Keywords perinatal mortality - stillbirth - infant death - assisted reproductive technology
Data from 2011 to 2015 National Survey of Family Growth (NSFG) suggest that almost
7% of married women of reproductive age (15–44 years) were unable to achieve pregnancy
after at least 12 consecutive months of trying to conceive, and that 12% of women
of reproductive age (over 7 million) had received some type of infertility service,
including testing, medical advice, or treatment to become pregnant.[1 ] Assisted reproductive technology (ART), defined as any procedure in which oocytes
or embryos are handled in the laboratory for the purpose of establishing a pregnancy,
has been increasingly used in the United States since 1981.[2 ] Today, ART-conceived infants account for approximately 1.8% of all infants born
in the United States, with about a third being multiples (twins or higher-order).[3 ]
[4 ]
Although ART treatments are generally considered safe, adverse pregnancy outcomes
such as low birth weight, preterm birth, birth defects, stillbirths, infant deaths,
and maternal deaths may occur at a higher rate compared with spontaneously conceived
pregnancies.[2 ]
[3 ]
[5 ] Researchers in Scandinavia found a two- to fourfold increased risk of stillbirth
delivery among women who conceived using ART as compared with women who conceived
naturally.[6 ]
[7 ]
[8 ]
[9 ] This increased risk of stillbirth delivery was particularly prominent among singleton
births occurring at <28 weeks of gestation.[6 ] Chughtai et al[9 ] also found a significantly higher overall perinatal mortality rate among ART-conceived
births (16.5/1,000 births) than non-ART births (11.3/1,000 births). However, the perinatal
mortality rate was lower for ART births at <32 weeks of gestation compared with non-ART
births. While the latest available data in the United States show that <1% of ART
births resulted in stillbirth in 2016, no information is available on early infant
deaths following ART.[2 ] The purpose of this study was to exam whether ART is associated with increased risk
of perinatal mortality.
Materials and Methods
We conducted a retrospective cohort study using data from the States Monitoring Assisted
Reproductive Technology (SMART) Collaborative. The SMART Collaborative database is
a linked database of the Centers for Disease Control and Prevention's (CDC's) National
ART Surveillance System (NASS) and standardized state vital records (including birth,
fetal death, and infant death certificates) for three states (Florida, Massachusetts,
and Michigan).[10 ]
[11 ]
[12 ] Briefly, the linkage was conducted by using Link Plus, a probabilistic record linkage
program developed by the U.S. CDC with maternal date of birth, infant date of birth
or fetal death, plurality, gravidity, and ZIP codes as linking variables.[10 ]
[11 ]
[12 ] Linkage method for live births has been previously described and validated[10 ]
[11 ]
[12 ]; a similar method was used to link fetal deaths. The SMART Collaborative database
yielded linkage rates of 91% for both birth certificate and infant death certificate
records and 74% for fetal death records. In compliance with the Fertility Clinic Success
Rate and Certification Act, each medical center in the United States that performs
ART procedures is required to report data on every ART procedure and its resultant
pregnancy outcome annually to the CDC.[2 ]
[13 ] The NASS database contains ART cycle-specific information on patient demographics,
medical history, pregnancy history, infertility diagnoses, clinical information pertaining
to the ART procedure, treatment outcomes, and if applicable, pregnancy outcomes.[2 ] State vital records provided the information on live born infants (birth certificate
data), stillborn infants (fetal death of ≥20 weeks of gestation from fetal death report
data), and early neonatal deaths (death of an infant within the first 7 days of life
from infant death certificate data).
We included all ART cycles performed during 2006 to 2011 (the most recent linked SMART
Collaborative data available) that resulted in livebirths, stillbirths, or early infant
deaths in a participating state and that were linked with state vital record data.[10 ]
[11 ]
[12 ] We excluded ART cycles in which gestational carriers were used and pregnancies with
unknown plurality. Stillbirths and infant deaths in the state vital records were classified
as non-ART perinatal deaths if they could not be linked to the NASS database. For
the purpose of this study, we used definition III of perinatal deaths described by
Barfield and the American Academy of Pediatrics Committee on Fetus and Newborn, which
includes fetal death of ≥20 weeks of gestation and infant deaths <7 days of age to
better capture deaths that occurred at early gestational age and around the time of
delivery.[14 ] Fetus or infant was the unit of analysis for the study.
