1
Sex di e ences in heal hy li e expec ancy among nonagena ians: A mul is a e su i al
model using da a om he Vi ali y 90+ s udy
Emiel O. Hoogendijk, PhD1, Maaike an de Noo d , MSc1, B egje D. Onwu eaka-Philipsen, PhD2,
Do ly J.H. Deeg, PhD1, Ma ijn Huisman, PhD1,3, Linda En o h, PhD4, & Ma ja Jylhä, MD, PhD4
1Depa men o Epidemiology & Bios a is ics, Ams e dam Public Heal h esea ch ins i u e,
Ams e dam UMC – loca ion VU Uni e si y Medical Cen e , Ams e dam, he Ne he lands
2Depa men o Public and Occupa ional Heal h, Ams e dam Public Heal h esea ch ins i u e,
Ams e dam UMC – loca ion VU Uni e si y Medical Cen e , Ams e dam, he Ne he lands
3Depa men o Sociology, V ije Uni e si ei , Ams e dam, he Ne he lands
4Facul y o Social Sciences (Heal h Sciences), Tampe e Uni e si y and Ge on ology Resea ch Cen e ,
Finland
Co esponding au ho : Emiel O. Hoogendijk, PhD, Depa men o Epidemiology & Bios a is ics,
Ams e dam Public Heal h esea ch ins i u e, Ams e dam UMC – loca ion VU Uni e si y Medical
Cen e , P.O. Box 7057, 1007MB Ams e dam, he Ne he lands. Email: e.hoogendijk@ umc.nl,
elephone: +31204443146.
This is he pos p in e sion o he a icle, which has been published in
Expe imen al Ge on ology, 2019, Volume 116, 80-85.
h ps://doi.o g/10.1016/j.exge .2018.12.015
2
ABSTRACT
Objec i es: Li le is known abou sex di e ences in heal hy li e expec ancy among he oldes old, he
as es g owing segmen o he olde popula ion. This s udy examines sex di e ences in o al, heal hy
and unheal hy li e expec ancy among nonagena ians.
Me hods: Longi udinal da a o 884 olde adul s aged 90 and o e pa icipa ing in he Vi ali y 90+
s udy (Tampe e, Finland) we e used, including 2,501 obse a ions (heal h o dea h s a es) om 5
measu emen wa es be ween 2001 and 2014. Using he MSM and ELECT packages in R, mul is a e
su i al models we e pe o med o es ima e he ansi ion p obabili ies o olde adul s h ough he
di e en heal h s a es and o calcula e li e expec ancies. The analyses we e done sepa a ely o wo
heal h indica o s (disabili y and mul imo bidi y) o see whe he pa e ns we e consis en .
Resul s: Women had highe o al li e expec ancies han men (abou 8 mon hs), bu also highe
unheal hy li e expec ancies. Men had a highe disabili y- ee li e expec ancy be ween he age o 90
and 95 compa ed o women. Fo mul imo bidi y, no sex di e ences in heal hy li e expec ancy we e
ound.
Conclusions: This s udy showed ha he male- emale heal h-su i al pa adox emains a e y old
age. Women aged 90+ li e longe han men, and spend mo e ime in poo heal h.
Keywo ds: nonagena ians; disabili y; li e expec ancy; mul is a e modeling
3
1. In oduc ion
The male- emale heal h-su i al pa adox is well-es ablished in many coun ies wo ldwide [1-4]. This
is he phenomenon ha highe li e expec ancy is accompanied by highe a es o poo heal h in women
compa ed o men. This is seen as a pa adox, since poo heal h is usually associa ed wi h lowe
su i al. The ac ha men die a younge ages han women, despi e hei be e heal h, has been
a ibu ed o a ious biological, beha io al, social and con ex ual ac o s [5].
Despi e he la ge numbe o s udies ocused on sex di e ences in heal h and mo ali y,
ela i ely li le is known abou he male- emale heal h-su i al pa adox a e y old age, i.e. people
aged 90 and o e , o se e al easons. Fi s , al hough he oldes old cons i u e one o he as es
g owing segmen s o he olde popula ion [6], he inc easing numbe o people in his age g oup is s ill
a qui e ecen de elopmen [7]. Second, popula ion-based s udies o olde adul s a ely include
su icien da a on people aged 90 and o e , o be able o calcula e eliable es ima es on heal hy li e
expec ancy. The e o e, li le is known abou he ex en o which sex di e ences in heal hy li e
expec ancy con inue o exis a e y old age.
