Journal of Gerontology: MEDICAL SCIENCES
2003, Vol. 58A, No. 6, 566–572
Copyright 2003 by The Gerontological Society of America
Communication Between Identical Twins:
Health Behavior and Social Factors Are Associated With
Longevity That Is Greater Among Identical
Than Fraternal U.S. World War II Veteran Twins
Malcolm D. Zaretsky
Department of Molecular and Cell Biology, University of California at Berkeley.
Background. Longevity is greater for identical twins than for fraternal twins from the same population. Factors that are
explanatory for this difference are not known.
Methods. Multivariate survival analysis is applied to current mortality data for 26,974 male twins with known
zygosities of the National Academy of Science–National Research Council World War II Veteran Twins Registry, and this
analysis is applied to their health and social behavior and personal histories, as collected from two survey questionnaires
distributed in 1967 and 1983 (with 14,300 and 9475 responses received, respectively). To explain this difference in
longevity, social, health, and personal history factors are evaluated for associations with longevity.
Results. Survival functions of identical and fraternal twins differed significantly ( p , .0001). Median lifetimes were 82
years for identical and 80.5 years for fraternal twins. The correlation between lifetimes of identical twin partners was
greater than that of fraternal twins. For identical but not for fraternal twins, the risk of mortality was significantly lower for
twin partners who communicated 1 or more times per month, compared with those who communicated less frequently
(relative risk .80, 95% confidence interval 0.68–0.94, p ¼ .008, with control for other factors associated with longevity:
smoking, exercise, a current marriage, living with both parents until age 15 or older, and having a live co-twin). Distributions of communication, exercise level, and smoking prevalence were more beneficial with regard to longevity for
identical than for fraternal twins as a group.
Conclusions. Frequent communication between identical but not fraternal twin partners, and both level of exercise and
prevalence of smoking, distributed more beneficially in terms of longevity for identical compared with fraternal twins, are
explanatory for the greater longevity of identical than fraternal twins.
I
DENTICAL or monozygotic (MZ) twins were shown
to live longer than fraternal or dizygotic (DZ) twins in
studies with the Swedish Twin Registry (1,2). The
factors that affect this difference in longevity are not
well known, with the exception of the mortality status of
the twin partner. Consistent differences between MZ and
DZ twins have not been found for factors known to
affect health and longevity from studies that did not
involve twins [i.e., smoking, physical activity, alcohol
abuse, marital status (1,3–5), and social support networks
(6–8)]. The study now reported analyzes the difference
between longevity of MZ and DZ twins in the National
Academy of Sciences–National Research Council World
War II Veterans Twins Registry (NAS–NRC WW II
Twins Registry) with regard to explanatory behavioral
and social factors. This study tests the hypothesis that
that there are associations of longevity with health and
social behavior that differ significantly between MZ and
DZ twins, thereby accounting for the greater longevity of
MZ twins. Filling this gap in knowledge regarding the
factors that underlie the difference in longevity between
MZ and DZ twins contributes to our understanding of
human longevity and aging, as have other studies of
human twins (9–13).
566
METHODS
The NAS–NRC WW II Twins Registry, the largest twin
registry to be studied, contains mortality and zygosity (MZ
or DZ) data for 26,894 World War II U.S. veteran twins
(11,832 MZ and 15,062 DZ twins) born between 1917 and
1927. The assembly of the registry has been described (13).
The registrants’ mortality statuses are periodically obtained
from the computerized records of the U.S. Department of
Veterans Affairs (DVA), formerly the Veterans Administration, which is notified of the death of approximately 98% of
World War II veterans by relatives or morticians, who seek
to claim a burial allowance. Cause of death information is
obtained from DVA records of death certificate information.
The results now reported are based on mortality data through
November 1999 and on 2 survey questionnaires distributed
in 1967–1969 and 1983–1985 to registry members with
14,300 (6301 MZ, 7399 DZ) and 9475 (4292 MZ, 5183
DZ) identified responses (80% and 71% response rates,
respectively), when corrected for mortality (14). Registry
members who returned both 1967 and 1983 questionnaires
numbered 7034 (3405 MZ, 3629 DZ), of which 6635 (3244
MZ, 3391 DZ) had live twin partners (Figure 1).