We used the Cochrane-Armitage test to assess trends in ART and non-ART perinatal mortality
rates (number of perinatal deaths/1,000 live births) during 2006 to 2011 by plurality.
We used Pearson's Chi-squared test to examine bivariate associations between selected
characteristics (maternal sociodemographic factors, smoking status during pregnancy,
parity, pre-pregnancy body mass index [BMI], gestational weight gain according to
Institute of Medicine (IOM) recommendations,[15 ] preexisting diabetes and hypertension, and infant sex), and the method of conception
(ART vs. non-ART) among all births (live births and infant deaths) and perinatal deaths
by plurality. All states and territories require reporting of standard maternal demographic
characteristics including pregnancy history, fetal characteristics, medical risk factors
associated with pregnancy, obstetrics, and delivery information. However, pre-pregnancy
BMI and gestational weight gain during pregnancy were not uniformly collected in Massachusetts
during the years of study; therefore, analyses involving these variables are restricted
to Florida and Michigan only.[16 ] For variables that were available in both NASS and vital records, we used information
from vital records.
To address the impact of imperfect linked observations, we weighted the observations
by the probability of successful linkage; we used weighted log-binomial regression
models to calculate crude and adjusted risk ratios (cRR and aRR, respectively) with
95% confidence intervals (CIs) for the association between method of conception (ART
vs. non-ART) and perinatal death, stratified by plurality (singletons vs. multiples)
and gestational age (<28 and ≥28 weeks).[16 ] We selected a priori potential confounders associated with perinatal deaths and
included in the adjusted models: state, age, race, marital status, parity, smoking
status, and preexisting conditions of diabetes and hypertension. In addition, we conducted
subanalyses using data from Florida and Michigan only to determine the effects of
pre-pregnancy BMI and gestational weight gain by including and excluding them from
the model. We also applied weighted propensity scores using ART as the outcome and
maternal characteristics (age, race, education, marital status, smoking status, alcohol
use, and parity) as predictors to correct for potential population bias between the
ART and non-ART groups. We used a generalized estimating equations approach to account
for the correlation of infants born to the same mother. Due to high frequency of missing
values (>10%) for certain variables (pre-pregnancy BMI and gestational weight gain),
we used multiple imputation to estimate these data points, assuming these missing
values were missing at random (MAR).
Statistical analyses were conducted by using SAS, version 9.4 (SAS Institute), SUDAAN
11.0.3 (RTI International), and all p -values <0.05 were considered statistically significant. The institutional review
boards (IRB) of the CDC and the Massachusetts Department of Public Health approved
this study; the Michigan Department of Health and Human Services IRB and the Florida
Department of Health IRB determined that their institutions were not engaged in human
subject research. Table cells with counts less than 10, and cells allowing for calculation
of counts less than 10 were suppressed to protect patient confidentiality.
Results
We identified a total of 39,862 ART-conceived births and 2,495,710 non-ART births
born to Florida, Massachusetts, and Michigan resident mothers during 2006 to 2011.
After excluding cycles using gestational carrier and pregnancies with unknown plurality,
there were 39,824 and 2,263,633 ART and non-ART births, respectively. During 2006
to 2011, there were 570 ART-conceived perinatal deaths and 25,158 non-ART conceived
perinatal deaths in the three states. [Fig. 1 ] presents ART and non-ART perinatal mortality rates by plurality and year. Overall,
for both singleton and multiple deliveries, annual ART perinatal mortality rates were
lower than that for non-ART perinatal mortality. Non-ART perinatal deaths among singleton
gestations generally decreased over time from 10.2 in 2006 to 9.7 per 1,000 live births
in 2011 (p = 0.003), while no significant changes were observed in other groups.
Fig. 1 Assisted reproductive technology and nonassisted reproductive technology perinatal
mortality rates by plurality and year.