The ew s udies ha ha e been conduc ed among he oldes old show ha men ha e a highe
isk o mo ali y han women in popula ions aged 85 yea s and olde [8, 9]. These s udies also ound
ha disabili y inc eases he isk o mo ali y mo e in men han in women [8, 9]. Howe e , wha his
exac ly means o he numbe o yea s li ed in good and poo heal h a e y old age emains unclea .
The only longi udinal s udy ha p o ides de ailed insigh in o heal hy li e yea s among people aged 90
yea s and olde was conduc ed in Denma k [10]. This s udy calcula ed a e age li e imes o Danish
oldes old be ween 1998 and 2005, and obse ed be ween he ages o 92 and 100 an a e age li e ime
o 2.7 yea s o men and 3.3 yea s o women. This s udy also indica ed ha a la ge p opo ion o he
emaining yea s was spen in good heal h, and ha his did no di e much be ween men and women
[10]. These esul s ha e no been eplica ed so a .
Mul is a e models acili a e a be e unde s anding o ansi ions be ween heal h s a es and
dea h among he oldes old. Recen de elopmen s in mul is a e modeling make i possible o s udy he
ole o isk ac o s, such as sex, in ansi ions be ween hese s a es [11]. Mo eo e , based on mul is a e
models, o e all, heal hy, and unheal hy li e expec ancies (LEs) may be es ima ed o speci ic g oups
4
[12]. The objec i e o his s udy was o examine sex di e ences in o al, heal hy and unheal hy li e
expec ancy in a popula ion o nonagena ians (i.e., people aged 90 yea s and o e ). Fi s , we in es iga e
he ole o sex in ansi ions be ween heal h s a es and dea h, using wo di e en heal h indica o s.
Then, we es ima e LEs o men and women o gain insigh in o he male- emale heal h-su i al
pa adox a e y old age.
2. Me hods
2.1 S udy sample and design
Fo his longi udinal s udy, da a we e used om he Vi ali y 90+ S udy, a popula ion-based s udy o
nonagena ians in he ci y o Tampe e, Finland [7, 9]. A baseline in 2001, all indi iduals aged 90 and
olde , i espec i e o heal h o place o li ing, we e included in a mailed su ey. The da a collec ion
was epea ed in 2003, 2007, 2010, and 2014, e e y ime including he whole age g oup in he a ea. As
all hose who pa icipa ed once we e also included in he nex wa es, longi udinal da a is a ailable o
all people who en e ed he s udy in 2001. A baseline, a ques ionnai e was mailed o all inhabi an s
aged 90 yea s and o e (n = 1,129), o which 892 (79%) e u ned he ques ionnai e. The esponse a e
was 86% among hose s ill ali e du ing he da a collec ion, as 87 indi iduals had died be ween
sampling and sending ou o he ques ionnai es.
In he cu en s udy, we included pa icipan s ha e u ned he mailed ques ionnai e in 2001,
and who had a leas one heal h s a e a baseline (2001) and one heal h o dea h s a e a ollow-up
a ailable (2003-2014). Tha was he case o 884 people, who p o ided a maximum o 2,501
obse a ions. The equencies o numbe o obse a ions we e: wo obse a ions (n = 354), h ee
obse a ions (n = 399), ou obse a ions (n = 89) and i e obse a ions (n = 48). I pe sons we e no
able o ill ou he ques ionnai e, hey we e ins uc ed o ask help om a amily membe , ca egi e , o
iend. In ce ain cases, when pa icipan s we e no able o selec answe s, hese helpe s pa icipa ed as
p oxy. In his s udy, he pe cen age o p oxy pa icipan s was 23.6%. P oxy a es we e 15.3% o men
and 25.6% o women (p di e ence <0.01).
5
The e hics commi ee o he Pi kanmaa Hospi al Dis ic o he e hics commi ee o he
Tampe e Heal h Cen e , depending on he s udy yea , app o ed he s udy. All pa icipan s o hei
legal ep esen a i es p o ided w i en in o med consen .
