Considerations for data production > Healthy Canadians >

37-HLT: Health adjusted life expectancy (HALE)

37a-HLT: Health adjusted life expectancy (HALE) for overall population

Definition
Rationale and notes for interpretation
Technical specifications
Data availability
Considerations for indicator quality and comparability
Responsibility to produce the data

Definition

Health Adjusted Life Expectancy (HALE) is an indicator of overall population health.  It combines measures of both age- and sex-specific health status, and age- and sex-specific mortality into a single statistic. HALE represents the number of expected years of life equivalent to years lived in full health, based on the average experience in a population.  In this sense, HALE is not only a measure of quantity of life but also a measure of quality of life.

Top of Page

Rationale and notes for interpretation

Canadians have been experiencing continuing increases in life expectancy for many decades. However, with the increasing prevalence of chronic disease, there has been an international debate as to whether or not these added years of life expectancy are years spent in good or poor health. By examining trends in HALE in conjunctions with trends in life expectancy (LE), it is possible to assess whether Canadians' increasing life expectancy is associated with a "compression or an expansion of morbidity". In other words, if HALE increases more over time than LE, we can then conclude that added years of life expectancy are indeed more often years in better health. This example is one illustration of the use and interpretation of HALE.

As with life expectancy, HALE is a standardized statistical indicator. It is not the number of full health equivalent years a particular newborn (or person currently age 65) can actually expect to live. The reason is that mortality rates and levels of health status only for the observation period (e.g. 2001) are used, and these are averages for the entire population. Historically, mortality rates in Canada have been falling, so that the mortality rates individuals are likely to face in future years as they age may be lower. Canada does not yet have consistent data over a sufficiently long period to know what the trends in health status have been, or are likely to be in the future. Moreover, individuals' circumstances vary so that, for example, if they had become chronically ill at an early age, their particular health-adjusted life expectancy would be less.

This indicator should be presented by sex since HALE of women and men differs so much.

Coefficients of variation, confidence intervals (both of which are provided by Statistics Canada), and tests of significance must be analyzed before differences between provinces can be interpreted as real (statistically significant).

Top of Page

Technical specifications

The following outline the sources used as well as the calculation of Coefficient of variation for HALE:

Data sources used

• 1996 Enumeration Areas by Income Tercile File
• 2000-2001 Abridged Life Tables by Income Tercile
• Canadian Community Health Survey, Cycle 1.1 (2000-2001)
• National Population Health Survey, Institutional File, Cycle 2 (1996-1997 cross-sectional file)
• 2001 Census Institutional Counts

1996 Enumeration Area by Income Tercile File

The enumeration area (EA) by income tercile file assigned each EA to a tercile based on income data from the 1996 Census. For each EA, the average household income was calculated and then assigned to a tercile. The first tercile contains EAs with the lowest incomes, the second tercile contains middle income EAs, and the highest income EAs are in the third tercile.

Wilkins et al (2002)¹ classified 1996 Census EAs into income terciles based on income per single-person equivalent, which is a household size-adjusted average household income (pre-tax, post-transfer) at the EA level. Because of different costs of living, the income terciles were derived separately for each Census Metropolitan Area (CMA) or Census Agglomeration (CA), as well as for rural areas within each province.

Life Tables

A special abridged life table for Canada and the provinces was created based on 2000 and 2001 death data for each income tercile and by age group, sex, and province. Income terciles were assigned by the ecological approach using the work of Wilkins et al, where deaths as of 1995-97 were linked to 1996 Census EA terciles based on postal code of place of residence. The ratios of deaths by age group, sex, and province in 1995-97 were used to assign deaths in 2000 and 2001 to income terciles. This was done in the absence of postal codes tied to the latest deaths which would have made it possible to assign income terciles to 2001 geographic units (dissemination areas).

The tercile assigned deaths of 1995-97 excluded the institutional component of the population since the ecological approach to assign income levels was considered inappropriate for the population living in institutions. There were approximately 5% of deaths overall that were unassigned to terciles (mostly associated with the highest age groups) due to this exclusion of institutions. The 2000-2001 ratios were formed by taking the proportion of deaths in each tercile over the sum of deaths in all terciles. These ratios were then used to assign all deaths of 2000-2001, to account for the whole population (institutionalized component included). The distribution of deaths of institutionalized residents across terciles is therefore evenly distributed, instead of excluded.