Communication between co-twins (either by telephone
or letter) included only in the 1983 questionnaire were
COMMUNICATION AND LONGEVITY OF TWINS
Figure 1. Flow diagram indicating numbers of monozygotic (MZ) and
dizygotic (DZ) twins for the following: NAS–NRC WW II Twins Registry,
1967 and 1983 questionnaires returned by those in the registry and those in the
registry who returned both 1967 and 1983 questionnaires with both twin
partners alive.
stratified into two categories in this analysis: greater than or
equal to once per month and less than once per month. The
amounts of exercise outside of work after the age of 35
years included in the 1967 questionnaire were ranked by
quartiles. The 1983 questionnaire included three exercise
questions: number of flights of stairs climbed per day, number of city blocks walked per day, and walking speed. Responses to these questions were ranked and summed, and
the sums were ranked by quartiles. Alcoholic beverage
consumption by drinks per day of beer, wine, and liquor, as
well as smoking habits, were included in both 1967 and
1983 questionnaires. Alcoholic beverage consumption was
categorized by grams per day of ethanol consumed: very
light ¼ greater than 0 to 1.49 g/day, light ¼ 1.50 to 4.99 g/
day, moderate ¼ 5 to 29.99 g/day, and heavy ¼ 30.00 g/day
or greater. The average amount of alcohol consumed per
day was calculated from the sum of alcohol content in beer
(13.2 g ethanol/12-oz bottle), wine (10.8 g ethanol/4-oz
glass), and liquor (15.1 g ethanol/2-oz shot) (15). Smokers
were categorized as smokers or nonsmokers. Body-mass
index, measured as weight (kg)/(height [m])2, was categorized either as normal, when it measured from 18.5 to
24.9 kg/m2, as proposed by the World Health Organization,
or as outside of the normal range (16).
Statistics
Explanatory factors were included as indicator or dummy
variables in the Cox proportional hazards model (17). The
time variable in the model was lifetime or current age. Twins
remaining alive at the end of the follow-up period, in
567
November 1999, were right censored. Exercise quartiles for
both the 1967 and 1983 survey responses were included in
the Cox proportional hazards model as three indicator
variables for the highest three quartiles, with the least
exercise quartile as the reference category. Alcoholic
beverage consumption was indicated by three variables—
light, moderate, and heavy drinker—referred to very light
drinker. Marital status was indicated by two variables—
currently married and formerly married—referred to never
married. Frequency of communication between co-twins,
smoker versus nonsmoker status, co-twin mortality status,
body-mass index, and having lived with parents to age 15
years or older were dichotomous indicator variables. Timedependent variables, including values provided by both 1967
and 1983 questionnaires—current smoking status, exercise
level, alcohol consumption, and body-mass index—changed
values at the mid-point between the ages of the registrant on
the dates of responses to the two questionnaires. Timedependent co-twin mortality status changed value on the date
of death of the co-twin. Calculations were based on nonmissing data.
The Kaplan–Meier estimate of the survivor function, S(ti),
the survival probability at time ti, is constructed as the product of successive probabilities of survival from ti to tiþ1 at
distinct times of death t1, t2, . . . ti, assuming independence
of each death (18):
Sðti Þ ¼ p1 p2 . . . pi ;
where pj ¼ (1 dj /nj), nj is the number of individuals in the
risk set, alive and uncensored, just prior to tj, and dj is the
number of deaths in the interval tj to tjþ1 ( j ¼ 1, 2 . . . i).
Because survival functions are skewed toward larger
values of time, the median is preferred over the mean as the
measure of central location. The median survival time is that
value of time of death at which the survival function is equal
to .5. Since the survival function is a step function, there
may not be a time of death at which the survival function is
exactly .5. In that case, the median survival time is defined
as the smallest time of death corresponding to a value of the
survival function less than .5.
The standard error (SE) of the survival function at some
value of t in the interval tk to tkþ1 is given by an approximation, known as Greenwood’s Formula (18):
(
)1=2
Xk
½dj =nj ðnj dj Þ
SEfSðtÞg ¼ ½SðtÞ
j¼1
The 95% confidence interval of the Kaplan–Meier
estimate of S(t) for any value of t is given by S(t) 6 z0.025
SEfS(t)g, where z0.025 ¼ 1.96 is the upper (1 0.95)/2 ¼
.025 point of the normal distribution, assuming a normal
distribution for the Kaplan–Meier estimate of S(t) at all
values of t. The SAS LIFETEST procedure was used to
calculate Kaplan–Meier survival statistics (19).