2.2 Measu es
Two heal h indica o s we e used in he cu en s udy, disabili y and mul imo bidi y, o see whe he
esul s we e consis en ac oss heal h indica o s. Heal h indica o s we e iden ically measu ed a each
ollow-up. Disabili y was measu ed wi h i e mobili y ac i i ies and ac i i ies o daily li ing (ADL):
o mo e abou indoo s, o walk 400 me e s, o use s ai s, o d ess and und ess and o ge in and ou o
bed. Fo each i em, iden ical ques ions we e asked in he mailed ques ionnai e: “A e you able o..?”.
Response ca ego ies we e “Yes, wi hou di icul y”, “Yes, wi h di icul y”, “Only i someone helps”,
and “No”. Disabili y was de ined as being dependen o wo o mo e ac i i ies, whe e dependence
was conside ed p esen i a pa icipan was no able o pe o m a ce ain ac i i y o only wi h help [13,
14].
Mo bidi y was measu ed by sel - epo . Pa icipan s we e asked in he mailed ques ionnai e
whe he a doc o had old hem ha hey had any o he ollowing en condi ions: hype ension,
a he oscle osis, hea disease, cance , demen ia, s oke, diabe es, heuma ic diso de , os eoa h i is,
and Pa kinson`s disease. Mul imo bidi y was conside ed p esen i wo o mo e diseases we e epo ed
[15].
2.3 Mo ali y
All-cause mo ali y s a us, including da e o dea h, was e ie ed om he Finnish Na ional Popula ion
Regis e un il May 2014, and linked o he da ase wi h pe sonal iden i ie codes.
2.4 S a is ical analysis
A baseline, cha ac e is ics o he s udy popula ion we e epo ed o he o al sample and by sex, wi h
means and s anda d de ia ion o con inuous a iables and pe cen ages o ca ego ical a iables.
Baseline di e ences be ween men and women we e de e mined using Chi squa e es s and - es s.

6
Mul is a e modeling was used o assess ansi ions be ween heal h s a es and dea h du ing 13
yea s o ollow-up. Wi h mul is a e modeling i is possible o simul aneously model ansi ions
be ween heal h s a es and o examine he ole o co a ia es on all ansi ions. A h ee-s a e model was
applied, whe e s a e 1 was he heal hy s a e (absence o disabili y o mul imo bidi y), s a e 2 was he
unheal hy s a e (disabili y o mul imo bidi y p esen ), and s a e 3 was dea h as abso bing s a e (Figu e
1). Mul is a e su i al models wi h age and sex as co a ia es we e es ima ed sepa a ely o disabili y
and mul imo bidi y. Age was included as ime- a ying co a ia e. Da e o dea h was used o calcula e
he exac age o dea h. When pa icipan s had missing s a es be ween wo known s a es, in e al
censo ing was applied. Righ censo ing was applied when he las s a e was missing and he pa icipan
was known o be s ill ali e. Howe e , his was done o e y ew cases (n = 8), as mos people had
died a he end o he s udy pe iod (May 2014). Ini ially, we allowed backwa d ansi ions om s a e 2
o 1, bu due o low numbe s o backwa d ansi ions we had o ix his ansi ion in he analyses on
mul imo bidi y o bo h co a ia es, and we had o ix he sex pa ame e o his ansi ion in he
analyses on disabili y. Haza d a ios and 95% con idence in e als o s a e ansi ions a e epo ed o
age and sex.
The mul is a e su i al models we e es ima ed using he MSM package o R [16]. Based on
he pa ame e s o he mul is a e models, LEs we e calcula ed using he ELECT (Es ima ing Li e
Expec ancies in Con inuous Time) package o R [12]. ELECT es ima es o al and ma ginal LEs based
on mul inomial eg ession models o s a e p e alence. To al, unheal hy, and heal hy LEs we e
es ima ed o men and women sepa a ely, o bo h disabili y and mul imo bidi y as heal h a iables.
LEs we e es ima ed o indi iduals aged 90 o 100 yea s. To examine sex di e ences in o al,
unheal hy, and heal hy li e expec ancy, 95% con idence in e als o LEs we e es ima ed. All
s a is ical analyses we e pe o med using R e sion 3.4.2.
