Clusters within a general adult population of alcohol abstainers

P Colin Cryera, Jean Saundersb, Linda M Jenkinsa, Helen Nealeb, Adam C Cooka and Timothy J Petersc

a Centre for Health Services Studies, University of Kent at Canterbury, Kent, UK.
b Applied Statistics Research Unit (asru), St. Augustine's Business Park, Whitstable, Kent, UK.
c Department of Clinical Biochemistry, Guy's, King's and St Thomas' School of Medicine, London, UK.

Dr Colin Cryer, CHSS at Tunbridge Wells, Oak Lodge, David Salomons' Estate, Broomhill Road, Tunbridge Wells, Kent TN3 0TG, UK. E-mail: colin.cryer{at}hhc.umds.ac.uk


    Abstract
 Top
 Abstract
 Introduction
 Method
 Results
 Discussion
 References
 
Background Our previous study found that alcohol abstainers use acute services more and preventative services less than safe level drinkers. The observed relationships between four categories of alcohol consumption and service use were J-shaped for acute services and inverted J-shaped for preventive services. The aim of this paper was to further investigate these relationships.

Methods The design was a health and lifestyle survey of 41 000 randomly-sampled adults in SE England. The response rate was 60%. Distinctive subgroups within the alcohol abstainer group were investigated using cluster analysis, based on socio-demographic and health status variables. Odds ratios for services use for the abstainer clusters, and three alcohol consumption groups were estimated from a logistic regression model which included age, social class, ethnic group, employment status, household composition, whether the respondent was a carer, smoking habit, use of private health insurance, and health status.

Results Two clusters were formed for both males and females. Cluster 1 comprised, on average, older, frailer, and more disabled people. Cluster 2 comprised younger, healthier people, a greater proportion of whom belonged to ethnic minority groups. Cluster 2 had similar rates of use of Accident & Emergency, GP, optician, and dental services compared with safe level drinkers. Cluster 1's rates differed from those of both Cluster 2 and safe level drinkers in almost all instances.

Conclusions The J- and inverted J-shaped relationships between alcohol consumption and service use are partly explained by a subgroup of abstainers who are older, of less good health, and who use hospital, clinic, and domiciliary healthcare services much more than safe level drinkers.

Keywords Alcohol drinking, temperance, health status, health surveys, health care surveys

Accepted 1 August 2000


    Introduction
 Top
 Abstract
 Introduction
 Method
 Results
 Discussion
 References
 
In a previous study,1 associations were found between alcohol consumption and use of health services. The results showed that harmful drinkers use preventative services less than safe limit drinkers, and that they are higher than normal users of acute medical services. This work also found that abstainers are less healthy than safe level drinkers in that they use acute services more and preventative services less. The observed relationships were J-shaped for acute services and inverted J-shaped for preventive services.

J- or U-shaped relationships for the association between alcohol consumption and mortality have been reported in a number,28 but not all,9 studies. Such relationships have also been shown between alcohol consumption and coronary events or heart disease,4,915 self-rated health,7,1618 limiting long-term illness, and psychological distress.18 Reduced risk of diabetes mellitus19 and reported better health20 amongst the mild and moderate drinkers relative to those who never or rarely drink has also been found.

Relatively little work has been carried out to examine the characteristics of those who do not drink. It has been postulated that J-shaped curves may be the result of complex associations between demographic, physical, psychosocial, other confounding factors and health; the characteristics of the alcohol abstainers (other than their non-use of alcohol) may account for their poorer health outcomes.6,21,22

It has been suggested that the group of non-drinkers does not share the same characteristics as the alcohol consuming groups and is therefore not suitable as a reference group in analyses of relationships between alcohol and health outcomes.13,23 On this basis, it was also considered an unsuitable comparison group in our previous work.1

The characteristics of the alcohol abstainer group have not been fully investigated. Our previous observations led to the following proposed hypotheses:

  1. Alcohol abstainers are made up of two main groups:
    1. lifelong abstainers
    2. those who gave up drinking because of age and/or ill health.