In general, the differences of maternal and infant characteristics by ART status for
singleton deliveries were consistent whether looking at all births or perinatal deaths
([Table 1 ]). Among singleton deliveries, most ART perinatal deaths occurred among women older
than 35 years (41.7%), while most non-ART perinatal deaths occurred among women aged
35 years and younger (84.6%). The majority of ART perinatal deaths occurred among
non-Hispanic white women (70.9%), while the majority of non-ART perinatal deaths occurred
among women of other and unknown race/ethnicity (56.1%). Additionally, mothers who
conceived using ART and who experienced perinatal deaths were less likely to smoke
during pregnancy than mothers of non-ART perinatal deaths (<6 and 16.3%, respectively).
Overall, the distribution of maternal education, marital status, pre-pregnancy BMI,
and preexisting health conditions among multiple deliveries (data not shown) were
similar to that of singleton deliveries for ART and non-ART perinatal deaths. However,
among multiples, gestational weight gain was mostly above the IOM recommendations
(41.4 and 39.0% for ART and non-ART, respectively).
Table 1
Maternal and fetal characteristics of all births (live births and fetal deaths) and
perinatal deaths by method of conception (ART vs. non-ART) among singleton deliveries
(2006–2011)
All births
Perinatal deaths
ART
n = 21,426 (1.0%)
Non-ART
n = 2,194,493 (99.0%)
ART
n = 151 (0.7%)
Non-ART
n = 22,307 (99.3%)
State[a ]
[b ]
Florida
7,091
(33.1)
1,167,042
(53.2)
50
(33.1)
12,058
(54.1)
Massachusetts
9,989
(46.6)
405,573
(18.5)
58
(38.4)
2,976
(13.3)
Michigan
4,346
(20.3)
621,878
(28.3)
43
(28.5)
7,273
(32.6)
Maternal age[a ]
[b ]
< 26
367
(1.7)
705,073
(32.1)
NR
NR
7,575
(34.0)
26–30
2,980
(13.9)
680,413
(31.0)
NR
NR
6,302
(28.3)
31–35
7,578
(35.4)
529,609
(24.1)
51
(33.8)
4,988
(22.4)
> 35
10,501
(49.0)
279,398
(12.7)
63
(41.7)
3,442
(15.4)
Maternal race/ethnicity[a ]
[b ]
Non-Hispanic White
16,644
(77.7)
1,245,344
(56.8)
107
(70.9)
9,783
(43.9)
Other/unknown
4,782
(22.3)
949,149
(43.3)
44
(29.1)
12,524
(56.1)
Maternal education[a ]
[b ]
High school or less
4,139
(19.3)
1,254,078
(57.2)
NR
NR
11,383
(51.0)
College
7,009
(32.7)
563,591
(25.7)
36
(23.8)
4,996
(22.4)
More than college
10,145
(47.4)
361,230
(16.5)
82
(54.3)
3,393
(15.2)
Unknown/missing
133
(0.6)
15,594
(0.7)
NR
NR
2,535
(11.4)
Marital status[a ]
[b ]
Married
20,279
(94.5)
1,350,791
(61.6)
138
(91.4)
10,584
(47.5)
Unmarried
NR
841,771
(38.4)
NR
NR
11,130
(49.9)
Unknown/missing
NR
NR
1,931
(0.1)
NR
NR
593
(2.7)
Smoking status during pregnancy[a ]
[b ]
Yes
273
(1.3)
214,586
(9.8)
NR
NR
3,645
(16.3)
No
21,103
(98.5)
1,971,079
(89.8)
142
(94.0)
17,652
(79.1)
Unknown/missing
50
(0.2)
8,828
(0.4)
NR
NR
1,009
(4.5)
Parity[a ]
[b ]
1
13,724
(64.1)
834,835
(38.0)
100
(66.2)
8,089
(36.3)
2
5,573
(26.0)
741,471
(33.8)
24
(15.9)
5,603
(25.1)
≥3
1,997
(9.3)
606,469
(27.6)
NR
NR
7,358
(33.0)
Unknown/missing
132
(0.6)
11,718
(0.5)
NR
NR
1,257
(5.6)
Pre-pregnancy body mass index[a ]
[b ]
[c ]
Underweight
331
(2.9)
63,574
(3.6)
NR
NR
552
(2.9)
Normal weight
5,582
