Social network interaction and mortality

J Chroa Dis Vol. 40, No. 10,pp. 949-957,1987 0021-9681/87$3.00+ 0.00 Printed in Great Britain. All rights reserved Copyrrght 0 1987Pergamon Journals Ltd SOCIAL NETWORK INTERACTION AND MORTALITY A SIX YEAR FOLLOW-UP STUDY OF A RANDOM SAMPLE OF THE SWEDISH POPULATION KRISTINA ORTH-GOM~R’ and JEFFREY V. JOHNSON* ‘National Institute for Psychosocial Factors and Health and Karolinska Institute, Stockholm, Sweden and *Department of Behavioral Sciences and Health Education, The Johns Hopkins School of Hygiene and Public Health, Baltimore, Maryland, U.S.A. (Received in revised form 17 December 1986) Abstract-This study examined the relationship between social network interactron and total and cardiovascular mortality in 17,433 Swedish men and women between the ages of 29 and 74 durmg a 6 year follow-up period. The study group was interviewed concerning their social network interactions and a total score was formed which summarized the availability of social contact. A number of sociodemographic and health related background variables known to be associated with mortality risk were also considered. Mortality was examined by linking the interview material with the Swedish National Mortality Registry. In the 6-year follow-up period 841 deaths occurred. The crude relative risk of dying during this period was 3.7 (95% CL 3.2; 4.3) when the lower social network tertile was compared to the upper two tertiles. When controlling for potential confounding effects, only age had a maJor influence on the association between social network interaction and mortality (RR age- adjusted = 1.46, 95% CL 1.25; 1.72). Controlling for age and sex, age and educational level, age and employment status, age and immigrant status, age and smoking, age and exercise habits and age and chronic disease at interview left the relative risk virtually unchanged. Controlling simultaneously for age, smoking, exercise and chronic illness yielded a risk estimate of 1.36 (95% CL 1.06; 1.69). Similar results were obtained when separately analyzing for cardiovascular disease mortality in an identical manner. Social network Social support Mortality Cardiovascular disease Population study INTRODUCTION was suggested that a possible common denomi- nator of these social conditions was lack of The concept of social support has attracted an support to indivuals from primary groups of increasing amount of attention since the midst importance to them [l, 31. of the 1970’s. The theoretical work of Cassel Over the past decade considerable evidence and others [l-5] suggesting that positive social has accumulated which indicates that there is a environmental influences would increase the in- positive relationship between low levels of social dividual’s resistance to disease agents, inspired support on the one hand and various adverse several investigators to test this hypotheses in health outcomes on the other. Several of these general population studies. Cassel’s formulation studies have used a randomly selected popu- [I] was based on earlier observations that dis- lation, a prospective design and an uncon- ease risk was associated with such diverse phe- founded outcome variable, such as mortality nomena as social disorganization, lack of accul- incidence. These investigations have yielded rel- turation and lack of psychosocial resources. It atively reliable evidence as to the health effects of social support and to the time sequence of events (i.e. does lack of social support precede Correspondence should be addressed to: Kristina Orth- Gamer, National Institute for Psychosocial Factors and or follow an illness episode). Many of these Health, Box 60 210, S-104 01 Stockholm, Sweden. studies have also been able to account for 949 950 KRISTINA ORTH-Got&~ and JEFFREYV. JOHNSON possible confounders and background variables in another context homogenous populations which provide alternative explanations for the may show so little variation in the measures association between social support and mor- studies that true associations cannot be un- tality risk. covered [14]. Among the first large investigations of the To our knowledge none of the investigations relationship between health and human ties was published so far have been able to study the the study performed in Alameda County, Cali- association of social ties and mortality in an fornia. This investigation found that social net- entire national population sample. Such a study work interaction had a definite relationship to provides the basis from which to generalize