The world's population is aging. Persons alive at age 85 years or older, often designated the “oldest old,” are the fastest-growing age group in most industrialized countries and are among the largest consumers of health care resources.¹ Identifying strategies for remaining healthy, vigorous, and disability-free at older ages has become a major priority, as reflected by the US Surgeon General's Healthy People 2010 objectives,² the US National Institute on Aging Strategic Plan,³ and increased study of healthy “successful”⁴ or “effective”⁵ aging.

In many ways, remarkable progress has been made in creating healthier seniors.⁶ Seniors are living longer and with less morbidity and disability than in past years; however, health and survival benefits have been less apparent in men than in women.⁷ Relatively few men actually live to oldest-old age. By age 85 years, women outnumber men by 2.2 to 1 in the United States.⁸ This ratio exceeds 3 to 1 in nonagenarians and 4 to 1 in centenarians.⁹ While the survival gap has narrowed recently in the United States,⁸ some data suggest that healthier lifestyles in both men and women may further extend healthy years and narrow the gender gap.¹⁰

A challenge in research on aging and health is defining the phenotype. What is healthy aging and how do we measure it? A recent literature review identified more than 500 studies that examined some aspect of healthy aging,¹¹ but only 28 of these studies had operationalized definitions (using categorical or continuous variables) as an outcome measure. Fewer than half of these 28 were prospective studies. Prospective epidemiological studies with substantial numbers of long-lived participants and phenotypic information useful for gerontological research are rare but essential to shape definitions and identify risk factors for health and survival at older ages.^11- 13

The Honolulu Heart Program/Honolulu Asia Aging Study (HHP/HAAS) has followed a cohort of US men of Japanese ancestry living in Hawaii for 40 years. The HHP/HAAS provides a valuable opportunity to define aging phenotypes and prospectively assess risk factors linked to healthy aging in men.¹⁴ The HHP/HAAS participants have been well characterized in terms of functional status and incidence of chronic age-associated diseases, including cardiovascular disease and dementia.^15- 18 These men provide an important window for understanding what is realistically possible for healthy aging in men.

For this study, we focused on a phenotype of healthy aging referred to by a recent National Institute on Aging expert panel as exceptional survival.¹² This phenotype is characterized by absence of morbidity and absence of physical/cognitive impairment in persons who survive to an advanced age. We focused on the age of 85 years to be consistent with the current definition of oldest old,¹⁹ but since this choice is arbitrary, some of our analyses include survival to ages 75, 80, 85, and 90 years. We examined potential biological, lifestyle, and sociodemographic risk factors present at the baseline assessment in 1965-1968, when the study participants were middle-aged. Our goal was to identify risk factors for healthy survival that are easily measured in both clinical and research settings and may be modifiable for clinical, health policy, and epidemiological purposes.

The HHP is a population-based, prospective study of cardiovascular disease among 8006 Japanese American men (recruited from 9877 men with valid contact information) who were born between 1900 and 1919 and lived on the island of Oahu in 1965.¹⁵ The HAAS is an ongoing study of dementia in the HHP cohort that began in 1991.¹⁷ The HHP cohort recruitment, design, and procedures have been described in detail elsewhere.^20,21 At the time of study enrollment (1965-1968), participants were aged 45 to 68 years (mean age, 54 years). Approximately 12% of these men were born in Japan and 88% were born in the United States. From the commencement of the HHP, information on the development of incident coronary heart disease and stroke, as well as mortality from all causes, has been obtained by monitoring obituaries in local newspapers (English and Japanese) and through surveillance of hospital discharge records.¹⁵ A follow-up survey in the 1991-1993 examination found that only 5 men could not be traced for mortality information.²²

For the purposes of this study, total mortality, physical function, cognitive function, and incidence of 6 major chronic diseases were assessed during 8 follow-up examinations, which were conducted through 2005. Of the 8006 original HHP/HAAS study participants, we excluded 1501 because they either died within 1 year of the study onset or had presence of clinical morbidity at the baseline examination based on self-reported history or clinical findings, which included coronary heart disease, stroke, cancer, diabetes, gastrectomy, chronic lung diseases, or kidney or liver diseases. An additional 685 participants who were missing information on physical function at baseline or follow-up examinations were excluded, leaving a total of 5820 participants in this analysis. Procedures were in accordance with institutional guidelines and approved by the Institutional Review Board of Kuakini Medical Center. Written informed consent was obtained from all study participants or family representatives if participants were unable to provide consent.

