Changes in health status, symptoms, and functioning during aging defy simple description. Despite the inevitability of death, no orchestrated or predictable decrements in health or types of sickness uniformly precede it, and individuals vary widely in how and how successfully they age. Certain groups may experience aging and health differently from others. The most obvious instance is gender: older men and women have been observed to have different lifespans [
Successful aging is a multidimensional construct, which initially centered on absence of disease and objective physical and social functioning [
There is utility in systematically scrutinizing the changes in different aspects of health status that occur with advancing age among men and women. First, evidence from research can help to counter ageist or sexist stereotypes about health during the aging process, which may influence social expectations, clinical care, and patient decision making [
We sought to characterize a variety of measures of successful aging among a group of older adults, with particular attention to the differences between men and women of different ages. To explore changes in health systematically, and to allow age and gender comparisons, we analyzed transition probabilities for 12 different health variables which capture different aspects of psychological, physical, cognitive, and functional status. We hypothesized, for these 12 different domains of successful aging, that (1) the prevalence of health decreases because both the probability of remaining healthy and the probability of returning to a healthy state decline with increasing age, and (2) the prevalence and transition probabilities are different for men and women.
Data came from the Cardiovascular Health Study (CHS), a population-based longitudinal study of risk factors for heart disease and stroke in 5888 adults aged 65 and older at baseline [
The 12 variables used in this study were measures of health based on self-report or observation. They were selected in order to capture various aspects of psychological, physical, cognitive, and functional status. All were measured annually. Each value was dichotomized into “Healthy” and “Sick”, as shown in Table
12 Measures of successful aging.
Category | Abbreviation | Question | Definition of healthy |
---|---|---|---|
Not hospitalized | HOSP | “Did you stay overnight in the hospital in the last 6 months?” | No report of being hospitalized |
No bed days | BED | “During the past two weeks, how many days have you stayed |
No days in bed reported |
Life satisfaction | SPL | “How satisfied are you with the meaning and purpose of your |
Score of 1 to 4 |
Life as a whole | FLW | “How do you feel about life as a whole?” (1: delighted; 3: |
Score of 1–3 |
Not depressed | DEP | 10 questions of the center for epidemiologic studies short |
Score < 10, out of a possible 30 points |
No limitations in activities of daily living | ADL | “Do you have any difficulty performing this activity?” from a list of walking, transferring, eating, dressing, bathing, or toileting | No difficulties reported |
No limitations in independent activities of daily living | IADL | “Do you have any difficulty performing this activity?” |
No difficulties reported |
Intact extremity strength | EXSTR | “Do you have any difficulty with this activity” from a list |
No difficulty reported |
Self-rated health | SRH | “How would you rate your health in general: excellent, |
Excellent, very good, or good self-reported health |
Intact cognition | COG | Modified minimental state examination [ |
Score above 89 |
Ability to ambulate | TWLK | Timed 15 foot walk | Less than 10 seconds |
Frequent ambulation | BLK | “During the last week, how many city blocks did you walk?” | >4 blocks per day, on average |
In order to simplify comparisons, age was divided into three categories—65–74, 75–84, and 85–94—in accordance with the common definitions of “young old,” “old old,” and “oldest old”. Persons could contribute data to more than one age category, depending on their age at the start of each transition.
Missing data were imputed, after a transformation to recode death as zero health, by interpolating over time between existing data points for each person. Any data that remained missing at the end of a sequence were extrapolated as an average of the last observed value and of transformed self-rated health (which was measured every 6 months and is thus well characterized) [
In order to represent and compare prevalence and incidence of healthy and sick states in various domains, and to account for death, we used a transition probability approach. This technique has been used to examine other changes in health among older adults, including self-rated health [
Transition probability model, showing the probability of remaining in a state or moving to another state during specific time intervals (such as one year).
When evaluated at two time points, there are thus six possible transitions among these states: remaining healthy (P(HtoH)), becoming sick (P(HtoS)), remaining sick (P(StoS)), becoming healthy (P(StoH)), dying from a state of health (P(HtoD)), and dying from a state of sickness (P(StoD)). Persons move among those states with certain probabilities, which may vary by age, gender, or other characteristics. The equilibrium—or steady state—prevalence of a system can be calculated directly from the transition probabilities [
The prevalence of the healthy state of each variable in each wave was calculated as the percentage of living persons who were healthy. The one-year probabilities of transitioning from state to state were estimated from crosstabulation of data collected one-year apart. All cases where a beginning state and a starting state were available were used, for all 5888 participants. General patterns by age and gender groups were described. We estimated one-year transition probabilities for participants starting in each health state (sick or healthy). These transition probabilities, shown in Figure
The level of significance for individual comparisons was set at
Analyses were conducted in Stata (StataCorp, College Station, Texas, version 11.2).
