexamination period the whole cohort was over 60 years of age. Health status Subjects with self-assessed poor or fair health status were less likely to attend. Similar percentages reporting poor or fair health also reported an activity-limiting disability, suggesting a relationship between these two factors and non-attendance. Indeed, almost twice as many non-attenders (25 per cent) reported both poor or fair health and a limiting disability as attenders (14 per cent). Conditions impairing physical mobility are influential determinants of attendance and will therefore be underestimated in re-examination data. Subjects with illnesses specifically related to the focus of a study may prefer not to attend because they already have a diagnosis and do not wish to undergo additional medical assessment. On the other hand, it has been found that those with a vested interest in the subject of a study will be more likely to complete a questionnaire. 15,16 Recall of doctor diagnosed IHDrelated illness in this study was non-significantly higher in reexamination non-attenders than attenders. The two conditions reported significantly more often by non-attenders were stroke and peripheral vascular disease. Reporting of physically disabling symptoms using standardized and validated questionnaires (long-standing disability, a recent fall, intermittent claudication and bronchitis) were also greater amongst nonattenders. It seems reasonable to conclude that the higher prevalence of stroke and PVD reported amongst non-attenders relates to their disabling nature, rather than their status as cardiovascular conditions, and that it is conditions that render attendance physically difficult that prevent attendance. In addition, those subjected to frequent hospital or medical consultations may feel disinclined to attend non-essential examinations. Medication The use of more medication, and less consumption of vitamin or mineral supplements by non-attenders, is consistent with poorer health. However, despite vitamin and mineral intake being lower amongst non-attenders, there were still 23.9 per cent who did report usage, providing evidence that a quarter of nonattenders display a degree of positive health behaviour, but fail to participate in re-examination. Smoking and drinking Smoking status at Q96, as well as smoking status at screening, was the strongest predictor for non-attendance to re-examination, and is likely to contribute to explaining the higher reporting of respiratory symptoms amongst non-attenders. 14 Indeed, non-attenders included twice the proportion of men with both respiratory symptoms and currently smoking (11.3 per cent) compared with attenders (5.6 per cent). The public information available on the relationship between smoking and cardiorespiratory disease could have discouraged current smokers from participating in the re-examination because of fear of adverse comment or fear of what the examination may detect. It follows that health-conscious individuals will be more likely to abstain from publicly acknowledged health risk behaviour and have a positive approach to health-related research studies. It has been found in other studies that smokers are less likely to participate in research projects. 7,8 Indeed, poorer health and non-participation have frequently been shown to go hand in hand. 6,9 Lower attendance rates by subjects considering their

6 290 JOURNAL OF PUBLIC HEALTH MEDICINE state of health as only fair or poor may be explained by the fear that attending a health check would be likely to confirm this. Non-attenders are more likely to be heavy drinkers, but also more likely to abstain from drinking alcohol. Over threequarters of these non-drinkers are ex-drinkers, many having given up drinking because of ill-health. Earlier work on characteristics associated with ex-drinkers 17 highlighted the same characteristics as found to be associated with non-attenders to re-examination. Socio-economic factors The prevalence of several advantageous socio-economic factors was found to be significantly higher in attenders and suggests that standards of living associated with domestic stability, financial security and material resources are reflected in the better health and health-related behaviour that these individuals display. Earlier findings published on deceased non-participants 14 suggested that single men, manual workers and men without close family support were over-represented in the non-participant group. Manual workers have been found to feature more amongst non-participants in other studies. 