3. Resul s
Table 1 shows he cha ac e is ics a baseline o he o al sample and by sex. The sample included 172
men and 712 women. The mean age was 92.2 yea s (SD = 2.5), wi h a ange om 90 o 106 yea s. In
he o al sample, mul imo bidi y p e alence a baseline was 67.3%, wi h hea disease (53.4%),
7
demen ia (42.9%), os eoa h i is (35.8%), hype ension (31.8%) and a he oscle osis (16.7%) as he
mos common condi ions. Mul imo bidi y was highe in women (70.6%) han in men (53.5%).
Hype ension and os eoa h i is we e mo e o en p esen among women, whe eas cance was mo e
equen ly obse ed among men. Disabili y was p esen in 43.3% o he sample a baseline. All
disabili y i ems showed a highe p e alence in women han in men (p<0.01).
Du ing 13 yea s o ollow-up, 98.9% o he sample died (n=874). Among he 48 obse a ions
a he las ollow-up measu emen (in 2014), he e we e wo alid heal h s a e obse a ions, 8 we e
igh -censo ed and 38 we e dea h s a es (deceased be ween 2010 and 2014). The esul s o he
mul is a e models a e p esen ed in Table 2. Age and sex we e no associa ed wi h ansi ioning om a
heal hy s a e o an unheal hy s a e, o bo h heal h indica o s. In he analyses using disabili y as heal h
indica o , a highe age was associa ed wi h an inc eased isk o ansi ioning om a heal hy s a e o
dea h (HR = 1.15, 95% CI = 1.03-1.28) and wi h ansi ioning om an unheal hy s a e o dea h (HR =
1.04, 95% CI = 1.02-1.07). Female sex was associa ed wi h a lowe likelihood o ansi ion om an
unheal hy s a e o dea h (HR = 0.52, 95% CI = 0.40-0.67), bu no wi h ansi ioning om a heal hy
s a e o dea h (HR = 0.52, 95% CI = 0.23-1.16). When using mul imo bidi y as heal h indica o , a
highe age was again associa ed wi h a highe isk o ansi ioning om a heal hy s a e o dea h (HR =
1.14, 95% CI = 1.08-1.21) and om an unheal hy s a e o dea h (HR = 1.08, 95% CI = 1.06-1.11). Fo
emales, he e was a lowe isk o ansi ioning om a heal hy s a e o dea h (HR = 0.51, 95% CI =
0.31-0.84) and om an unheal hy s a e o dea h (HR = 0.73, 95% CI = 0.58-0.92).
LEs we e calcula ed o he ages 90 o 100 (Table 3 and Figu e 2). The es ima es o o al li e
expec ancy sligh ly di e be ween he wo heal h indica o s, due o small di e ences in he pa ame e s
and obse a ions included in he analyses o each indica o . Howe e , his does no a ec he
in e p e a ion o esul s, as sex di e ences o o al li e expec ancy a e consis en ac oss heal h
indica o s. To al li e expec ancy a age 90 was a ound 3.7 yea s o women and 3 yea s o men,
co esponding wi h a di e ence o 8 mon hs in o al li e expec ancy. This dec eased o 1.8 yea s
(women) and 1.3 yea s (men) a he age o 100, which is s ill a di e ence o 6 mon hs in o al li e
expec ancy.
8
Fo bo h heal h indica o s, sex di e ences in unheal hy li e expec ancy we e obse ed,
showing ha women spend mo e ime in poo heal h, in bo h absolu e (yea s) and ela i e (p opo ion)
e ms. The a e age li e expec ancy wi h disabili y o women was 2 yea s a age 90, 1.7 yea s a age
95 and 1.5 yea s a age 100, which co esponds o 53%, 66%, and 78% o he o al li e expec ancy,
espec i ely. Fo men, li e expec ancy wi h disabili y a he age o 90, 95 and 100 was 26%, 36% and
48% o o al li e expec ancy, espec i ely. Figu e 2 p o ides an illus a ion o hese di e ences.
