Skip to content
Back to search results

Healthy Life Expectancy Decomposition

Introduced by
Wilma J. Nusselder & Caspar W. N. Looman; Daniel F. Sullivan
Year
2004
Type
Demographic decomposition pipeline for a health-expectancy difference
Family
Demography / health-expectancy decomposition
Font size

What does it do?

Healthy (or disability-free) life expectancy combines how long people live with how much of that life is spent in good health, and differences in it between groups or over time reflect two distinct forces: changes in mortality and changes in the prevalence of disability.

Healthy-life-expectancy decomposition separates these forces. Building on the Sullivan method — which weights life-table person-years by the age-specific share of life lived without disability — Wilma Nusselder and Caspar Looman's 2004 method splits the gap in health expectancy between two populations into an additive mortality component and a disability component for each age, and can further attribute each to specific causes. This resolves the central interpretive ambiguity of health expectancy: a population can have higher healthy life expectancy because its people die later, because they are less disabled at each age, or both, and only a decomposition can tell which.

The intuition behind the method

Imagine two populations with the same headline healthy life expectancy. They could have reached it very differently: one might have low mortality but high disability (people live long but spend many years impaired), the other high mortality but low disability (shorter lives, but healthier while they last). If you only know healthy life expectancy you cannot distinguish these worlds, yet they call for opposite policies — one needs to prevent disability, the other to prevent death. The decomposition pulls the two levers apart. It asks, age band by age band, how much of the difference comes from people surviving longer (the mortality effect) versus from a smaller fraction of survivors being disabled at that age (the disability effect). The two effects add up exactly to the total gap, so you get a clean ledger: this many years of the difference are about living longer, those many are about living healthier.

History

Daniel Sullivan's 1971 HSMHA Health Reports paper introduced the single index combining mortality and morbidity that became health expectancy, computed by weighting life-table person-years with age-specific disability prevalence. As health expectancy spread through population-health surveillance, the need arose to explain differences in it, not just measure them. Wilma Nusselder and Caspar Looman provided the standard solution in their 2004 Demography paper, extending Arriaga-style life-expectancy decomposition to the Sullivan framework and partitioning differences in health expectancy into additive mortality and disability components, with a further breakdown by cause. Their method is now the reference approach in the European Health Expectancy Monitoring tradition and is cross-linked here to the Sullivan-method and healthy-life-expectancy pages.

When to use it

Use healthy-life-expectancy decomposition when two populations (groups, sexes, regions, or time points) differ in disability-free or self-rated-healthy life expectancy and you need to know whether the difference reflects mortality, morbidity, or both, and through which causes. It requires, for each population, an abridged life table and cross-sectional age-specific prevalence of the chosen health state, and the cause breakdown additionally needs cause-specific mortality and ideally cause-specific disability data. It is the natural extension when a plain life-expectancy decomposition is insufficient because the question concerns years of healthy life rather than total years. It inherits the cross-sectional assumptions of the Sullivan method and is descriptive rather than causal; for purely mortality-based gaps, the simpler life-expectancy decomposition suffices.

Limitations and considerations

  • Inherits the Sullivan method's cross-sectional assumption: prevalence at a point in time stands in for a cohort's experience, which can bias estimates when disability trends are changing.
  • Sensitive to how the health state (disability, activity limitation, self-rated health) is defined and measured, and to comparability of that definition across populations.
  • Requires reliable age-specific prevalence data, which are often noisier and less standardized than mortality data.
  • Descriptive, not causal, and the mortality/disability split depends on the choice of an averaged reference schedule, so apportionment at the margins is not unique.

Source conditions do not replace independent scientific review or validation by an analysis engine.

Strengths

  • Resolves the key ambiguity of health expectancy by separating the mortality (living longer) and disability (living healthier) channels of a difference.
  • Additive and exact: the mortality and disability components sum to the observed gap in healthy life expectancy, enabling a complete accounting.
  • Builds directly on the widely used Sullivan method and standard abridged life tables, so it uses data agencies already collect.
  • Extends to a cause-of-death and cause-of-disability breakdown, supporting public-health priority setting between preventing death and preventing disability.

Applications

  • Determining whether sex or socioeconomic gaps in disability-free life expectancy stem from differences in mortality or in disability.
  • Tracking whether gains in healthy life expectancy over time reflect compression or expansion of morbidity.
  • Attributing differences in health expectancy to specific causes of death and types of disability for priority setting.
  • Comparing health expectancy across countries to locate the mortality and morbidity sources of an advantage or disadvantage.
  • Evaluating population-health policies by decomposing changes in healthy life expectancy into survival and morbidity channels.

How it works

How it works

1

The pipeline begins by computing Sullivan-based healthy life expectancy for each population. Sullivan's method takes the person-years lived in each age interval from an abridged life table and multiplies them by the complement of the age-specific disability prevalence, so that each interval contributes only the years lived in good health. Cumulating these health-weighted person-years from age x upward and dividing by the survivors at x gives healthy life expectancy at that age. This requires two ingredients per population: a life table (mortality) and cross-sectional age-specific prevalence of disability or ill health.