45,297 transition pairs (starting and ending state, separated by one year) from the 5888 participants were analyzed. During the study, 1684 participants died. For the whole sample, 13.5% of observations were missing and imputed, with some variability across the 12 domains of health. In the year prior to death, 7% of all observations were missing, which were part of the imputed fraction. The median number of imputed observations per participant was one.
As an illustration of the analytic approach, the prevalences and selected transition probabilities for ADL are shown in Figure
Prevalence, maintenance, and recovery for ADL Health by age and sex. The prevalence estimates are the proportion with no ADL difficulties. The transition probability estimates are the likelihoods of remaining healthy (P(HtoH)) and of returning to health (P(HtoS)) over one-year intervals.
The first two lines of Table
Prevalence of a healthy state among men and women, with health defined separately for each domain. Bolded entries represent significantly higher prevalence of health in women (left columns) or men (right columns). The differences between groups based on age are described in the text.
Female | Male | |||||
---|---|---|---|---|---|---|
65–74 | 75–84 | 85–94 | 65–74 | 75–84 | 85–94 | |
Number of observations | 12261 | 12433 | 2183 | 7801 | 8867 | 1752 |
Mean age | 71.0 | 78.6 | 87.5 | 71.1 | 78.6 | 87.7 |
HOSP: no hospital days |
|
|
|
88.0 | 85.6 | 81.7 |
BED: no bed days | 94.3 | 92.3 | 89.3 |
|
|
|
SPL: Satisfied with purpose of life | 75.7 | 69.8 | 59.6 |
|
|
|
DEP: not depressed | 80.2 | 73.8 | 64.3 |
|
|
|
ADL: no ADL difficulties | 86.3 | 76.1 | 56.7 |
|
|
|
FLW: feel life is worthwhile | 94.4 | 91.0 | 83.6 |
|
91.7 | 85.6 |
EXSTR: good extremity strength | 67.6 | 57.6 | 42.1 |
|
|
|
SRH: high self-rated health | 79.0 | 70.8 | 60.7 | 80.0 |
|
|
TWLK: walk 10 feet < 10 seconds | 64.3 | 44.5 | 16.9 |
|
|
|
IADL: no IADL difficulties | 71.8 | 58.7 | 39.5 |
|
|
|
COG: 3 MSE > 90 |
|
|
|
67.2 | 53.1 | 25.7 |
BLK: walked 4+ blocks per day | 33.3 | 20.9 | 8.5 |
|
|
|
Prevalence by Age and Gender. All of the prevalence values in Table
Table
One-year transition probabilities for those starting in a healthy state for 12 different variables. Bolded entries indicate a significantly healthier transition (more likely remaining healthy, less likely remaining sick, or less likely dying) among women compared to men (top half) or men compared to women (bottom half). The differences between groups based on age are described in the text.
Age 65–74 | Age 75–84 | Age 85–94 | |||||||
---|---|---|---|---|---|---|---|---|---|
P(HtoH) | P(HtoS) | P(HtoD) | P(HtoH) | P(HtoS) | P(HtoD) | P(HtoH) | P(HtoS) | P(HtoD) | |
Female | |||||||||
| |||||||||
HOSP |
|
|
|
|
|
|
|
|
|
BED | 0.95 | 0.05 |
|
0.92 | 0.06 |
|
0.86 | 0.08 |
|
SPL | 0.85 | 0.15 |
|
0.80 | 0.18 |
|
0.70 | 0.27 |
|
DEP | 0.88 | 0.11 |
|
0.84 | 0.14 |
|
0.77 | 0.19 |
|
ADL | 0.90 | 0.09 |
|
0.85 | 0.14 |
|
0.72 | 0.24 |
|
FLW |
|
0.04 |
|
0.92 | 0.06 |
|
|
0.11 |
|
EXSTR | 0.83 | 0.17 |
|
0.77 | 0.21 |
|
0.66 | 0.30 |
|
SRH |
|
|
|
0.85 | 0.37 |
|
0.73 |
|
|
TWLK | 0.83 | 0.17 |
|
0.72 | 0.27 |
|
0.54 | 0.45 |
|
IADL | 0.85 | 0.15 |
|
0.77 | 0.21 |
|
0.61 | 0.35 | 0.05 |
COG |
|
|
|
|
|
|
0.67 | 0.30 | 0.04 |
BLK | 0.65 | 0.01 |
|
0.55 |
|
|
0.49 | 0.03 | 0.03 |
| |||||||||
Male | |||||||||
| |||||||||
HOSP | 0.88 | 0.10 | 0.02 | 0.85 | 0.12 | 0.03 | 0.76 | 0.15 | 0.09 |
