Table 1 Characteristic of the 26 included reports. Abbreviations explained in footnote

Educational differences

 Hayward & Lichter 1998 [37]^a

45–59 years old men;

N = 5 020;

1966–1983; U.S.

The National Longitudinal Survey of Older Men;

Self-reports.

In the labour force, retirement, disability, died.

Increment–decrement model with transition rates derived from multivariate hazard models for each type of transition. Cohort data. No age censoring.

WLE at age 45 and 55 by education

The estimates of time in other three states by education are also available.

 Millimet et al. 2003 [38]²

17–86 years old;

N = 200 916;

1992–2000; U.S.

Bereau of Labour Statistics and Current Population Surveys 1990–2000;

Self-reports.

Employed, unemployed, inactive.

Increment–decrement model using relative frequencies (transitional probabilities calculated for entire sample) and econometric model (transitional probabilities calculated for subgroups). Period data. No age censoring.

WLE for the 21–70 age range by gender, education and initial status (employed, unemployment and inactivity).

 Karlsson et al. 2009 [39]

0–90 years old;

N = 9 865, 78% (N = 7 735) of the 26–60 years old; 1991–2004;

U.K.

First fourteen waves of the British Household

Panel Survey (BHPS);

Self-reports.

Employment, health status, retirement, died.

Simulation of life trajectories based on estimated hazard rates for mortality, employment, and retirement, utilizing econometric approach. Period data.

Life expectancy (LE), healthy life expectancy (HLE) and WLE at age 50 by gender and combination of employment, education, health status at initial state.

 Millimet et al. 2010 [40]^b

17–71 years old;

1992–2001;

U.S.

Bereau of Labour Statistics and Current Population Surveys 1990–2000;

Self-reports.

Employed, unemployed, inactive.

Increment–decrement model using relative frequencies (transitional probabilities calculated for entire sample) and econometric model (transitional probabilities calculated for subgroups).

Period data. No age censoring.

WLE for the 30–60 age range by gender, initial status (employed, unemployment and inactivity), education, race, marital status

 Skoog et al. 2011 [41]³

16–75 + years old;

N = 471 722;

2005–2009;

U.S.

Current Population Survey;

Self-reports.

Economically active, inactive, dead.

Markov increment-decrement worklife expectancy model.

Period data. Censoring at age of 76.

WLE expectancy for the 16–75 age range conditional on initial state (active/ inactive) by gender and education.

 Nurminen 2012 [42]

15–74 years old;

2000–2010;

Finland

Finnish Labour Force Survey; Self-reports.

Employed, unemployed, economically inactive, dead.

Multistate life table method. data. Censoring at age of 65.

Partial WLE, unemployment and inactivity expectancy for the 15–60 age range (5-year interval) by gender and education.

 Krueger & Slesnick 2014 [43]^c

18–70 years old non-institutional population;

N = 635 555;

2009–2013;

U.S.

Current Population Survey (CPS); Self-reports.

In the labour force (active, in market work), active, in non-market work, inactive, died.

Markov increment-decrement worklife expectancy model.

Period data. Censoring at age of 71.

WLE (years active in the market) for the 25–70 age range, conditional on initial state (active in the labour market work/active in non-market work) by gender and education.

 Özer 2014 [44]

16 + years old women;

N = 11 762; 2009–2010;

Turkey

Income and Living Conditions Surveys,

TURKSTAT; Self-reports.

Being active (employed or unemployed), being inactive (out of labor force).

Multistate working life tables based on increment–decrement model. Period data. Censoring at age of 75.

WLE for the 20–70 age range, by education, conditional on initial state (active or inactive).

The expected time being inactive by education, conditional on initial state is also shown.

 Loichinger & Weber 2016 [1]

15–74 years old;

28 EU countries;

1983–2013

Eurostat database based on European Labor Force Survey and the Human Mortality Database (based on registers); Self-reports.

Economically active, inactive.

Sullivan method, with prevalence rates aggregated by 5-year age-groups. Period data. Censoring at age of 75.

WLE (remaining active life expectancy) at age 50 in 2009 by country, gender and education for 11 EU countries.