The hazard function, h(t), the probability of dying at time
t conditional on having survived to that time, may be
expressed in terms of the derivative with respect to time of
the logarithm of the survivor function:
hðtÞ ¼ d=dt flog SðtÞg:
ZARETSKY
568
The Cox proportional hazards model expresses the hazard
function for the ith individual as a product of a nonparametric
baseline hazard function with a factor that is an exponential
of a sum of products of k model parameters b1, b2, . . . bk,
and values of the variables X1, X2, . . . Xk for that individual:
(
)
Xk
hi ¼ h0 exp
bj xij ;
j¼1
where the baseline hazard function, h0, is the hazard
function of an individual with values of zero for all k
variables in the model. The sum of products of the significant model parameters with their respective variables
for the model with MZ and DZ twins in aggregate is:
X
bi Xi ¼ 0:24 ðco-twin mortalityÞ
0:38 ðzygosity co-twin mortalityÞ
þ 0:32 ðzygosityÞ
0:22 ðzygosity co-twin communicationÞ
0:48 ðexercise quartile 2Þ
0:50 ðexercise quartile 3Þ
0:65 ðexercise quartile 4Þ
þ 0:44 ðcurrent smokerÞ
0:29 ðcurrently marriedÞ
0:23 ðlived with both parents to age 15Þ:
The model parameters b1, b2, . . . bk are determined by the
maximum likelihood method. It is not necessary to provide
the baseline hazard function to determine the hazard ratio, or
risk ratio, of 2 individuals with different values of any of the
explanatory variables, fXkg:
(
)
Xk
bj ðxij xi9j Þ
hi =hi9 ¼ exp
j¼1
The variables in the models in these analyses are all
factors with values 0 or 1. The risk ratio for two individuals
differing only in the values of 1 factor, Xj, xij ¼ 1, xi9j ¼ 0, is,
therefore, ebj. The Wald chi-square statistic, the square of
the regression coefficient divided by its standard error,
referred to a chi-square distribution with 1 degree of freedom, evaluated the significance of the regression coefficients (20). The 95% confidence interval for bj is given by
bj 6 z0.025 SE(bj). The 95% confidence interval for the risk
ratio ebj is exp fbj 6 z0.025 SE(bj)g.
Risk ratios for the factors communication frequency and
co-twin mortality, which had a significant interaction with
zygosity, were calculated from the regression coefficients of
the main effects and interaction terms in the multivariate Cox
proportional hazards model. The 95% confidence intervals of
these risk ratios were calculated with the SEs of the sums of
the regression coefficients, derived from their variances and
covariances. The SAS procedure PHREG performs the
calculations for the Cox proportional hazards model and
provides the SEs of the regression coefficients, SE(bj), and
their covariances (20).
Missing Data
Values were missing for the explanatory variables tested
in the Cox proportional hazards model for some of the 6635
registry members who had returned both the 1967 and 1983
questionnaires and whose co-twins were alive in 1983. The
frequencies of missing values were relatively low. The
numbers and percentages of missing values were as follows:
exercise, 1967—200 (3%); exercise, 1983—592 (9%);
smoker, 1967—99 (1%); smoker, 1983—40 (0.6%); alcoholic drinks, 1967—600 (9%); alcoholic drinks, 1983—944
(14%); body mass index, 1967—136 (2%); body mass
index, 1983—179 (3%); communication with co-twin—537
(8%); and lived together with parents to age 15 or older—
155 (2%). Lifetimes of those with missing values did not
differ significantly from the members whose values were not
missing (Wilcoxon test, two-sided, p ¼ .23).
