Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Social Ties and Change in Social Ties in Relation to Subsequent Total and Cause-specific Mortality and Coronary Heart Disease Incidence in Men

Social Ties and Change in Social Ties in Relation to Subsequent Total and Cause-specific... Abstract The authors prospectively examined the effects of social ties and change in social ties, as measured by a well-known social network index, on total and cause-specific mortality and on coronary heart disease incidence in 28,369 US male health professionals aged 42–77 years in 1988. Over 10 years, the relative risk of total mortality for men in the lower two levels of social integration compared with more socially integrated men was 1.19 (95% confidence interval: 1.06, 1.34) after controlling for age, occupation, health behaviors, general physical condition, coronary risk factors, and dietary habits. In multivariate analysis, deaths from accidents and suicide and from other noncancer, noncardiovascular causes were significantly increased among less socially connected men. Socially isolated men also had an increased risk of fatal coronary heart disease (multivariate relative risk = 1.82, 95% confidence interval: 1.02, 3.23). An increase in the overall social network index between 1988 and 1996 was not significantly associated with subsequent 2-year mortality. In analyses of change in social network components restricted to older men, each categorical unit increase in number of close friends was significantly associated with a 29% decrease in risk of death. Increase in religious service attendance over time was also significantly predictive of decreased mortality. aged, coronary disease, incidence, longitudinal studies, men, mortality, social isolation, social support CI, confidence interval, ICD-9, International Classification of Diseases, Ninth Revision, RR, relative risk To date, 14 known cohort studies have demonstrated a protective association between social interaction and mortality (1–14). Overall, socially isolated persons have two to four times the risk of all-cause mortality compared with those with more ties to friends, relatives, and community. While there is abundant evidence that social interaction protects against overall mortality, several issues remain unresolved. Relatively little is known of the impact of social ties on specific endpoints. A few studies have detected protective effects of social interaction on cardiovascular (7, 12) and cancer (15) mortality and on the incidence of coronary heart disease and stroke (12, 16). In a previous analysis of the Health Professionals Follow-up Study (12), socially isolated men had higher age-adjusted risks of death from cardiovascular disease and from accidents and suicide compared with well-integrated men. Over 4 years, social ties were also related to the incidence of fatal coronary heart disease. However, interpretation was limited for certain endpoints because of small case numbers. In the present study, we reexamined the relations between social ties, as measured by the Berkman-Syme social network index (1), and cause-specific mortality and coronary heart disease in the Health Professionals Follow-up Study. With extended follow-up, we sought to further clarify the associations with specific endpoints. Another question requiring closer study is the potential mechanisms by which social interaction affects health. Theoretically, social relationships may affect health through behavioral and/or physiologic pathways (17). Social support appears to influence negative health behaviors such as smoking, heavy alcohol consumption, poor dietary habits, sedentary lifestyle, and suboptimal health service utilization (18–23). Social ties may buffer against stress and thereby reduce activation of the neuroendocrine system, leading to diminished progression of atherosclerosis and other pathophysiologic processes (24–26). In the current analyses, health behavior, medical, and dietary factors were controlled for as confounders and were simultaneously evaluated for their mediating potential. Finally, whether change in social networks affects subsequent mortality has not been established. The relation between change in social ties and health has been examined primarily in terms of marital transition, particularly recent widowhood, and all-cause mortality (27–29). Generally, mortality rates have been higher for bereaved persons than for their married counterparts. Yet, the effects of social isolation and grief cannot be easily disentangled in bereavement studies. To our knowledge, only one study has examined change in more extensive ties; change in a social network index was not predictive of mortality in multivariate analysis (30). To address this issue, we measured social networks at two points 8 years apart, and we examined the impact of change in social ties on subsequent 2-year mortality. MATERIALS AND METHODS Study population The Health Professionals Follow-up Study is a prospective cohort study of chronic disease among 51,529 male health professionals aged 40–75 years in 1986. Cohort members are dentists (58 percent), veterinarians (20 percent), pharmacists (8 percent), optometrists (7 percent), osteopaths (4 percent), and podiatrists (3 percent). At enrollment, data on medical history and risk factors were obtained from the participants by mailed questionnaire. Every 2 years, follow-up questionnaires have been mailed to update information on exposures and newly diagnosed diseases. Response rates for the questionnaires have been over 90 percent. In 1988 and 1996, the Berkman-Syme social network index was incorporated into the questionnaire. Because preexisting disease could affect a person's social network, we excluded 12,052 participants who reported any of the following diagnoses prior to the return of the 1988 questionnaire: myocardial infarction, coronary artery bypass graft, percutaneous transluminal coronary angioplasty, angina pectoris, heart rhythm disorder, stroke, cancer (except nonmelanoma skin cancer), and chronic renal failure; we also omitted 434 men who died before 1988. We also excluded 10,674 men who did not provide social network data in 1988. The main study population therefore consisted of 28,369 men. Nonrespondents had a higher risk of dying over the 10-year follow-up (7.2 vs. 6.0 percent), and they were more likely to have smoked in 1986. Other health behaviors and medical conditions such as hypertension were distributed similarly among the respondent groups. Mean age, and distributions of marital status, employment status, and occupation, in 1986 were also similar. Because the social network index was updated in 1996, participants who reported any of the above-mentioned conditions prior to 1996 were excluded from the consideration of person-time beyond 1996. For our analysis of change in social networks, a subgroup of 17,769 men who completed the social network indices in 1988 and 1996 and were free of these conditions in 1996 was followed from 1996 to 1998. In 1988, these subjects had the same network index score (mean, 3.2) as the main group but were slightly younger, and they had fewer negative health behaviors and conditions. The social network index The Berkman-Syme social network index is based on four types of social connections: 1) marital status (married vs. unmarried); 2) sociability (frequency and contact with close friends and relatives measured as a subscale with levels 1, 2, and 3; lower values indicate fewer numbers and less contact); 3) religious group affiliation (yes vs. no); and 4) membership in other social or community organizations (yes vs. no). Participants provided data on marital status, number of close friends and relatives, number of friends and relatives seen at least once a month, frequency of religious service attendance, and hours spent each week participating in other social or community groups. The index measures degree of social integration, taking into account number as well as relative importance of different ties. Therefore, intimate ties are given more weight than group affiliations. The index has four levels: low, medium, medium-high, and high (respectively, I, II, III, and IV). Persons with low levels of social ties can be characterized as unmarried, having few friends or relatives, and not being involved in community groups. Further details on construction of the index have been described elsewhere (31). Covariates Information on covariates was updated every 2 years except as noted. Data were collected on health behaviors, including smoking, alcohol intake, combined leisure-time and routine physical activity, body mass index, and routine physical examination in the last 2 years. We obtained information on general physical condition, such as the ability to climb several flights of stairs and do heavy housework, and on employment status. Participants also reported diagnoses of hypertension, diabetes, and hypercholesterolemia and antidepressant use. Parental history of myocardial infarction before age 60 years was reported in 1986. Intakes of alcohol and energy-adjusted total and saturated fats, folate, and fiber were measured every 4 years by using a semiquantitative food frequency questionnaire (32). Vitamin supplement use was also reported. Occupational data on health profession (e.g., dentist) were obtained in 1986. Mortality and morbidity endpoints All endpoints occurred between return of the 1988 questionnaire and January 31, 1998, for the main analysis and between return of the 1996 questionnaire and January 31, 1998, for the analysis of change in social ties. In most instances, we were informed of a subject's death by next of kin, postal authorities, or work associates. The National Death Index was used to ascertain vital status of repeat nonresponders to the mailed questionnaires. It is estimated that 98 percent or more of cohort deaths are ascertained with the National Death Index (33). To determine cause of death, death certificates and hospital or pathology reports were reviewed by physicians blinded to exposure status. The following mortality outcomes were of interest: all-cause mortality; cardiovascular disease mortality, including coronary heart disease (International Classification of Diseases, Ninth Revision (ICD-9), codes 410–414, 798) and stroke (ICD codes 430–438); cancer mortality (ICD-9, codes 140–209); mortality due to external causes of injury (ICD “E” codes including accidents, suicides, poisonings, and other trauma); and mortality from all other causes. Medical records for reported coronary heart disease were reviewed by a physician in a blinded manner. Nonfatal myocardial infarction was classified as “definite” if World Health Organization criteria (34) were met, and required symptoms with typical changes in electrocardiograph readings or cardiac enzyme levels were noted. Nonfatal myocardial infarction was classified as “probable” when hospital records were not available but the event required hospital admission and the diagnosis was corroborated in an additional contact via letter or telephone. Deaths from myocardial infarction were confirmed by using hospital records and autopsy reports. Additionally, fatal coronary heart disease was confirmed by using death certificates if coronary heart disease was listed as the most plausible cause of death and preexistence of coronary heart disease was supported by hospital records or interviews with next of kin. Sudden cardiac death was defined as death occurring within 1 hour of symptom onset with no serious antecedent condition and no other likely cause of death reported. Approximately 12 percent of nonfatal myocardial infarctions and 13 percent of fatal cases were confirmed with supplementary information other than medical records and were deemed “probable.” We included both definite and probable cases in counts, since exclusion of probable events did not alter our results. All incident first events of nonfatal myocardial infarction, fatal coronary heart disease, sudden cardiac death, coronary artery bypass graft surgery, and angioplasty were included in counts for total coronary heart disease. Statistical analysis For each level of social integration, baseline proportions of potential confounders were age standardized by using the overall cohort's age distribution as the standard. Each subject contributed person-time from return of the 1988 questionnaire (or, for the analysis of change in social ties, from return of the 1996 questionnaire) until the time of the first event or death, or January 31, 1998. Relative risk estimates were adjusted for age (5-year categories) by using the Mantel-Haenszel summary estimator (35). Pooled logistic regression was used to examine the impact of social networks, controlling for multiple confounders. All covariates were time varying except occupation and parental history of myocardial infarction. The relative risk obtained from pooled logistic regression closely approximates the hazard rate ratio from Cox proportional hazards models with time-varying covariates under conditions met by this study: short (e.g., 2-year) follow-up intervals for grouping events and low incidence of an event within each interval (36). The Mantel extension test was used to examine trends in risks across levels of the social network index (37). The impact of an increase in social interaction on mortality, adjusted for overall level, was examined by including both average level and change in social ties between 1988 and 1996 as continuous predictors in multiple logistic regression models. Since most subjects experienced only incremental shifts in social network variables over time (e.g., level II to III of the social network index) rather than extreme changes (e.g., level I to IV), we used original categorical scores as values. Scores 1–4 were used for levels I–IV of the overall index, and 1–3 were used for the sociability subscale; a higher score indicated increased integration. For questions on friends and relatives, it was hypothesized that the effect of change would diminish with increasing underlying number; for example, an increase from 0 to 1–2 friends would have the same effect as an increase from 3–5 to 6–9 friends. Hence, we sought to indirectly account for potential differences in intimacy and support per relation. For religious service attendance, each unit increase in monthly attendance was hypothesized to have the same effect, regardless of underlying frequency. Dichotomous variables were used to examine the effects of increased social group participation over time (e.g., 10 or fewer to 11 or more hours per week) and becoming married. Because change in physical function may be linked to change in social ties, we controlled for overall levels, including change between 1988 and 1996, in the ability to climb several flights of stairs and do heavy housework as well as employment status in 1988 and 1996. We also controlled for occupation in 1986 and the following variables in 1996: age, smoking, alcohol intake, physical activity, and antidepressant use. Since the elderly may be particularly vulnerable to changes in both social networks and health status (38–40), change in social interaction could have more pronounced effects among older persons. Accordingly, we conducted analyses within strata of men aged less than 65 years and 65 years or