 Dudel & Myrskylä 2017 [34]^d

50–99 years old;

N = 30 254;

1992–2011;

U.S.

The Health and Retirement Study (HRS); Self-reports.

Employed, retired (retired or 70 + years old and out of the labor force, out of the labor force (under 70 yrs. old and not working) or unemployed, died.

Multistate Markov model, annual transitions.

WLE was weighted but not conditioned on the initial state

Period data. No age censoring.

WLE at age 50 (years and %) by time period (1993–1997, 1998–2002, 2003–2007, and 2008–2011); and by gender/race/ education

 Stanek & Requena 2019 [46]

50 + years old;

N ~ 180 000 individuals;

2000–2014;

Spain

An Economically Active Population Survey and life tables for the Spanish population (the Spanish National Statistics;

Self-reports.

Employed, unemployed, retired, being economically inactive (other than retirement and unemployment).

Sullivan method.

Period data. No age censoring.

WLE at age 50 by gender and education in 2012

The expected time in other three states by gender and education are also shown.

 van der Noordt et al. 2019 [24]

55–65 years old;

N = 1074;

1992–1996; 2002–2006;

2012–2016;

The Netherlands

Three population-based cohorts; Self-reports.

In the workforce without disability, in the workforce with disability, out of the workforce.

Multistate Markov model based on annual transitions.

Cohort data. Censoring at age of 68.

Total WLE, WLE with disability and WLE without disability at age 58 by education.

 Robroek et al. 2020 [6]

16–66 years old;

N = 4 999 947;

2001–2014;

The Netherlands

Social statistical database (Statistics Netherlands); monthly information on the main income components,

Mortality statistics.

Registers.

Paid employment, disability benefits, unemployment, no income through paid employment, early retirement, being a student, emigration, died.

Multistate model biased on monthly transitions.

Period data. Censored at age of 66.

WLE between ages 16–66, 30–66 and 50–66 by gender and education.

The expected time in other three states by gender and education are also shown.

 Lorenti et al. 2020 [47]^d

50 + years old;

N = 19 062;

2008–2014;

U.S.

Health and Retirement Study (HRS) - a longitudinal survey;

Self-reports.

Working, disabled, not working (includes inactive, unemployed, retired), died or age 100 years

Multistate Markov model based on annual transitions.

Period data. No age censoring.

Working and disability LE at age 50 by gender, childhood disadvantage and education.

 Weber & Loichinger

2020 [2]

50–69 years old;

2017;

26 European countries

Eurostat database based on the European Labour Force Survey (EU-LFS);

Period life tables for ages 0–85 + by sex and country from Eurostat;

Data from SHARE Survey wave 1 (2004) - wave 7 (2017); Self-reports.

Economically active, inactive.

Sullivan method.

Period data. Censoring at age of 60 (for 50–59 years old group) and 70 (for 60–69 years old group).

The number of years a person is expected to be economically active and generally, physically, and cognitively healthy (called working, healthy, physical, and

cognitive life expectancy, respectively) between age 50–59 and 60–69 by country and gender.

 Nexø et al. 2021 [49]

18–65 years old;

N = 115 118 (with diabetes) and N = 663 656 (without diabetes); 2000–2016; Denmark

Several linked Danish national registers.

Work, long-term sickness absence, temporary absence, unemployment, disability pension, died.

Multistate Cox proportional hazard model with weekly transitions between the states.

Period data. Censoring at age of 65.

WLE for the 30–60 age range by gender, education and cohabitant status.

The expected time in other states by gender, education and cohabitant status is also shown.

 Tetzlaff et al. 2022 [52]

18–69 years old;

N = 1 193 061;

2011–2013; Germany

The statutory health insurance provider (AOK Lower Saxony, AOKN);

Registers.

In the labour force (employed and unemployed) and not in the labour force (other).

Markov increment-decrement worklife expectancy model using two states life table.

Period data. Censoring at age of 70.

Years in the labour market at age 18 and 50 by gender and education

 Schram et al. 2022 [30]

50–66 years old;

N = 11 800; 2010, 2015; The Netherlands

The Dutch

longitudinal Study on Transitions in Employment, Ability and Motivation (STREAM).

Registers.