Simila pa e ns we e obse ed o li e expec ancy wi h mul imo bidi y, al hough sex di e ences we e
a bi smalle when using his heal h indica o . Table 3 also p o ides heal hy li e expec ancy es ima es.
Be ween he age o 90 and 95, men had highe disabili y- ee li e expec ancy han women. Fo
example, a he age o 90 he sex di e ence in disabili y- ee li e expec ancy was 6 mon hs, in a o o
men. A he age o 100, sex di e ences in disabili y- ee li e expec ancy we e no longe obse ed. Fo
mul imo bidi y, no sex di e ences in heal hy li e expec ancy we e ound.
4. Discussion
The main aim o his s udy was o examine sex di e ences in o al, heal hy, and unheal hy li e
expec ancy among nonagena ians, using es ima es based on mul is a e su i al models. The esul s
indica ed ha sex di e ences in LEs we e p esen up o e y high age. Di e ences we e mainly
obse ed in o al and unheal hy li e expec ancy: women aged 90 yea s and o e li e longe han men,
and spend mo e ime in poo heal h, o bo h disabili y and mul imo bidi y as heal h indica o . Based
on hese esul s we conclude ha he male- emale heal h-su i al pa adox is s ill p esen in he e y
old.
The esul s o his s udy e ealed ha he o al li e expec ancy a he age o 90 was 3.7 yea s
o women and 3 yea s o men. This is no e y di e en om a p e ious s udy in a Danish
popula ion, ha ound an a e age li e ime among nonagena ians o 3.3 yea s o women and 2.7 yea s
o men [10]. Howe e , his s udy only p o ided a e ages o he o al g oup aged 90 and o e , and
did no calcula e LEs o speci ic ages. Ou indings p o ide a mo e de ailed insigh in o dec eases in
li e expec ancy wi h ad ancing age. Fo a pe son aged 100 yea s, he o al li e expec ancy dec eased o
1.8 yea s o women and 1.3 yea s o men.
9
Pa e ns wi h ega d o unheal hy li e expec ancy we e simila o hose p e iously obse ed in
younge age g oups [1], as we ound ha women li e mo e yea s wi h disabili y and mul imo bidi y
han men. A he age o 90, women spend on a e age 14 mon hs mo e wi h disabili y and 12 mon hs
mo e wi h mul imo bidi y han men. A la ge p opo ion o he o al li e expec ancy be ween he ages
o 90 and 100 is es ima ed o be in he unheal hy s a e. These indings a e con a y o hose om he
p e iously men ioned s udy in Denma k, whe e nonagena ians we e es ima ed o spend 75% o hei
emaining li e ime in physical independence. Mo eo e , his Danish s udy did no ind subs an ial sex
di e ences in unheal hy li e expec ancy. This sugges s ha es ima es should be in e p e ed in he
con ex o he coun y whe e he esea ch is conduc ed. Wi h ega d o heal hy li e expec ancy, we
obse ed be ween he age o 90 and 95 ha men had a highe disabili y- ee li e expec ancy o abou 5
o 6 mon hs. This is line wi h esul s om he Newcas le 85+ s udy, in which men we e es ima ed o
li e 6 mon hs longe wi hou disabili y compa ed o women [8].
Women we e less likely o ansi ion om a heal hy o unheal hy s a e o dea h han men. This
is p obably jus an exp ession o he lowe mo ali y a es among emale nonagena ians [9]. The
inding ha age and sex we e no associa ed wi h ansi ioning om a heal hy s a e o an unheal hy
s a e in he mul is a e models, may be due o he ac ha p e alence o disabili y and mul imo bidi y
was al eady high a baseline. I is known ha among nonagena ians he p e alence o disabili y and
mul imo bidi y is high [17, 18].
This was one o he i s s udies o use longi udinal da a o es ima e LEs in he oldes old.
S eng hs o he s udy include he la ge sample o nonagena ians, he use o an en i e age coho om
one a ea, he long ollow-up pe iod, and he no el analy ical app oach. The mul is a e su i al model
app oach is among he ew s a is ical me hods ha does no equi e na ional li e able da a o calcula e
LEs [12, 19], as opposed o he commonly used Sulli an me hod [20]. Using he ELECT package in
R, LEs may be calcula ed in any longi udinal da ase wi h in o ma ion on heal h s a es and mo ali y.