more. RESULTS At the start of follow-up, 6.0 percent of the study population was socially isolated (level I of the social network index). The majority (51.3 percent) of subjects were well integrated (level IV). In table 1, we present age-standardized characteristics by social network index level in 1988. Socially isolated men were more likely to smoke, consume more alcohol, and be less active. TABLE 1. Age-standardized characteristics,* according to social network index level, of Health Professionals Follow-up Study participants in 1988, United States   Berkman-Syme social network index†     IV (high)  III  II  I (low)  No. (%)  14,555 (51.3)  5,329 (18.8)  6,798 (24.0)  1,687 (6.0)  Mean age (years)  55.2  55.4  55.1  55.4  Current smoking (%)  7.9  9.4  10.0  13.0  Alcohol intake ≥20 g/day (%)  14.1  20.2  20.6  23.4  High body mass index‡ (%)  20.1  18.9  19.0  17.8  Low level of physical activity§ (%)  16.8  19.8  21.3  24.6  Routine physical examination in the last 2 years (%)  66.7  63.3  60.0  55.6  Full-time employment (%)  82.1  81.4  80.2  76.8  Ability to climb flights of stairs (%)  82.1  79.7  80.6  79.4  Ability to do heavy housework (%)  97.6  96.8  96.8  95.9  Hypertension (%)  17.3  18.2  18.3  17.7  High serum cholesterol (%)  15.1  15.9  15.5  14.7  Diabetes (%)  2.4  2.2  2.4  2.6  Myocardial infarction in a parent aged <60 years (%)  11.2  11.9  12.4  11.3  Antidepressant use in 1990 (%)  0.9  0.8  1.5  1.6  Multivitamin use (%)  40.3  40.3  42.6  44.7  Vitamin E use (%)  19.5  20.5  22.0  22.5  High‡ intake of (%)           Total fat  20.2  21.0  21.0  19.6   Saturated fat  20.4  21.4  21.9  22.0  Low§ intake of (%)           Folate  18.3  21.3  22.5  24.3   Fiber  17.9  22.1  23.8  27.1    Berkman-Syme social network index†     IV (high)  III  II  I (low)  No. (%)  14,555 (51.3)  5,329 (18.8)  6,798 (24.0)  1,687 (6.0)  Mean age (years)  55.2  55.4  55.1  55.4  Current smoking (%)  7.9  9.4  10.0  13.0  Alcohol intake ≥20 g/day (%)  14.1  20.2  20.6  23.4  High body mass index‡ (%)  20.1  18.9  19.0  17.8  Low level of physical activity§ (%)  16.8  19.8  21.3  24.6  Routine physical examination in the last 2 years (%)  66.7  63.3  60.0  55.6  Full-time employment (%)  82.1  81.4  80.2  76.8  Ability to climb flights of stairs (%)  82.1  79.7  80.6  79.4  Ability to do heavy housework (%)  97.6  96.8  96.8  95.9  Hypertension (%)  17.3  18.2  18.3  17.7  High serum cholesterol (%)  15.1  15.9  15.5  14.7  Diabetes (%)  2.4  2.2  2.4  2.6  Myocardial infarction in a parent aged <60 years (%)  11.2  11.9  12.4  11.3  Antidepressant use in 1990 (%)  0.9  0.8  1.5  1.6  Multivitamin use (%)  40.3  40.3  42.6  44.7  Vitamin E use (%)  19.5  20.5  22.0  22.5  High‡ intake of (%)           Total fat  20.2  21.0  21.0  19.6   Saturated fat  20.4  21.4  21.9  22.0  Low§ intake of (%)           Folate  18.3  21.3  22.5  24.3   Fiber  17.9  22.1  23.8  27.1  * Intakes of total fat, saturated fat, folate, and fiber were energy adjusted. † Am J Epidemiol 1979;109:186–204. ‡ “High” defined as being in the highest quintile of the baseline population. § “Low” defined as being in the lowest quintile of the baseline population. View Large Over the 10-year follow-up, 1,365 deaths occurred in 256,684 person-years. Of these deaths, 320 were due to cardiovascular disease, 504 to cancer, 81 to accidents and suicides, and 272 to other causes. In table 2, we present the age-adjusted and multivariate relative risks of total and cause-specific mortality by level of social network index. Risk of total mortality was similarly elevated only for those men in the lower two levels, suggesting a threshold effect. In additional analyses, data for men in the two most isolated levels were combined and were compared with those for men in higher levels; men in the combined lower levels had a significantly increased risk of total mortality (multivariate relative risk (RR) = 1.19, 95 percent confidence interval (CI): 1.06, 1.34). When socially isolated men were compared with the most socially integrated men, the age-adjusted relative risk of cardiovascular death was 1.53 (95 percent CI: 1.02, 2.29). The corresponding multivariate estimate was attenuated and was not statistically significant. Men with a moderately low number of social ties (level II) had more than twice the risk of death from accidents and suicides relative to men with the most social connections (multivariate RR = 2.32, 95 percent CI: 1.39, 3.86). The p value for the multivariate trend test was 0.01. The multivariate relative risk of mortality from other causes was 1.55 (95 percent CI: 1.04, 2.32) for men in the lowest level of the index compared with socially well-integrated men. In the category of other causes, the most frequent causes of death were chronic obstructive pulmonary disease (10.7 percent) and chronic liver disease and cirrhosis (8.1 percent). Social ties were not associated with risk of cancer mortality. TABLE 2. Age-adjusted and multivariate relative risks (95% confidence intervals) of total and cause-specific mortality, by social network index level, Health Professionals Follow-up Study, United States, 1988–1998*   Berkman-Syme social network index†   p for trend    IV (high)  III  II  I (low)  Total mortality             No. of cases  620  244  385  116     Age-adjusted RR‡  1.00  1.00 (0.86, 1.16)  1.28 (1.13, 1.45)  1.49 (1.23, 1.82)  <0.001   Multivariate RR§  1.00  0.97 (0.83, 1.13)  1.18 (1.03, 1.34)  1.20 (0.97, 1.48)  0.009  Cardiovascular disease             No. of cases  146  60  86  28     Age-adjusted RR  1.00  1.04 (0.77, 1.41)  1.21 (0.93, 1.58)  1.53 (1.02, 2.29)  0.03   Multivariate RR§  1.00  1.07 (0.79, 1.44)  1.16 (0.88, 1.52)  1.37 (0.91, 2.08)  0.11  Total cancer             No. of cases  248  91  133  32     Age-adjusted RR  1.00  0.94 (0.74, 1.19)  1.11 (0.90, 1.37)  1.03 (0.72, 1.49)  0.47   Multivariate RR¶  1.00  0.91 (0.71, 1.16)  1.02 (0.82, 1.27)  0.86 (0.59, 1.25)  0.71  Accidents and suicides             No. of cases  28  15  33  5     Age-adjusted RR  1.00  1.38 (0.74, 2.59)  2.44 (1.48, 4.03)  1.44 (0.55, 3.72)  0.003   Multivariate RR#  1.00  1.40 (0.74, 2.63)  2.32 (1.39, 3.86)  1.27 (0.49, 3.33)  0.01  Other causes             No. of cases  118  46  74  34     Age-adjusted RR  1.00  0.99 (0.70, 1.39)  1.29 (0.96, 1.73)  2.29 (1.56, 3.36)  <0.001   Multivariate RR**  1.00  0.95 (0.67, 1.35)  1.12 (0.83, 1.51)  1.55 (1.04, 2.32)  0.07    Berkman-Syme social network index†   p for trend    IV (high)  III  II  I (low)  Total mortality             No. of cases  620  244  385  116     Age-adjusted RR‡  1.00  1.00 (0.86, 1.16)  1.28 (1.13, 1.45)  1.49 (1.23, 1.82)  <0.001   Multivariate RR§  1.00  0.97 (0.83, 1.13)  1.18 (1.03, 1.34)  1.20 (0.97, 1.48)  0.009  Cardiovascular disease             No. of cases  146  60  86  28     Age-adjusted RR  1.00  1.04 (0.77, 1.41)  1.21 (0.93, 1.58)  1.53 (1.02, 2.29)  0.03   Multivariate RR§  1.00  1.07 (0.79, 1.44)  1.16 (0.88, 1.52)  1.37 (0.91, 2.08)  0.11  Total cancer             No. of cases  248  91  133  32     Age-adjusted RR  1.00  0.94 (0.74, 1.19)  1.11 (0.90, 1.37)  1.03 (0.72, 1.49)  0.47   Multivariate RR¶  1.00  0.91 (0.71, 1.16)  1.02 (0.82, 1.27)  0.86 (0.59, 1.25)  0.71  Accidents and suicides             No. of cases  28  15  33  5     Age-adjusted RR  1.00  1.38 (0.74, 2.59)  2.44 (1.48, 4.03)  1.44 (0.55, 3.72)  0.003   Multivariate RR#  1.00  1.40 (0.74, 2.63)  2.32 (1.39, 3.86)  1.27 (0.49, 3.33)  0.01  Other causes             No. of cases  118  46  74  34     Age-adjusted RR  1.00  0.99 (0.70, 1.39)  1.29 (0.96, 1.73)  2.29 (1.56, 3.36)  <0.001   Multivariate RR**  1.00  0.95 (0.67, 1.35)  1.12 (0.83, 1.51)  1.55 (1.04, 2.32)  0.07  * All covariates were time varying except as noted. † Am J Epidemiol 1979;109:186–204. ‡ RR, relative risk. § Multivariate relative risks were adjusted for age (5-year categories) in 1988, time period (1988–1990, 1990–1992, 1992–1994, 1994–1996, 1996–1998), occupation in 1986, smoking history (never, past, and current in categories of 1–14, 15–24, and ≥25 cigarettes/day), daily alcohol intake (0, 0.01–9.9, 10–19.9, 20–29.9, and ≥30 g/day), quintiles of body mass index, quintiles of physical activity, routine physical examination in the last 2 years (yes/no), ability to climb several flights of stairs (yes/no), ability to do heavy housework (yes/no), employment status (full time, part time, retired, disabled), history of hypertension, diabetes, high serum cholesterol, history of myocardial infarction in a parent aged <60 years (yes/no) in 1986, quintiles of energy-adjusted intakes of total fat, saturated fat, folate, and fiber, and multivitamin and vitamin E supplement use (yes/no). ¶ Multivariate model same as specified in § but excluding hypertension, diabetes, high serum cholesterol, parental history of myocardial infarction in 1986, and all dietary variables. # Multivariate model adjusted for age, time period, occupation in 1986, smoking history, daily alcohol intake, physical activity, routine physical examination, employment status, and antidepressant use. ** Multivariate model same as specified in § but excluding parental history of myocardial infarction in 1986 and all dietary variables. View Large Confounding effects were examined further by adding covariates in conceptually related groups to a model of social network index and all-cause mortality adjusted for age and occupation only (table 3). Adding health behaviors to this base model resulted in the greatest change in the relative risk estimate comparing socially isolated with socially well-integrated men. Inclusion of general physical condition variables attenuated the age-and-occupation-adjusted relative risk to a moderate degree. In contrast, adding comorbidity or dietary variables to the base model did not substantially alter the relative risk estimates. Therefore, covariates related to health behavior and physical condition appeared to be the most influential confounders or, alternatively, to demonstrate the most potential as mediators between social networks and mortality. When we excluded health behaviors from the fully adjusted model for total mortality (table 2), the multivariate relative risk comparing lowest with highest levels of integration increased from 1.20 (95 percent CI: 0.97, 1.48) to 1.34 (95 percent CI: 1.09, 1.64). TABLE 3. Multivariate relative risks (95% confidence intervals) of total mortality, by social network index level from incremental models, Health Professionals Follow-up Study, United States, 1988–1998*   Berkman-Syme social network index†     IV (high)  III  II  I (low)  Age + occupation‡  1.00  1.01 (0.87, 1.18)  1.30 (1.14, 1.48)  1.53 (1.25, 1.87)  Age + occupation + health behavior variables§  1.00  0.97 (0.84, 1.13)  1.18 (1.03, 1.34)  1.28 (1.05, 1.58)  Age + occupation + general physical condition variables¶  1.00  0.99 (0.85, 1.15)  1.26 (1.10, 1.43)  1.37 (1.11, 1.68)  Age + occupation + medical history variables#  1.00  1.02 (0.88, 1.19)  1.29 (1.14, 1.47)  1.51 (1.24, 1.85)  Age + occupation + dietary variables**  1.00  1.01 (0.87, 1.18)  1.30 (1.14, 1.48)  1.50 (1.23, 1.84)    Berkman-Syme social network index†     IV (high)  III  II  I (low)  Age + occupation‡  1.00  1.01 (0.87, 1.18)  1.30 (1.14, 1.48)  1.53 (1.25, 1.87)  Age + occupation + health behavior variables§  1.00  0.97 (0.84, 1.13)  1.18 (1.03, 1.34)  1.28 (1.05, 1.58)  Age + occupation + general physical condition variables¶  1.00  0.99 (0.85, 1.15)  1.26 (1.10, 1.43)  1.37 (1.11, 1.68)  Age + occupation + medical history variables#  1.00  1.02 (0.88, 1.19)  1.29 (1.14, 1.47)  1.51 (1.24, 1.85)  Age + occupation + dietary variables**  1.00  1.01 (0.87, 1.18)  1.30 (1.14, 1.48)  1.50 (1.23, 1.84)  * All covariates were time varying except as noted. † Am J Epidemiol 1979;109:186–204. ‡ Occupation was assessed in 1986. § Health behavior variables include smoking history (never, past, and current in categories of 1–14, 15–24, and ≥25 cigarettes/day), daily alcohol intake (0, 0.01–9.9, 10–19.9, 20–29.9, and ≥30 g/day), quintiles of body mass index, quintiles of physical activity, routine physical examination in the last 2 years (yes/no). ¶ General physical condition variables include ability to climb several flights of stairs (yes/no), ability to do heavy housework (yes/no), employment status (full time, part time, retired, disabled). # Medical history variables include history of hypertension, diabetes, high serum cholesterol, history of myocardial infarction in a parent aged <60 years (yes/no) in 1986. ** Dietary variables include quintiles of energy-adjusted intakes of total fat, saturated fat, folate, and fiber, and multivitamin and vitamin E supplement use (yes/no). View Large Analysis of the overall index may have masked differential effects of underlying network components. Accordingly, results from multivariate analyses of individual components and of total and cause-specific mortality are presented in table 4. In these analyses, network components were mutually adjusted. Unmarried men had increased risks of total mortality and of death from accidents and suicide and from other causes. Religious service attendance and social group participation were protective against all-cause mortality. Having more close friends and contact with relatives were each associated with a decreased risk of dying from other causes, while more close relatives protected against accidental death and suicide. TABLE 4. Multivariate relative risks (95% confidence intervals) of total and cause-specific mortality, by social network index components,* Health Professionals Follow-up Study, United States, 1988–1998†   Total mortality‡  CVD§ mortality‡  Accidents, suicide¶  Other mortality#  Sociability subscale           1 = low  1.05 (0.87, 1.26)  1.03 (0.70, 1.53)  1.54 (0.75, 3.17)  0.99 (0.66, 1.47)   2 = medium  1.04 (0.91, 1.19)  1.21 (0.91, 1.60)  1.14 (0.63, 2.04)  0.92 (0.68, 1.24)   3 = high  1.00  1.00  1.00  1.00  Marital status           1 = unmarried  1.27 (1.07, 1.50)  1.14 (0.79, 1.63)  2.40 (1.40, 4.12)  1.63 (1.16, 2.27)   0 = married  1.00  1.00  1.00  1.00  Religious service attendance per month           1 = never/almost never  1.15 (1.02, 1.30)  1.21 (0.95, 1.53)  1.16 (0.73, 1.86)  1.23 (0.95, 1.60)   0 = once per year or more  1.00  1.00  1.00  1.00  Social group participation           1 = 10 or fewer hours per week  1.56 (1.04, 2.34)  1.76 (0.72, 4.30)  3.34 (0.82, 13.7)**  1.11 (0.51, 2.40)   0 = 11 or more hours per week  1.00  1.00  1.00  1.00  No. of close relatives           1 = none  1.01 (0.83, 1.24)  0.97 (0.64, 1.48)  2.93 (1.47, 5.82)  0.70 (0.44, 1.12)   0 = 1 or more  1.00  1.00  1.00  1.00  No. of relatives seen per month           1 = none  1.14 (1.00, 1.29)  1.18 (0.91, 1.52)  0.83 (0.48, 1.42)  1.42 (1.08, 1.87)   0 = 1 or more  1.00  1.00  1.00  1.00  No. of close friends           1 = none  0.94 (0.60, 1.46)  1.39 (0.59, 3.26)  0.96 (0.54, 1.71)**  2.84 (1.09, 7.36)   0 = 1 or more  1.00  1.00  1.00  1.00  No. of friends seen per month           1 = none  0.87 (0.64, 1.18)  0.82 (0.43, 1.57)  0.31 (0.07, 1.34)  0.44 (0.19, 1.00)   0 = 1 or more  1.00  1.00  1.00  1.00    Total mortality‡  CVD§ mortality‡  Accidents, suicide¶  Other mortality#  Sociability subscale           1 = low  1.05 (0.87, 1.26)  1.03 (0.70, 1.53)  1.54 (0.75, 3.17)  0.99 (0.66, 1.47)   2 = medium  1.04 (0.91, 1.19)  1.21 (0.91, 1.60)  1.14 (0.63, 2.04)  0.92 (0.68, 1.24)   3 = high  1.00  1.00  1.00  1.00  Marital status           1 = unmarried  1.27 (1.07, 1.50)  1.14 (0.79, 1.63)  2.40 (1.40, 4.12)  1.63 (1.16, 2.27)   0 = married  1.00  1.00  1.00  1.00  Religious service attendance per month           1 = never/almost never  1.15 (1.02, 1.30)  1.21 (0.95, 1.53)  1.16 (0.73, 1.86)  1.23 (0.95, 1.60)   0 = once per year or more  1.00  1.00  1.00  1.00  Social group participation           1 = 10 or fewer hours per week  1.56 (1.04, 2.34)  1.76 (0.72, 4.30)  3.34 (0.82, 13.7)**  1.11 (0.51, 2.40)   0 = 11 or more hours per week  1.00  1.00  1.00  1.00  No. of close relatives           1 = none  1.01 (0.83, 1.24)  0.97 (0.64, 1.48)  2.93 (1.47, 5.82)  0.70 (0.44, 1.12)   0 = 1 or more  1.00  1.00  1.00  1.00  No. of relatives seen per month           1 = none  1.14 (1.00, 1.29)  1.18 (0.91, 1.52)  0.83 (0.48, 1.42)  1.42 (1.08, 1.87)   0 = 1 or more  1.00  1.00  1.00  1.00  No. of close friends           1 = none  0.94 (0.60, 1.46)  1.39 (0.59, 3.26)  0.96 (0.54, 1.71)**  2.84 (1.09, 7.36)   0 = 1 or more  1.00  1.00  1.00  1.00  No. of friends seen per month           1 = none  0.87 (0.64, 1.18)  0.82 (0.43, 1.57)  0.31 (0.07, 1.34)  0.44 (0.19, 1.00)   0 = 1 or more  1.00  1.00  1.00  1.00  * Components of the Berkman-Syme social network index (Am J Epidemiol 1979;109:186–204) were mutually adjusted for each other. † All covariates were time varying except as noted. ‡ Multivariate relative risks were adjusted for age (5-year categories), time period (1988–1990, 1990–1992, 1992–1994, 1994–1996, 1996–1998), occupation in 1986, smoking history (never, past, and current in categories of 1–14, 15–24, and ≥25 cigarettes/day), daily alcohol intake (0, 0.01–9.9, 10–19.9, 20–29.9, and ≥30 g/day), quintiles of body mass index, quintiles of physical activity, routine physical examination in the last 2 years (yes/no), ability to climb several flights of stairs (yes/no), ability to do heavy housework (yes/no), employment status (full time, part time, retired, disabled), history of hypertension, diabetes, high serum cholesterol, history of myocardial infarction in a parent aged <60 years (yes/no) in 1986, quintiles of energy-adjusted intakes of total fat, saturated fat, folate, and fiber, and multivitamin and vitamin E supplement use (yes/no). § CVD, cardiovascular disease. ¶ Multivariate model adjusted for age, time period, occupation in 1986, smoking history, daily alcohol intake, physical activity, routine physical examiniation, employment status, and antidepressant use. # Multivariate model same as specified in ‡ but excluding parental history of myocardial infarction in 1986 and all dietary variables. ** Because of sparse deaths by accident/suicide, categories were collapsed as follows: social group participation (0 = 6 or more hours per week, 1 = 5 or fewer hours per week), number of close friends (0 = 3 or more, 1 = 2 or less). View Large In table 5, we present age-adjusted and multivariate relative risks of coronary heart disease. Over 10 years, we observed 1,816 incident cases of total coronary heart disease; 618 cases of nonfatal myocardial infarction, 142 cases of fatal coronary heart disease (excluding sudden cardiac death), and 97 sudden cardiac deaths occurred. Socially isolated men had an increased risk of fatal coronary heart disease compared with socially well-integrated men (multivariate RR = 1.82, 95 percent CI: 1.02, 3.23). Of individual network components, only the number of and contact with close friends were significantly associated with fatal coronary heart disease (data not shown). The incidence of total coronary heart disease, nonfatal myocardial infarction, and sudden cardiac death was not significantly increased among socially isolated men. TABLE 5. Age-adjusted and multivariate relative risks* (95% confidence intervals) of coronary heart disease, by social network index level, Health Professionals Follow-up Study, United States, 1988–1998†   Berkman-Syme social network index‡   p for trend    IV (high)  III  II  I (low)  Nonfatal myocardial infarction             No. of cases  306  139  130  43     Age-adjusted RR§  1.00  1.18 (0.96, 1.44)  0.88 (0.72, 1.08)  1.13 (0.82, 1.56)  0.80   Multivariate RR  1.00  1.18 (0.96, 1.44)  0.86 (0.70, 1.06)  1.11 (0.80, 1.53)  0.61  Fatal coronary heart disease             No. of cases  62  29  36  15     Age-adjusted RR  1.00  1.19 (0.77, 1.85)  1.20 (0.79, 1.80)  1.93 (1.10, 3.39)  0.06   Multivariate RR  1.00  1.26 (0.80, 1.96)  1.15 (0.76, 1.75)  1.82 (1.02, 3.23)  0.10  Sudden cardiac death             No. of cases  50  20  22  5     Age-adjusted RR  1.00  1.02 (0.61, 1.72)  0.91 (0.55, 1.50)  0.80 (0.32, 2.01)  0.60   Multivariate RR  1.00  1.02 (0.61, 1.73)  0.87 (0.52, 1.44)  0.71 (0.28, 1.81)  0.43  Total coronary heart disease             No. of cases  917  383  400  116     Age-adjusted RR  1.00  1.08 (0.96, 1.21)  0.90 (0.80, 1.02)  1.02 (0.84, 1.23)  0.34   Multivariate RR  1.00  1.07 (0.95, 1.21)  0.88 (0.78, 0.99)  0.99 (0.81, 1.20)  0.14    Berkman-Syme social network index‡   p for trend    IV (high)  III  II  I (low)  Nonfatal myocardial infarction             No. of cases  306  139  130  43     Age-adjusted RR§  1.00  1.18 (0.96, 1.44)  0.88 (0.72, 1.08)  1.13 (0.82, 1.56)  0.80   Multivariate RR  1.00  1.18 (0.96, 1.44)  0.86 (0.70, 1.06)  1.11 (0.80, 1.53)  0.61  Fatal coronary heart disease             No. of cases  62  29  36  15     Age-adjusted RR  1.00  1.19 (0.77, 1.85)  1.20 (0.79, 1.80)  1.93 (1.10, 3.39)  0.06   Multivariate RR  1.00  1.26 (0.80, 1.96)  1.15 (0.76, 1.75)  1.82 (1.02, 3.23)  0.10  Sudden cardiac death             No. of cases  50  20  22  5     Age-adjusted RR  1.00  1.02 (0.61, 1.72)  0.91 (0.55, 1.50)  0.80 (0.32, 2.01)  0.60   Multivariate RR  1.00  1.02 (0.61, 1.73)  0.87 (0.52, 1.44)  0.71 (0.28, 1.81)  0.43  Total coronary heart disease             No. of cases  917  383  400  116     Age-adjusted RR  1.00  1.08 (0.96, 1.21)  0.90 (0.80, 1.02)  1.02 (0.84, 1.23)  0.34   Multivariate RR  1.00  1.07 (0.95, 1.21)  0.88 (0.78, 0.99)  0.99 (0.81, 1.20)  0.14  * Multivariate relative risks were adjusted for age (5-year categories), time period (1988–1990, 1990–1992, 1992–1994, 1994–1996, 1996–1998), occupation in 1986, smoking history (never, past, and current in categories of 1–14, 15–24, and ≥25 cigarettes/day), daily alcohol intake (0, 0.01–9.9, 10–19.9, 20–29.9, and ≥30 g/day), quintiles of body mass index, quintiles of physical activity, routine physical examination in the last 2 years (yes/no), ability to climb several flights of stairs (yes/no), ability to do heavy housework (yes/no), employment status (full time, part time, retired, disabled), history of hypertension, diabetes, high serum cholesterol, history of myocardial infarction in a parent aged <60 years (yes/no) in 1986, quintiles of energy-adjusted intakes of total fat, saturated fat, folate, and fiber, and multivitamin and vitamin E supplement use (yes/no). † All covariates were time varying except as noted. ‡ Am J Epidemiol 1979;109:186–204. § RR, relative risk. View Large Prior to analyzing the effect of change in social ties on mortality, we examined the stability of social networks over time. When we compared 1988 and 1996 social network indices, level of social interaction did not remain fixed; the Spearman correlation coefficient was 0.57. Correlation coefficients for network components were as follows: religious service attendance, r = 0.79; social group participation, r = 0.47; sociability subscale, r = 0.52; number of close friends, r = 0.61 and relatives, r = 0.58; and monthly contact with friends, r = 0.57 and relatives, r = 0.60. An increase in the number of social ties between 1988 and 1996 was not predictive of subsequent 2-year mortality for men less than age 65 years (table 6). For older men, we noted a nonsignificant decrease in mortality associated with an increase in the overall index between 1988 and 1996. For several facets of the elderly person's social network, an increase in social interaction over time was associated with survival. Because of a limited number of deaths of elderly persons (n = 72), confidence intervals were wide and often included 1. In age-adjusted analysis, a within-person unit increase on the sociability subscale was associated with a 36 percent decrease in mortality, although this decrease was attenuated and was not significant in multivariate analysis. Friendships as opposed to interaction with relatives appeared to drive the latter association. Each categorical unit increase in the number of close friends was significantly associated with a 29 percent decrease in mortality. In addition, every additional religious service attendance per month was associated with a significant, although modest decline of 7 percent. In contrast, becoming married was linked to markedly, albeit nonsignificantly increased mortality risk among older men. TABLE 6. Age-adjusted and multivariate relative risks* (95% confidence intervals) of total mortality, per unit increase† in social ties‡ between 1988 and 1996, for men aged <65 and ≥65 years, Health Professionals Follow-up Study, United States, 1996–1998   Cases     Aged <65 years (n = 87)  Aged ≥65 years (n = 72)  Social network index (1 = I, 2 = II, 3 = III, 4 = IV)       Age-adjusted RR§  1.02 (0.80, 1.28)  0.88 (0.69, 1.14)   Multivariate RR  1.03 (0.81, 1.31)  0.91 (0.69, 1.20)  Sociability subscale (1, 2, 3)       Age-adjusted RR  1.05 (0.73, 1.50)  0.64 (0.44, 0.93)   Multivariate RR  1.02 (0.70, 1.48)  0.66 (0.43, 1.01)  Marital status (unmarried to married status)       Age-adjusted RR  1.16 (0.54, 2.50)  2.04 (0.88, 4.70)   Multivariate RR  1.21 (0.55, 2.67)  2.70 (0.99, 7.39)  Religious service attendance per month       Age-adjusted RR  0.98 (0.94, 1.03)  0.94 (0.89, 1.00)   Multivariate RR  0.99 (0.95, 1.03)  0.93 (0.87, 0.99)  Social group participation       Age-adjusted RR  1.41 (0.42, 4.72)  0.63 (0.25, 1.62)   Multivariate RR  1.29 (0.38, 4.35)  0.76 (0.27, 2.17)  Close relatives¶       Age-adjusted RR  0.97 (0.80, 1.19)  1.00 (0.81, 1.25)   Multivariate RR  0.96 (0.78, 1.18)  0.99 (0.79, 1.25)  Close relatives seen per month¶       Age-adjusted RR  0.98 (0.79, 1.22)  1.06 (0.83, 1.37)   Multivariate RR  1.00 (0.80, 1.25)  1.03 (0.78, 1.37)  Close friends¶       Age-adjusted RR  0.95 (0.76, 1.18)  0.71 (0.56, 0.90)   Multivariate RR  0.94 (0.75, 1.18)  0.71 (0.55, 0.92)  Close friends seen per month¶       Age-adjusted RR  0.90 (0.73, 1.10)  0.83 (0.67, 1.03)   Multivariate RR  0.90 (0.73, 1.12)  0.85 (0.66, 1.08)    Cases     Aged <65 years (n = 87)  Aged ≥65 years (n = 72)  Social network index (1 = I, 2 = II, 3 = III, 4 = IV)       Age-adjusted RR§  1.02 (0.80, 1.28)  0.88 (0.69, 1.14)   Multivariate RR  1.03 (0.81, 1.31)  0.91 (0.69, 1.20)  Sociability subscale (1, 2, 3)       Age-adjusted RR  1.05 (0.73, 1.50)  0.64 (0.44, 0.93)   Multivariate RR  1.02 (0.70, 1.48)  0.66 (0.43, 1.01)  Marital status (unmarried to married status)       Age-adjusted RR  1.16 (0.54, 2.50)  2.04 (0.88, 4.70)   Multivariate RR  1.21 (0.55, 2.67)  2.70 (0.99, 7.39)  Religious service attendance per month       Age-adjusted RR  0.98 (0.94, 1.03)  0.94 (0.89, 1.00)   Multivariate RR  0.99 (0.95, 1.03)  0.93 (0.87, 0.99)  Social group participation       Age-adjusted RR  1.41 (0.42, 4.72)  0.63 (0.25, 1.62)   Multivariate RR  1.29 (0.38, 4.35)  0.76 (0.27, 2.17)  Close relatives¶       Age-adjusted RR  0.97 (0.80, 1.19)  1.00 (0.81, 1.25)   Multivariate RR  0.96 (0.78, 1.18)  0.99 (0.79, 1.25)  Close relatives seen per month¶       Age-adjusted RR  0.98 (0.79, 1.22)  1.06 (0.83, 1.37)   Multivariate RR  1.00 (0.80, 1.25)  1.03 (0.78, 1.37)  Close friends¶       Age-adjusted RR  0.95 (0.76, 1.18)  0.71 (0.56, 0.90)   Multivariate RR  0.94 (0.75, 1.18)  0.71 (0.55, 0.92)  Close friends seen per month¶       Age-adjusted RR  0.90 (0.73, 1.10)  0.83 (0.67, 1.03)   Multivariate RR  0.90 (0.73, 1.12)  0.85 (0.66, 1.08)  * Multivariate relative risks were adjusted for overall level of the social tie between 1988 and 1996, overall levels including change between 1988 and 1996 in ability to climb several flights of stairs (yes/no) and do heavy housework (yes/no), employment status (full time, part time, retired/disabled) in 1988 and 1996, occupation in 1986, and the following variables in 1996: age (5-year categories), smoking history (never, past, and current in categories of 1–14, 15–24, and ≥25 cigarettes/day), daily alcohol intake (0, 0.01–9.9, 10–19.9, 20–29.9, and ≥30 g/day), quintiles of physical activity, and antidepressant use. † Except for the dichotomous variables marital status (0 = unmarried; 1 = married) and social group participation (0 = 10 or fewer hours per week; 1 = 11 or more hours per week); to examine the within-person effect of marrying and increased social group participation, binary coding was inverted compared with coding in table 4. ‡ Includes the Berkman-Syme social network index and its components (Am J Epidemiol 1979;109:186–204). § RR, relative risk. ¶ Scores were as follows: 1 = 0 relatives or friends, 2 = 1–2, 3 = 3–5, 4 = 6–9, 5 = ≥10. View Large DISCUSSION With an extended follow-up period of 10 years, we confirmed that men in the Health Professionals Follow-up Study who have fewer social ties have increased risks of mortality from all causes and from accidents and suicide. In addition, social isolation was associated with an increased risk of death from other causes. Prior studies in which the Berkman-Syme index or similar measures were used have demonstrated stronger associations with mortality (5, 8). In the Evans County study (5), men with a low level of social interaction had a 50 percent increased risk of death compared with socially well-integrated men (multivariate RR = 1.5, 95 percent CI: 0.8, 2.6). Because they are of a relatively high socioeconomic status and have greater health awareness, health professionals may be less vulnerable to the detrimental effects of social isolation. Hence, the present findings cannot be generalized to populations of lower socioeconomic status or educational attainment, in which a lack of social ties may have more profound effects. Nonresponse bias is a potential limitation, since only 73 percent of eligible subjects responded to questions on social ties in 1988. In our analysis of change in social ties, we found that only 81 percent of eligible men updated their information on social ties in 1996. While nonresponse may have been related to poorer health status, it did not appear to be associated with level of social integration. Therefore, nonresponse bias is unlikely to account for our findings. In terms of generalizability, caution is advised when extrapolating to less healthy populations that might be more adversely affected by social isolation. In addition, mortality risk estimates may be conservative because we excluded participants with preexisting disease, which precluded potentially important effects of social ties on disease prognosis. One of the cohort's strengths is its homogeneity in terms of socioeconomic status, which is a well-established predictor of health and is related to level of social ties. Furthermore, we adjusted for occupation and employment status in multivariate analyses. Although