mortality, which held true, even when control- concerning the impact of social support across ling for biological risk factors, health practices, the entire social spectrum, including ethnic and and sociodemographic indicators [6]. These urban-rural variations. findings have been partially confirmed in later investigations of social relationships and mor- tality in Tecumseh, Michigan [7] and in Durham County, North Carolina [8]. More recently, in THE PRESENT INVESTIGATION the Evans County, Georgia Study, it was found The data base used in the present in- that social network interaction had a moderate vestigation was the Swedish National Survey of effect on mortality risk, but only for white males Living Conditions which was designed to de- PI- scribe aspects of social well being in the popu- Outside the United States the relationship lation that could not be measured by economic between social support and health has been indicators. This survey was originally inspired documented only in the past few years. A by concern with the unequal distribution of beneficial effect of social participation and so- economic and social resources in Sweden. The cial integration on survival has been reported in Swedish Parliament proposed that a variety of 50 year old men from Gothenburg, Sweden [lo]. characteristics concerning human well being In an analysis of data from the North Karelia should be continuously surveyed. These in- Study in Finland lack of social ties was associ- cluded objective measures of individual living ated with a substantial increase in mortality risk and working conditions that had the potential [ill. for evaluating the distribution of hardships and Although the overall picture seems to indicate social resources in the population [15]. a relationship between social support and phys- In accordance with these intentions two of the ical health outcomes, there are substantial in- annual Surveys of Living Conditions of the consistencies between studies. For example, the Swedish population included an interview ques- magnitude of the observed mortality risk is tionnaire on social relations and social inter- quite different from study to study [ 121.Further- action. This instrument was intended to quan- more, the measurement of social support itself tify the range and frequency of social interaction has not been consistent between studies. Often, with all members of the respondent’s social the measures used were not developed a priori, network. Also included in the Survey were a but were composed of whatever questions were number of other welfare indicators, such as available concerning the social sphere in on- current health status, physical and mental dis- going surveys. The measures obtained have been ability, educational and occupational level, and almost invariably quantitative descriptions of risk behaviours, such as smoking and lack of social ties, social networks, social integration physical exercise [ 161. and participation. The fact that these measures Through linkage of the Survey with the Swed- vary between studies may partially explain the ish National Mortality Registry it was possible differences in the magnitude of risk [13]. to address the following questions: Another possible explanation for inconsistent findings is that the characteristics of the various (1) Is low social support a predictor of mor- populations which have been studied also vary tality in the Swedish population ? widely. Some investigators have looked at fairly (2) If so, does the mortality risk of low social homogenous populations of white, stable, mid- support remain after adjusting for potential dle class communities, whereas others have in- confounding factors and effect modifiers such as cluded different ethnic groups and a broader age, initial chronic illness, socioeconomic fac- spectrum of social strata. As suggested by Rose, tors, and health behaviors?  