Ne e heless, some limi a ions should be conside ed when in e p e ing he esul s o his s udy. Fi s ,
he numbe o men included in his s udy is small as compa ed o he numbe o women. Howe e , his
e lec s ha mo ali y a es a e highe among men. The pe cen age o men co esponds wi h hei sha e
in he o al popula ion. In he Vi ali y 90+ s udy, he e we e no di e ences in pa icipa ion a es
16
Figu e 1. Design: h ee-s a e model
1. Heal hy
(no disabili y o no
mul imo bidi y)
2. Unheal hy
(disabili y o
mul imo bidi y)
3. Dea h
Table 2. Haza d a ios and 95% con idence in e als o he e ec o age and sex on ansi ions
h ough he di e en heal h s a es
Disabili y
Mul imo bidi y
Age
Sex
Age
Sex
T ansi ions
HR (95% CI)
HR (95% CI)
HR (95% CI)
HR (95% CI)
S a e 1 –S a e 2
1.07 (0.99, 1.14)
1.37 (0.82, 2.27)
0.97 (0.91, 1.04)
1.61 (0.97, 2.70)
S a e 1 - Dea h
1.15 (1.03, 1.28)*
0.52 (0.23, 1.16)
1.14 (1.08, 1.21)*
0.51 (0.31, 0.84)*
S a e 2 – S a e 1
0.89 (0.74, 1.08)
--
--
--
S a e 2 - Dea h
1.04 (1.02, 1.07)*
0.52 (0.40, 0.67)*
1.08 (1.06, 1.11)*
0.73 (0.58, 0.92)*
S a e 1 = heal hy (no disabili y o no mul imo bidi y); S a e 2 = unheal hy (disabili y o
mul imo bidi y); Sex (0 = men, 1 = women); *p<0.05; N obse a ions disabili y = 2,501; N
obse a ions mul imo bidi y = 2,488

17
Table 3. Li e expec ancy es ima es o men and women by heal h indica o
Disabili y
Mul imo bidi y
Men
Women
Men
Women
Age 90
To al li e expec ancy in yea s (95% CIs)
3.13 (2.68, 3.51)
3.77 (3.53, 3.99)*
2.96 (2.60, 3.34)
3.70 (3.46, 3.92)*
Unheal hy li e expec ancy in yea s (95% CIs)
0.84 (0.64, 1.06)
2.01 (1.81, 2.18)*
1.68 (1.41, 1.98)
2.66 (2.46, 2.85)*
Heal hy li e expec ancy in yea s (95% CIs)
2.28 (1.84, 2.66)
1.76 (1.58, 1.95)*
1.28 (0.99, 1.26)
1.04 (0.89, 1.19)
Age 95
To al li e expec ancy in yea s (95% CIs)
2.01 (1.57, 2.30)
2.64 (2.42, 2.91)*
1.98 (1.69, 2.28)
2.62 (2.41, 2.83)*
Unheal hy li e expec ancy in yea s (95% CIs)
0.72 (0.52, 0.92)
1.74 (1.54, 1.94)*
1.05 (0.86, 1.27)
1.78 (1.60, 1.96)*
Heal hy li e expec ancy in yea s (95% CIs)
1.29 (0.96, 1.54)
0.91 (0.76, 1.08)*
0.93 (0.72, 1.17)
0.85 (0.69, 1.00)
Age 100
To al li e expec ancy in yea s (95% CIs)
1.27 (0.87, 1.56)
1.91 (1.63, 2.22)*
1.25 (0.98, 1.59)
1.78 (1.55, 2.05)*
Unheal hy li e expec ancy in yea s (95% CIs)
0.62 (0.44, 0.83)
1.49 (1.22, 1.78)*
0.63 (0.47, 0.85)
1.14 (0.95, 1.38)*
Heal hy li e expec ancy in yea s (95% CIs)
0.64 (0.34, 0.89)
0.42 (0.29, 0.59)
0.62 (0.39, 0.90)
0.64 (0.47, 0.84)
*Sex di e ence = p <0.05
18
Figu e 2. Li e expec ancy in good and poo heal h by gende and heal h indica o (in yea s)