Association Between Subjective Well-being and Living Longer Without Disability or Illness

Key Points Question Longitudinal observational studies indicate that greater subjective well-being is associated with longer survival, but are these additional years spent in good health? Findings In this survey study of 9761 older men and women from the English Longitudinal Study of Ageing who were followed up for a maximum of 10 years, higher affective well-being was associated not only with longer life expectancy at older ages, but also with a greater proportion of additional years in good health without chronic disease or disability. Meaning Subjective well-being is associated with healthier aging as well as greater longevity, but it is not yet known whether programs to enhance well-being will extend healthy life expectancy.


eAppendix. Computation of healthy life expectancy
The computation of health expectancy using the Sullivan method is usually applied to cross-sectional data, and requires life tables and information on age specific proportions of the population in healthy or unhealthy stages. These proportions are prevalence measures of the actual and current health status of a real population and are used to divide years lived in the life table population. With panel data reliable estimates of life table inputs cannot be obtained. The multi-state life table (MSLT) model has been developed to analyse stochastic processes that involve multiple and recurrent events (typical of longitudinal data), in order to estimate expected duration in various states. MSLT method uses a set of transition schedules from healthy, unhealthy and to death estimated using longitudinal data.
The possible transitions among the health states for disability are represented in the following chart And the transitions among health states for chronic condition in the following, note that recovery is not allowed (from unhealthy to healthy).
The advantages of multistate life table method are: it is based on incidence measures representing current health transitions; it allows movement in both directions between all surviving health states; it allows death rates to differ by health state so it takes into account the different mortality profiles by health status. The estimation of transition schedules is very important and can be done using logistic regression, multinomial logistic regression or hazard regression.

Death
Healthy Unhealthy

Healthy Unhealthy
We used the Stochastic Population Analysis for Complex Events (SPACE) 1 program in SAS 9.2 to estimate MSLT functions. There are two main components to this program: the data component which prepares the input datasets and the statistical component in which transition probabilities and the MSLT functions and their variances are estimated. Specifically, during the data component age-specific transition probabilities for all possible transitions are estimated from the data using multinomial logistic regression conditional on age, sex, well-being factors, wealth and living with a partner. The package also allows the use of survival analysis, however, given that for disability and chronic conditions we do not have exact dates of diagnosis, while we have exact dates of mortality, this would produce and imbalance, therefore multinomial logistic regression is recommended 1 . Health expectancies from the age of 50 and over are then calculated based on these estimated transition probabilities using a stochastic (micro-simulation) approach. By using micro-simulation it is possible to simulate the life paths of the members of the population in order to derive several summary statistics of the population dynamics (an example is available on page 136 of reference 4). The program generates individual trajectories for a simulated cohort of 100,000 persons with distributions of covariates at the starting point based on the observed study-specific prevalence by five year age group and sex. Variability for these MSLT estimates (variances, standard errors and corresponding confidence intervals) are computed using a bootstrap method with 500 replicates for the whole analysis process (multinomial analysis and simulation steps). This method takes account of attrition from the study under the missing at random assumption 1 . More information can be found at http://www.cdc.gov/nchs/data_access/space.htm.