Social Networks and Mortality in Sweden 951 (3) Is there a graded dose-response re- Non-response rates were 21% in the first and lationship of social support and mortality, or is 19% in the second year. Mortality rates could there a critical level beyond which no further not be directly assessed for non-respondents. effect is seen? However, in a separate study of half the sample, (4) Does social support have a similar re- a statistically non-significant but consistent ten- lationship to both total and cardiovascular mor- dency towards a higher frequency of long-term tality? illness was observed for non-respondents [17]. In another Swedish population study, non- MATERIAL AND METHODS participants were found to have a significantly higher mortality than participants [I 81. Study group As expected, there were few deaths in younger The target population of the Swedish Survey age groups. Since mortality incidence is minimal of Living Conditions was the adult Swedish until age 35, the study group was limited to population between 16 and 74 years of age. A those who were 35 or older by the end of the 6 random sample of 14,000 men and women was year follow-up period. Also, we were mainly obtained in 1976 and of 14,500 in 1977. The interested in the effects of relatively stable social sample was systematically drawn from all indi- network systems on mortality. Since the social viduals born on the 15th of each month. For contacts of young adults are known to be each individual the spouse or cohabitant was variable, we wanted to limit our sub-sample to also interviewed. In the first year (1976) the an age at which most people would have estab- latter group constituted 40% of the sample. In lished a household and a relatively enduring set the second year (1977) additional sampling was of social ties [193. The original survey did not performed among single persons with at least include those over the age of 74, because of one child living at home. Two thousand such problems with the understanding and the appli- persons were chosen. The remaining 12,500 cability of many of the questions. Thus, the were chosen using the same procedure as in the total sample for examination of social support first year. In the combined sample couples were in the two annual surveys consisted of 17,433 overly represented in comparison to the total men and women aged 29-74. population, but this was partially counteracted It was possible to compare the study sample by the separate selection of single persons in the with the total population on a number of im- second year [16]. portant variables (Table 1). The respective dis- Table 1. Comparison of the study sample with the total population of the same age (29-74 years). Figures for the total population were obtained by assigning weights, based on population census scores, to all individuals of the study sample Study sample Total population n = 17,433 n = 4,463,871 Males Females Males Females Age distribution 29-44 years 20.3% 21.5% 19.6% 18.6% 45-64 20.9% 21.1% 21.7% 22.2% 65-74 8.6% 7.6% 8.3% 9.6% Sex distribution Men 49.8% 49.6% Women 50.2% 50.4% Marital status Married 77.9% 70.9% Single, divorced or widowed 22.1% 29.1% Educational grade Mandatory 54.0% 55.0% High school or college 33.3% 32.4% University 12.6% 12.5% Immigrant status Swedish origin 89.5% 89.5% Non-Swedish origin 10.5% 10.5% Employment status- Employed 70.2% 68.2% Unemployed 29.8% 31.8% 6 year mortality 5.9% 6.6% 952 KRI~TINAORTH-GO~~%R and JEFFREYV. JOHNSON Table 2. Items in the social network interaction index. Means, standard devi- ations and Pearson correlation coefficients. (p < 0.001 for all correlation coefficients) Item index correlation Variable Mean SD coefficients Parent contact frequency 2.24 2.26 0.56 Parent last visit 2.02 2.31 0.57 Child contact frequency 4.69 2.09 0.49 Child last 4.68 2.09 0.49 Sibling contact frequency 3.12 1.76 0.32 Sibling last visit 3.14 2.09 0.34 Marital status 4.90 2.07 0.24 Score 0 = non-married 6 = married Youth friend contact frequency 3.02 2.46 0.39 Youth friend phone or letter contact 2.88 2.42 0.33 Neighbor contact frequency 3.29 1.93 0.45 Neighbor last visit 3.74 2.10 0.45 Neighbor exchange of assistance 3.26 2.22 0.50 Casual neighborhood interaction 4.92 1.44 0.37 Friend contact frequency 4.03 1.29 0.34 Friend last visit 4.53 1.56 0.37 Co-worker contact frequency 1.47 1.79 0.44 Co-worker last visit 1.53 1.95 0.45 Social contact at work 2.61 2.35 0.42 Total social interaction index 59.69 15.60 tributions of age, sex, educational level, immi- index was formed by summing across items. grant and employment status were found to be Scores varied between 0 and 106 with a mean of nearly identical. In the study sample, married 59.69 and a standard deviation of 15.60. The persons were somewhat more common (7%) items and their correlations with the total scale and mortality was slightly lower (0.7%) as score are shown in Table 2. In addition to the compared to the total population [19]. reported frequency of contacts with the various members of the social network, marital status Measures of social network interaction was also included in the interaction index, for The Survey of Living Conditions contained we considered it to be a crucial