Years in paid employment, involuntary work exit (unemployment, disability benefits), voluntary work exit (economic inactivity or (early) retirement) died.

Multistate model biased on monthly transitions.

Period data. Censoring at age of 66.

WLE between ages 50–66 by gender and education.

The expected time in other states by gender and education is also shown.

Occupational class differences

 Dudel et al. 2018 [4]

15–99 years old individuals who pay taxes or receive social security benefits;

N = 1 272 695:

2004–2013;

Spain

Spanish Continuous Working Life Sample,

register-based social security benefits.

Eurostat life tables based on the EU Labor Force Survey; Registers.

Employed, unemployed, retired, inactive, died or age 99 years.

Multistate Markov model based on annual transitions.

Period data. No age censoring.

Remaining LE in employment at age 15, 2004 to 2012 by gender and occupational class.

Remaining LE at age 15 in other labour force state by gender, occupational class and year is also shown.

 Leinonen et al. 2018 [5]

50–69 years old;

N = 5 170 689 unweighted person-years; 1988–2007;

Finland

Administrative registers with annual information through the end of 2012;

Registers.

Employment, unemployed, disability retirement, other forms of early retirement, statutory retirement, outside of the labor force.

Sullivan method.

Period and cohort data. No age censoring

Working and retirement LE at age 50 by gender and social class.

 Lorenti et al. 2019 [45]

15–99 years old;

N = 880 000;

2003–2013;

Italy

Nationally representative Longitudinal Sample INPS (National Institute of Social Security); Registers.

Employed, unemployed, retired, economically inactive, died.

Multistate Markov model based on annual transitions.

Period data. No age censoring

Remaining LE in employment (WLE) at age 15 for years from 2003 to 2013 by gender and occupational class.

 Schram et al. 2021 [50]

50–63 years old;

N = 415 105;

2004–2014;

Finland

Nationally representative register-based cohort;

Registers.

Work, time-restricted work disability, unemployment, economic inactivity, disability retirement, retirement, died.

Multistate Cox regression model based on daily transitions.

Cohort data. Censoring at age of 63.

WLE between ages 50–63 by gender and occupational class.

The expected time in other states by gender and occupational class is also shown.

 Lynch et al. 2022 [51]^e

50–102 years old;

N = 15 284; 2002–2013;

U.K.

The English Longitudinal Study of Ageing, data from six cross-sectional samples; Self-reports.

Healthy and in work, not healthy and in work, healthy and not in work, not health and not in work, died.

Interpolated Markov chain multistate modelling of panel data using multinomial logistic regression based on annual transitions.

Period data. No age censoring.

WLE (healthy and in work, and not healthy and in work), healthy life expectancies (healthy and in work and healthy not in work, HLE), at age 50. HWLE for the 50–75 age range by occupational class.

Educational and occupational class differences

 Hayward & Grady 1990 [36]^a

55 years old men in the labour force;

N = 2 816; 1966–1983;

U.S.

The National Longitudinal Survey of Older Men;

Self-reports.

In the labour force, retirement, disability, died.

Increment–decrement model with transition rates derived from multivariate hazard models for each type of transition. Cohort data. No age censoring.

WLE at age 55 by education and occupation.

 Parker et al. 2020 [21]^e

50–102 years old;

N = 15 284;

2002–2013;

U.K

The English Longitudinal Study of Ageing, data from six waves of cross-sectional samples;

Self-reports.

Healthy and in work, not healthy and in work, healthy and not in work, not health and not in work died.

Interpolated Markov chain multistate modelling of panel data using multinomial logistic regression based on annual transitions. Period data. No age censoring.

LE, HLE (healthy and in work and healthy not in work), WLE (healthy and in work, and not healthy and in work) and healthy WLE (HWLE) at age 50 by education and occupational class.

 Dudel et al. 2021 [48]

55–64 years old (born 1941 and 1955);

1996–2019;

Germany

The annual German Microcensus survey;

Self-reports

Employed, unemployed retired/inactive.

Modified Sullivan’s method.

Cohort data. Censoring at age of 65.

WLE adjusted for working time (measured in full-time equivalent years) for the 55 to 64 age range by region, gender, education and occupational class.