Transitions among health states, using 12 different measures of health
7-8-2011 recup_44.doc
Paula Diehr and Stephen Thielke
Abstract
Introduction
A population may be conceptualized as comprising 3 states: healthy, sick, and (over time) dead. Persons move among those states with certain probabilities, which may vary by age and gender. There are many ways of defining the states, and transition probabilities are seldom available.
Methods
From a longitudinal study of older adults, we analyzed twelve different variables: self-rated health, ADL, IADL, depression (CESD), cognition (3MSE), time to walk 10 feet, number of days spent in bed, number of blocks walked, extremity strength, recent hospitalizations, feeling that life is worthwhile, and satisfaction with the purpose of life. We dichotomized responses for each variable into “healthy” or “sick”, and estimated transition probabilities among the states. We examined patterns in the prevalences and the probabilities by age, sex and variable type.
Findings
All 12 health variables had similar behavior. There were monotonic age effects, in the expected directions, which were non-linear except for the probability of resilience or of remaining sick. Men had a higher death rate than women, but men also had a higher healthy prevalence for most variables. Women were more likely to become and remain sick, and were less likely to recover on variables measuring aspects of physical health.
Discussion
The age patterns were expected. Women were usually favored for mortality, and men for the other variables. The only measures that favored women, COG and HOSP, were determined externally rather than self-reported. Perhaps men are clueless (stoic) while women are whiney (in touch with their feelings/bodies/etc.). Men are more rectangular.
Conclusion
Older adults were likely to be healthy and to maintain their health or to recover from the sick state, no matter how health was measured. There were substantial differences in the gender patterns of prevalence and transition probabilities for the differing health measures. By most measures, men were healthier, in spite of having a higher mortality rate. Women may self-report worse health than is appropriate.
(315 words)
Transitions among health states, using 12 different measures of health
1 Introduction
A population may be conceptualized as comprising 3 states, healthy, sick, or (if followed over time) dead. For example, for Activities of Daily Living (ADL), “healthy” may be defined as having no ADL difficulties, and “sick” may be defined as having one or more difficulties. Persons move among those states with certain probabilities, which may vary by age and gender. Figure 1 shows the 6 one-year transition probabilities for ADL (from a dataset described later), by age, for women. In order from the top are the probability of staying healthy from one year to the next, P(H to H); the probability of staying sick, P(S to S); the probability of getting sick, P(H to S); the probability of dying from the sick state, P(S to D); and the lowest is the probability of dying from the healthy state, P(H to D). Not surprisingly, P(H to H) and P(S to H) decline with age, while the other probabilities increase with age. At the oldest ages, P(S to H) is lower than P(S to D), meaning that a sick person is then more likely to die than to recover. Men had similar patterns (not shown).
[Figure 1 about here]
Figure 2 shows the prevalence of the healthy state (healthy prevalence) and two of the transition probabilities. The two heaviest lines represent the healthy prevalence (proportion of the living who had no ADL difficulties), with a solid line for males and a dotted line for females. The healthy prevalence is higher for men than for women. It declines over time, and the slope becomes steeper with age. The lowermost two lines represent the probability of recovering from the sick state to be in the healthy state one year later, P(S to H). The probability is initially near 0.4, declines approximately linearly with age, and is higher for women than for men. The two topmost lines in the graph represent the probability of staying in the healthy state, P(H to H). This probability is initially about 0.9, but declines with age and is higher for men than for women. Figure 1 and Figure 2 suggest some hypotheses of interest about healthy prevalence and transition probabilities.
[Figure 2 about here]
“Healthy” may be operationalized in many other ways, such as having no IADL difficulties, or having fewer than 10 points on a scale of depressive symptoms. Transition probabilities are often used in life-table calculations and other analyses that model longitudinal health. The probabilities themselves, however, are rarely published. Transition probabilities based on ADL, IADL, or self-rated health have been published. [1] [2] [3] [4] [5] [6] [7] Transition probabilities are also available for disability, [8] [9] [10] depression, [11] exhaustion, [12] and body mass index.[2] To explore transition probabilities further, we will present and describe the transition probabilities for 12 different health variables commonly used in studies of older adults.
[Stephen to write something about sex and old age here?]
We hypothesize that healthy prevalence and transition probabilities will become less favorable with older age; that the changes will not be linear in age, but steeper at older ages; that transition probabilities will be different for men and women; that P(S to S) may not be monotone in age , based on relationships seen for self-rated health. [nsick paper 3] ; and that age and sex patterns for probabilities will depend on the particular definition of “Healthy” or “Sick”.
2 Methods
2.1 Data
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.[13] Participants were recruited from a random sample of Medicare eligibles in four U.S. communities, and extensive data were collected during annual clinic visits and telephone calls. The original cohort of 5201 participants, recruited in about 1990, had up to ten annual clinic examinations. A second cohort of 687 African Americans, enrolled in about 1993, had up to seven annual examinations. Follow-up was virtually complete for surviving participants.[14]
The twelve variables used in this study, common descriptors of aging that were measured annually, are shown in Table 1. Each value was dichotomized into “Healthy” and “Sick”, using standard thresholds where available, or choosing new thresholds that ensured sufficient data for estimating transition probabilities. Definitions of “healthy” are as follows: not being hospitalized in the previous year (HOSP); no days spent in bed in the previous two weeks (BED); a score of 1 to 4 on a 10-point scale rating satisfaction with the purpose of life (SPL); a Center for Epidemiologic Studies Short Depression score < 10 (DEP);[15] [16] no difficulties with activities of daily living - walking, transferring, eating, dressing, bathing, or toileting (ADL); a score of 1 to 3 on a 6-point scale rating whether life was worthwhile; no problems with extremity strength - lifting, reaching, or gripping (EXSTR); being in excellent, very good, or good self-reported health (EVGG); able to walk 15 feet in less than 10 seconds (measured) (TWLK); no difficulties with instrumental activities of daily living—heavy or light housework, shopping, meal preparation, money management, or telephoning (IADL); having a Modified Mini Mental State Examination score above 89 (COG);[17] and walking more than 4 blocks per day, on average (BLK). Missing data were imputed, as explained in Appendix A.