aspect of social 18 items concerning different aspects of the interaction. respondent’s social network. These question- naire items were identically worded and scored Measures of other risk factors for mortality for both years. The questions were constructed The Survey of Living Conditions included to elicit objective responses regarding particular other items of potential importance for evalu- situations in the respondent’s life. For example, ating health and longevity. A number of ques- subjects were asked to indicate how often they tions on medically diagnosed long-term illness see a particular person such as a parent, sibling, and disability were used in combination as a friend or neighbor, and to indicate the prox- single background variable. Response alterna- imity of the last visit. Respondents were not tives were combined and dichotomized into questioned concerning satisfaction or emotional presence or absence of any chronic illness at the attitude toward any source of contact. This is time of interview. Questions on specific diag- consistent with the general purpose of the Swed- noses and treatment were included in the inter- ish Survey of Living Conditions to construct an view and coded according to the International objective picture of the resources available to Classification of Diseases (8th revision) by two the Swedish population [20]. Therefore, our independent consultant physicians. This pro- social network scale estimates the number of cedure was validated for accuracy by examining sources of social contact and the degree of the predictive value of cardiovascular disease contact with each source. for cardiovascular mortality. The correlation Each item was scored from 0 (no availability) between having the diagnosis of cardiovascular to 6 (high availability) with intermediate levels. disease at the time of interview and later dying A total availability of social network interaction of this condition was 0.84 (gamma).  Social Networks and Mortality in Sweden Questions on smoking were combined and 30 , subjects were divided into three categories (non- LEGEND smokers, smokers of 1-19 cigarettes/day and 20 25 .. 0 Low Medium or more cigarettes/day). Questions on physical 6 2. _. g High exercise in leisure time were combined and subjects were assigned to one of three groups % (no exercise, occasional exercise, and regular 5 l5 exercise at least twice a week). 2 10 -. Level of education was trichotomized into mandatory (9 year) schooling, high school or 5 ‘. college/university. Respondents were rated as working or not working at the time of the 01 Jza interview. Finally immigrants included all first 29-44 45-54 55-64 65-74 generation non-Swedes. Fig. 1. Total mortality in social network tertiles for Swedish men. Analysis of mortality an estimate of mortality risk was obtained for For the analysis of mortality the Survey of the lowest social network tertile compared to Living Conditions was linked to the National the remaining two thirds of the population. Death Registry over a 6 year follow-up period. Ninety-five percent confidence limits were Total mortality and mortality from arte- obtained using the following expression; riosclerotic heart disease and cerebrovascular B + SE (1.96), where B = logistic coefficient, and peripheral vascular disease (ICD-codes SE = standard error. 400-404, 410-414, 421, 430-436, 440-445) were analyzed. Eight hundred and forty-one deaths occurred in the study group (562 men, RESULTS 279 women). Of these deaths, 414 (292 men and Social network interaction scores were found 122 women) were ascribed to cardiovascular to be normally distributed in the study popu- diseases as defined above. In Table 1, the 6 year lation. Subjects were divided into three equally mortality experience in the study group is com- large groups according to their individual index pared to that of the total population within the scores. Total and cardiovascular mortality rates same period of time. Mortality was somewhat were compared in low, medium and high index lower in the study group. This was possibly due tertiles for men and women. to poorer health of those not sampled and Figure 1 shows the mortality rates of men in non-married persons [ 17, 181. the social network tertiles in each of four age classes (29-44, 45-54, 55-64 and 65-74 years). Statistical methods In all four age categories mortality rates were The relative risk ratio, estimated by the odds highest in the lowest network tertile. Further- ratio, was the principal measure of association more, in all but the oldest age