Canadian Community Health Survey

The CCHS was used for calculating the average health utility index (HUI) by age groups, sex, province, and income tercile. As well, household population counts were also collected from the CCHS.

In order to calculate HUI and population counts by income tercile, respondents' enumeration areas were assigned to income terciles based on the 1996 EA by Income Tercile File mentioned above. While income data are collected by the survey, there is no way in which the survey income data could be linked directly to deaths. Therefore, the ecological approach for assigning respondents to terciles was applied to be consistent with the life table terciles.

Cycle 1.1 was used to be consistent with the 2000-2001 abridged life table as well as the institutionalized population count from the 2001 Census.

Persons under age fifteen were excluded from the calculations. The CCHS only collects data for respondents aged 12 and over. Persons 12 to 14 had to be excluded to match the age groupings in the abridged life tables. All respondents under age 15 were given a HUI score of 1.00.

National Population Health Survey

The 1996-1997 NPHS Institutional Component was used to produce the HUI for institutional residents since they were not part of the CCHS sampling frame. The average HUI was calculated by age groups (under 65 and 65+), sex, and region (Atlantic Provinces, Quebec, Ontario, the Prairie Provinces, and British Columbia). Because of the small sample size, especially among younger respondents, the HUI was only calculated for two broad age groups. The NPHS does not collect data at the provincial level, thus the regional grouping.

The ecological approach was considered inappropriate for the institutional population since their place of residence has more to do with health needs than choice of neighbourhood. Therefore, individual respondents could not be matched to an EA tercile and HUI could not be calculated by tercile for each respondent.

In order to assign terciles to persons living in institutions in a manner consistent with the assignment of terciles for household residents, the NPHS longitudinal file was analyzed. Respondents who lived in the community during the 1994 NPHS but moved into an institution in any subsequent cycle were coded to an EA tercile based on the postal code of their household in 1994. As a result, 39.1% of institutional residents were placed in the first tercile, 37.8% in the second tercile, and 23.2% in the third tercile. The above proportions were only used for assigning institutional population counts to terciles. The HUI was held constant across terciles for institutionalized residents.

In order to match the age groupings of the CCHS and life table, respondents under the age of 15 living in institutions were excluded and given a HUI value of 1.00.

Because of the small sample size, HUIs by region for the institutionalized population cannot be released.

Census Institutional Counts

A number of different data sources for the institutional counts were considered. The NPHS Institutional file could not be used because of different sampling frames between some of the provinces. As well, the NPHS would only provide counts regionally not provincially. The Residential Care Facility Survey could not be used because the age groupings did not match the age groupings of the abridged life table. As a result, the Census was judged to be the preferred data source for institutional population counts. In order to match the sampling frame of the NPHS (used for the HUI), only persons living in nursing homes, residences for seniors (persons less than 65 were excluded), and treatment centres and institutions for persons with a disability were included in the counts. Persons living in smaller treatment centres (group homes, for example) with less than 12 residents are included in the CCHS sampling frame.

The proportion of the population living in institutions or households for each age group, sex, province, and tercile were calculated in order to weigh the institutional and non-institutional HUIs.

Computation of the precision for the Health adjusted life expectancy (HALE):

This appendix explains how the different measures of precision were computed for the HALE. The primary objective is to obtain HALE standard errors in order to derive coefficients of variation (CVs) and confidence intervals (CIs).

HALE itself is computed using two sources of data; first mortality data is used to obtain the life expectancy part of the equation, then survey data is used to derive the health status-adjusted part. Both of these sources bring some sort of variability in the computation of the HALE. Therefore, when computing standard errors for HALE, the variability from both sources must be accounted for.

Mathers (1991)² presents the details of standard error calculations in the context of Disability free life expectancy (DFLE) which is arithmetically closely related to HALE. Equation C.14 from Appendix C of Mathers was used as the starting point for deriving the variance formulae for HALE. The equation was adapted and consists of the following:

where,
i = index representing the age group
w = total number of age groups used for the derivation of HALE (w=20 in our case)
a = specific age group for which the HALE is computed. Values of a=0 (HALE at birth) and a=15 (HALE at 65 yrs old) were used.
l = number of survivors
LL = number of Life years lived in this age group
n = length of the age interval (most age intervals used were 5 years)
HUI = global health utility index derived from both household and institution data sources
HALE = health adjusted life expectancy
f = fraction of age interval lived by individuals who die in the interval. This was calculated using l and LL as follows:

The variances were therefore computed using this formula for each sub-population of interest, which were based mainly on the province * sex * income tercile variables.