A physical examination was performed at baseline, which included height and weight, grip strength, seated blood pressure, and forced expiratory volume in the first second (FEV₁). Levels of total cholesterol, uric acid, glucose (1 hour after a 50-g glucose load), triglycerides, and hematocrit were determined from nonfasting blood samples. Routine urinalysis was also performed. A medical history including lifestyle factors such as smoking status, alcohol consumption, and physical activity²³ was obtained. Information was collected on sociodemographic characteristics including occupation.²⁴

Initially, we classified participants into 1 of 4 phenotypes: (1) nonsurvivors—men who died before a specified age (75, 80, 85, or 90 years); (2) so-called “usual survivors but disabled”—men who survived until the specified age but with physical or cognitive disability and with or without a major chronic disease; (3) usual survivors with major chronic diseases but no disability; and (4) exceptional survivors—men who survived to the specified age without major chronic disease and also without cognitive or physical impairment. Since univariate analyses showed that the distributions of risk factors for the 2 usual survival phenotypes were very similar, these 2 groups were combined in the multivariate analysis.

Chronic diseases of interest in our analysis included coronary heart disease, stroke, cancer (excluding nonmelanoma skin cancer), chronic obstructive pulmonary disease, Parkinson disease, and treated diabetes. These 6 diseases were chosen on the basis of good phenotypic information in the HHP/HAAS cohort and the fact that they are among the most common age-associated chronic diseases. Presence of these diseases was identified by either the HHP/HAAS surveillance program or the HHP/HAAS follow-up examinations to the end of 2005. Screening for cognitive impairment was with the Cognitive Abilities Screening Instrument (score <74)¹⁸ and diagnosed by cognitive function tests and clinical findings using standard Clinical Dementia Rating Scale criteria for all dementia subtypes.¹⁷ Physical impairment was defined as difficulty walking half-mile.

Analysis of covariance was used to compare the baseline risk factors across the survival phenotypes adjusting for age at baseline. Odds ratios for mortality vs survival (for each specified age) and, among survivors, for having at least 1 morbid condition vs being free of these conditions were estimated using logistic regression models. For this analysis, continuous variables were dichotomized as high or low based on conventional cutoff points or median values.

A large number of variables were considered in the analysis. Backward stepwise logistic regression was used to select a subset of variables in the final model. This method starts with all variables in the model and reduces the model 1 variable at a time. This procedure continues until all variables in the model have P values smaller than the preselected level, such as .10.

All P values reported are for 2-tailed tests. A P value of .05 or less was considered statistically significant in univariate analyses and less than .10 was considered statistically significant in stepwise logistic regression. Including variables significant at the .10 level increased the size of the standard errors of the other regression coefficients somewhat but made comparison of the models for the 2 types of outcomes easier. Likelihood ratio tests were used to test for association between a set of independent variables and an outcome variable after adjusting for age. A difference of −2 log likelihood between a model with age plus other independent variables and that of a model with age only is asymptotically distributed as a χ² distribution.^25,26

Our intention was to estimate the probability of overall survival or exceptional (“healthy”) survival based on total number of risk factors. Risk scores based on simple numbers of risk factors are easily understood, can help guide clinical decision making, and may act as a motivator for both clinicians and patients. To assess cumulative effects of multiple risk factors on health outcomes, a survival risk score (SRS) was created. The risk factors used in our SRS were selected from variables significant in univariate analyses by backward stepwise logistic regression, deleting variables from the model that had P>.10.

Separate models were estimated for overall survival to age 85 years and exceptional survival at age 85 years as outcomes. Age at baseline was forced into all models. Overweight, hypertension, and high triglyceride level were defined using specific criteria from national expert panels.^27- 29 Hematocrit and uric acid were dichotomized using the median value. High alcohol intake was dichotomized as 3 or more drinks/d (based on an increased risk of mortality in the HHP/HAAS cohort)³⁰ or less than 3 drinks/d. Smoking was dichotomized as ever or never. Education was dichotomized as high for achievement of graduation from high school or low otherwise.

Variables included in the SRS were overweight at midlife (body mass index ≥25; calculated as weight in kilograms divided by height in meters squared), low grip strength (<39 kg), hypertension (blood pressure ≥140/90 or prescription of an antihypertensive medication), hyperglycemia (≥200 mg/dL [11.1 mmol/L] 1 hour after glucose load), high triglyceride level (≥150 mg/dL [1.70 mmol/L]), high hematocrit (≥45%), high uric acid level (≥5.9 mg/dL), ever smoking, 3 or more alcohol drinks/d, low education (<12 years in school), and not married. The FEV₁ measure was significantly associated with outcome but was not included in the multivariate models because of invalid measurements for 20% of the participants.