category there used in the analyses. Both multiple contingency was little difference in mortality rates between table and logistic regression procedures were medium and high social network tertiles. performed. In controlling for confounding fac- Figure 2 shows the same comparison in tors the data were first stratified into categories women. In all age categories except the oldest of the confounding factor and a weighted aver- one (those aged 65-74) mortality rates were age of the stratum specific rate ratios was ob- highest in the lowest network tertile. As in the tained using Mantel-Haenszel procedures [21]. analysis of men there was little difference in Next, multiple logistic regression analyses were mortality between medium and high tertiles performed with maximum likelihood estimates within each age group. computed by the Newton-Raphson method In the oldest age classes the picture was less [22]. The estimates of adjusted relative risks clear. In men there seemed to be a graded from both procedures were compared, and relationship between the degree of social net- found to be almost identical. In this paper we work interaction and mortality, whereas for only report the findings from the logistic re- older women those with the highest network gression estimates. score actually had the greatest risk of dying. By taking the antilog of the logit coefficient, However, it should be noted that the high 954 KR~INA ORTH-&I&I and JEFFREYV. JOHNSON had no effect at all on the risk estimate. Control- 1 20 LEGEND ling for age and chronic illness or disability at 16 interview reduced the risk to 1.38, which was still statistically significant. Similarly, control- 14 VI ling for unemployment reduced the risk to 1.36. $ 12 d In the second half of the study group (those e 10 s interviewed in 1977), data on smoking and 2 6 exercise was also available. Controlling for age 6 and smoking habits yielded a relative risk esti- mate of 1.53 (95% CL 1.20; 1.96) and control- ling for age and sedentary lifestyle an estimate of 1.43 (95% CL 1.12; 1.84). Thus, the mag- 29-44 45 -54 55-64 65-74 nitude of the risk ratios were not affected to any Fig. 2. Total mortality in social network tertiles for Swedish great extent. However, due to a relatively small women. number of deaths, and the considerably reduced sample size confidence limits were slightly wider and statistical significance somewhat weaker network tertile of the oldest female age group than in the previous analyses (Table 3). was extremely small (n = 84) as compared to In the final model the available well known medium (n = 392) and low (n = 1028) tertiles. mortality predictors as well as cardiovascular When calculating the crude relative mortality risk factors were controlled for simultaneously. risk of all men and women in the lower as Thus, including age, chronic illness at time of compared to the upper two tertiles, an estimated interview, smoking and sedentary life style relative risk of 3.71 (95% CL 3.21; 4.29) was yielded a risk estimate of 1.34 (95% CL 1.06; obtained, which was statistically significant 1.69), which was somewhat lower than the (p < 0.001). estimates obtained in the previous analyses As indicated above, the frequency of social yet continued to be statistically significant interaction and social contacts decreased with (p < 0.02) (Table 3). increasing age. Thus controlling for the poten- Given the size of the study population and the tial confounding effect of age was the first step length of the follow-up period, a detailed anal- in examining possible alternative explanatory yses of specific causes of death was not feasible. factors. The age adjusted relative risk estimate However, it was possible to analyze separately was 1.46 (95% CL 1.25; 1.72), which was lower for the most common cause of death, cardio- than the crude estimate but still statistically vascular disease mortality. The analyses were significant. carried out in the same way as for total mor- In the subsequent analyses the effects of other tality. The crude relative risk estimate was 4.05 potential confounders on the age adjusted rela- (95% CL 3.28; 4.98). As in the analyses of total tive mortality risk were examined (Table 3). mortality, this estimate was substantially re- Controlling for age and sex, for age and edu- duced, when adjusting for age (RR = 1.38 95% cational level and for age and immigrant status CL 1.12; 1.72) (Table 4). Table 3. Relative mortality risk estimates of the lower as compared to the upper two social network tertiles (n = 17,433, number of deaths = 841) RR 95% CL P Crude 3.72 3.21; 4.29 0.0000 Adjusted for: Age 1.46 1.25; 1.72 o.oooo Age + sex 1.49 1.26; 1.76 0.0000 Age + chronic illness 1.38 1.18; 1.63 0.0001 Age + employment status I .36 1.16; 1.61 0.0001 Age + educational level 1.46 1.25; 1.71 0.0000 Age + immigrant status 1.46 1.24; 1.71 0.0000 Age + smoking habits* 1.53 I .20; 1.96 0.0008 Age + exercise habits* 1.43 1.12; 1.84 0.0047 Age + smoking + exercise + chronic illness* 1.34 I .06: 1.69 0.0243 *Information on smoking and exercise habits available in only half of the study population (n = 8718, number of deaths = 35 1).  