The standard error can be obtained by taking the square root of the variance, while the CV for a specific estimate of HALE can be derived by taking the ratio between the standard error of the estimate and the estimate itself. Finally, the boundaries of the CI are obtained by adding and subtracting 1.96 times the standard error to the HALE estimate.

Details of computation

All parameters used in the variance equation presented above were already all defined for the calculation of the HALE itself. Only the variance of the HUI was left to compute in order to be able to compute the variance of HALE. The computation is examined in the next section below.

Variance of the HUI

The HUI used in the calculation of the HALE variance is computed using two survey data sources; the CCHS Cycle 2.1 survey data for the portion of the population living in households, and the NPHS Cycle 2 Institution survey data for the population that is institutionalized. In fact, the HUI is defined as:

where,

is the HUI computed specifically for age group i for the population living in households, using CCHS Cycle 2.1 data

is the proportion of the whole population (households + health institutions) that is living in households.

is the HUI computed specifically for age group i for the population living in health institutions, using NPHS Cycle 2 Institution data

is the proportion of the whole population (households + health institutions) that is living in health institutions.

Although technically both proportions and are associated with some sampling error, for simplicity of calculation, they were both considered as constant when deriving the variance formula of the HUI. The variance equation was therefore defined as:

where,
is the variance computed for the household population
is the variance computed for the institutionalized population

Both survey data sources used are designed according to two different sampling plans, which must be reflected in the HUI variances calculations. Since CCHS uses a multistage stratified design, was computed using the bootstrap technique. This resampling technique is the method adopted by the survey in order to compute accurate sampling error measures (for more details about the method, consult the CCHS public-use microdata files (PUMF) users guides). As for, it was obtained using an exact variance formula since the NPHS Institution survey relies on a simpler design, that is, a simple stratified design.

Finally, a few technical points should be noted:

• For the institution population, since the HUI value for younger age groups could not be computed (because these age groups are not covered by the survey), a value of 1 was set for the HUI of these age groups. Consequently, for variance calculation purposes, the variance of these cases was set to 0.
• For a small number of cases, the variance of the HUI could not be computed since the sample available for the specific age group of the sub-population was 1. For those cases, a variance was imputed from a neighbouring group.
• Similarly, for some cases related to small sub-populations, the f parameter could not be estimated since the number of survivors (l) had not changed between the two age groups implicated in the calculation (resulting in a division by 0). In these cases, the f parameter was imputed by the mean of similar sub-populations.

Using the computed CV, HALE estimates were finally validated against the quality assurance guidelines. The guidelines stipulate that an estimate with a CV between 16.5% and 33.3% is marginal, and should be identified as such in the publication. As well, if the CV is greater than 33.3%, the estimate is considered to be of poor quality and should not be considered for publication. For further details about the quality assurance guidelines, consult any CCHS public-use microdata file users guide.

Top of Page

Data availability

• HALE will be calculated at birth and at age 65, by province and sex only.
• Data are not available for the territories

Top of Page

Considerations for indicator quality and comparability

HALE will be calculated at birth and age 65. However, HALE at birth will be based on data for those aged 15 and over.

Because of the small sample size for the institutional component of the NPHS the average HUI for institutional residents will be calculated for people under 65 and people aged 65 and over. As well, the NPHS provides only regional data so the average HUI for institutional residents will be calculated for the Atlantic Provinces, Quebec, Ontario, the Prairie Provinces, and British Columbia.

Top of Page

Responsibility to produce the data

Statistics Canada

Footnotes

¹ Wilkins, Russell, Edward Ng, Jean-Marie Berthelot, and Francine Mayer. (2002). "Provincial Differences in Disability-Free Life Expectancy by Neighbourhood Income and Education in Canada, 1996". Technical Report to the Performance Indicators Reporting Committee (PIRC) of the Federal-Provincial-Territorial Conference of Deputy Ministers of Health.

² Mathers, C. (1991). Health Expectancies in Australia, 1981 and 1988. Canberra : Australian Institute of Health/Australian Government Publishing Service.