5 Implications for epidemiological studies These data suggest that non-participation or non-attendance following a survey invitation will lead to some underestimation of disease prevalence. However, the extent of underestimation depends markedly on the disease indicator being used. For the assessment of subjective poor general health and disability, nonresponse bias is likely to be marked. For a diagnosis of coronary heart disease, bias appears to be small, although it is more marked for other forms of vascular disease, notably stroke and peripheral vascular disease, conditions more seriously affecting mobility. Among major cardiovascular risk factors, the prevalence of cigarette smoking and possibly high blood pressure would be underestimated. Among social markers, the prevalence of greater degrees of deprivation and isolation are likely to be reduced by non-response. Baseline non-participation would also tend to reduce estimates of cardiovascular disease incidence; again, this would be particularly marked for severe and disabling forms of cardiovascular disease and relatively mild for diagnoses of heart attack, angina and high blood pressure. Although it is possible that non-response bias might affect estimates of exposure disease relationships, this would occur only if both exposure and disease were related differently among non-participants a possibility of which there is so far limited evidence. 18 Either the collection of information on postal questionnaire or taking steps to facilitate the attendance of disabled people would help limit some of the effects of non-participation or non-attendance bias. Conclusion Non-attenders have an increased prevalence of a wide range of disadvantageous characteristics and an extremely high early mortality rate. It is therefore important, irrespective of attendance rates achieved, that information on baseline non-participants and subsequent cohort non-attenders in epidemiological studies be collected at all stages of follow-up to highlight areas of possible bias. Acknowledgements The British Regional Heart Study, initially funded by the Medical Research Council ( ), has subsequently been supported by programme grants from the British Heart Foundation ( ) and the Department of Health, England ( ). The opinions expressed in this paper are those of the authors and not necessarily those of the sponsors. References 1 O Neill TW, Marsden D, Silman AJ. Differences in the characteristics of responders and non-responders in a prevalence survey of vertebral osteoporosis. Osteoporosis Int 1995; 5(5): Jooste PL, Yach D, Steenkamp HJ, et al. Drop-out and newcomer bias in a community cardiovascular follow-up study. Int J Epidemiol 1990; 19(2): Woodruff SI, Conway TL, Edwards CC. Increasing response rates to a smoking survey for US Navy enlisted women. Evaluation Hlth Professions 2000; 23(2): Ohlson CG, Ydreborg B. Participants and non-participants of different health categories in a health survey. A cross-sectional register study. Scand J Epidemiol 1993; 21(2): Ronmark E, Lundqvist A, Lundback B, et al. Non-responders to a postal questionnaire on respiratory symptoms and diseases. Eur J Epidemiol 1999; 15(3): Hill A, Roberts J, Ewings P, et al. Non-response in a lifestyle survey. J Publ Hlth Med 1997; 19: Etter J, Perneger TV. Analysis of non-response bias in a mailed health survey. J Clin Epidemiol 1997; 50(10): Macera CA, Jackson KL, Davis DR, et al. Patterns of non-response to a mail survey. J Clin Epidemiol 1990; 43(12): Hoeymans N, Feskens EJM, Van Den Bos GAM, et al. Non-response bias in a study of cardiovascular diseases, functional status and self-rated health among elderly men. Age Ageing 1998; 27: Bakke P, Gulsvik A, Lilleng P, et al. Postal survey on airborne occupational exposure and respiratory disorders in Norway. Causes and consequences of non-response. J Epidemiol Commun Hlth 1990; 44(4): Shaper AG, Pocock SJ, Walker M, et al. British Regional Heart Study: cardiovascular risk factors in middle-aged men in 24 towns. Br Med J 1981; 283: Walker M, Shaper AG, Lennon L, et al. Twenty year follow-up of a cohort based in general practices in 24 British towns. J Pub Hlth Med 2000; 22(4): Elford J, Phillips A, Thomson AG, et al. Migration and geographic variations in blood pressure in Britain. Br Med J 1990; 300: Walker M, Shaper AG, Cook DG. Non-participation and mortality in a prospective study of cardiovascular disease. J Epidemiol Commun Hlth 1987; 41(4): Cartwright A. Who responds to postal questionnaires? J Epidemiol Commun Hlth 1986; 40:

7 NON-ATTENDANCE AT RE-EXAMINATION Cartwright A, Windsor J. Who else responds to postal questionnaires? Are those involved in the subject of the study more likely to do so? J Commun Med 1989; 11(4): Wannamethee G, Shaper AG. Men who do not drink: a report from the British Regional Heart Study. Int J Epidemiol 1988; 17(2): Rothman KJ, Greenland S. Modern epidemiology, 2nd edn. Philidelphia, PA: Lippincott Raven, Accepted on 14 June 2002