Page 88 - IJPS-3-1

Modeling trajectories of long-term care needs at old age: A Japanese-Swedish comparison

was prepared using individual, longitudinal SNAC data for the population part of the
Kungsholmen study (SNAC-K) from the baseline study (2001–2004) and the three-
year follow-up (2004–2007) (Lagergren, 2004). The basic design of the population
part is to survey a sample of persons in the age groups 60, 66, 72, 78, 81, 84, 87, 90,
93, and 96 years. The younger people (aged 60–72 years) are re-examined after six
years, the older (78 years and above) after three years. The data collection is very
broad and involves medical examinations, questionnaires, and interviews concerning
living circumstances, style, disability, provision of informal and formal care, and
physical and mental performance tests. From these data, IADL dependency and
ADL dependency were defined in the same way as in the Japanese study. The IADL
variables of preparing meals, purchasing household items or medication, doing laundry
(not in Japan), doing light household work, and taking a bus or train were used, as
were the ADL variables of taking a bath or shower, dressing, eating, standing up from
a bed or chair, going to the bathroom, and using the toilet.
In the SNAC study, home-related LTC provisions are registered in terms of number
of home-help services provided per week. Four levels were used in the original
analysis (no home help, <2 hours/week, 2–12 hours/week, and >12 hours/week).
For institutional care there were two levels: sheltered housing and nursing home.
However, for reasons of comparability with the Japanese results, where no division
into levels of home-related LTC was possible, the three levels of home help will be
summarized in the presentation of results. Likewise, the two levels of institutional care
will be summarized. Using the SNAC data, a dataset was prepared containing 1,233
observations of men and women from the age groups 78, 81, 84, 87, and 90 years at
baseline (2001–2004) and the same persons three years later at first follow-up (2004–
2007). The data thus contained five age groups for each gender, and for each age-group
fifteen states at baseline (three levels of dependency x five levels of LTC = 15 states).
In the follow-up, death is added as a state.
Transition probabilities were calculated by a series of logistic regression analyses
in both cases. The calculations referred to transitions in three-year time steps. The
first calculation step involved three-year probability of death using age group,
gender, initial dependency (three levels), and initial level of LTC (three levels in the
case of Japan and five levels in the Swedish case) as independent variables. In the
next step, three-year transition probabilities between states of dependency for the
survivors were calculated using multinomial logistic regression analysis and age
group, gender, initial dependency, and initial level of LTC as independent variables.
In the last step, transition probabilities between levels of LTC were calculated, again
using multinomial logistic regression analysis and age group, gender, initial as well as
updated dependency, and initial level of LTC as independent variables. Tables showing
the results of the regression analyses are found in the Appendix.
In this way, transition probabilities between the states, including death, were
calculated for men and women from 78 to 81 years. Then, using the same regression
results, transition probabilities were calculated from 81 to 84, from 84 to 87, from 87
to 90, and finally from 90 to 93 years. By successive multiplication of the resulting
stepwise transition, probability matrices corresponding accumulated matrices for
transition of states from 78 years to 81, 84, 87, 90, and 93 years were calculated.
The calculations were made separately for men and women and were based on the
Markov assumption of independency between time steps (see Section 4: Discussion).
For Sweden, the results were calibrated to agree with national distributions of death,
dependency, and LTC provision in 2003. For Japan, the calibration was made to agree
with the dependency and LTC distributions in wave 4 as no national distributions
were available. Calibration was done by age group and gender for both dependency
and LTC-level distributions. In both cases, the technique was to adjust the intercept
coefficients in the regression analysis in order to achieve a certain distribution of the
target variable. This means that all relations between variables remain the same in
terms of odds ratios—only the levels are adjusted.

82 International Journal of Population Studies 2017, Volume 3, Issue 1