  2. The rates of service use for lifelong abstainers are similar to those of safe drinkers.
  3. The rates of service use of those who gave up drinking due to age or ill health are dissimilar to those for safe drinkers.

The investigation of these hypotheses is the focus of this paper.


    Method
 Top
 Abstract
 Introduction
 Method
 Results
 Discussion
 References
 
Data from a health and lifestyle survey, which had a target sample of 41 000 people aged 16 and over, have been used. Details of the survey have been described previously.24 The survey covered the geographical area of the previous South East Thames Regional Health Authority area, which comprised Kent, East Sussex and south east London. This area covers all types of area from inner city to rural areas, and from deprived to affluent areas. A random sample of people aged 16 and over, stratified by Health Authority, was taken from the registers of patients held by the then Family Health Service Authorities. Each person selected received a self-completion postal questionnaire containing questions on perceived health status, health-related behaviour and use of services, as well as the socio-demographic characteristics of the responders. Four weeks later, a postcard reminder was sent to all non-respondents. This was followed by a further reminder letter and questionnaire 4 weeks later to those who had not responded.

The survey took place in 1992 and produced an overall response rate of 60%. The response rate varied from 65% in a non-metropolitan area to 41% in inner London. An analysis was carried out to test the effect of non-response. Compared to the 1991 Census, the profile of responders matched reasonably well on a number of variables, but exhibited patterns often found in postal surveys: namely a better response from women than men, a poorer response rate from younger than older people, and a better response rate from married than from single people. The responder profile was similar to that in the census for housing tenure and for household composition.

Measurement
The variables used in this analysis include a range of socio-demographic variables, health status, alcohol and cigarette consumption, and service use measures. The socio-demographic variables were age, social class, ethnic group, employment status, whether the respondent lives with children or other adults, whether he/she is a carer, and use of private health insurance.

Health status was measured using SF-3625 and by limiting long-term illness, using a question similar to that used in the 1991 census. The SF-36 is a health measurement tool developed in the US, but which has been anglicized and validated for use in Britain. It comprises 36 questions, from which eight domain scores are derived, namely: physical functioning; role limitations due to physical problems; social functioning; role limitations due to emotional problems; bodily pain; mental health; vitality/ fatigue; and general health perceptions. Each domain score is scaled so that it takes values between 0 and 100, where 0 represents the worst and 100 the best possible state.

Alcohol consumption was measured in units consumed per week and banded to reflect drinking levels included in The Health of the Nation targets.26 It was calculated from questions which asked how often respondents drank and how much they drank on typical week and weekend days. Safe drinking was up to 21 units per week for men, and up to 14 units per week for women. Intermediate drinking levels were 22–50 units per week for men, and 15–35 units per week for women. Harmful drinking was defined as more than 50 units for men, and more than 35 units for women.

The use of services was recorded in differing ways for the differing services. This included how long since the respondent last consulted their GP, visited their dentist, had their eyes tested, had their blood pressure measured, had their blood cholesterol measured, had a cervical smear test, or had a mammogram. For use of Accident & Emergency (A&E) and outpatient visits, the number of attendances in the previous 3 months was recorded. For domiciliary visits (district nurse, health visitor, chiropodist, occupational therapist, physiotherapist, community psychiatric nurse, social worker, home help, meals on wheels, your GP, voluntary agency worker, midwife, environmental health officer), and hospital clinic attendances (dietician, chiropodist, OT, physiotherapist, psychotherapist), respondents recorded whether they had been seen in the last 3 months. Respondents were also asked whether, and how many times, they had been admitted to hospital as an inpatient or a day case in the last 12 months.

This survey did not ask any direct questions to determine the reason for not consuming alcohol. Consequently, the approach taken in this work was to identify distinctive subgroups within the alcohol abstainer group, using cluster analysis, based on socio-demographic and health status variables, and to consider both the characteristics of these subgroups and their relationship with rates of service use.