Social Networks and Mortality in Sweden 955 Table 4. Relative cardiovascular mortality risk estimates of the lower as compared to upper two social network tertiles (n = 17,433, number of cardiovascular deaths = 414) RR 95% CL p Crude 4.05 3.28; 4.98 0.0000 Adjusting for: Age 1.38 1.12; 1.72 0.0043 Age + sex 1.41 1.14; 1.75 0.0025 Age + cardiovascular disease 1.36 1.09; 1.69 0.0076 Age + employment status 1.28 1.03; 1.60 0.0310 Age + educational level 1.38 1.11; 1.73 0.0041 Age + immigrant status 1.39 1.12; 1.73 0.0041 Age + smoking* 1.49 1.05; 2.12 0.0245 Age + exercise* 1.41 0.99; 2.00 0.0555 Age + smoking + exercise + cardiovascular disease* 1.37 0.97; 1.96 0.0768 *Information on smoking and physical exercise available in only one-half of the sample (n = 8718, number of cardiovascular deaths = 175). Controlling for other available confounders remained relatively unchanged. Because these had very little effect on the age-adjusted relative two health behaviors could only be assessed in risk estimate (Table 4). Only two factors one half of the study group, the statistical reduced the risk. Controlling for age and significance of the social support risk estimate cardiovascular illness at interview yielded a risk was somewhat diminished. Nevertheless, even estimate of 1.36 (95% CL 1.09; 1.69) and after taking into account the combined effects of controlling for unemployment at interview chronic illness, smoking and physical inactivity, yielded an estimate of 1.28 (95% CL 1.12; 1.73). there was still a moderate, independent, excess Both risk estimates were statistically significant mortality risk of approximately 30% associated (p < 0.01). with low social support. A very similar pattern In the multifactorial model, including age, was found in the association between low social presence of cardiovascular illness, smoking and support and the major specific cause of death, exercise habits a relative risk estimate of 1.37 cardiovascular disease mortality. These results (95% CL 0.97; 1.96) was obtained. This was suggest that lack of social support is associated almost equal in magnitude to the estimate ob- with a moderate risk for both total and cardio- tained for total mortality, but due possibly to vascular mortality. the smaller number of cardiovascular deaths Furthermore, there was some evidence of a together with restricted sample size, it was critical level of social contacts needed, above of only borderline statistical significance which no further effect could be detected. The (p c 0.077) (Table 4). third of the study population with the lowest social support had an increased mortality risk, DISCUSSION but there was little or no difference between the medium and high support groups. To further In the present study the health related effects explore this suggested threshold effect we ana- of low social support were analyzed in a repre- lyzed the same risks in quartiles and quintiles of sentative sample of the adult Swedish popu- the social support measure. These analyses also lation. Low social support, as indicated by a indicated that the critical increase in mortality relatively low level of social interaction and few risk was to be found in the lower 20-30% of the social ties, was associated with an excess mor- population. tality risk of approximately 50%) after adjust- Although a number of researchers have ment for the effects of age and sex had been reached the conclusion that there is an associ- performed. Controlling for other factors, which ation between lack of social support and mor- were considered to influence both the avail- tality there are a number of discrepancies that ability of social contacts and mortality risk, merit attention [23,24]. The Alameda County (presence of chronic illness and disability, un- study, for example, demonstrated that there was employment, low education and immigrant sta- an independent mortality risk for these with low tus) did not substantially alter the strength of social