Methods of dealing with missing values in data, introducing values that complete a data set with missing values,
imputation, have been discussed extensively (21,22). The
most commonly applied method is to assign the mean values
of the nonmissing values of each variable to the missing
values of that variable. That is the method applied here but
with median values assigned rather than mean values,
because the variables were either categorical or skewed. In
the case of variables expected to have the same value for both
members of a twin pair, the pair member’s nonmissing values
were assigned as the pair member value that had been
missing. Those variables include co-twin communication (63
[1%]) missing values and living together with both parents to
age 15 or older (29 [0.5%]).
The resulting Cox proportional hazards model parameters
did not differ significantly, well within the 95% confidence
limits, from the model for MZ and DZ twins in aggregate
based on the incomplete data, given above. The reported
results appearing in Tables 1A–1C were those calculated
with nonmissing data.
The sum of products of the significant model parameters
with their respective variables for the Cox proportional
hazards model with MZ and DZ twins in aggregate
calculated with assigned values in place of missing values is:
X
bi Xi ¼ 0:23 ðco-twin mortalityÞ
0:37 ðzygosity co-twin mortalityÞ
þ 0:31 ðzygosityÞ
0:24 ðzygosity co-twin communicationÞ
0:45 ðexercise quartile 2Þ
0:47 ðexercise quartile 3Þ
0:63 ðexercise quartile 4Þ
þ 0:46 ðcurrent smokerÞ
0:29 ðcurrently marriedÞ
0:22 ðlived with both parents to age 15Þ:
RESULTS
Kaplan–Meier survival analysis that included all those in
the NAS–NRC WW II Twins Registry who had lived to age
35 years or more demonstrated that longevity of MZ twins
COMMUNICATION AND LONGEVITY OF TWINS
569
Table 1. Risk Ratios of Factors for Survival in the Cox Proportional Hazards Model
Factor
Regression Coefficient (b)
A. Monozygotic (MZ) and Dizygotic (DZ) Twins in Aggregate
Co-twin communication:
MZ
0.22
Frequent (1/month) versus infrequent (,1/month)
DZ
0.05
Zygosity, MZ:DZ, by frequency of verbal or written
All
0.17
communication with co-twin
1/month
0.28
,1/month
0.064
MZ
0.62
Co-twin mortality:
Live versus deceased co-twin§
DZ
0.24
Exercise level:
2nd
0.47
Nth quartile versus least exercise (1st) quartilejj
3rd
0.49
4th
0.63
Smoker versus nonsmokerjj
0.44
Lived with both parents to age 15 versus ,15
0.22
Currently married versus never married
0.29
Formerly married versus never married
0.03
light
0.12
Alcoholic beverage consumption level versus
moderate
0.03
very light consumptionjj
heavy
0.01
Body mass index: normal versus not normal rangejj
0.019
p Value*
.0078
.53
.0080
,.0001
.30
,.0001
.0085
,.0001
,.0001
,.0001
,.0001
.010
.0002
.82
.086
.87
.85
.45
Risk Ratio (CI )
0.80
0.95
0.84
0.75
0.93
0.54
0.79
0.62
0.61
0.53
1.56
0.80
0.75
0.97
0.88
1.03
0.99
1.02
(0.68–0.94)à
(0.80–1.12)
(0.74–0.95)à
(0.66–0.86)à
(0.83–1.06)
(0.45–0.65)à
(0.67–0.95)à
(0.54–0.71)à
(0.51–0.73)à
(0.45–0.61)à
(1.39–1.75)à
(0.67–0.95)à
(0.64–0.88)à
(0.71–1.31)
(0.75–1.02)
(0.88–1.04)
(0.84–1.16)
(0.99–1.02)
B. MZ Twins
Co-twin communication: Frequent (1/month) versus
infrequent (,1/month)
Co-twin mortality: Live versus deceased co-twin§
Exercise level:
Nth quartile versus least exercise (1st) quartilejj
Smoker versus nonsmokerjj
Currently married versus never married
Formerly married versus never married
Alcoholic beverage
consumption level versus
very light consumptionjj
Body mass index: normal versus not normal rangejj
Lived with both parents to age 15 versus ,15
0.22
2nd
3rd
4th
light
moderate
heavy
.0082
0.64
0.57
0.44
0.66
0.36
0.030
0.045
0.13
0.012
0.10
0.016
0.22
,.0001
,.0001
.0005
,.0001
,.0001
.0095
.84
.24
.90
.37
.20
.088
0.041
.53
0.23
0.39
0.57
0.63
0.52
0.28
0.15
0.11
.017