socioeconomic status can vary within occupational categories, we minimized such variation by restricting the cohort to men of high status. Previous validation studies have found dietary and alcohol intake, body mass index, and physical activity levels to be reported accurately by the cohort (32, 41–43). Self-reports of smoking habits are generally accurate in observational studies of adults (44). Accordingly, any residual confounding due to covariate misclassification should be minimal. Regarding endpoint ascertainment, death certificates were used to classify deaths from accidents and suicide and from other causes. Cancer and cardiovascular deaths (as well as incident coronary events) were physician-reviewed using medical records and standardized criteria. Since all classifications were blinded, any remaining error would likely have been random with respect to exposure, attenuating rather than exaggerating true effects. With specific coronary endpoints, we confirmed that social interaction is related primarily to fatal coronary heart disease and not to nonfatal myocardial infarction or sudden cardiac death. Social networks seem to affect prognosis rather than initial development of disease. A caveat is the possibility of missing cases, particularly silent myocardial infarctions. If socially isolated men were less likely to be diagnosed after suffering nonfatal infarction, then a null association might be observed despite a true underlying effect. Other investigators have directly examined the effect of social relationships on survival of coronary patients. Controlling for disease severity, functional status, and comorbidity, Berkman et al. reported a threefold increased risk of 6-month mortality for patients with less emotional support (45). In our analysis of network components, intimate ties such as close friendships were protective against fatal coronary heart disease. Our findings support the existence of mechanistic pathways independent of a range of controlled covariates. Social isolation may be directly related to disease via physiologic mechanisms such as accelerated aging (46), increased cardiovascular reactivity (24), and impaired immune function (47, 48). On the other hand, if social networks impact health via behavioral pathways, analyses including health and dietary behaviors would be inappropriate. In incremental models of the overall index and total mortality, health behaviors were the most influential confounders or potential mediators, while inclusion of dietary and comorbidity variables did not alter relative risk estimates. If health behaviors are truly mediators, then removing them from the fully adjusted model should result in a more accurate estimate of the underlying effect. The associations with multiple endpoints suggest a general susceptibility to disease but may also reflect different cause-specific mechanisms. To discern how social relationships impact health, recent studies have begun to incorporate functional measures of social ties (13, 14). The Berkman-Syme social network index measures structural aspects of relationships and does not assess functional social support. We were able to examine emotional support only indirectly by using items pertaining to marital status and to close friends and relatives. Although confidence intervals did not consistently exclude 1, the direction of effects for an increase in the overall index and several of its components suggested a health benefit associated with increasing social ties among elderly men. Underlying behavioral modifications or physiologic alterations presumably have more impact on survival among the more vulnerable aged. An increased number of close friends over time was particularly protective against subsequent death. Notably, ties with friends and relatives may exert greater influence on survival with age (8). The observation that men who became married had a nonsignificantly increased mortality risk does not concur with well-established reports of lower mortality among the married (49, 50), and it may be due to chance. Our findings cannot be readily attributed to social interaction acting as a proxy for health status, since men with serious illness were excluded prior to the assessment of social networks in 1988 and 1996. Furthermore, in multivariate models, we controlled for overall level including change in physical condition between 1988 and 1996, as well as age, antidepressant use, and key health behaviors in 1996. Because the temporal relation between change in social ties and change in health and behavior remains unclear, adjustment may have resulted in underestimation of effects. In the only known previous study examining change in social networks, change in an overall index was not predictive of mortality among 2,153 rural elders in multivariate analysis (30). Increase in social group participation and any change in church attendance were associated with an increased risk of mortality among men. Discrepant results between studies may be due to differences in population characteristics (rural vs. nonrural), length of follow-up (9 vs. 2 years), and interval for change (3 vs. 8 years). In summary, level of social ties was predictive of a variety of health outcomes over an extended follow-up period. Our findings suggest that within-person change in social interaction may play a role in short-term survival among elderly men. It is unclear whether change predicts mortality over a longer follow-up or whether long-term or acute changes are more detrimental. Such information would be invaluable in planning interventions over the life course. Additional studies are needed to examine social network change in relation to development of specific health endpoints in different populations. Reprint requests to Dr. Ichiro Kawachi, Department of Health and Social Behavior, Harvard School of Public Health, 677 Huntington Avenue, Boston, MA 02115 (e-mail: ichiro.kawachi@channing.harvard.edu). This study was supported by research grants HL 35464 and CA 55075 from the National Institutes of Health. Dr. P. M. Eng was supported by a National Research Service Award (3 T32 ES 07069-16A1) from the National Institute for Environmental Health Sciences. Dr. I. Kawachi was supported by a Career Development Award from the National Heart, Lung, and Blood Institute. REFERENCES 1. Berkman LF, Syme SL. Social networks, host resistance, and mortality: a nine-year follow-up study of Alameda County residents. Am J Epidemiol  1979; 109: 186–204. Google Scholar 2. House JS, Robbins C, Metzner HL. The association of social relationships and activities with mortality: prospective evidence from the Tecumseh Community Health Study. Am J Epidemiol  1982; 116: 123–40. Google Scholar 3. Blazer DG. Social support and mortality in an elderly community population. Am J Epidemiol  1982; 115: 684–94. Google Scholar 4. Welin L, Tibblin G, Svardsudd K, et al. Prospective study of social influences on mortality: the study of men born in 1913 and 1923. Lancet  1985; 1: 915–18. Google Scholar 5. Schoenbach VJ, Kaplan BH, Fredman L, et al. Social ties and mortality in Evans County, Georgia. Am J Epidemiol  1986; 123: 577–91. Google Scholar 6. Orth-Gomer K, Johnson J. Social network interaction and mortality: a six year follow-up of a random sample of the Swedish population. J Chronic Dis  1987; 40: 949–57. Google Scholar 7. Kaplan GA, Salonen JT, Cohen RD, et al. Social connections and mortality from all causes and from cardiovascular disease: prospective evidence from eastern Finland. Am J Epidemiol  1988; 128: 370–80. Google Scholar 8. Seeman TE, Kaplan GA, Knudsen L, et al. Social network ties and mortality among the elderly in the Alameda County study. Am J Epidemiol  1987; 126: 714–23. Google Scholar 9. Hanson BS, Isacsson SO, Janzon L, et al. Social network and social support influence mortality in elderly men. Am J Epidemiol  1989; 130: 100–11. Google Scholar 10. Vogt TM, Mullooly JP, Ernst D, et al. Social networks as predictors of ischemic heart disease, cancer, stroke and hypertension—incidence, survival and mortality. J Clin Epidemiol  1992; 45: 659–66. Google Scholar 11. Seeman TE, Berkman LF, Kohout F, et al. Intercommunity variations in the association between social ties and mortality in the elderly: a comparative analysis of three communities. Ann Epidemiol  1993; 3: 325–35. Google Scholar 12. Kawachi I, Colditz GA, Ascherio A, et al. A prospective study of social networks in relation to total mortality and cardiovascular disease in men in the USA. J Epidemiol Community Health  1996; 50: 245–51. Google Scholar 13. Sugisawa H, Liang J, Liu X. Social networks, social support, and mortality among older people in Japan. J Gerontol  1994; 49: S3–S13. Google Scholar 14. Pennix BW, van Tilburg T, Kriegsman DM, et al. Effects of social support and personal coping resources on mortality in older age: the Longitudinal Aging Study Amsterdam. Am J Epidemiol  1997; 146: 510–19. Google Scholar 15. Reynolds P, Kaplan GA. Social connections and risk for cancer: prospective evidence from the Alameda County study. Behav Med  1990; 16: 101–10. Google Scholar 16. Orth-Gomer K, Rosengren A, Wilhelmsen L. Lack of social support and incidence of coronary heart disease in middle-aged Swedish men. Psychosom Med  1993; 55: 37–43. Google Scholar 17. Berkman LF. The role of social relations in health promotion. Psychosom Med  1995; 57: 245–54. Google Scholar 18. Hanson BS, Isacsson SO, Janzon L, et al. Social support and quitting smoking for good: is there an association? Results from the population study “Men Born in 1914,” Malmo, Sweden. Addict Behav  1990; 15: 221–33. Google Scholar 19. Broman CL. Social relationships and health-related behavior. J Behav Med  1993; 16: 335–50. Google Scholar 20. McIntosh WA, Shifflett PA, Picou JS. Social support, stressful events, strain, dietary intake, and the elderly. Med Care  1989; 27: 140–53. Google Scholar 21. Donkin AJ, Johnson AE, Lilley JM. Gender and living alone as determinants of fruit and vegetable consumption among the elderly living at home in urban Nottingham. Appetite  1998; 30: 39–51. Google Scholar 22. Trieber FA, Batanowski T, Broden DS, et al. Social support for exercise: relationship to physical activity for young adults. Prev Med  1991; 20: 737–50. Google Scholar 23. Gotay CC, Wilson ME. Social support and cancer screening in African American, Hispanic, and Native American women. Cancer Pract  1998; 6: 31–7. Google Scholar 24. Kamarck TW, Manuck SB, Jennings JR. Social support reduces cardiovascular reactivity to psychological challenge: a laboratory model. Psychosom Med  1990; 52: 42–58. Google Scholar 25. Thomas PD, Goodwin JM, Goodwin JS. Effect of social support on stress-related changes in cholesterol levels, uric acid level and immune function in an elderly sample. Am J Psychiatry  1985; 142: 735–7. Google Scholar 26. Uchino BN, Cacioppo JT, Kiecolt-Glaser JK. The relationship between social support and physiological processes: a review with emphasis on underlying mechanisms and implications for health. Psychol Bull  1996; 119: 488–531. Google Scholar 27. Ebrahim S, Wannamethee G, McCallum A, et al. Marital status, change in marital status, and mortality in middle-aged British men. Am J Epidemiol  1995; 142: 834–42. Google Scholar 28. Kaprio J, Koskenvuo M, Rita H. Mortality after bereavement: a prospective study of 95,647 widowed persons. Am J Public Health  1987; 77: 283–6. Google Scholar 29. Mellstrom D, Nilsson A, Oden A, et al. Mortality among the widowed in Sweden. Scand J Soc Med  1982; 10: 33–41. Google Scholar 30. Cerhan JR, Wallace RB. Change in social ties and subsequent mortality in rural elders. Epidemiology  1997; 8: 475–81. Google Scholar 31. Berkman LF. Social networks, host resistance, and mortality: a follow-up study of Alameda County residents. Doctoral dissertation. Berkeley, CA: University of California at Berkeley, 1977. Google Scholar 32. Rimm EB, Giovannucci EL, Stampfer MJ, et al. Repro-ducibility and validity of an expanded self-administered semiquantitative food frequency questionnaire among male health professionals. Am J Epidemiol  1992; 135: 1114–26. Google Scholar 33. Stampfer MJ, Willett WC, Speizer FE, et al. Test of the National Death Index. Am J Epidemiol  1984; 119: 837–9. Google Scholar 34. Rose GA, Blackburn H. Cardiovascular survey methods. 2nd ed. Geneva, Switzerland: World Health Organization, 1982. Google Scholar 35. Mantel N, Haenszel W. Statistical aspects of the analysis of data from retrospective studies of disease. J Natl Cancer Inst  1959; 22: 719–48. Google Scholar 36. D'Agostino RB, Lee ML, Belanger AJ, et al. Relation of pooled logistic regression to time dependent Cox regression analysis: the Framingham Heart Study. Stat Med  1990; 9: 1501–15. Google Scholar 37. Mantel N. Chi-square tests with one degree of freedom: extensions of the Mantel-Haenszel procedure. J Am Stat Assoc  1963; 58: 690–700. Google Scholar 38. Cerhan JR, Wallace RB. Predictors of decline in social relationships in the rural elderly. Am J Epidemiol  1993; 137: 870–80. Google Scholar 39. Berkman LF. The assessment of social networks and social support in the elderly. J Am Geriatr Soc  1983; 31: 743–9. Google Scholar 40. Minkler M. Social support and health of the elderly. In: Cohen S, Syme SL, eds. Social support and health. New York, NY: Academic Press, 1985:199–216. Google Scholar 41. Giovannucci E, Colditz G, Stampfer MJ, et al. The assessment of alcohol consumption by a simple self-administered questionnaire. Am J Epidemiol  1991; 133: 810–17. Google Scholar 42. Rimm EB, Stampfer MJ, Colditz GA, et al. Validity of self-reported waist and hip circumferences in men and women. Epidemiology  1990; 1: 466–73. Google Scholar 43. Chasan-Taber S, Rimm EB, Stampfer MJ, et al. Repro-ducibility and validity of a self-administered physical activity questionnaire for male health professionals. Epidemiology  1996; 7: 81–6. Google Scholar 44. Patrick DL, Cheadle A, Thompson DC, et al. The validity of self-reported smoking: a review and meta-analysis. Am J Public Health  1994; 84: 1086–93. Google Scholar 45. Berkman LF, Leo-Summers L, Horwitz RI. Emotional support and survival after myocardial infarction: a prospective, population-based study of the elderly. Ann Intern Med  1992; 117: 1003–9. Google Scholar 46. Berkman LF, Glass T. Social integration, social networks, social support, and health. In: Berkman LF, Kawachi I, eds. Social epidemiology. New York, NY: Oxford University Press, 2000:137–73. Google Scholar 47. Kiecolt-Glaser JK, Garner W, Speicher CE, et al. Psychosocial modifiers of immunocompetence in medical students. Psychosom Med  1984; 46: 7–14. Google Scholar 48. Kiecolt-Glaser JK, Fisher LD, Ogrocki P, et al. Marital quality, marital disruption and immune function. Psychosom Med  1987; 49: 13–34. Google Scholar 49. Gove W. Sex, marital status, and mortality. Am J Sociol  1973; 79: 45–67. Google Scholar 50. Hu Y, Goldman N. Mortality differentials by marital status: an international comparison. Demography  1990; 27: 233–50. Google Scholar http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png American Journal of Epidemiology Oxford University Press

Social Ties and Change in Social Ties in Relation to Subsequent Total and Cause-specific Mortality and Coronary Heart Disease Incidence in Men

Loading next page...