The probability of exceptional survival and overall survival from age 55 years to age 75, 80, 85, or 90 years was computed using coefficients and standard errors estimated from multivariate logistic regression, adjusting for baseline age. Dummy variables were created to correspond to levels of SRS (0-5 and ≥6). The statistical software SAS, version 9 (SAS Institute Inc, Cary, NC) was used in the statistical analysis.

A total of 5820 individuals met the inclusion criteria for this study. Of these men, 3369 (58%) died before age 85 years and were classified as nonsurvivors. Seven hundred fifty-eight men with disease but no disability (13%) and 1038 men with disability without regard to disease status (18%) survived to age 85 years and were classified as usual survivors. The remaining 655 men (11%) survived to age 85 years without any of the 6 selected chronic diseases and without cognitive or physical impairment. These men were classified as exceptional survivors.

Table 1 displays the age-adjusted baseline characteristics of the 4 survival phenotypes: exceptional survivors, diseased usual survivors, disabled usual survivors, and nonsurvivors. The distributions of the risk factors were very similar in both usual survival groups. The P values for comparison of these 2 groups were greater than .05 for the vast majority of comparisons. Therefore, we combined these 2 groups to form a single group of usual survivors for multivariate analysis. Having redefined the survival groups as usual, exceptional, and nonsurvivors, we found marked differences across the groups for most risk factors.

Assessment of anthropometric and physiological variables revealed that exceptional survivors had stronger grip strength, suggesting increased physiological reserve and/or higher physical fitness at midlife. Exceptional survivors also tended to be leaner at both young (25 years) and middle (55 years) adulthood. Several hematological and biochemical measures were associated with exceptional survival, including lower levels of serum triglycerides, glucose, and uric acid, important markers of insulin sensitivity.³¹

Apart from age (and excluding FEV₁ because of its large number of missing values), there were 29 baseline variables in total from Table 1. To assess whether the entire set of variables included variables that would be valid for later, more refined logistic regression analyses, we used likelihood ratio χ² tests for the entire set of 19 independent variables. After adjustment for age, the likelihood ratio χ² tests, when tested jointly in multiple logistic regression models, were χ²₁₉ = 319.27 (P<.001) for survival to age 85 years and χ²₁₉ = 114.51 (P<.001) for exceptional survival status (ie, survival to age 85 years free of the designated illnesses or disabilities). Therefore, there is extremely strong evidence that at least some of the original 19 independent variables were truly associated with our outcomes. Although each model as a whole was highly significant, it is obvious that many of the variables were not associated with the outcomes.

Table 2 shows age-adjusted odds ratios for variables that were significantly associated with either usual or exceptional survival in the initial logistic regression analyses. Consistently high ORs for nonsurvival vs overall survival or usual survival vs exceptional survival were evident in overweight and lower-educated men. Marital status was associated with overall survival but not healthy survival.

Associations of particular variables with overall survival and exceptional survival are shown in Table 3, which displays the results of backward stepwise logistic regression that incorporated significant variables from the univariate analyses. It is evident that many common phenotypic measures are associated with both exceptional survival and overall survival. Higher ORs for usual survival vs exceptional survival were evident for all risk factors except marital status. A similar pattern, with the exception of education, was observed for nonsurvival. High blood glucose level, in particular, appears associated with both mortality and survival with morbidity or impairment. Interestingly, overweight at midlife is highly associated with exceptional survival but at best only moderately associated with overall survival; conversely, ever smoking is associated with overall survival but has only a borderline association with exceptional survival.

Figure 1 illustrates the estimated probabilities that a 55-year-old study participant who had no major disease at baseline would survive to ages 75, 80, 85, and 90 years according to cumulative risk factors from the baseline examination. The vertical lines are likelihood ratio–based 95% confidence intervals for the estimated probabilities. For all age outcome categories, the probability of survival declines approximately linearly as the number of risk factors increases. Among men with no risk factors at baseline, the estimated probabilities of surviving to ages 75, 80, 85, and 90 years are 0.89, 0.79, 0.69, and 0.33, respectively. The corresponding probabilities for men with 6 or more risk factors are 0.47, 0.37, 0.22, and 0.06, respectively.

Figure 2 shows the estimated probability of being free of the 6 major chronic diseases and major cognitive or physical impairment among survivors. Although the confidence intervals are wide, there is a fairly regular decline in the probability of being an exceptional survivor as the number of risk factors increases; the results for age 90 years are unclear because of the small number of men who were in such good health at age 90 years. Among men with no baseline risk factors, the estimated probabilities of being among the exceptional survivors at ages 75, 80, and 85 are 0.72, 0.70, and 0.55, respectively; the corresponding values for men with 6 or more risk factors are considerably lower at 0.43, 0.22, and 0.09.