Statistical analysis
Cluster analysis was used, initially, to investigate the optimum number of the alcohol abstainer clusters separately for males and females. Cluster analysis is concerned with identification of groups of similar respondents. There are two main approaches to the production of clusters i.e. hierarchic techniques and partitioning techniques, where respondents are allowed to move in and out of groups at different stages of the analysis. Techniques of this latter type have been used in this study. Initially, some arbitrary cluster centres are chosen and individuals are allocated to the nearest one. New centres are then calculated for each of the clusters formed in this way. An individual is then moved to a new cluster if it is closer to that cluster's centre than it is to the centre of its present group. During each iteration, clusters close together are merged and dispersed clusters are split. The process continues iteratively until stability is achieved with a predetermined number of clusters. We considered a range of values for the final number of clusters.

Ward's Minimum Variance Cluster Analysis27 was used to estimate the optimum number of clusters. This resulted in the calculation of the Pseudo F statistic, which has been suggested by Calinski and Harabasz28 for estimating the optimum number of clusters. It is a measure of the separation between clusters, and is calculated by the formula:

where R2 is the percentage of the variance explained by the cluster groups, c is the number of clusters, and n is the number of observations. The higher the value of this statistic the greater the separation between groups.

The SAS procedure FASTCLUS27 was then used to determine the clusters. This is more suitable when using very large data sets since the use of many other routines results in excessive computing times. The variables included in the cluster analysis were age, social class, ethnic origin, marital status, employment status, household composition, type of accommodation, use of private health insurance, smoking status, and health status as measured by the SF-36 domain scores. Dummy variables were used to enable categorical variables to be used in the cluster analysis (and also in the logistic regression analyses).

For the clusters that were formed in this way, the socio-demographic characteristics and health status were compared between clusters. For males and females separately, utilization rates were estimated for each of the two abstainer clusters and for the three alcohol consumption groups for the following service usage: A&E, GP, hospital inpatient and outpatient, domiciliary healthcare service, opticians, dental services, mammography (women only), and cervical cytology (women only) services. Linear modelling, both for ordered categorical outcomes and for dichotomized outcomes, was carried out using the SAS statistical procedure, LOGISTIC,27 to explore the relationship between the abstainer categories, alcohol use categories and service use, firstly unadjusted and subsequently adjusted for potential confounding by other variables. The potential confounding variables included in the logistic regression models were: age, social class, ethnic group, employment status, whether lives with children, whether lives with other adults, whether is a carer, limiting long-term illness, depression status, smoking, and use of private health insurance. For each service, the null hypothesis being tested was that there was no difference in the rate of service attendance between each abstainer cluster, and the safe, intermediate and harmful drinkers. This was tested against the alternative hypothesis that there were some real differences in the rates, wherever these may occur. The results are presented as odds ratios (OR) with 95% CI. We repeated this analysis but controlling only for socio-demographic variables; this additional analysis did not include the variables limiting long-term illness and depression status. This was to avoid adjusting for health status variables, which were hypothesized to be important in the definition of the abstainer clusters.


    Results
 Top
 Abstract
 Introduction
 Method
 Results
 Discussion
 References
 
Assuming less than ten clusters, the evidence was strongest for the existence of two main clusters for females, and for two or three clusters for males (Table 1Go). The measure of separation for two or three clusters amongst the males was very similar, as measured by the Pseudo F statistic. When this peaks the separation between clusters is at its highest. We proceeded with two clusters for males and two for females since our hypothesis is that the two main subgroups are ‘sick quitters’ and ‘lifelong abstainers’.21,29 Limiting our analysis to just two clusters for males still allowed us to examine the characteristics of these main clusters, and was also chosen in order to be consistent with the results of the cluster analysis for females. The two clusters that were identified were labelled Cluster 1 and Cluster 2. These clusters explain 7.2% of the variation for males, and 9.4% of the variation for females. Even a ten-cluster model would explain only 29.5% of the variance for males and 25.8% of the variance for females.