 
/lp/oxford-university-press/social-ties-and-change-in-social-ties-in-relation-to-subsequent-total-e0iaoq0AWP

References (51)

Publisher
Oxford University Press
ISSN
0002-9262
eISSN
1476-6256
DOI
10.1093/aje/155.8.700
Publisher site
See Article on Publisher Site

Abstract

Abstract The authors prospectively examined the effects of social ties and change in social ties, as measured by a well-known social network index, on total and cause-specific mortality and on coronary heart disease incidence in 28,369 US male health professionals aged 42–77 years in 1988. Over 10 years, the relative risk of total mortality for men in the lower two levels of social integration compared with more socially integrated men was 1.19 (95% confidence interval: 1.06, 1.34) after controlling for age, occupation, health behaviors, general physical condition, coronary risk factors, and dietary habits. In multivariate analysis, deaths from accidents and suicide and from other noncancer, noncardiovascular causes were significantly increased among less socially connected men. Socially isolated men also had an increased risk of fatal coronary heart disease (multivariate relative risk = 1.82, 95% confidence interval: 1.02, 3.23). An increase in the overall social network index between 1988 and 1996 was not significantly associated with subsequent 2-year mortality. In analyses of change in social network components restricted to older men, each categorical unit increase in number of close friends was significantly associated with a 29% decrease in risk of death. Increase in religious service attendance over time was also significantly predictive of decreased mortality. aged, coronary disease, incidence, longitudinal studies, men, mortality, social isolation, social support CI, confidence interval, ICD-9, International Classification of Diseases, Ninth Revision, RR, relative risk To date, 14 known cohort studies have demonstrated a protective association between social interaction and mortality (1–14). Overall, socially isolated persons have two to four times the risk of all-cause mortality compared with those with more ties to friends, relatives, and community. While there is abundant evidence that social interaction protects against overall mortality, several issues remain unresolved. Relatively little is known of the impact of social ties on specific endpoints. A few studies have detected protective effects of social interaction on cardiovascular (7, 12) and cancer (15) mortality and on the incidence of coronary heart disease and stroke (12, 16). In a previous analysis of the Health Professionals Follow-up Study (12), socially isolated men had higher age-adjusted risks of death from cardiovascular disease and from accidents and suicide compared with well-integrated men. Over 4 years, social ties were also related to the incidence of fatal coronary heart disease. However, interpretation was limited for certain endpoints because of small case numbers. In the present study, we reexamined the relations between social ties, as measured by the Berkman-Syme social network index (1), and cause-specific mortality and coronary heart disease in the Health Professionals Follow-up Study. With extended follow-up, we sought to further clarify the associations with specific endpoints. Another question requiring closer study is the potential mechanisms by which social interaction affects health. Theoretically, social relationships may affect health through behavioral and/or physiologic pathways (17). Social support appears to influence negative health behaviors such as smoking, heavy alcohol consumption, poor dietary habits, sedentary lifestyle, and suboptimal health service utilization (18–23). Social ties may buffer against stress and thereby reduce activation of the neuroendocrine system, leading to diminished progression of atherosclerosis and other pathophysiologic processes (24–26). In the current analyses, health behavior, medical, and dietary factors were controlled for as confounders and were simultaneously evaluated for their mediating potential. Finally, whether change in social networks affects subsequent mortality has not been established. The relation between change in social ties and health has been examined primarily in terms of marital transition, particularly recent widowhood, and all-cause mortality (27–29). Generally, mortality rates have been higher for bereaved persons than for their married counterparts. Yet, the effects of social isolation and grief cannot be easily disentangled in bereavement studies. To our knowledge, only one study has examined change in more extensive ties; change in a social network index was not predictive of mortality in multivariate analysis (30). To address this issue, we measured social networks at two points 8 years apart, and we examined the impact of change in social ties on subsequent 2-year mortality. MATERIALS AND METHODS Study population The Health Professionals Follow-up Study is a prospective cohort study of chronic disease among 51,529 male health professionals aged 40–75 years in 1986. Cohort members are dentists (58 percent), veterinarians (20 percent), pharmacists (8 percent), optometrists (7 percent), osteopaths (4 percent), and podiatrists (3 percent). At enrollment, data on medical history and risk factors were obtained from the participants by mailed questionnaire. Every 2 years, follow-up questionnaires have been mailed to update information on exposures and newly diagnosed diseases. Response rates for the questionnaires have been over 90 percent. In 1988 and 1996, the Berkman-Syme social network index was incorporated into the questionnaire. Because preexisting disease could affect a person's social network, we excluded 12,052 participants who reported any of the following diagnoses prior to the return of the 1988 questionnaire: myocardial infarction, coronary artery bypass graft, percutaneous transluminal coronary angioplasty, angina pectoris, heart rhythm disorder, stroke, cancer (except nonmelanoma skin cancer), and chronic renal failure; we also omitted 434 men who died before 1988. We also excluded 10,674 men who did not provide social network data in 1988. The main study population therefore consisted of 28,369 men. Nonrespondents had a higher risk of dying over the 10-year follow-up (7.2 vs. 6.0 percent), and they were more likely to have smoked in 1986. Other health behaviors and medical conditions such as hypertension were distributed similarly among the respondent groups. Mean age, and distributions of marital status, employment status, and occupation, in 1986 were also similar. Because the social network index was updated in 1996, participants who reported any of the above-mentioned conditions prior to 1996 were excluded from the consideration of person-time beyond 1996. For our analysis of change in social networks, a subgroup of 17,769 men who completed the social network indices in 1988 and 1996 and were free of these conditions in 1996 was followed from 1996 to 1998. In 1988, these subjects had the same network index score (mean, 3.2) as the main group but were slightly younger, and they had fewer negative health behaviors and conditions. The social network index The Berkman-Syme social network index is based on four types of social connections: 1) marital status (married vs. unmarried); 2) sociability (frequency and contact with close friends and relatives measured as a subscale with levels 1, 2, and 3; lower values indicate fewer numbers and less contact); 3) religious group affiliation (yes vs. no); and 4) membership in other social or community organizations (yes vs. no). Participants provided data on marital status, number of close friends and relatives, number of friends and relatives seen at least once a month, frequency of religious service attendance, and hours spent each week participating in other social or community groups. The index measures degree of social integration, taking into account number as well as relative importance of different ties. Therefore, intimate ties are given more weight than group affiliations. The index has four levels: low, medium, medium-high, and high (respectively, I, II, III, and IV). Persons with low levels of social ties can be characterized as unmarried, having few friends or relatives, and not being involved in community groups. Further details on construction of the index have been described elsewhere (31). Covariates Information on covariates was updated every 2 years except as noted. Data were collected on health behaviors, including smoking, alcohol intake, combined leisure-time and routine physical activity, body mass index, and routine physical examination in the last 2 years. We obtained information on general physical condition, such as the ability to climb several flights of stairs and do heavy housework, and on employment status. Participants also reported diagnoses of hypertension, diabetes, and hypercholesterolemia and antidepressant use. Parental history of myocardial infarction before age 60 years was reported in 1986. Intakes of alcohol and energy-adjusted total and saturated fats, folate, and fiber were measured every 4 years by using a semiquantitative food frequency questionnaire (32). Vitamin supplement use was also reported. Occupational data on health profession (e.g., dentist) were obtained in 1986. Mortality and morbidity endpoints All endpoints occurred between return of the 1988 questionnaire and January 31, 1998, for the main analysis and between return of the 1996 questionnaire and January 31, 1998, for the analysis of change in social ties. In most instances, we were informed of a subject's death by next of kin, postal authorities, or work associates. The National Death Index was used to ascertain vital status of repeat nonresponders to the mailed questionnaires. It is estimated that 98 percent or more of cohort deaths are ascertained with the National Death Index (33). To determine cause of death, death certificates and hospital or pathology reports were reviewed by physicians blinded to exposure status. The following mortality outcomes were of interest: all-cause mortality; cardiovascular disease mortality, including coronary heart disease (International Classification of Diseases, Ninth Revision (ICD-9), codes 410–414, 798) and stroke (ICD codes 430–438); cancer mortality (ICD-9, codes 140–209); mortality due to external causes of injury (ICD “E” codes including accidents, suicides, poisonings, and other trauma); and mortality from all other causes. Medical records for reported coronary heart disease were reviewed by a physician in a blinded manner. Nonfatal myocardial infarction was classified as “definite” if World Health Organization criteria (34) were met, and required symptoms with typical changes in electrocardiograph readings or cardiac enzyme levels were noted. Nonfatal myocardial infarction was classified as “probable” when hospital records were not available but the event required hospital admission and the diagnosis was corroborated in an additional contact via letter or telephone. Deaths from myocardial infarction were confirmed by using hospital records and autopsy reports. Additionally, fatal coronary heart disease was confirmed by using death certificates if coronary heart disease was listed as the most plausible cause of death and preexistence of coronary heart disease was supported by hospital records or interviews with next of kin. Sudden cardiac death was defined as death occurring within 1 hour of symptom onset with no serious antecedent condition and no other likely cause of death reported. Approximately 12 percent of nonfatal myocardial infarctions and 13 percent of fatal cases were confirmed with supplementary information other than medical records and were deemed “probable.” We included both definite and probable cases in counts, since exclusion of probable events did not alter our results. All incident first events of nonfatal myocardial infarction, fatal coronary heart disease, sudden cardiac death, coronary artery bypass graft surgery, and angioplasty were included in counts for total coronary heart disease. Statistical analysis For each level of social integration, baseline proportions of potential confounders were age standardized by using the overall cohort's age distribution as the standard. Each subject contributed person-time from return of the 1988 questionnaire (or, for the analysis of change in social ties, from return of the 1996 questionnaire) until the time of the first event or death, or January 31, 1998. Relative risk estimates were adjusted for age (5-year categories) by using the Mantel-Haenszel summary estimator (35). Pooled logistic regression was used to examine the impact of social networks, controlling for multiple confounders. All covariates were time varying except occupation and parental history of myocardial infarction. The relative risk obtained from pooled logistic regression closely approximates the hazard rate ratio from Cox proportional hazards models with time-varying covariates under conditions met by this study: short (e.g., 2-year) follow-up intervals for grouping events and low incidence of an event within each interval (36). The Mantel extension test was used to examine trends in risks across levels of the social network index (37). The impact of an increase in social interaction on mortality, adjusted for overall level, was examined by including both average level and change in social ties between 1988 and 1996 as continuous predictors in multiple logistic regression models. Since most subjects experienced only incremental shifts in social network variables over time (e.g., level II to III of the social network index) rather than extreme changes (e.g., level I to IV), we used original categorical scores as values. Scores 1–4 were used for levels I–IV of the overall index, and 1–3 were used for the sociability subscale; a higher score indicated increased integration. For questions on friends and relatives, it was hypothesized that the effect of change would diminish with increasing underlying number; for example, an increase from 0 to 1–2 friends would have the same effect as an increase from 3–5 to 6–9 friends. Hence, we sought to indirectly account for potential differences in intimacy and support per relation. For religious service attendance, each unit increase in monthly attendance was hypothesized to have the same effect, regardless of underlying frequency. Dichotomous variables were used to examine the effects of increased social group participation over time (e.g., 10 or fewer to 11 or more hours per week) and becoming married. Because change in physical function may be linked to change in social ties, we controlled for overall levels, including change between 1988 and 1996, in the ability to climb several flights of stairs and do heavy housework as well as employment status in 1988 and 1996. We also controlled for occupation in 1986 and the following variables in 1996: age, smoking, alcohol intake, physical activity, and antidepressant use. Since the elderly may be particularly vulnerable to changes in both social networks and health status (38–40), change in social interaction could have more pronounced effects among older persons. Accordingly, we conducted analyses within strata of men aged less than 