BED | 0.95 |
|
0.02 | 0.92 |
|
0.04 | 0.84 | 0.07 | 0.10 |
SPL |
|
|
0.02 |
|
|
0.03 | 0.71 |
|
0.08 |
DEP |
|
|
0.02 |
|
|
0.03 | 0.76 |
|
0.07 |
ADL |
|
|
0.02 |
|
|
0.03 | 0.75 |
|
0.07 |
FLW | 0.95 |
|
0.02 | 0.92 | 0.05 | 0.03 | 0.82 | 0.10 | 0.08 |
EXSTR |
|
|
0.02 |
|
|
0.03 | 0.74 |
|
0.06 |
SRH | 0.89 | 0.35 | 0.01 | 0.84 |
|
0.03 | 0.75 | 0.45 | 0.06 |
TWLK |
|
|
0.02 |
|
|
0.02 | 0.61 |
|
0.05 |
IADL |
|
|
0.01 |
|
|
0.02 | 0.64 |
|
0.06 |
COG | 0.83 | 0.15 | 0.02 | 0.78 | 0.20 | 0.03 | 0.62 | 0.33 | 0.05 |
BLK |
|
0.02 | 0.01 |
|
0.02 | 0.02 | 0.54 | 0.04 | 0.03 |
In all three comparisons (remaining healthy, becoming sick, and dying from a healthy state), there was a significant decline with advancing age for almost all the domains of health. Men were more likely than women to remain healthy and less likely to become sick in the majority of comparisons, and men were more likely to die.
Table
One-year transition probabilities for those starting in a sick state for 12 different variables. Bolded entries indicate a significantly healthier transition (more likely becoming healthy, less likely remaining sick, or less likely dying) among women compared to men (top half) or men compared to women (bottom half). The differences between groups based on age are described in the text.
Age 65–74 | Age 75–84 | Age 85–94 | |||||||
---|---|---|---|---|---|---|---|---|---|
P(StoH) | P(StoS) | P(StoD) | P(StoH) | P(StoS) | P(StoD) | P(StoH) | P(StoS) | P(StoD) | |
Female | |||||||||
| |||||||||
HOSP | 0.68 | 0.26 | 0.06 |
|
0.29 |
|
0.54 | 0.27 |
|
BED | 0.55 | 0.37 |
|
|
0.41 |
|
0.34 | 0.37 |
|
SPL | 0.37 | 0.60 |
|
0.30 | 0.64 |
|
0.24 | 0.61 |
|
DEP | 0.35 | 0.62 |
|
0.29 | 0.64 |
|
0.20 | 0.63 |
|
ADL |
|
0.59 |
|
0.25 | 0.67 |
|
0.15 | 0.70 |
|
FLW |
|
0.54 |
|
|
0.56 |
|
0.18 | 0.53 |
|
EXSTR | 0.31 | 0.66 |
|
0.23 | 0.72 |
|
0.14 | 0.74 |
|
SRH | 0.31 | 0.64 |
|
0.25 | 0.67 |
|
|
0.61 |
|
TWLK | 0.29 | 0.68 |
|
0.16 | 0.80 |
|
0.06 | 0.84 |
|
IADL | 0.30 | 0.67 |
|
0.22 | 0.73 |
|
0.16 | 0.73 |
|
COG | 0.27 | 0.70 |
|
0.17 | 0.78 |
|
0.07 | 0.83 |
|
BLK | 0.14 | 0.84 |
|
0.08 | 0.88 |
|
0.03 | 0.88 |
|
| |||||||||
Male | |||||||||
| |||||||||
HOSP | 0.63 | 0.29 | 0.08 | 0.56 | 0.28 | 0.16 | 0.49 | 0.23 | 0.28 |
BED | 0.49 | 0.35 | 0.15 | 0.33 | 0.36 | 0.32 | 0.26 | 0.33 | 0.41 |
SPL |
|
|
0.06 | 0.32 |
|
0.12 | 0.24 |
|
0.22 |
DEP | 0.37 |
|
0.08 | 0.27 |
|
0.16 | 0.18 | 0.58 | 0.25 |
ADL | 0.31 | 0.59 | 0.10 | 0.23 |
|
0.17 | 0.13 |
|
0.24 |
FLW | 0.29 | 0.55 | 0.16 | 0.20 | 0.52 | 0.28 | 0.15 | 0.46 | 0.39 |
EXSTR |
|
|
0.07 |
|
|
0.14 |
|
|
0.24 |
SRH | 0.32 |
|
0.08 | 0.25 |
|
0.13 | 0.18 | 0.59 | 0.24 |
TWLK |
|
|
0.05 |
|
|
0.09 |
|
|
0.16 |
IADL |
|
|
0.08 |
|
|
0.12 | 0.17 |
|
0.19 |
COG | 0.27 | 0.69 | 0.04 | 0.18 |
|
0.08 | 0.07 |
|
0.15 |
BLK |
|
|
0.04 |
|
|
0.07 |
|
|
0.15 |
In almost all domains of health, the probability of recovery from a sick state declined significantly with age, while the probability of dying from a sick state increased with age. Women were significantly more likely than men to recover from a sick state in six of 36 total groups; men were more likely to recover in 12. Men were less likely to remain in a sick state than women in 23 of the 36 groups; in no cases were women significantly less likely than men to remain sick. For every health variable, men were significantly more likely to die from a state of sickness than women were.