0.083
0.014
0.24
.0096
,.0001
,.0001
,.0001
,.0001
.009
.46
.30
.84
.47
.26
.051
0.80 (0.68–0.95)à
0.53
0.57
0.64
0.52
1.44
0.74
1.05
0.88
0.99
1.11
1.02
0.80
(0.44–0.63)à
(0.46–0.69)à
(0.50–0.83)à
(0.42–0.64)à
(1.21–1.71)à
(0.60–0.93)à
(0.67–1.62)
(0.71–1.09)
(0.82–1.19)
(0.89–1.38)
(0.99–1.04)
(0.62–1.03)
C. DZ Twins
Co-twin communication: Frequent (1/month) versus
infrequent (,1/month)
Co-twin mortality: Live versus deceased co-twin§
Exercise level:
Nth quartile versus least exercise (1st) quartilejj
Smoker vs. non-smokerjj
Currently married versus never married
Formerly married versus never married
Alcoholic beverage consumption
versus very light consumptionjj
Body mass index: normal versus not normal rangejj
Lived with both parents to age 15 versus ,15
2nd
3rd
4th
light
moderate
heavy
0.96 (0.81–1.13)
0.79
0.68
0.56
0.53
1.69
0.76
0.85
0.90
0.98
0.92
0.99
0.79
(0.67–0.95)à
(0.57–0.81)à
(0.43–0.73)à
(0.43–0.65)à
(1.44–1.98)à
(0.61–0.93)à
(0.56–1.29)
(0.73–1.10)
(0.83–1.16)
(0.74–1.15)
(0.96–1.01)
(0.62–1.00)
Note: WW II Registry twins who responded to both the 1967 and 1983 questionnaires with live co-twins in 1983 were included in the model calculations:
3244 MZ twins and 3391 DZ twins. For the aggregate model, Section A, separate MZ and DZ table entries are shown for the 2 factors with significant interactions with zygosity, co-twin mortality and co-twin communication. Mortality data are that obtained from Department of Veterans Affairs (DVA) records. Data
for explanatory factors were based on both 1967 and 1983 questionnaires as indicated (jj) or only on the 1983 questionnaire, if not so noted. Cox proportional
hazards models significances: p , .0001 (likelihood ratio, Wald’s, and score test) (20). Table entries of all factors, significant or nonsignificant, are those for
models that include all significant explanatory factors.
*Significance of model regression coefficient (Wald Chi-square) (20).
95% Confidence interval.
à
Significant factor.
§
Source: DVA death records, time-dependent factor.
jj
Source: Both 1967 and 1983 questionnaires, time-dependent factor.
570
ZARETSKY
Figure 2. Kaplan–Meier survival functions. In each graph, the p value is the significance of the log-rank test for the homogeneity of the survival functions for
monozygotic (MZ) and dizygotic (DZ) twins, and RR is the unadjusted Cox proportional hazards model risk ratio of MZ to DZ mortality, with its 95% confidence
interval. Adjusted Cox proportional hazards model MZ:DZ risk ratios by frequency of verbal or written communication, corresponding to parts B–D of the figure, are
given in Table 2A. A, Survival is significantly greater for all MZ than for all DZ twins in the WW II Twins Registry who have lived to at least age 35 years, updated 11/
30/99 (N ¼ 11,456 MZ, 14,564 DZ). The median lifetimes for MZ and DZ twins with their 95% confidence intervals were 82.0 (81.5–82.5) years and 80.5 (79.9–81.1)
years, respectively. The survival probability of MZ twins was significantly greater than that of DZ twins for ages 50 and above. P values (t test, two-sided) shown
below the X-axis indicate the significance of the difference between MZ and DZ survival probabilities at the respective ages. B–D, Twins responding to 1983
questionnaire with both twins of a pair alive. B, All co-twin communication frequencies: Survival is significantly greater for MZ than DZ twins (N: MZ ¼ 3244, DZ ¼
3391). C, Frequent (1 time/month) communication stratum: MZ twins are at lower risk of mortality than DZ twins (N: MZ ¼ 1801, DZ ¼ 1287). Survival as of
November 1999: MZ ¼ 81.0%, DZ ¼ 75.6% ( p ¼ .0005, t test, two-sided). D, Infrequent (,1 time/month) communication stratum: MZ and DZ twins do not differ
significantly in risk of mortality (N: MZ ¼ 1443, DZ ¼ 2104). Survival as of November 1999: both MZ and DZ ¼ 75.3%.