View this table:
[in this window]
[in a new window]
 
Table 1 Results of Ward's Minimum Variance cluster analysis
 
Table 2Go shows the characteristics of these two identifiable abstainer clusters for each gender, as well as the characteristics of the alcohol consumption groups. For both males and females, Cluster 1 contains, on average, older people with poorer health status, more disabled people, a large proportion of retired people, a greater proportion living in a flat, shared home, or elderly accommodation, and who were living on their own, and a greater proportion of ex-smokers (male only). These are consistent with some of the characteristics of a ‘sick quitter’ group. In contrast, for both males and females, Cluster 2 contains, on average, younger, healthier people, a larger proportion of people who had private health insurance, who were not living on their own, a greater proportion who belong to ethnic minorities, were more likely to be working, and were more likely to be non-smokers (males only). These are more consistent with the characteristics of a ‘lifelong abstainer group’.


View this table:
[in this window]
[in a new window]
 
Table 2 Characteristics of the two abstainer clusters and the other drinking groups
 
Table 3Go shows the associations between alcohol use and hospital/clinic use, and use of domiciliary healthcare services. These results for both males and females indicate that the rates for Cluster 2 are similar to those for safe level drinkers, but that the rates for Cluster 1 are much higher than those for Cluster 2, except for day case rates (females only). For females, the rates for Cluster 1 for inpatient and outpatient use are similar to those for harmful drinkers, but were much higher for hospital/ clinic use and use of domiciliary services. For males, the rates of service use for Cluster 1, for each of the services shown, are much higher than for all other alcohol consumption groups including harmful drinkers.


View this table:
[in this window]
[in a new window]
 
Table 3 Associations between alcohol consumption and hospital, clinic, and domiciliary healthcare service use
 
Tables 4 and 5GoGo show the pattern of service use by abstainer clusters and alcohol consumption groups, for those services investigated in our previous paper:1 A&E, GP, dental, optician, mammography (females only) and cervical cytology (females only) services. For both males and females, it shows that the abstainer Cluster 2 has similar rates of use for most of the above services compared with safe level drinkers, but that Clusters 1 and 2 differ in almost all instances. For females, this pattern is not observed for mammography, where the rates for Clusters 1 and 2 are similar to one another but are different from the safe level drinker group. For cervical cytology, the rates for Cluster 2 are intermediate to Cluster 1 and safe level drinker rates. The OR quoted reflect this pattern.


View this table:
[in this window]
[in a new window]
 
Table 4 Associations between alcohol use and service use—males
 

View this table:
[in this window]
[in a new window]
 
Table 5 Associations between alcohol use and service use—females
 
The OR estimates for A&E service use for males and females show that an estimated 2.7 to 3 times the number of people in abstainer Cluster 1 used A&E services more than once in the previous 3 months compared with Cluster 2. The proportion of people using A&E services more than once was similar for Cluster 2 and for safe level drinkers, and was estimated to be 2.3 (males) and 1.8 (females) times greater for harmful level drinkers. The proportion of people who had used GP services within the last 3 months was substantially greater for Cluster 1 than Cluster 2. The proportion who had used GP services in the past 3 months was similar for Cluster 2 and safe level drinkers, was slightly less for intermediate level drinkers and less again for harmful drinkers. The proportion of men and women in Cluster 1 who had visited their dentist in the previous 2 years was substantially less than for each of the other groups including Cluster 2. The proportion of men who had seen their optician in the previous 2 years was similar for both abstainer groups, it was slightly less for the intermediate level drinkers and was substantially less for the harmful drinkers. For women, the proportion who had had an eye test in the previous 2 years was greater for Cluster 1 than Cluster 2. This proportion of women was similar for Cluster 2 and safe level drinkers, and was slightly less for intermediate and harmful level drinkers.