interaction scores, which, although the association. Even when controlling for life slightly higher for women than for men, re- style factors (smoking or lack of physcial exer- mained significantly elevated in both sexes [6]. A cise) the estimate of the magnitude of risk similar finding was reported by the Durham 956 KRISTINAORTH-GO&R and JEFFREYV. JOHNSON County investigators [8]. In the Tecumseh study, either alone, or in combination with stressors, on the other hand, the investigators reported a was not found to have an effect on incidence of significant independent mortality effect for men CHD during a follow-up period [30]. but not for women [7]. Moreover, in the Evans In conclusion, although a number of ques- County study, it was found that the excess tions still remain, the present study does provide mortality risk associated with relative social additional support for the hypothesis that there isolation was confined to white males, for no is an independent association between lack of significant risk was found in either white fe- social support and mortality. The demonstra- males, or in blacks [9]. Findings from the Finn- tion that this association exists in the entire ish North Karelia study also indicate that social Swedish population strengthens the overall network interaction is significantly associated conclusion that social isolation has an adverse with mortality among men but not women [1 11. impact on health. We have found that the effects Our finding of an equally strong association of social support are similar for both cardio- between social interaction and mortality risk vascular and total mortality. This suggests there among both men and women is consistent with is a general, rather than cause-specific, effect of the Alameda County and Durham County stud- social support on health status. Therefore, these ies, but not consistent with the findings of findings support the theoretical insights of Tecumseh, Evans County and North Karelia. It Cassel [ 1] who suggested that-regardless of the is conceivable that the considerable ethnic and nature of the disease agent-social support may sociodemographic differences between these have the potential of increasing host resistance various study populations may account for at and of thereby improving human health and least part of this inconsistency. In order to well-being. resolve these issues it may be necessary to Acknowledgements-The authors wish to thank Ellen M. further investigate the social interaction and Hall for her careful editorial work: Anders Ahlbom for his mortality relationship in a number of national valuable suggestions on statistical procedures; and Per-Olof Fredricksson, Gudrun Lindberg, Joachim Vogel, and Ann- population samples which are representative of Marie Bolander for their advice and practical support with both men and women and of different ethnic, the data analysis. socioeconomic and residential groups. The study was financially supported by the Swedish Work There is also some disagreement between Environment Fund, the Swedish National Board of Health and Welfare and the Commission for Social Research. studies which have examined various cardio- vascular outcomes. In both the North Karelia REFERENCES and Alameda County investigations, it was 1. Cassel J: The contribution of the social environment found that both total and cardiovascular mor- to host resistance. Am J Euidemiol 104: 107-123, 1976 tality was associated with social support [6, 111. 2. Antonovsky A: Breakdown: A needed fourth step in Other investigators who have focused on car- the conceptual armamentarium of modem medicine. See Sci Med 6: 537-544, 1972 diovascular outcomes have reported that a lack 3. Syme SL: Behavioral factors associated with the of social interaction and support from various etiology of physical disease: A social epidemiological sources was found to be associated with in- approach. Am J Public Health 64: 1043-1045, 1974 4. Weiss RS: The provisions of social relationships. In creased risk of cardiovascular disease and mor- Doing unto Others. Englewood Cliffs, N.Y.: Prentice tality in the Framingham study [25]; cardio- Hall, 1974 vascular disease prevalence among 5. Bowlby J: The making and breaking of affectional bonds. Aetiology and psychopathology in the light of Japanese-American men in San Francisco [26]; attachment theory. Br J Psychiat 130: 210, 1977 and increased risk of angina pectoris among 6. Berkman LF, Syme SL: Social networks, host re- Israeli civil servants [27]. In a recent study of sistance and mortality: A nine-year