was greater than that of DZ twins, with unadjusted risk ratio
(RR), MZ:DZ, of 0.92 ( p , .0001). Survival probabilities
of MZ and DZ twins differed significantly at all ages 50
years (Figure 2A). For all 26,974 twins in the registry,
including 43% that were censored, the median difference
between lifetimes of MZ twin partners was 2.7 years
(interquartile range [IQR] ¼ 12.6 years) and 5 years (IQR ¼
15.4 years) for DZ twin partners, which differed significantly ( p , .0001, Wilcoxon test, two-sided). The lifetimes
of MZ twin pairs are more highly correlated than those of
DZ pairs. For NAS–NRC WW II Twins of all ages who had
lived together any number of years, the intraclass correlation
coefficients (23) for MZ and DZ twins were .17 and .07,
respectively, which were significantly different ( p , .0001,
t test, two-sided). Broad sense heritability (24), calculated
with these correlations, is 0.20. For twins having lived
together at least 15 years, intraclass correlations for all ages
were MZ ¼ 0.15 and DZ ¼ 0.06 ( p , .0001), with
heritability of .18. For twins having lived together any
number of years or having lived together at least 15 years
with at least one twin of a pair with lifetime within the age
group, heritability was highest, .24, at ages 30 or 40 and
not significant at ages 70 or 80, for which 83% and 95%
of the observations, respectively, were censored.
Multivariate Cox proportional hazards survival analysis
for MZ and DZ twins in aggregate assessed the relative risks
of mortality for health and social behaviors (Table 1A).
There were significant interactions between zygosity, the
variable indicating the MZ or DZ twin type, and communication between twin partners ( p ¼ .008) as well as for
zygosity and the presence of a live twin ( p ¼ .003). Frequent
communication between twin partners was associated with
reduced mortality for MZ but not for DZ twins. Kaplan–
Meier survival analysis by strata of frequency of communication between twin partners demonstrated a significantly
greater survival for MZ than for DZ twins in the frequent
(1 time/month) communication stratum but not in the infrequent (,1 time/month) communication stratum (Figures
2B–2D). This difference was confirmed by the multivariate
Cox proportional hazards analysis, which controlled for
other significant factors (Table 1A). Other sociability
measures—active memberships in church or community
COMMUNICATION AND LONGEVITY OF TWINS
Table 2. Distributions of Health and Social Behavior Factors in
Monozygotic (MZ) and Dizygotic (DZ) Twins
Proportion (%) (N*)
Factor
Communication with co-twin
frequently (1/month)à
MZ
DZ
p Value 48.8 (1584) 29.6 (1003) ,.0001
571
that were not significant in the Cox survival analysis did not
differ significantly in their distributions between MZ and
DZ twins. The distribution of alcoholic drinks consumption,
not significant in the Cox survival analysis as a timedependent variable, differed significantly between MZ and
DZ twins in 1967 but not in 1983.
Exercise level above median:§
1967
1983
45.7 (2794) 43.1 (3113)
44.6 (1383) 41.6 (1367)
.003
.02
79.2 (4998) 81.7 (6117)
70.0 (2382) 72.6 (2630)
.004
.01
Alcoholic beverage
consumption: 1967
light
16.3 (926) 17.6 (1260)
moderate 19.8 (1123) 17.6 (1254)
heavy
15.7 (891) 17.0 (1216)
.04
.002
.04
Alcoholic beverage
consumption: 1983
light
20.2 (710) 19.8 (821)
moderate 37.5 (1321) 37.1 (1534)
heavy
16.5 (594) 16.97 (681)
.7
.7
.6
50.0 (3090) 50.3 (3682)
42.2 (1399) 42.0 (1480)
.7
.9
77.8 (5590) 77.6 (6612)
88.6 (2988) 88.4 (3195)
.8
.8
91.0 (3220) 90.1 (3212)
.5
Smoker or former smoker:
1967
1983
Body-mass index
in normal range:
1967
1983
Current marriage:
1967
1983
Lived with both
parents to age 15
*Number in category.
