mortality of twins and singletons yin bun cheung, ph.d. paul yip, ph.d. johan karlberg, m.d., ph.d....
TRANSCRIPT
Mortality of Twins and Singletons
Yin Bun Cheung, Ph.D.
Paul Yip, Ph.D.
Johan Karlberg, M.D., Ph.D.
Readings
Objectives: mortality patterns
• Understand the patterns of gestational age-specific neonatal mortality in twins and singletons.
• Appreciate the (non-) comparability between twins and singletons.
Objectives: statistical models
• Learn a flexible way to handle interaction / effect-modification.
• Interpretation of a (logistic) regression model in different ways.
Mortality by gestational age
• Does the mortality difference between twins & singletons depend on gestational age?
• Do twins and singletons have the same gestational age pattern of mortality?
Difficulties in research
• Sample size
• Referral bias
• Statistical adjustment / matching
A study based on Swedish Medical Birth Registry
• All births born in Sweden in 1982-1995.
• Includes 32,942 twins and 1.5 million singletons.
• Details in Reading 1.
Cumulative distribution of GA
0
25
50
75
100
24 26 28 30 32 34 36 38 40 42
Gestational age (weeks)
Per
cen
t
.
Singletons
Twins
Odds ratios of short GA in twins and singletons
Risk factors Twins Singletons
Mother smoke 1.16* 1.39*
Prior still-birth 0.87 1.57*
Goodness-of-fit P=0.08 P=0.14
* P<0.05
Varying-coefficient model
Mortality impact of twin pregnancy as a function of GA:
Log odds=a1+b1GA+b2GA2+ (a2+c1GA+c2GA2)Twin+...
Varying-coefficient model
As a model of separate curves:
Log odds in singletons:
a1+b1GA+b2GA2
Log odds in twins:
(a1+a2)+(b1+c1)GA+(b2+c2)GA2
Birth weight SDS by GA
-3-2-10123
28 30 32 34 36 38 40 42
Gestational age (weeks)
Mea
n SD
S
SingletonsTwins
Size at birth in twins
Growth restriction in twins concentrates in late pregnancy.
•Physical constraints in utero?
•Limited placental function?
•Selection bias?
Neonatal deaths per 1000
020406080
100120140160180200
28 30 32 34 36 38 40 42
Gestational age (weeks)
Neo
nat
al d
eath
SingletonsTwins
A varying-coefficient logistic regression model
• The impact of twin pregnancy as a quadratic function of GA.
• Adjusted for confounders, e.g. prior still-birth, smoking.
• With & without adjustment for size at birth.
Log OR, fully adjusted
-2
-1
0
1
2
3
4
5
28 30 32 34 36 38 40 42
Gestational age (weeks)
Log
odd
s ra
tio
Log odds, fully adjusted
-6
-4
-2
0
2
4
28 30 32 34 36 38 40 42
Gestational age (weeks)
Log
odd
s
Singletons
Twins
Is twin birth hazardous? Why?
• Twins had lower odds of death prior to 36 weeks of GA; higher odds thereafter.
• Longer GA was related to lower mortality. But the decline was sharper in singletons than in twins.
Log OR, not adjusted for size at birth
-2
-1
0
1
2
3
4
5
28 30 32 34 36 38 40 42
Gestational age (weeks)
Log
odd
s ra
tio
Log odds, not adjusted for size at birth
-6
-4
-2
0
2
4
28 30 32 34 36 38 40 42
Gestational age (weeks)
Log
odd
s
Singletons
Twins
The role of fetal growth and size at birth?
• Without adjustment for size at birth the log OR climbed up faster after around 34 weeks.
• Twins’ mortality reached it’s lowest point at 38 weeks; it turned upward after that.
Clinical interpretations
• Twins have an earlier development
• In utero environment in late pregnancy not good for twins
• Residual confounding
Optimal GA for twins
The optimal gestational age for twins appeared to be 37-39 weeks in terms of neonatal mortality.
This is earlier than singletons’.
Readings1 Cheung YB, Yip P, Karlberg J. Mortality of
twins and singletons by gestational age. Am J Epidemiol 2000;152:1107-16.
2 Lie RT. Intersecting perinatal mortality curves by gestational age -- are appearances deceiving? Am J Epidemiol 2000;152:1117-19.
3 Cheung YB, Yip P, Karlberg J. Respond to “Are appearances deceiving”. Am J Epidemiol 2000;152:1120.