65 years and 65 years or more. RESULTS At the start of follow-up, 6.0 percent of the study population was socially isolated (level I of the social network index). The majority (51.3 percent) of subjects were well integrated (level IV). In table 1, we present age-standardized characteristics by social network index level in 1988. Socially isolated men were more likely to smoke, consume more alcohol, and be less active. TABLE 1. Age-standardized characteristics,* according to social network index level, of Health Professionals Follow-up Study participants in 1988, United States   Berkman-Syme social network index†     IV (high)  III  II  I (low)  No. (%)  14,555 (51.3)  5,329 (18.8)  6,798 (24.0)  1,687 (6.0)  Mean age (years)  55.2  55.4  55.1  55.4  Current smoking (%)  7.9  9.4  10.0  13.0  Alcohol intake ≥20 g/day (%)  14.1  20.2  20.6  23.4  High body mass index‡ (%)  20.1  18.9  19.0  17.8  Low level of physical activity§ (%)  16.8  19.8  21.3  24.6  Routine physical examination in the last 2 years (%)  66.7  63.3  60.0  55.6  Full-time employment (%)  82.1  81.4  80.2  76.8  Ability to climb flights of stairs (%)  82.1  79.7  80.6  79.4  Ability to do heavy housework (%)  97.6  96.8  96.8  95.9  Hypertension (%)  17.3  18.2  18.3  17.7  High serum cholesterol (%)  15.1  15.9  15.5  14.7  Diabetes (%)  2.4  2.2  2.4  2.6  Myocardial infarction in a parent aged <60 years (%)  11.2  11.9  12.4  11.3  Antidepressant use in 1990 (%)  0.9  0.8  1.5  1.6  Multivitamin use (%)  40.3  40.3  42.6  44.7  Vitamin E use (%)  19.5  20.5  22.0  22.5  High‡ intake of (%)           Total fat  20.2  21.0  21.0  19.6   Saturated fat  20.4  21.4  21.9  22.0  Low§ intake of (%)           Folate  18.3  21.3  22.5  24.3   Fiber  17.9  22.1  23.8  27.1    Berkman-Syme social network index†     IV (high)  III  II  I (low)  No. (%)  14,555 (51.3)  5,329 (18.8)  6,798 (24.0)  1,687 (6.0)  Mean age (years)  55.2  55.4  55.1  55.4  Current smoking (%)  7.9  9.4  10.0  13.0  Alcohol intake ≥20 g/day (%)  14.1  20.2  20.6  23.4  High body mass index‡ (%)  20.1  18.9  19.0  17.8  Low level of physical activity§ (%)  16.8  19.8  21.3  24.6  Routine physical examination in the last 2 years (%)  66.7  63.3  60.0  55.6  Full-time employment (%)  82.1  81.4  80.2  76.8  Ability to climb flights of stairs (%)  82.1  79.7  80.6  79.4  Ability to do heavy housework (%)  97.6  96.8  96.8  95.9  Hypertension (%)  17.3  18.2  18.3  17.7  High serum cholesterol (%)  15.1  15.9  15.5  14.7  Diabetes (%)  2.4  2.2  2.4  2.6  Myocardial infarction in a parent aged <60 years (%)  11.2  11.9  12.4  11.3  Antidepressant use in 1990 (%)  0.9  0.8  1.5  1.6  Multivitamin use (%)  40.3  40.3  42.6  44.7  Vitamin E use (%)  19.5  20.5  22.0  22.5  High‡ intake of (%)           Total fat  20.2  21.0  21.0  19.6   Saturated fat  20.4  21.4  21.9  22.0  Low§ intake of (%)           Folate  18.3  21.3  22.5  24.3   Fiber  17.9  22.1  23.8  27.1  * Intakes of total fat, saturated fat, folate, and fiber were energy adjusted. † Am J Epidemiol 1979;109:186–204. ‡ “High” defined as being in the highest quintile of the baseline population. § “Low” defined as being in the lowest quintile of the baseline population. View Large Over the 10-year follow-up, 1,365 deaths occurred in 256,684 person-years. Of these deaths, 320 were due to cardiovascular disease, 504 to cancer, 81 to accidents and suicides, and 272 to other causes. In table 2, we present the age-adjusted and multivariate relative risks of total and cause-specific mortality by level of social network index. Risk of total mortality was similarly elevated only for those men in the lower two levels, suggesting a threshold effect. In additional analyses, data for men in the two most isolated levels were combined and were compared with those for men in higher levels; men in the combined lower levels had a significantly increased risk of total mortality (multivariate relative risk (RR) = 1.19, 95 percent confidence interval (CI): 1.06, 1.34). When socially isolated men were compared with the most socially integrated men, the age-adjusted relative risk of cardiovascular death was 1.53 (95 percent CI: 1.02, 2.29). The corresponding multivariate estimate was attenuated and was not statistically significant. Men with a moderately low number of social ties (level II) had more than twice the risk of death from accidents and suicides relative to men with the most social connections (multivariate RR = 2.32, 95 percent CI: 1.39, 3.86). The p value for the multivariate trend test was 0.01. The multivariate relative risk of mortality from other causes was 1.55 (95 percent CI: 1.04, 2.32) for men in the lowest level of the index compared with socially well-integrated men. In the category of other causes, the most frequent causes of death were chronic obstructive pulmonary disease (10.7 percent) and chronic liver disease and cirrhosis (8.1 percent). Social ties were not associated with risk of cancer mortality. TABLE 2. Age-adjusted and multivariate relative risks (95% confidence intervals) of total and cause-specific mortality, by social network index level, Health Professionals Follow-up Study, United States, 1988–1998*   Berkman-Syme social network index†   p for trend    IV (high)  III  II  I (low)  Total mortality             No. of cases  620  244  385  116     Age-adjusted RR‡  1.00  1.00 (0.86, 1.16)  1.28 (1.13, 1.45)  1.49 (1.23, 1.82)  <0.001   Multivariate RR§  1.00  0.97 (0.83, 1.13)  1.18 (1.03, 1.34)  1.20 (0.97, 1.48)  0.009  Cardiovascular disease             No. of cases  146  60  86  28     Age-adjusted RR  1.00  1.04 (0.77, 1.41)  1.21 (0.93, 1.58)  1.53 (1.02, 2.29)  0.03   Multivariate RR§  1.00  1.07 (0.79, 1.44)  1.16 (0.88, 1.52)  1.37 (0.91, 2.08)  0.11  Total cancer             No. of cases  248  91  133  32     Age-adjusted RR  1.00  0.94 (0.74, 1.19)  1.11 (0.90, 1.37)  1.03 (0.72, 1.49)  0.47   Multivariate RR¶  1.00  0.91 (0.71, 1.16)  1.02 (0.82, 1.27)  0.86 (0.59, 1.25)  0.71  Accidents and suicides             No. of cases  28  15  33  5     Age-adjusted RR  1.00  1.38 (0.74, 2.59)  2.44 (1.48, 4.03)  1.44 (0.55, 3.72)  0.003   Multivariate RR#  1.00  1.40 (0.74, 2.63)  2.32 (1.39, 3.86)  1.27 (0.49, 3.33)  0.01  Other causes             No. of cases  118  46  74  34     Age-adjusted RR  1.00  0.99 (0.70, 1.39)  1.29 (0.96, 1.73)  2.29 (1.56, 3.36)  <0.001   Multivariate RR**  1.00  0.95 (0.67, 1.35)  1.12 (0.83, 1.51)  1.55 (1.04, 2.32)  0.07    Berkman-Syme social network index†   p for trend    IV (high)  III  II  I (low)  Total mortality             No. of cases  620  244  385  116     Age-adjusted RR‡  1.00  1.00 (0.86, 1.16)  1.28 (1.13, 1.45)  1.49 (1.23, 1.82)  <0.001   Multivariate RR§  1.00  0.97 (0.83, 1.13)  1.18 (1.03, 1.34)  1.20 (0.97, 1.48)  0.009  Cardiovascular disease             No. of cases  146  60  86  28     Age-adjusted RR  1.00  1.04 (0.77, 1.41)  1.21 (0.93, 1.58)  1.53 (1.02, 2.29)  0.03   Multivariate RR§  1.00  1.07 (0.79, 1.44)  1.16 (0.88, 1.52)  1.37 (0.91, 2.08)  0.11  Total cancer             No. of cases  248  91  133  32     Age-adjusted RR  1.00  0.94 (0.74, 1.19)  1.11 (0.90, 1.37)  1.03 (0.72, 1.49)  0.47   Multivariate RR¶  1.00  0.91 (0.71, 1.16)  1.02 (0.82, 1.27)  0.86 (0.59, 1.25)  0.71  Accidents and suicides             No. of cases  28  15  33  5     Age-adjusted RR  1.00  1.38 (0.74, 2.59)  2.44 (1.48, 4.03)  1.44 (0.55, 3.72)  0.003   Multivariate RR#  1.00  1.40 (0.74, 2.63)  2.32 (1.39, 3.86)  1.27 (0.49, 3.33)  0.01  Other causes             No. of cases  118  46  74  34     Age-adjusted RR  1.00  0.99 (0.70, 1.39)  1.29 (0.96, 1.73)  2.29 (1.56, 3.36)  <0.001   Multivariate RR**  1.00  0.95 (0.67, 1.35)  1.12 (0.83, 1.51)  1.55 (1.04, 2.32)  0.07  * All covariates were time varying except as noted. † Am J Epidemiol 1979;109:186–204. ‡ RR, relative risk. § Multivariate relative risks were adjusted for age (5-year categories) in 1988, time period (1988–1990, 1990–1992, 1992–1994, 1994–1996, 1996–1998), occupation in 1986, smoking history (never, past, and current in categories of 1–14, 15–24, and ≥25 cigarettes/day), daily alcohol intake (0, 0.01–9.9, 10–19.9, 20–29.9, and ≥30 g/day), quintiles of body mass index, quintiles of physical activity, routine physical examination in the last 2 years (yes/no), ability to climb several flights of stairs (yes/no), ability to do heavy housework (yes/no), employment status (full time, part time, retired, disabled), history of hypertension, diabetes, high serum cholesterol, history of myocardial infarction in a parent aged <60 years (yes/no) in 1986, quintiles of energy-adjusted intakes of total fat, saturated fat, folate, and fiber, and multivitamin and vitamin E supplement use (yes/no). ¶ Multivariate model same as specified in § but excluding hypertension, diabetes, high serum cholesterol, parental history of myocardial infarction in 1986, and all dietary variables. # Multivariate model adjusted for age, time period, occupation in 1986, smoking history, daily alcohol intake, physical activity, routine physical examination, employment status, and antidepressant use. ** Multivariate model same as specified in § but excluding parental history of myocardial infarction in 1986 and all dietary variables. View Large Confounding effects were examined further by adding covariates in conceptually related groups to a model of social network index and all-cause mortality adjusted for age and occupation only (table 3). Adding health behaviors to this base model resulted in the greatest change in the relative risk estimate comparing socially isolated with socially well-integrated men. Inclusion of general physical condition variables attenuated the age-and-occupation-adjusted relative risk to a moderate degree. In contrast, adding comorbidity or dietary variables to the base model did not substantially alter the relative risk estimates. Therefore, covariates related to health behavior and physical condition appeared to be the most influential confounders or, alternatively, to demonstrate the most potential as mediators between social networks and mortality. When we excluded health behaviors from the fully adjusted model for total mortality (table 2), the multivariate relative risk comparing lowest with highest levels of integration increased from 1.20 (95 percent CI: 0.97, 1.48) to 1.34 (95 percent CI: 1.09, 1.64). TABLE 3. Multivariate relative risks (95% confidence intervals) of total mortality, by social network index level from incremental models, Health Professionals Follow-up Study, United States, 1988–1998*   Berkman-Syme social network index†     IV (high)  III  II  I (low)  Age + occupation‡  1.00  1.01 (0.87, 1.18)  1.30 (1.14, 1.48)  1.53 (1.25, 1.87)  Age + occupation + health behavior variables§  1.00  0.97 (0.84, 1.13)  1.18 (1.03, 1.34)  1.28 (1.05, 1.58)  Age + occupation + general physical condition variables¶  1.00  0.99 (0.85, 1.15)  1.26 (1.10, 1.43)  1.37 (1.11, 1.68)  Age + occupation + medical history variables#  1.00  1.02 (0.88, 1.19)  1.29 (1.14, 1.47)  1.51 (1.24, 1.85)  Age + occupation + dietary variables**  1.00  1.01 (0.87, 1.18)  1.30 (1.14, 1.48)  1.50 (1.23, 1.84)    Berkman-Syme social network index†     IV (high)  III  II  I (low)  Age + occupation‡  1.00  1.01 (0.87, 1.18)  1.30 (1.14, 1.48)  1.53 (1.25, 1.87)  Age + occupation + health behavior variables§  1.00  0.97 (0.84, 1.13)  1.18 (1.03, 1.34)  1.28 (1.05, 1.58)  Age + occupation + general physical condition variables¶  1.00  0.99 (0.85, 1.15)  1.26 (1.10, 1.43)  1.37 (1.11, 1.68)  Age + occupation + medical history variables#  1.00  1.02 (0.88, 1.19)  1.29 (1.14, 1.47)  1.51 (1.24, 1.85)  Age + occupation + dietary variables**  1.00  1.01 (0.87, 1.18)  1.30 (1.14, 1.48)  1.50 (1.23, 1.84)  * All covariates were time varying except as noted. † Am J Epidemiol 1979;109:186–204. ‡ Occupation was assessed in 1986. § Health behavior variables include smoking history (never, past, and current in categories of 1–14, 15–24, and ≥25 cigarettes/day), daily alcohol intake (0, 0.01–9.9, 10–19.9, 20–29.9, and ≥30 g/day), quintiles of body mass index, quintiles of physical activity, routine physical examination in the last 2 years (yes/no). ¶ General physical condition variables include ability to climb several flights of stairs (yes/no), ability to do heavy housework (yes/no), employment status (full time, part time, retired, disabled). # Medical history variables include history of hypertension, diabetes, high serum cholesterol, history of myocardial infarction in a parent aged <60 years (yes/no) in 1986. ** Dietary variables include quintiles of energy-adjusted intakes of total fat, saturated fat, folate, and fiber, and multivitamin and vitamin E supplement use (yes/no). View Large Analysis of the overall index may have masked differential effects of underlying network components. Accordingly, results from multivariate analyses of individual components and of total and cause-specific mortality are presented in table 4. In these analyses, network components were mutually adjusted. Unmarried men had increased risks of total mortality and of death from accidents and suicide and from other causes. Religious service attendance and social group participation were protective against all-cause mortality. Having more close friends and contact with relatives were each associated with a decreased risk of dying from other causes, while more close relatives protected against accidental death and suicide. TABLE 4. Multivariate relative risks (95% confidence intervals) of total and cause-specific mortality, by social network index components,* Health Professionals Follow-up Study, United States, 1988–1998†   Total mortality‡  CVD§ mortality‡  Accidents, suicide¶  Other mortality#  Sociability subscale           1 = low  1.05 (0.87, 1.26)  1.03 (0.70, 1.53)  1.54 (0.75, 3.17)  0.99 (0.66, 1.47)   2 = medium  1.04 (0.91, 1.19)  1.21 (0.91, 1.60)  1.14 (0.63, 2.04)  0.92 (0.68, 1.24)   3 = high  1.00  1.00  1.00  1.00  Marital status           1 = unmarried  1.27 (1.07, 1.50)  1.14 (0.79, 1.63)  2.40 (1.40, 4.12)  1.63 (1.16, 2.27)   0 = married  1.00  1.00  1.00  1.00  Religious service attendance per month           1 = never/almost never  1.15 (1.02, 1.30)  1.21 (0.95, 1.53)  1.16 (0.73, 1.86)  1.23 (0.95, 1.60)   0 = once per year or more  1.00  1.00  1.00  1.00  Social group participation           1 = 10 or fewer hours per week  1.56 (1.04, 2.34)  1.76 (0.72, 4.30)  3.34 (0.82, 13.7)**  1.11 (0.51, 2.40)   0 = 11 or more hours per week  1.00  1.00  1.00  1.00  No. of close relatives           1 = none  1.01 (0.83, 1.24)  0.97 (0.64, 1.48)  2.93 (1.47, 5.82)  0.70 (0.44, 1.12)   0 = 1 or more  1.00  1.00  1.00  1.00  No. of relatives seen per month           1 = none  1.14 (1.00, 1.29)  1.18 (0.91, 1.52)  0.83 (0.48, 1.42)  1.42 (1.08, 1.87)   0 = 1 or more  1.00  1.00  1.00  1.00  No. of close friends           1 = none  0.94 (0.60, 1.46)  1.39 (0.59, 3.26)  0.96 (0.54, 1.71)**  2.84 (1.09, 7.36)   0 = 1 or more  1.00  1.00  1.00  1.00  No. of friends seen per month           1 = none  0.87 (0.64, 1.18)  0.82 (0.43, 1.57)  0.31 (0.07, 1.34)  0.44 (0.19, 1.00)   0 = 1 or more  1.00  1.00  1.00  1.00    Total mortality‡  CVD§ mortality‡  Accidents, suicide¶  Other mortality#  Sociability subscale           1 = low  1.05 (0.87, 