Fisher’s exact test, 2-sided, for significance of the difference between
MZ and DZ proportions.
à
1983 Questionnaire, both twins alive. Not included in the 1967 questionnaire.
§
Median is that of MZ and DZ registry members in aggregate with nonmissing data.
groups and close relationships with other relatives or
friends—were not significant factors in the Cox proportional
hazards model.
A significant interaction with zygosity indicated that the
relative risk of mortality for having a live compared with
a deceased twin partner was significantly lower for MZ than
for DZ twins. Smoking, exercise at all levels, having lived
with both parents to age 15 or older, and a current marriage
in 1983 were associated with longevity for MZ and DZ twins
in aggregate, with no significant difference in risk ratios
by zygosity. Alcoholic beverage consumption, body-mass
index, and a former marriage were not significant factors
when tested in the model. Cox proportional hazards survival
analysis by zygosity strata, separately for MZ and DZ twins,
produced results similar to that of the aggregate analysis
(Tables 1B and 1C). Notably, frequent communication was
a significant factor for longevity for MZ twins (Table 1B) but
not for DZ twins (Table 1C).
The significant health and social behavior factors for
survival in the Cox proportional hazards analysis—smoking, exercise, and communication between co-twins, with
the exception of a current marriage—are distributed with
greater benefit in terms of longevity for MZ twins than for
DZ twins (Table 2). The distributions of behavioral factors
DISCUSSION
Evidence has been presented that communication
between identical twin partners and health and social
behavior are explanatory for the greater longevity of MZ
than of DZ twins. Frequent communication between twin
partners stands out among health and social behavior
factors as being associated with survival for MZ but not for
DZ twins in this study. Among twin pairs who did not communicate frequently, MZ and DZ survival did not differ
significantly. Further, the distributions by MZ and DZ
groups of communication between co-twins, smoking prevalence and exercise level, significant factors for longevity,
were more beneficial for MZ than for DZ twins in terms
of longevity. The heritability of longevity in this cohort is
similar to that found in studies of other twin registries, 20%–
30% (2,10,12), indicating that longevity is affected by
environmental factors to a greater degree than genetic
factors.
Caution is necessary when inferring causality from
associations. However, greater frequency of communication
between MZ than between DZ twin partners is indicative of
stronger ties between the former and argues that genetic
relatedness for these twins has an environmental influence.
Although not shown in this study, frequent communication
could provide greater support and competition for a healthier
lifestyle that lasts into later life and could effect reduced
susceptibility to disease or improved outcomes for illnesses.
Frequent communication interacting with relatedness could
have similar effects on non-twins as well. Similar studies in
other twin cohorts that include female pairs of twins are
needed to test whether the conclusions of this study are more
widely applicable. These results call for further behavioral
and physiological investigations to provide understanding of
the mechanisms that underlie the influence of communication on the lifestyle and greater longevity of MZ compared
with DZ twins.
ACKNOWLEDGMENTS
I thank Dr. Walter M. Bortz II and Prof. Richard Strohman for their
intellectual guidance and encouragement. The Institute of Medicine of the
U.S. National Academy of Sciences-National Research Council, Medical
Follow-up Agency provided the data of the NAS–NRC World War II
Veterans Twins Registry. I wish to acknowledge their helpfulness and
cooperation as well as their dedication in providing this highly useful
source of information. The content of this article represents solely the work
of the author. None of the contents, including methods, results, discussion,
and conclusions, reflect the views or opinions of the National Academy of
Sciences–National Research Council Twins Committee, the Medical
Follow-Up Agency, which provided these data, the Institute of Medicine,
the National Academy of Sciences, or the National Research Council.
Address correspondence to Dr. Malcolm D. Zaretsky, University of
California, Department of Molecular and Cell Biology, 237 Hildebrand Hall
3206, Berkeley, CA 94720-3206. E-mail: [email protected]
572
ZARETSKY
REFERENCES
1. Hrubec Z, Floderus-Myrhed B, de Faire U, Sarna S. Familial factors in
mortality with control of epidemiological covariables. Swedish twins
born 1886–1925. Acta Genet Med Gemellol (Roma). 1984;33:403–412.