The results of the linear modelling of service use are shown in Table 6Go. These results have been adjusted for the potential confounding variables listed in Methods. The results of the logistic regressions using ordered categorical outcomes were similar to those after the outcomes had been dichotomized. Only the latter results are shown in Table 6Go. Adjustment for confounding reduced the strength of many of the relationships shown in Tables 4 and 5GoGo. Significant associations were sustained for use of GP (female only), dentist (male and female), optician (male only), mammography and cervical cytology services.


View this table:
[in this window]
[in a new window]
 
Table 6 Odds ratios (OR) for the use of clinical services
 
The results of the logistic regression analysis, which excluded limiting long-term illness and depression status from the model, were almost identical to those shown in Table 6Go.


    Discussion
 Top
 Abstract
 Introduction
 Method
 Results
 Discussion
 References
 
It has been suggested that alcohol abstainers include two distinct groups: lifelong abstainers and ex-drinkers. It is the ex-drinkers that have characteristics likely to be associated with increased morbidity and mortality.21 Consistent with this, it has been found that former drinkers report poorer mental and physical health.5,30,31 A subgroup of the ex-drinkers are ‘sick quitters’. It has been found that self-perception of bad health was predictive of alcohol cessation amongst middle-aged men.32 On theoretical grounds, our hypotheses seem justified: that the rates of service use for lifelong abstainers are similar to those for safe drinkers, but that the rates for those who gave up drinking due to age or ill-health are dissimilar to those for safe drinkers. Evidence running counter to this, however, is the increased risk of circulatory disease amongst both lifelong abstainers and former drinkers compared with moderate drinkers that has been found in several studies.33

In the current study, our data were consistent with the existence of two clusters in the alcohol abstainer group for both males and females. The first abstainer cluster (Cluster 1) has characteristics similar to a group of ‘sick quitters’, while the second abstainer cluster (Cluster 2) have characteristics more similar to those of ‘lifelong abstainers’.

The pattern of hospital and clinic use, and the use of domiciliary services, found in our study, are consistent with abstainer Cluster 1 consisting of people with poorer health and/or older people. People belonging to Cluster 1 used the following services substantially more than either abstainer Cluster 2 or the alcohol use groups: district nurse, health visitor, chiropodist, and home help. The proportion of abstainer Cluster 1 using any domiciliary healthcare service was 34% for both males and females compared with 5–6% (males) and 6–13% (females) for the other groups.

It appears that these J- and inverted J-shaped curves, which were found in our previous paper,1 are due primarily to higher (or lower) rates for Cluster 1 relative to both Cluster 2 and the safe level drinkers. This is with the exception of mammography and cervical cytology for which Clusters 1 and 2 are more similar. This will be due in part to the fact that it was appropriate to analyse only females aged 50–64 (mammography) and females 20–64 (cervical cytology) for these services, since it was for these ages that these services were offered to women.

For these restricted age groups, Cluster 1 and Cluster 2 were more similar for some variables (age, social class, ethnicity, smoking status, and whether covered by private health insurance) but not for others (marital status, employment status, whether living alone, and health status). For example, the average age was 58.8 and 56.9 years in abstainer Clusters 1 and 2 for the mammography subset, and 52.4 and 41.3 years in Clusters 1 and 2 for the cervical cytology subset. In the total sample, the average ages of females in Clusters 1 and 2 were 68.7 and 43.2 years. On the other hand, for these restricted age groups those in Cluster 1, relative to Cluster 2, were much more likely to be single or divorced/separated, more likely to be unemployed due to disability or ill health, much more likely to be living alone, and to have poorer health status. Across all variables, however, abstainer Clusters 1 and 2 were much more similar for the age groups used in the analysis of utilization of mammography and cervical cytology services than in the female sample as a whole.