follow-up study of Alameda County Residents. Am J Epidemiol 109: work related social support which used a repre- 186-204, 1979 sentative sample of the Swedish employed pop- 7. House JS, Robbins C, Metzner HL: The association ulation, Johnson [29] found a multiplicative of social relationships and activities with mortality: Prospective evidence from the Tecumseh Community effect of lack of social support from co-workers Health Study. Am J Epidemiol 116: 123-140, 1982 on the one hand and job strain as defined by 8. Blazer DG: Social support and mortality in an elderly Karasek et al. [28] on the other. The combined community population. Am J Epidemiol 115: 684-694, 1982 effects of high job demands, low work control, 9. Schoenbach V, Kaplan BH, Fredman L, Kleinbaum and low work related social support yielded an D: Social ties and mortality in Evans County, excess prevalence risk of over 100% [29]. Georgia. Am J Epidemiol 123: 577-591, 1986 10. Welin L, Tibblin G, Svardsudd K et al.: Prospective The Honolulu heart study, however, did not study of social influences on mortality. Lancet 1: replicate the above findings. Social support, 915-918, 1985  Social Networks and Mortality m Sweden 951 Il. Kaplan GA, Salonen JT, Cohen RD, Brand RJ, Syme 21. Rothman KJ, Boice JD: Epidemiologic analysis with SL. Puska P: Social connecttons and mortality from a programmable calculator. Epidemiology Resources, all causes and cardiovascular disease: Prospective Inc., Boston, Mass., 1982 evidence from eastern Finland. Paper given at the 22. Walker SH, Duncan DB: Estimation of the proba- NIH Workshop on Social Support and Cardiovascular bility of an event as a function of several independent Disease. Irvine, April, 1986 variables. 54: 167-179, 1967 12. Berkman L: Social networks, support and health: 23. Broadhead WU, Kaplan BH, James SA, Wagner EH, Taking the next step forward. Am J Epidemiol 123(4): Schoenback VA, Grimson R, Heyden S, Tibbhn G, 559-562. 1986 Gehlbach S: The epidemiologic evidence for a re- 13. Orth-Gomer K, Unden A-L: The measurement of lationship between social support and health. Am J social support m population surveys. SIX Sci Med Epidemiol 117: 521-537, 1983 1987, 24: 83-94, 1987 24. Berkman LF: Assessing the physical health effects of 14. Rose G: Sick individuals and sick populations. Am J social networks and social support. Ann Rev Public Epidemiol 14: 32-38, 1985 Health 5. 413-432, 1984 15. Living Conditions: Report on Inequality in Sweden. 25. Haynes S, Feinleib M: Women, work and coronary Distributton of welfare at the end of the 1970’s. SCB heart disease: Prospecttve findings from the Fram- (Swedish Central Statisttcal Bureau). Stockholm, 1981 ineham Heart Studv. Am J Public Health 70: 16. Levnadsforhallanden. Ensamhet och gemenskap. (Liv- 133-141, 1980 . mg conditions. Isolatton and togetherness. An out- 26. Joseph J: Social affiliation, risk factor status, and look of social parttcipation. English summary). Re- coronary heart disease: A cross-sectional study of nort No. 18. Official Statistics of Sweden. Stockholm: Japanese-American men. PhD Thesis, University of Nattonal Central Bureau of Statistics, 1980 Calif, Berkeley, 1980 17. Lindstrom H: Rortfallsfel vid uppskattning av sjukf- 27. Medalie J, Snyder M, Groen JJ et al.: Angina pectoris ranvaro. (Errors in disability estimates, due to non- among 10,000 men: 5-year incidence and univariate response). Report no 24. Stockholm: Central Bureau analysis. Am J Med 55: 583-594, 1973 of Stattsttcs. 1981 28. Karasek R, Baker D, Marxer F, Ahlbom A, Theorell 18. Wilhelmsen L, Ljungberg S, Wedel H, Werko L: A T: Job decision latttude, job demands, and cardio- comparison between participants and non- vascular disease: A prospective study of Swedish men. participants in a primary preventive trial. J Chron Dis Am J Public Health 71: 694-705, 1981 29: 331-339, 1976 29. Johnson JV: The impact of work place social support, 19. Statistical Abstracts of Sweden 1976-1981. Stock- Job demands and work control upon cardiovascular holm: Offictal Stattstics of Sweden, Statistics Sweden, disease in Sweden. University of Stockholm: Environ- 1976- 1981 mental and Organizational Psychology Monograph 20. Vogel J: The Swedish annual Level of Living Surveys: Series 1: I-204, 1986 Soctal indicators and social reportmg as an official 30. Reed D, McGee D, Yano K et al.: Social networks stattsttcs program. Paper presented at the Tenth and coronary heart disease among Japanese men in World Congress of Sociology, Mexico City, August, Hawaii. Am J Epidemiol 117. 384-396, 1983 1982