This analysis, unlike previous approaches that have focused on one or a few domains of health, examined the prevalence of and transitions in 12 measures of successful aging among a cohort of older adults. The 12 health-related variables, while related, were diverse and included measures of physical and mental health, quality of life, and health behaviors. Some were self-reported and some were measured through objective tests. Some were subjective single item questions while others were based on structured responses. Despite these differences, most of the measures of successful aging performed in very similar ways with regard to their associations with age and gender.
Overall, the trends in prevalence suggest both high initial rates of health and consistent incremental declines in health with advancing age. First, the prevalence of emotional and functional health shown in Table
Men were observed to have a higher prevalence of a healthy state except for hospitalization and cognitive status. In Table
We found several key differences in aging based on gender. Men died more often than women from a state of either sickness or health, remained more healthy than women while alive, and were more likely to recover from being sick. These differences were not characterized by similar transitions offset by a period of years, as might happen if men's health trajectories were simply premature or accelerated versions of women'strajectories. Comparing the adjacent age categories for men and women (for instance, 65–7-year-old men compared to 75–84-year-old women) shows that younger men's transitions were more similar to older women'stransitions than younger women's transitions to older men's transitions for 58 (60%) of the 96 possible comparisons. This is seen graphically for ADL status in Figure
These results demonstrate qualitative differences in health transitions between men and women during aging, not just that men “age faster”, as has been suggested [
The observed differences in incidence and prevalence of health between men and women encourage speculation about their etiology in human gender dimorphism and argue for fundamental differences between how men and women age. Historical analyses suggest that the environmental pressures of infectious disease and resource availability have caused women to live longer [
First, about one-tenth of the data were missing and had to be imputed. The ascertainment of death, however, was essentially complete. The approach we used for imputation, using interpolation and extrapolation, may have minimized changes by assuming that the missing health status was mainly consistent with the data before and after the missing measurement and declined close to death. There was no difference in missingness between men and women, and it is unlikely that this imputation method would intensify group differences. Second, the individual significance tests that are reported here should be considered as descriptive rather than definitive, due to the large number of comparisons that were made in this analysis. Third, the category of “healthy” in the various outcome measures should not be interpreted literally, since the cutoffs for healthy and sick states were assigned by categorizing each variable individually, and not by cross-validating them with other metrics for subjective or objective health. Most of the cutoffs have face validity as markers of health. Fourth, in the interests of generating straightforward descriptive results, we did not adjust for other patient-level health or sociodemographic characteristics which might confound the associations between age, gender, and health. Investigating these could help understand how men and women differ in the aging process.
Transition probabilities among various states of health during aging, and separately for men and women, are rarely reported in ways that can be compared. We calculated the gender-specific transition probabilities for 12 health-related variables across different domains of health and believe that these have utility for organizing future research and for characterizing the changes that happen during aging. In general, the 12 variables all behaved similarly. Recuperation from sickness declined with age, but still occurred frequently. Men and women experienced different types of change in health over time, with men showing more health and less sickness, but greater likelihood of dying. Men did not simply age faster compared to women. They exhibited a more “square” pattern of health status over time, with dropoff at younger ages than women. There is no simple explanation for the differences observed between men and women; they may partly relate to women's health-related behaviors or perceptions of health, or to more biologically fundamental gender dimorphisms. Future study can help to clarify how health is constructed and changes in different groups during aging.
The research reported in this paper was supported by contracts N01-HC-85239 and N01-HC-85079 through N01-HC-85086, N01-HC-35129, N01 HC-15103, N01 HC-55222, N01-HC-75150, and N01-HC-45133 and Grant HL080295 from the National Heart, Lung, and Blood Institute (NHLBI), with additional contribution from the National Institute of Neurological Disorders and Stroke (NINDS). Additional support was provided through AG-023629, AG-15928, AG-20098, and AG-027058 from the National Institute on Aging (NIA). A full list of principal CHS investigators and institutions can be found at