Age was divided into three categories 65-74, 76-84, and 85-94, in accordance with the common definitions of “young old”, “old old”, and “oldest old”. These groups are sometimes referred to by their midpoints, 70, 80, and 90. Persons could contribute data to more than one age category. The unit of analysis was the transition pair, defined as two measures of a variable, one year apart, for the same person.
2.2 Analysis
The healthy prevalence of each variable was calculated as the percent 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, combining data from all years. General patterns were described. The associations of the prevalences and probabilities with age and sex were tested using cross-sectional time series logistic regression (XTLOGIT in Stata), which accounts for persons having contributed multiple observations to the prevalence and transition data.
3 Findings
3.1 Prevalence
The first two lines of Table 1 show the sample size (number of transition pairs) and mean age, by age and sex. Mean age was similar for men and women. The following 12 lines show the healthy prevalence of each variable. For example, on line 3 (HOSP), 91.3% of the women aged 65-74 were “healthy”, defined as “having no hospital days”. For men in the same state, healthy prevalence was 88.0%. Over the three age groups, women’s prevalence for HOSP declined from 91.3% to 87.8% to 84.9%. Not surprisingly, all of the prevalences in Table 1 decline with age. There was usually a significantly steeper (non-linear) decline from age 80 to 90 than from age 70 to 80. This non-linearity was statistically significant for EXSTR, TWLK, COG, and BLK (men and women) and BED and IADL (men only).
The bolded entries in Table 1 show the situations where women had a higher healthy prevalence than men. This occurred only for COG and HOSP. The XTLOGIT analysis of prevalence for individual variables confirmed that women were significantly healthier than men only for HOSP and COG. (See also appendix table 1).
[Table 1 about here]
3.2 Transition probabilities
3.2.1 Transition probabilities for Sick persons
Table 2 shows the transition probabilities for persons who were initially in the sick state. For example, P(S to H) for the youngest women is .677 when the health measure is HOSP, and the corresponding P(S to S)= .262 and P(S to D) = .061. The three probabilities sum to 1.0. Columns 4-6 show those probabilities for age 75-84, and columns 7-9 show the probabilities for age 85-94. The lower half of the table shows the same quantities for men. The bolded entries in the top half indicate probabilities of resilience or death that were favorable to women (more likely to recover, or less likely to die).
[Table 2 about here]
3.2.1.1 P(S to D)
P(S to D) always increased monotonically with age. As suggested above in Figure 1, in the oldest age group, especially for men, P(S to D) is often greater than P(S to H). Women were always less likely to die than men, no matter how the health states were defined.
3.2.1.2 P(S to H)
P(S to H) always declined with age. The trend was fairly linear, but there was significantly more decline after age 80 (non-linearity) for TWLK, COG and BLK for women, and for EXSTR, TWLK, and COG for men.
Women were more likely than men to recover from the sick state for about half the variables. Based on the XTLOGIT analysis, P(S to H) was significantly higher for women than for men in at least one age category for HOSP, BED, DEP, ADL, FLW, and EVGG. Men had significantly higher P(S to H) in at least one age group for EXSTR, TWLK, and BLK. (See also appendix table 2a).
3.2.1.3 P(S to S)
We also analyzed the probability of staying sick, P(S to S). (See appendix table 2b). The age relationships for P(S to S) were complex. In half (12) of the comparisons there was no significant age relationships. In the other 12, age 70 was always significantly less than 80, meaning that P(S to S) increased from 70 to 80. However, compared to age 80, P(S to S) at age 90 was significantly lower for 2 variables, significantly higher 3 times, and not significantly different for 7 variables. This lack of monotonicity is not easy to summarize.
There was a strong gender effect, with women significantly more likely to stay sick then men in most cases. However, there was not a significant gender effect for HOSP, BED, and FLW, nor for ADL and COG at age 70.
3.2.2 Transitions probabilities for healthy persons
Table 3 shows the transition probabilities for persons who were initially in the healthy state. For example, P(H to H) for the youngest women is .917 when the health measure is HOSP (column 1), and the corresponding P(H to S)= .075 and P(H to D) = .008 (columns 2 and 3). Columns 4-6 show those probabilities for age 75-84, and 7-9 show the probabilities for age 85-94. The lower half of the table shows the corresponding values for men. The shaded entries represent probabilities that favored women (higher probability of remaining healthy or lower probability of becoming sick or dying).
[Table 3 about here]
3.2.2.1 P(H to D)
P(H to D) increased with age, was always higher for men, no matter how the healthy state was defined.
3.2.2.2 P(H to H)
P(H to H) declined with age. The decline usually accelerated significantly with older age, but was not significantly non-linear for HOSP and TWLK (women) and for ADL, EVGG, TWLK, and BLK (men).
Women generally had less favorable P(H to H) than men. The XTLOGIT analysis found significant effects favoring women in at least one age category for HOSP, FLW, and COG. Significant effects favoring men were found in at least one age category for DEP, ADL, EXSTR, TWLK, IADL, and BLK (See also appendix table 3A).