1.26)  1.03 (0.70, 1.53)  1.54 (0.75, 3.17)  0.99 (0.66, 1.47)   2 = medium  1.04 (0.91, 1.19)  1.21 (0.91, 1.60)  1.14 (0.63, 2.04)  0.92 (0.68, 1.24)   3 = high  1.00  1.00  1.00  1.00  Marital status           1 = unmarried  1.27 (1.07, 1.50)  1.14 (0.79, 1.63)  2.40 (1.40, 4.12)  1.63 (1.16, 2.27)   0 = married  1.00  1.00  1.00  1.00  Religious service attendance per month           1 = never/almost never  1.15 (1.02, 1.30)  1.21 (0.95, 1.53)  1.16 (0.73, 1.86)  1.23 (0.95, 1.60)   0 = once per year or more  1.00  1.00  1.00  1.00  Social group participation           1 = 10 or fewer hours per week  1.56 (1.04, 2.34)  1.76 (0.72, 4.30)  3.34 (0.82, 13.7)**  1.11 (0.51, 2.40)   0 = 11 or more hours per week  1.00  1.00  1.00  1.00  No. of close relatives           1 = none  1.01 (0.83, 1.24)  0.97 (0.64, 1.48)  2.93 (1.47, 5.82)  0.70 (0.44, 1.12)   0 = 1 or more  1.00  1.00  1.00  1.00  No. of relatives seen per month           1 = none  1.14 (1.00, 1.29)  1.18 (0.91, 1.52)  0.83 (0.48, 1.42)  1.42 (1.08, 1.87)   0 = 1 or more  1.00  1.00  1.00  1.00  No. of close friends           1 = none  0.94 (0.60, 1.46)  1.39 (0.59, 3.26)  0.96 (0.54, 1.71)**  2.84 (1.09, 7.36)   0 = 1 or more  1.00  1.00  1.00  1.00  No. of friends seen per month           1 = none  0.87 (0.64, 1.18)  0.82 (0.43, 1.57)  0.31 (0.07, 1.34)  0.44 (0.19, 1.00)   0 = 1 or more  1.00  1.00  1.00  1.00  * Components of the Berkman-Syme social network index (Am J Epidemiol 1979;109:186–204) were mutually adjusted for each other. † All covariates were time varying except as noted. ‡ Multivariate relative risks were adjusted for age (5-year categories), time period (1988–1990, 1990–1992, 1992–1994, 1994–1996, 1996–1998), occupation in 1986, smoking history (never, past, and current in categories of 1–14, 15–24, and ≥25 cigarettes/day), daily alcohol intake (0, 0.01–9.9, 10–19.9, 20–29.9, and ≥30 g/day), quintiles of body mass index, quintiles of physical activity, routine physical examination in the last 2 years (yes/no), ability to climb several flights of stairs (yes/no), ability to do heavy housework (yes/no), employment status (full time, part time, retired, disabled), history of hypertension, diabetes, high serum cholesterol, history of myocardial infarction in a parent aged <60 years (yes/no) in 1986, quintiles of energy-adjusted intakes of total fat, saturated fat, folate, and fiber, and multivitamin and vitamin E supplement use (yes/no). § CVD, cardiovascular disease. ¶ Multivariate model adjusted for age, time period, occupation in 1986, smoking history, daily alcohol intake, physical activity, routine physical examiniation, employment status, and antidepressant use. # Multivariate model same as specified in ‡ but excluding parental history of myocardial infarction in 1986 and all dietary variables. ** Because of sparse deaths by accident/suicide, categories were collapsed as follows: social group participation (0 = 6 or more hours per week, 1 = 5 or fewer hours per week), number of close friends (0 = 3 or more, 1 = 2 or less). View Large In table 5, we present age-adjusted and multivariate relative risks of coronary heart disease. Over 10 years, we observed 1,816 incident cases of total coronary heart disease; 618 cases of nonfatal myocardial infarction, 142 cases of fatal coronary heart disease (excluding sudden cardiac death), and 97 sudden cardiac deaths occurred. Socially isolated men had an increased risk of fatal coronary heart disease compared with socially well-integrated men (multivariate RR = 1.82, 95 percent CI: 1.02, 3.23). Of individual network components, only the number of and contact with close friends were significantly associated with fatal coronary heart disease (data not shown). The incidence of total coronary heart disease, nonfatal myocardial infarction, and sudden cardiac death was not significantly increased among socially isolated men. TABLE 5. Age-adjusted and multivariate relative risks* (95% confidence intervals) of coronary heart disease, by social network index level, Health Professionals Follow-up Study, United States, 1988–1998†   Berkman-Syme social network index‡   p for trend    IV (high)  III  II  I (low)  Nonfatal myocardial infarction             No. of cases  306  139  130  43     Age-adjusted RR§  1.00  1.18 (0.96, 1.44)  0.88 (0.72, 1.08)  1.13 (0.82, 1.56)  0.80   Multivariate RR  1.00  1.18 (0.96, 1.44)  0.86 (0.70, 1.06)  1.11 (0.80, 1.53)  0.61  Fatal coronary heart disease             No. of cases  62  29  36  15     Age-adjusted RR  1.00  1.19 (0.77, 1.85)  1.20 (0.79, 1.80)  1.93 (1.10, 3.39)  0.06   Multivariate RR  1.00  1.26 (0.80, 1.96)  1.15 (0.76, 1.75)  1.82 (1.02, 3.23)  0.10  Sudden cardiac death             No. of cases  50  20  22  5     Age-adjusted RR  1.00  1.02 (0.61, 1.72)  0.91 (0.55, 1.50)  0.80 (0.32, 2.01)  0.60   Multivariate RR  1.00  1.02 (0.61, 1.73)  0.87 (0.52, 1.44)  0.71 (0.28, 1.81)  0.43  Total coronary heart disease             No. of cases  917  383  400  116     Age-adjusted RR  1.00  1.08 (0.96, 1.21)  0.90 (0.80, 1.02)  1.02 (0.84, 1.23)  0.34   Multivariate RR  1.00  1.07 (0.95, 1.21)  0.88 (0.78, 0.99)  0.99 (0.81, 1.20)  0.14    Berkman-Syme social network index‡   p for trend    IV (high)  III  II  I (low)  Nonfatal myocardial infarction             No. of cases  306  139  130  43     Age-adjusted RR§  1.00  1.18 (0.96, 1.44)  0.88 (0.72, 1.08)  1.13 (0.82, 1.56)  0.80   Multivariate RR  1.00  1.18 (0.96, 1.44)  0.86 (0.70, 1.06)  1.11 (0.80, 1.53)  0.61  Fatal coronary heart disease             No. of cases  62  29  36  15     Age-adjusted RR  1.00  1.19 (0.77, 1.85)  1.20 (0.79, 1.80)  1.93 (1.10, 3.39)  0.06   Multivariate RR  1.00  1.26 (0.80, 1.96)  1.15 (0.76, 1.75)  1.82 (1.02, 3.23)  0.10  Sudden cardiac death             No. of cases  50  20  22  5     Age-adjusted RR  1.00  1.02 (0.61, 1.72)  0.91 (0.55, 1.50)  0.80 (0.32, 2.01)  0.60   Multivariate RR  1.00  1.02 (0.61, 1.73)  0.87 (0.52, 1.44)  0.71 (0.28, 1.81)  0.43  Total coronary heart disease             No. of cases  917  383  400  116     Age-adjusted RR  1.00  1.08 (0.96, 1.21)  0.90 (0.80, 1.02)  1.02 (0.84, 1.23)  0.34   Multivariate RR  1.00  1.07 (0.95, 1.21)  0.88 (0.78, 0.99)  0.99 (0.81, 1.20)  0.14  * Multivariate relative risks were adjusted for age (5-year categories), time period (1988–1990, 1990–1992, 1992–1994, 1994–1996, 1996–1998), occupation in 1986, smoking history (never, past, and current in categories of 1–14, 15–24, and ≥25 cigarettes/day), daily alcohol intake (0, 0.01–9.9, 10–19.9, 20–29.9, and ≥30 g/day), quintiles of body mass index, quintiles of physical activity, routine physical examination in the last 2 years (yes/no), ability to climb several flights of stairs (yes/no), ability to do heavy housework (yes/no), employment status (full time, part time, retired, disabled), history of hypertension, diabetes, high serum cholesterol, history of myocardial infarction in a parent aged <60 years (yes/no) in 1986, quintiles of energy-adjusted intakes of total fat, saturated fat, folate, and fiber, and multivitamin and vitamin E supplement use (yes/no). † All covariates were time varying except as noted. ‡ Am J Epidemiol 1979;109:186–204. § RR, relative risk. View Large Prior to analyzing the effect of change in social ties on mortality, we examined the stability of social networks over time. When we compared 1988 and 1996 social network indices, level of social interaction did not remain fixed; the Spearman correlation coefficient was 0.57. Correlation coefficients for network components were as follows: religious service attendance, r = 0.79; social group participation, r = 0.47; sociability subscale, r = 0.52; number of close friends, r = 0.61 and relatives, r = 0.58; and monthly contact with friends, r = 0.57 and relatives, r = 0.60. An increase in the number of social ties between 1988 and 1996 was not predictive of subsequent 2-year mortality for men less than age 65 years (table 6). For older men, we noted a nonsignificant decrease in mortality associated with an increase in the overall index between 1988 and 1996. For several facets of the elderly person's social network, an increase in social interaction over time was associated with survival. Because of a limited number of deaths of elderly persons (n = 72), confidence intervals were wide and often included 1. In age-adjusted analysis, a within-person unit increase on the sociability subscale was associated with a 36 percent decrease in mortality, although this decrease was attenuated and was not significant in multivariate analysis. Friendships as opposed to interaction with relatives appeared to drive the latter association. Each categorical unit increase in the number of close friends was significantly associated with a 29 percent decrease in mortality. In addition, every additional religious service attendance per month was associated with a significant, although modest decline of 7 percent. In contrast, becoming married was linked to markedly, albeit nonsignificantly increased mortality risk among older men. TABLE 6. Age-adjusted and multivariate relative risks* (95% confidence intervals) of total mortality, per unit increase† in social ties‡ between 1988 and 1996, for men aged <65 and ≥65 years, Health Professionals Follow-up Study, United States, 1996–1998   Cases     Aged <65 years (n = 87)  Aged ≥65 years (n = 72)  Social network index (1 = I, 2 = II, 3 = III, 4 = IV)       Age-adjusted RR§  1.02 (0.80, 1.28)  0.88 (0.69, 1.14)   Multivariate RR  1.03 (0.81, 1.31)  0.91 (0.69, 1.20)  Sociability subscale (1, 2, 3)       Age-adjusted RR  1.05 (0.73, 1.50)  0.64 (0.44, 0.93)   Multivariate RR  1.02 (0.70, 1.48)  0.66 (0.43, 1.01)  Marital status (unmarried to married status)       Age-adjusted RR  1.16 (0.54, 2.50)  2.04 (0.88, 4.70)   Multivariate RR  1.21 (0.55, 2.67)  2.70 (0.99, 7.39)  Religious service attendance per month       Age-adjusted RR  0.98 (0.94, 1.03)  0.94 (0.89, 1.00)   Multivariate RR  0.99 (0.95, 1.03)  0.93 (0.87, 0.99)  Social group participation       Age-adjusted RR  1.41 (0.42, 4.72)  0.63 (0.25, 1.62)   Multivariate RR  1.29 (0.38, 4.35)  0.76 (0.27, 2.17)  Close relatives¶       Age-adjusted RR  0.97 (0.80, 1.19)  1.00 (0.81, 1.25)   Multivariate RR  0.96 (0.78, 1.18)  0.99 (0.79, 1.25)  Close relatives seen per month¶       Age-adjusted RR  0.98 (0.79, 1.22)  1.06 (0.83, 1.37)   Multivariate RR  1.00 (0.80, 1.25)  1.03 (0.78, 1.37)  Close friends¶       Age-adjusted RR  0.95 (0.76, 1.18)  0.71 (0.56, 0.90)   Multivariate RR  0.94 (0.75, 1.18)  0.71 (0.55, 0.92)  Close friends seen per month¶       Age-adjusted RR  0.90 (0.73, 1.10)  0.83 (0.67, 1.03)   Multivariate RR  0.90 (0.73, 1.12)  0.85 (0.66, 1.08)    Cases     Aged <65 years (n = 87)  Aged ≥65 years (n = 72)  Social network index (1 = I, 2 = II, 3 = III, 4 = IV)       Age-adjusted RR§  1.02 (0.80, 1.28)  0.88 (0.69, 1.14)   Multivariate RR  1.03 (0.81, 1.31)  0.91 (0.69, 1.20)  Sociability subscale (1, 2, 3)       Age-adjusted RR  1.05 (0.73, 1.50)  0.64 (0.44, 0.93)   Multivariate RR  1.02 (0.70, 1.48)  0.66 (0.43, 1.01)  Marital status (unmarried to married status)       Age-adjusted RR  1.16 (0.54, 2.50)  2.04 (0.88, 4.70)   Multivariate RR  1.21 (0.55, 2.67)  2.70 (0.99, 7.39)  Religious service attendance per month       Age-adjusted RR  0.98 (0.94, 1.03)  0.94 (0.89, 1.00)   Multivariate RR  0.99 (0.95, 1.03)  0.93 (0.87, 0.99)  Social group participation       Age-adjusted RR  1.41 (0.42, 4.72)  0.63 (0.25, 1.62)   Multivariate RR  1.29 (0.38, 4.35)  0.76 (0.27, 2.17)  Close relatives¶       Age-adjusted RR  0.97 (0.80, 1.19)  1.00 (0.81, 1.25)   Multivariate RR  0.96 (0.78, 1.18)  0.99 (0.79, 1.25)  Close relatives seen per month¶       Age-adjusted RR  0.98 (0.79, 1.22)  1.06 (0.83, 1.37)   Multivariate RR  1.00 (0.80, 1.25)  1.03 (0.78, 1.37)  Close friends¶       Age-adjusted RR  0.95 (0.76, 1.18)  0.71 (0.56, 0.90)   Multivariate RR  0.94 (0.75, 1.18)  0.71 (0.55, 0.92)  Close friends seen per month¶       Age-adjusted RR  0.90 (0.73, 1.10)  0.83 (0.67, 1.03)   Multivariate RR  0.90 (0.73, 1.12)  0.85 (0.66, 1.08)  * Multivariate relative risks were adjusted for overall level of the social tie between 1988 and 1996, overall levels including change between 1988 and 1996 in ability to climb several flights of stairs (yes/no) and do heavy housework (yes/no), employment status (full time, part time, retired/disabled) in 1988 and 1996, occupation in 1986, and the following variables in 1996: age (5-year categories), smoking history (never, past, and current in categories of 1–14, 15–24, and ≥25 cigarettes/day), daily alcohol intake (0, 0.01–9.9, 10–19.9, 20–29.9, and ≥30 g/day), quintiles of physical activity, and antidepressant use. † Except for the dichotomous variables marital status (0 = unmarried; 1 = married) and social group participation (0 = 10 or fewer hours per week; 1 = 11 or more hours per week); to examine the within-person effect of marrying and increased social group participation, binary coding was inverted compared with coding in table 4. ‡ Includes the Berkman-Syme social network index and its components (Am J Epidemiol 1979;109:186–204). § RR, relative risk. ¶ Scores were as follows: 1 = 0 relatives or friends, 2 = 1–2, 3 = 3–5, 4 = 6–9, 5 = ≥10. View Large DISCUSSION With an extended follow-up period of 10 years, we confirmed that men in the Health Professionals Follow-up Study who have fewer social ties have increased risks of mortality from all causes and from accidents and suicide. In addition, social isolation was associated with an increased risk of death from other causes. Prior studies in which the Berkman-Syme index or similar measures were used have demonstrated stronger associations with mortality (5, 8). In the Evans County study (5), men with a low level of social interaction had a 50 percent increased risk of death compared with socially well-integrated men (multivariate RR = 1.5, 95 percent CI: 0.8, 2.6). Because they are of a relatively high socioeconomic status and have greater health awareness, health professionals may be less vulnerable to the detrimental effects of social isolation. Hence, the present findings cannot be generalized to populations of lower socioeconomic status or educational attainment, in which a lack of social ties may have more profound effects. Nonresponse bias is a potential limitation, since only 73 percent of eligible subjects responded to questions on social ties in 1988. In our analysis of change in social ties, we found that only 81 percent of eligible men updated their information on social ties in 1996. While nonresponse may have been related to poorer health status, it did not appear to be associated with level of social integration. Therefore, nonresponse bias is unlikely to account for our findings. In terms of generalizability, caution is advised when extrapolating to less healthy populations that might be more adversely affected by social isolation. In addition, mortality risk estimates may be conservative because we excluded participants with preexisting disease, which precluded potentially important effects of social ties on disease prognosis. One of the cohort's strengths is its homogeneity in terms of socioeconomic status, which is a well-established predictor of health and is related to level of