Formula
HLE_x = \frac{1}{l_x}\sum_{a \ge x} \bigl(1 - \pi_a\bigr)\,{}_{n}L_a
2

The target of the decomposition is the difference in healthy life expectancy between the two populations. Nusselder and Looman's central result is that this total difference can be written as the sum, over age intervals, of two additive pieces: a mortality contribution and a disability contribution. Because the components sum exactly to the observed gap, the method gives a complete and reconcilable accounting, and the analyst can read off how much of the difference is attributable to survival versus to morbidity at each age.

Formula
\Delta HLE = HLE^{(2)} - HLE^{(1)} = \sum_{x}\bigl(C_x^{\text{mort}} + C_x^{\text{disab}}\bigr)
3

The mortality component isolates the part of the health-expectancy gap that arises from differences in survival, holding the disability prevalence schedule fixed (at an average of the two populations). It extends the life-expectancy decomposition logic to a health-weighted table: where one population has lower mortality in an age band, more survivors accumulate health-weighted person-years there and beyond, contributing positively to healthy life expectancy. This is the 'living longer' channel, expressed in years of healthy life.

Formula
C_x^{\text{mort}} = \Bigl[\text{effect of } {}_{n}q_x^{(2)} - {}_{n}q_x^{(1)} \text{ on } HLE \mid \pi \text{ held at average}\Bigr]
4

The disability component isolates the part arising from differences in the prevalence of disability, holding survival fixed. Where one population has a lower share of disabled people at a given age, the same person-years count more toward healthy life, raising health expectancy. Operationally it weights the difference in age-specific disability prevalence by the (averaged) person-years exposed at that age. This is the 'living healthier' channel, and contrasting it with the mortality component reveals whether a population's advantage is driven by longevity or by lower morbidity.

Formula
C_x^{\text{disab}} = -\,\bar{L}_x\,\bigl(\pi_x^{(2)} - \pi_x^{(1)}\bigr)\big/ l_0
5

Finally, each age-specific component can be partitioned by cause. The mortality contribution at an age is split across causes of death in proportion to their contribution to the mortality difference there, exactly as in life-expectancy decomposition; the disability contribution can be split across causes or types of disability where cause-specific prevalence data exist. Summing over ages yields, for each cause, its net effect on the health-expectancy gap through mortality and through disability — the detailed output that makes the method valuable for public-health priority setting.

Formula
C_x^{\text{mort}} = \sum_i C_x^{\text{mort},i},\qquad C_x^{\text{disab}} = \sum_j C_x^{\text{disab},j}

Common pitfalls

  • Using non-comparable disability definitions or survey instruments across the two populations, which contaminates the disability component.
  • Interpreting the disability component as a change in incidence when prevalence reflects both onset and survival with disability.
  • Decomposing health expectancies built on mismatched life-table age intervals, which breaks the additive reconciliation with the total gap.
  • Reading a large mortality component as 'good news' without noting it may add disabled years if disability prevalence is high at the ages where survival improved.

Frequently asked questions

How is this different from the Sullivan method and from healthy life expectancy?

The Sullivan method and the healthy-life-expectancy page describe how to compute a single number — the years of life expected in good health — by weighting life-table person-years with disability prevalence. This page describes how to decompose a difference between two such numbers into a mortality part and a disability part, by age and cause. In short, Sullivan measures; this method explains. You compute Sullivan health expectancy for each population first, then apply the Nusselder-Looman decomposition to the gap.

Why separate the mortality and disability components at all?

Because the same difference in healthy life expectancy can arise from opposite underlying dynamics with opposite policy implications. A higher health expectancy driven by lower mortality means people live longer; driven by lower disability it means people live healthier; and falling mortality can even add disabled years. Nusselder and Looman's decomposition makes these channels explicit and additive, so analysts can distinguish compression of morbidity from mere postponement of death and target prevention of death or prevention of disability accordingly.

What are the main data and assumption caveats?

The method inherits Sullivan's cross-sectional assumption — point-in-time prevalence stands in for cohort experience — so it can mislead when disability is trending. It is also sensitive to how the health state is defined and to the comparability of disability measurement across the populations being compared. Prevalence data are typically noisier than mortality data, and the mortality/disability split uses an averaged reference schedule, so the exact apportionment is not unique. Results are descriptive and should be read as an accounting of where differences arise, not as causal effects.

Related methods

Sources

  1. Nusselder, W. J., & Looman, C. W. N. (2004). Decomposition of differences in health expectancy by cause. Demography, 41(2), 315-334.

    Open DOI reference
    Copied.
  2. Sullivan, D. F. (1971). A single index of mortality and morbidity. HSMHA Health Reports, 86(4), 347-354.

    Open publication reference
    Copied.