The probability of being alive at a specified age and also being an exceptional survivor at that age, given that the participant was alive, is the product of the 2 individual probabilities; for 55-year-old men with no risk factors at baseline, the probabilities that they would be alive and free of the defined morbidities at ages 75, 80, and 85 are 0.64, 0.55, and 0.38, respectively. The corresponding probabilities for men with 6 or more risk factors are 0.20, 0.08, and 0.02.

Remaining healthy and functional at older ages is an increasingly important public health goal. Why is it that some men live to advanced age in good health? Is it only luck? We hypothesized that healthy survivors within the HHP/HAAS cohort would share similar risk factors in midlife. This study demonstrates that while chance and circumstance are applicable to everyone, there were major differences in risk factors for healthy survival among elderly men that were discernible 40 years prior. Several interesting patterns emerged from these data that appear consistent with recent theories of how aging occurs^32- 34 and that also have important public health implications.

Anthropometric measures from this study, such as grip strength, suggest that it is important to be physically robust in midlife. This is consistent with theories of aging that suggest that better-built organisms last longer and that physiological reserve is an important determinant of survival.

Overall survival and exceptional survival were also more common among men with a lean body habitus in young and middle adulthood. This is a particularly relevant finding because there is increasing debate about whether or how much health risk is posed by overweight and obesity.^35- 37 In addition, several risk factors that reflect insulin resistance were associated with overall survival and healthy survival in this population. This is consistent with previous findings regarding the mortality risks of insulin resistance and current hypotheses regarding insulin as a modifier of the aging process.^34,38- 40

Lifestyle factors have received increasing attention as modifiers of morbidity, but little is known about their potential impact on healthy survival.^41- 44 Not surprisingly, smoking had a major impact on overall survival and exceptional survival that was dose-dependent,⁴² and overconsumption of alcohol was a significant risk factor for death³⁰ and for unhealthy survival. The most powerful sociodemographic attribute of exceptional survival was education. This is a recurring theme in studies of morbidity, in which higher-educated persons appear to fare better with regard to health outcomes than those less educated.⁴³ It would be interesting to study whether those who engaged in further education as adults (ie, “lifelong learners”) fared better than those who did not with regard to health and survival at very old ages. Men who had a marital partner in midlife also survived longer but did not appear to be healthier in very old age.

Cumulative effects of common risk factors, several linked to insulin resistance, were associated with both overall survival and exceptional survival. The SRS, which combined risk factors selected in multivariate models, showed a high probability for both survival and exceptional survival with fewer midlife risk factors. In fact, there was near 60% probability for exceptional survival at oldest-old age (85 years) with no risk factors in midlife, and this number decreased to less than 10% with 6 or more risk factors. Only 11% of men actually met the criteria for exceptional survivorship at oldest-old age in our cohort, suggesting possible room for improvement in an already long-lived population.

This study has several limitations. Since our study population consists of ethnic Japanese men, there may be genetic, sociocultural, or other factors operative in this population that limits generalizability to other populations. There may also be cohort effects that favor particular outcomes in this group of men. In addition, since we excluded men with chronic diseases at baseline, there may be different aging trajectories for men who already possess comorbidities or disability in midlife, and factors associated with these comorbidities or disabilities may be less modifiable with age. In addition, since this study focused on men, the results may not be entirely applicable to women. Further comparisons between this study and other studies of healthy survival might be of interest.^14,44

Strengths of the study include that it is among the largest, longest, and most complete follow-up studies on aging men. Our analyses were age-standardized to 55 years, which represents an age at which many men are facing significant midlife decisions about risk factor modification, including potential changes to physical activity, smoking, and other health habits. Therefore, information about the potential utility of such interventions for adding healthy years, rather than just additional years, could be highly motivating for men approaching retirement age. Disease incidence was determined by ongoing surveillance and participants were examined on multiple occasions, and since men with prevalent disease at baseline were excluded, this is a true incidence study. In this manner, we minimized potential effects of treatment for baseline diseases and changes in lifestyle on aging outcomes.

In summary, we have identified several potentially important risk factors for healthy survival in a large group of middle-aged men. These risk factors can be easily measured in clinical settings and are, for the most part, modifiable. This study suggests that common approaches that target multiple risk factors simultaneously, such as avoidance of smoking or hypertension, and approaches that enhance insulin sensitivity, such as maintaining a lean body weight, may improve the probability of better health at older ages. This may be especially important for men, few of whom survive to oldest-old age.