social ties. Furthermore, we adjusted for occupation and employment status in multivariate analyses. Although socioeconomic status can vary within occupational categories, we minimized such variation by restricting the cohort to men of high status. Previous validation studies have found dietary and alcohol intake, body mass index, and physical activity levels to be reported accurately by the cohort (32, 41–43). Self-reports of smoking habits are generally accurate in observational studies of adults (44). Accordingly, any residual confounding due to covariate misclassification should be minimal. Regarding endpoint ascertainment, death certificates were used to classify deaths from accidents and suicide and from other causes. Cancer and cardiovascular deaths (as well as incident coronary events) were physician-reviewed using medical records and standardized criteria. Since all classifications were blinded, any remaining error would likely have been random with respect to exposure, attenuating rather than exaggerating true effects. With specific coronary endpoints, we confirmed that social interaction is related primarily to fatal coronary heart disease and not to nonfatal myocardial infarction or sudden cardiac death. Social networks seem to affect prognosis rather than initial development of disease. A caveat is the possibility of missing cases, particularly silent myocardial infarctions. If socially isolated men were less likely to be diagnosed after suffering nonfatal infarction, then a null association might be observed despite a true underlying effect. Other investigators have directly examined the effect of social relationships on survival of coronary patients. Controlling for disease severity, functional status, and comorbidity, Berkman et al. reported a threefold increased risk of 6-month mortality for patients with less emotional support (45). In our analysis of network components, intimate ties such as close friendships were protective against fatal coronary heart disease. Our findings support the existence of mechanistic pathways independent of a range of controlled covariates. Social isolation may be directly related to disease via physiologic mechanisms such as accelerated aging (46), increased cardiovascular reactivity (24), and impaired immune function (47, 48). On the other hand, if social networks impact health via behavioral pathways, analyses including health and dietary behaviors would be inappropriate. In incremental models of the overall index and total mortality, health behaviors were the most influential confounders or potential mediators, while inclusion of dietary and comorbidity variables did not alter relative risk estimates. If health behaviors are truly mediators, then removing them from the fully adjusted model should result in a more accurate estimate of the underlying effect. The associations with multiple endpoints suggest a general susceptibility to disease but may also reflect different cause-specific mechanisms. To discern how social relationships impact health, recent studies have begun to incorporate functional measures of social ties (13, 14). The Berkman-Syme social network index measures structural aspects of relationships and does not assess functional social support. We were able to examine emotional support only indirectly by using items pertaining to marital status and to close friends and relatives. Although confidence intervals did not consistently exclude 1, the direction of effects for an increase in the overall index and several of its components suggested a health benefit associated with increasing social ties among elderly men. Underlying behavioral modifications or physiologic alterations presumably have more impact on survival among the more vulnerable aged. An increased number of close friends over time was particularly protective against subsequent death. Notably, ties with friends and relatives may exert greater influence on survival with age (8). The observation that men who became married had a nonsignificantly increased mortality risk does not concur with well-established reports of lower mortality among the married (49, 50), and it may be due to chance. Our findings cannot be readily attributed to social interaction acting as a proxy for health status, since men with serious illness were excluded prior to the assessment of social networks in 1988 and 1996. Furthermore, in multivariate models, we controlled for overall level including change in physical condition between 1988 and 1996, as well as age, antidepressant use, and key health behaviors in 1996. Because the temporal relation between change in social ties and change in health and behavior remains unclear, adjustment may have resulted in underestimation of effects. In the only known previous study examining change in social networks, change in an overall index was not predictive of mortality among 2,153 rural elders in multivariate analysis (30). Increase in social group participation and any change in church attendance were associated with an increased risk of mortality among men. Discrepant results between studies may be due to differences in population characteristics (rural vs. nonrural), length of follow-up (9 vs. 2 years), and interval for change (3 vs. 8 years). In summary, level of social ties was predictive of a variety of health outcomes over an extended follow-up period. Our findings suggest that within-person change in social interaction may play a role in short-term survival among elderly men. It is unclear whether change predicts mortality over a longer follow-up or whether long-term or acute changes are more detrimental. Such information would be invaluable in planning interventions over the life course. Additional studies are needed to examine social network change in relation to development of specific health endpoints in different populations. Reprint requests to Dr. Ichiro Kawachi, Department of Health and Social Behavior, Harvard School of Public Health, 677 Huntington Avenue, Boston, MA 02115 (e-mail: ichiro.kawachi@channing.harvard.edu). This study was supported by research grants HL 35464 and CA 55075 from the National Institutes of Health. Dr. P. M. Eng was supported by a National Research Service Award (3 T32 ES 07069-16A1) from the National Institute for Environmental Health Sciences. Dr. I. Kawachi was supported by a Career Development Award from the National Heart, Lung, and Blood Institute. REFERENCES 1. Berkman LF, Syme SL. Social networks, host resistance, and mortality: a nine-year follow-up study of Alameda County residents. Am J Epidemiol  1979; 109: 186–204. Google Scholar 2. House JS, Robbins C, Metzner HL. The association of social relationships and activities with mortality: prospective evidence from the Tecumseh Community Health Study. Am J Epidemiol  1982; 116: 123–40. Google Scholar 3. Blazer DG. Social support and mortality in an elderly community population. Am J Epidemiol  1982; 115: 684–94. Google Scholar 4. Welin L, Tibblin G, Svardsudd K, et al. Prospective study of social influences on mortality: the study of men born in 1913 and 1923. Lancet  1985; 1: 915–18. Google Scholar 5. Schoenbach VJ, Kaplan BH, Fredman L, et al. Social ties and mortality in Evans County, Georgia. Am J Epidemiol  1986; 123: 577–91. Google Scholar 6. Orth-Gomer K, Johnson J. Social network interaction and mortality: a six year follow-up of a random sample of the Swedish population. J Chronic Dis  1987; 40: 949–57. Google Scholar 7. Kaplan GA, Salonen JT, Cohen RD, et al. Social connections and mortality from all causes and from cardiovascular disease: prospective evidence from eastern Finland. Am J Epidemiol  1988; 128: 370–80. Google Scholar 8. Seeman TE, Kaplan GA, Knudsen L, et al. Social network ties and mortality among the elderly in the Alameda County study. Am J Epidemiol  1987; 126: 714–23. Google Scholar 9. Hanson BS, Isacsson SO, Janzon L, et al. Social network and social support influence mortality in elderly men. Am J Epidemiol  1989; 130: 100–11. Google Scholar 10. Vogt TM, Mullooly JP, Ernst D, et al. Social networks as predictors of ischemic heart disease, cancer, stroke and hypertension—incidence, survival and mortality. J Clin Epidemiol  1992; 45: 659–66. Google Scholar 11. Seeman TE, Berkman LF, Kohout F, et al. Intercommunity variations in the association between social ties and mortality in the elderly: a comparative analysis of three communities. Ann Epidemiol  1993; 3: 325–35. Google Scholar 12. Kawachi I, Colditz GA, Ascherio A, et al. A prospective study of social networks in relation to total mortality and cardiovascular disease in men in the USA. J Epidemiol Community Health  1996; 50: 245–51. Google Scholar 13. Sugisawa H, Liang J, Liu X. Social networks, social support, and mortality among older people in Japan. J Gerontol  1994; 49: S3–S13. Google Scholar 14. Pennix BW, van Tilburg T, Kriegsman DM, et al. Effects of social support and personal coping resources on mortality in older age: the Longitudinal Aging Study Amsterdam. Am J Epidemiol  1997; 146: 510–19. Google Scholar 15. Reynolds P, Kaplan GA. Social connections and risk for cancer: prospective evidence from the Alameda County study. Behav Med  1990; 16: 101–10. Google Scholar 16. Orth-Gomer K, Rosengren A, Wilhelmsen L. Lack of social support and incidence of coronary heart disease in middle-aged Swedish men. Psychosom Med  1993; 55: 37–43. Google Scholar 17. Berkman LF. The role of social relations in health promotion. Psychosom Med  1995; 57: 245–54. Google Scholar 18. Hanson BS, Isacsson SO, Janzon L, et al. Social support and quitting smoking for good: is there an association? Results from the population study “Men Born in 1914,” Malmo, Sweden. Addict Behav  1990; 15: 221–33. Google Scholar 19. Broman CL. Social relationships and health-related behavior. J Behav Med  1993; 16: 335–50. Google Scholar 20. McIntosh WA, Shifflett PA, Picou JS. Social support, stressful events, strain, dietary intake, and the elderly. Med Care  1989; 27: 140–53. Google Scholar 21. Donkin AJ, Johnson AE, Lilley JM. Gender and living alone as determinants of fruit and vegetable consumption among the elderly living at home in urban Nottingham. Appetite  1998; 30: 39–51. Google Scholar 22. Trieber FA, Batanowski T, Broden DS, et al. Social support for exercise: relationship to physical activity for young adults. Prev Med  1991; 20: 737–50. Google Scholar 23. Gotay CC, Wilson ME. Social support and cancer screening in African American, Hispanic, and Native American women. Cancer Pract  1998; 6: 31–7. Google Scholar 24. Kamarck TW, Manuck SB, Jennings JR. Social support reduces cardiovascular reactivity to psychological challenge: a laboratory model. Psychosom Med  1990; 52: 42–58. Google Scholar 25. Thomas PD, Goodwin JM, Goodwin JS. Effect of social support on stress-related changes in cholesterol levels, uric acid level and immune function in an elderly sample. Am J Psychiatry  1985; 142: 735–7. Google Scholar 26. Uchino BN, Cacioppo JT, Kiecolt-Glaser JK. The relationship between social support and physiological processes: a review with emphasis on underlying mechanisms and implications for health. Psychol Bull  1996; 119: 488–531. Google Scholar 27. Ebrahim S, Wannamethee G, McCallum A, et al. Marital status, change in marital status, and mortality in middle-aged British men. Am J Epidemiol  1995; 142: 834–42. Google Scholar 28. Kaprio J, Koskenvuo M, Rita H. Mortality after bereavement: a prospective study of 95,647 widowed persons. Am J Public Health  1987; 77: 283–6. Google Scholar 29. Mellstrom D, Nilsson A, Oden A, et al. Mortality among the widowed in Sweden. Scand J Soc Med  1982; 10: 33–41. Google Scholar 30. Cerhan JR, Wallace RB. Change in social ties and subsequent mortality in rural elders. Epidemiology  1997; 8: 475–81. Google Scholar 31. Berkman LF. Social networks, host resistance, and mortality: a follow-up study of Alameda County residents. Doctoral dissertation. Berkeley, CA: University of California at Berkeley, 1977. Google Scholar 32. Rimm EB, Giovannucci EL, Stampfer MJ, et al. Repro-ducibility and validity of an expanded self-administered semiquantitative food frequency questionnaire among male health professionals. Am J Epidemiol  1992; 135: 1114–26. Google Scholar 33. Stampfer MJ, Willett WC, Speizer FE, et al. Test of the National Death Index. Am J Epidemiol  1984; 119: 837–9. Google Scholar 34. Rose GA, Blackburn H. Cardiovascular survey methods. 2nd ed. Geneva, Switzerland: World Health Organization, 1982. Google Scholar 35. Mantel N, Haenszel W. Statistical aspects of the analysis of data from retrospective studies of disease. J Natl Cancer Inst  1959; 22: 719–48. Google Scholar 36. D'Agostino RB, Lee ML, Belanger AJ, et al. Relation of pooled logistic regression to time dependent Cox regression analysis: the Framingham Heart Study. Stat Med  1990; 9: 1501–15. Google Scholar 37. Mantel N. Chi-square tests with one degree of freedom: extensions of the Mantel-Haenszel procedure. J Am Stat Assoc  1963; 58: 690–700. Google Scholar 38. Cerhan JR, Wallace RB. Predictors of decline in social relationships in the rural elderly. Am J Epidemiol  1993; 137: 870–80. Google Scholar 39. Berkman LF. The assessment of social networks and social support in the elderly. J Am Geriatr Soc  1983; 31: 743–9. Google Scholar 40. Minkler M. Social support and health of the elderly. In: Cohen S, Syme SL, eds. Social support and health. New York, NY: Academic Press, 1985:199–216. Google Scholar 41. Giovannucci E, Colditz G, Stampfer MJ, et al. The assessment of alcohol consumption by a simple self-administered questionnaire. Am J Epidemiol  1991; 133: 810–17. Google Scholar 42. Rimm EB, Stampfer MJ, Colditz GA, et al. Validity of self-reported waist and hip circumferences in men and women. Epidemiology  1990; 1: 466–73. Google Scholar 43. Chasan-Taber S, Rimm EB, Stampfer MJ, et al. Repro-ducibility and validity of a self-administered physical activity questionnaire for male health professionals. Epidemiology  1996; 7: 81–6. Google Scholar 44. Patrick DL, Cheadle A, Thompson DC, et al. The validity of self-reported smoking: a review and meta-analysis. Am J Public Health  1994; 84: 1086–93. Google Scholar 45. Berkman LF, Leo-Summers L, Horwitz RI. Emotional support and survival after myocardial infarction: a prospective, population-based study of the elderly. Ann Intern Med  1992; 117: 1003–9. Google Scholar 46. Berkman LF, Glass T. Social integration, social networks, social support, and health. In: Berkman LF, Kawachi I, eds. Social epidemiology. New York, NY: Oxford University Press, 2000:137–73. Google Scholar 47. Kiecolt-Glaser JK, Garner W, Speicher CE, et al. Psychosocial modifiers of immunocompetence in medical students. Psychosom Med  1984; 46: 7–14. Google Scholar 48. Kiecolt-Glaser JK, Fisher LD, Ogrocki P, et al. Marital quality, marital disruption and immune function. Psychosom Med  1987; 49: 13–34. Google Scholar 49. Gove W. Sex, marital status, and mortality. Am J Sociol  1973; 79: 45–67. Google Scholar 50. Hu Y, Goldman N. Mortality differentials by marital status: an international comparison. Demography  1990; 27: 233–50. Google Scholar

Journal

American Journal of EpidemiologyOxford University Press

Published: Apr 15, 2002

There are no references for this article.