In conclusion, the J- and inverted J-shaped relationships in service use are partly explained by a subgroup of abstainers who are older or are retired people, of less good health, who use hospital, clinic, and domiciliary services much more than safe level drinkers. These findings support (but do not prove) the hypothesis that these may be a group of ‘sick quitters’, i.e. people who have given up drinking due to ill health or frailty. If so, it means that care must be taken in interpreting any J-shaped relationships between alcohol use and either service use or other health-related outcomes.


KEY MESSAGES

  • An adult general population of alcohol abstainers included two main clusters of people
  • Cluster 1 had characteristics similar to a group of ‘sick quitters’; Cluster 2 had characteristics more similar to those of ‘lifelong abstainers’.
  • The J- and inverted J-shaped relationships in service use found in our previous study are partly explained by a subgroup of abstainers (Cluster 1) who are older or are retired people, of less good health, who use hospital, clinic and domiciliary services much more than safe level drinkers.

 


    Acknowledgments
 
PCC, LMJ and ACC were funded by the NHS Executive, South East Regional Office.


    References
 Top
 Abstract
 Introduction
 Method
 Results
 Discussion
 References
 
1 Cryer PC, Jenkins LM, Cook AC et al. The use of acute and preventative medical services by a general population: relationship to alcohol consumption. Addiction 1999;94:1523–32.[ISI][Medline]

2 Shaper AG. Alcohol and mortality: a review of prospective studies. Br J Addict 1990;85:837–47.[ISI][Medline]

3 Klatsky AL, Armstrong MA, Friedman GD. Alcohol and mortality. Ann Intern Med 1992;117:646–54.[ISI][Medline]

4 Doll R, Peto R, Hall E, Wheatley K, Gray R. Mortality in relation to consumption of alcohol: 13 years' observations on male British doctors. Br Med J 1994;309:911–18.[Abstract/Free Full Text]

5 Thun MJ, Peto R, Lopez AD et al. Alcohol consumption and mortality among middle-aged and elderly US adults. N Engl J Med 1997;337: 1705–14.[Abstract/Free Full Text]

6 Fillmore KM, Golding JM, Graves KL et al. Alcohol consumption and mortality. I. Characteristics of drinking groups. Addiction 1998;93: 183–203.[ISI][Medline]

7 San Jose B, van de Mheen H, van Oers JA, Mackenbach JP, Garretsen HF. The U-shaped curve: various health measures and alcohol drinking patterns. J Stud Alcohol 1999;60:725–31.[ISI][Medline]

8 Hart CL, Davey Smith G, Hole DJ, Hawthorne VM. Alcohol consumption and mortality from all causes, coronary heart disease, and stroke: results from a prospective cohort study of Scottish men with 21 years of follow up. Br Med J 1999;318:1725–29.[Abstract/Free Full Text]

9 Moore RD, Pearson TA. Moderate alcohol consumption and coronary artery disease. A review. Medicine 1986;65:242–67.[ISI][Medline]

10 Jackson R, Scragg R, Beaglehole R. Alcohol consumption and risk of coronary heart disease. Br Med J 1991;303:211–16.[ISI][Medline]

11 Rimm EB, Giovannucci EL, Willett WC et al. Prospective study of alcohol consumption and risk of coronary disease in men. Lancet 1991;338:464–68.[ISI][Medline]

12 Maclure M. Demonstration of deductive meta-analysis: ethanol intake and risk of myocardial infarction. Epidemiol Rev 1993;15:328–51.[ISI][Medline]

13 Shaper AG, Wannamethee G, Walker M. Alcohol and coronary heart disease: a perspective from the British Regional Heart Study. Int J Epidemiol 1994;23:482–93.[Abstract]

14 Woodward M, Tunstall-Pedoe H. Alcohol consumption, diet, coronary risk factors, and prevalent coronary heart disease in men and women in the Scottish Heart Health Study. J Epidemiol Community Health 1995;49:354–62.[Abstract]