2. Ljungquist B, Berg S, Lanke J, McClearn GE, Pedersen NL. The effect
of genetic factors for longevity: a comparison of identical and fraternal
twins in the Swedish Twin Registry. J Gerontol Med Sci. 1998;53:
M441–M446.
3. Paffenbarger RS Jr, Hyde RT, Wing AL, Hsieh CC. Physical activity,
all-cause mortality, and longevity of college alumni. N Engl J Med.
1986;314:605–613.
4. Fuchs CS, Stampfer MJ, Colditz GA, et al. Alcohol consumption
and mortality among women [see comments] [published erratum
appears in N Engl J Med. 1997;336:523]. N Engl J Med. 1995;332:
1245–1250.
5. Vita AJ, Terry RB, Hubert HB, Fries JF. Aging, health risks, and
cumulative disability [see comments]. N Engl J Med. 1998;338:1035–
1041.
6. Berkman LF, Syme SL. Social networks, host resistance, and mortality:
a nine-year follow-up study of Alameda County residents. Am J
Epidemiol. 1979;109:186–204.
7. Ljungquist B, Sundstrom G. Health and social networks as predictors of
survival in old age. Scand J Soc Med. 1996;24:90–101.
8. Welin L, Tibblin G, Svardsudd K, et al. Prospective study of social
influences on mortality. The study of men born in 1913 and 1923.
Lancet. 1985;1:915–918.
9. Vaupel JW, Yashin AL, Holm N, Harvald B. Survival analysis in
genetics: Danish Twin Data applied to a gerontological question. In:
Klein JP, ed. Survival Analysis: State of the Art. Dordecht, The
Netherlands: Kluwer; 1992:121–382.
10. McGue M, Vaupel JW, Holm N, Harvald B. Longevity is moderately
heritable in a sample of Danish twins born 1870–1880. J Gerontol Biol
Sci. 1993;48:B237–B244.
11. McClearn GE, Svarntengren M, Pedersen NL, Heller DA, Plomin
R. Genetic and environmental influence on pulmonary function
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
in aging Swedish twins. J Gerontol Med Sci. 1994;49:M264–
M268.
Herskind AM, McGue M, Holm NV, Sorensen TI, Harvald B, Vaupel
JW. The heritability of human longevity: a population-based study of
2872 Danish twin pairs born 1870–1900. Human Genet. 1996;97:
319–323.
Hrubec Z, Neel JV. Familial factors in early deaths: twins followed 30
years to ages 51–61 in 1978. Human Genet. 1981;59:39–46.
Carmelli D, Heath AC, Robinette D. Genetic analysis of drinking
behavior in World War II veteran twins. Genet Epidemiol. 1993;10:201–
213.
Service NI. Human Provisional Table on Nutrient Content of Beverages. Washington, DC: U.S. Department of Agriculture; 1982.
Calle EE, Thun MJ, Petrelli JM, Rodriguez C, Heath CW Jr. Bodymass index and mortality in a prospective cohort of U.S. adults [see
comments]. N Engl J Med. 1999;341:1097–1105.
Cox DR, Oakes D. Analysis of Survival Data. London: Chapman and
Hall; 1984.
Collett D. Modelling Survival Data in Medical Research. London:
Chapman and Hall; 1994.
SAS Institute, Inc. SAS/STAT User’s Guide, Version 6. Cary, NC: SAS
Institute, Inc.; 1989.
SAS Institute, Inc. SAS/STAT Software Changes and Enhancement
through Release 6.11. Cary, NC; SAS Institute, Inc.; 1996.
Chapman DW. Proceedings of the American Statistical Association:
Social Statistics Section. 1976: 245–251.
Levy PS, Lemshow S. Sampling for Health Professionals. Belmont,
CA: Wadsworth, Inc.; 1980.
Fisher RA. Statistical Methods for Research Workers. New York:
Hafner; 1973.
Falconer DS, MacKay TFC. Introduction to Quantitative Genetics. 4th
ed. Essex, United Kingdom: Addison Wesley Longman; 1996.
Received October 1, 2002
Accepted November 13, 2002