15 McElduff P, Dobson AJ. How much alcohol and how often? Population based case-control study of alcohol consumption and risk of a major coronary event. Br Med J 1997;314:1159–64.[Abstract/Free Full Text]

16 Mackenbach JP, van den Bos J, Joung IMA, van de Mheen H, Stronks K. The determinants of excellent health: different from the determinants of ill-health? Int J Epidemiol 1994;23:1273–81.[Abstract]

17 Poikolainen K, Vartiainen E, Korhonen HJ. Alcohol intake and subjective health. Am J Epidemiol 1996;144:346–50.[Abstract]

18 Power C, Rodgers B, Hope S. U-shaped relation for alcohol consumption and health in early adulthood and implications for mortality. Lancet 1998;352:877.[ISI][Medline]

19 Ajani UA, Hennekens CH, Spelsberg A, Manson JE. Alcohol consumption and risk of type 2 diabetes mellitus among US male physicians. Arch Intern Med 2000;160:1025–30.[Abstract/Free Full Text]

20 Lyons RA, Lo SV, Monaghan S, Littlepage BN. Mortality and alcohol consumption. Moderate drinking also improves health. Br Med J 1995; 310:326.[Free Full Text]

21 Wannamethee G, Shaper AG. Men who do not drink: a report from the British Regional Heart Study. Int J Epidemiol 1988;17:307–16.[Abstract]

22 Andreasson S. Alcohol and J-shaped curves. Alcohol Clin Exp Res 1998;22:359S–64S.[ISI][Medline]

23 Wannamethee SG, Shaper AG. Lifelong teetotallers, ex-drinkers and drinkers: mortality and incidence of major coronary heart disease events in middle aged British men. Int J Epidemiol 1997;26:523–31.[Abstract]

24 Barker J, Cryer C, Maryon Davis A. HealthQuest SouthEast: Regional Report. Tunbridge Wells: South East Institute of Public Health, 1993.

25 Ware J, Snow K, Kasinski M, Gandek B. SF-36 Health Survey: A Manual and Interpretation Guide. Boston, MA: The Health Institute New England Medical Centre, 1993.

26 Secretary of State for Health. The Health of the Nation. London: HMSO, 1991.

27 SAS Institute Inc. SAS/STAT User's Guide. Version 6, Fourth Edition. Cary, NC: SAS Institute Inc., 1989, pp.519–64, 823–50, 1071–1126.

28 Calinski T, Harabasz J. A dendrite method for cluster analysis. Communications in Statistics 1974;3:1–27.

29 Armstrong MA, Midanik LT, Klatsky AL. Alcohol consumption and utilization of health services in a health maintenance organization. Med Care 1998;36:1599–605.[ISI][Medline]

30 Graham K. Alcohol abstention among older adults: reasons for abstaining and characteristics of abstainers. Addict Res 1998;6: 473–87.[ISI]

31 Graham K, Schmidt G. The effects of drinking on health of older adults. Am J Drug Alcohol Abuse 1998:24:465–81.[ISI][Medline]

32 Zins M, Carle F, Bugel I, Leclerc A, Di Orio F, Goldberg M. Predictors of change in alcohol consumption among Frenchmen of the GAZEL study cohort. Addiction 1999;94:385–95.[ISI][Medline]

33 Marmot M, Brunner E. Alcohol and cardiovascular disease: the status of the U shaped curve. Br Med J 1991;303:565–68.[ISI][Medline]





This Article
Abstract
FREE Full Text (PDF)
Alert me when this article is cited
Alert me if a correction is posted
Services
Email this article to a friend
Similar articles in this journal
Similar articles in ISI Web of Science
Similar articles in PubMed
Alert me to new issues of the journal
Add to My Personal Archive
Download to citation manager
Search for citing articles in:
ISI Web of Science (9)
Request Permissions
Google Scholar
Articles by Cryer, P C.
Articles by Peters, T. J
PubMed
PubMed Citation
Articles by Cryer, P C.
Articles by Peters, T. J