(48.8)
720,067
(40.3)
36
(38.7)
5,999
(31.0)
Overweight
2,357
(20.6)
380,444
(21.3)
22
(23.7)
3,920
(20.3)
Obese
1,495
(13.1)
334,028
(18.7)
19
(20.4)
5,075
(26.3)
Unknown/missing
1,672
(14.6)
290,807
(16.3)
NR
NR
3,785
(19.6)
Gestational weight gain[a ]
[c ]
Below IOM guidelines
3,690
(32.3)
586,575
(32.8)
32
(34.4)
8,192
(42.4)
Within IOM guidelines
3,203
(28.0)
440,936
(24.7)
20
(21.5)
3,392
(17.6)
Above IOM guidelines
3,973
(34.7)
665,179
(37.2)
NR
NR
4,977
(25.8)
Unknown/missing
571
(5.0)
96,230
(5.4)
NR
NR
2,770
(14.3)
Gestation weeks[b ]
< 28
194
(0.9)
19,874
(0.9)
62
(41.1)
11,501
(51.6)
≥28
21,173
(98.8)
2,169,883
(98.9)
89
(58.9)
10,655
(47.8)
Unknown/missing
59
(0.3)
4,736
(0.2)
0
0
116
(0.7)
Gender
Male
11,057
(51.6)
1,124,321
(51.2)
76
(50.3)
11,852
(53.1)
Female
10,369
(48.4)
1,069,901
(48.8)
75
(49.7)
10,198
(45.7)
Unknown/Missing
0
0
271
<0.1
0
0
257
(1.2)
Preexisting hypertension[a ]
Yes
420
(2.0)
31,456
(1.4)
NR
NR
983
(4.4)
No
20,882
(97.5)
2,151,510
(98.0)
147
(97.4)
20,667
(92.7)
Unknown/missing
124
(0.6)
11,527
(0.5)
NR
NR
657
(3.0)
Preexisting diabetes[a ]
Yes
1,625
(7.6)
117,496
(5.4)
NR
NR
1,353
(6.1)
No
19,677
(91.8)
2,065,473
(94.1)
144
(95.4)
20,297
(91.0)
Unknown/missing
124
(0.6)
11,524
(0.5)
NR
NR
401
(3.0)
Abbreviation: ART, assisted reproductive technology; IOM, Institute of Medicine; NR,
not reported to protect patient confidentiality.
a Statistically significant with p < 0.05 for all births.
b Statistically significant with p < 0.05 for perinatal deaths.
c Limited to Florida and Michigan only, information not available in Massachusetts.
Note: Analysis excludes maternal age <20 or >60 years, and unknown gestation type
(singletons vs. multiples).
[Table 2 ] presents the risk ratios for the association of perinatal death and ART use, stratified
by plurality (singletons vs. multiples), and gestational age (<28 and ≥28 weeks).
The crude or unadjusted risk of perinatal death was significantly lower among ART
births than among non-ART births for all singleton and multiple deliveries (crude
risk ratio [cRR] = 0.84, 95% CI: 0.71–0.98 and cRR = 0.68, 95% CI: 0.59–0.78, respectively),
as well as for singleton births <28 weeks (cRR = 0.42, 95% CI: 0.31–0.57) and multiple
births <28 weeks (cRR = 0.71, 95% CI: 0.58–0.87; [Table 2 ]). After adjusting for state of birth, maternal age, maternal race, parity, marital
status, smoking status, and preexisting diabetes and hypertension, the risk of perinatal
death remained significantly lower for singleton ART versus non-ART births <28 weeks
(aRR = 0.46, 95% CI: 0.26–0.85) and for multiple births <28 weeks (aRR = 0.64, 95%
CI: 0.45–0.89). When analyzing data from only Florida and Michigan, the results were
similar regardless of including or excluding pre-pregnancy BMI and gestational weight
gain in the adjusted model (results not shown).
Table 2
Association between perinatal deaths and ART use, by plurality and gestational age
(in weeks) among all births in Florida, Massachusetts, and Michigan (2006–2011)
ART births
n
ART perinatal deaths
n (%)
Non-ART births
n
Non-ART perinatal deaths
n (%)
cRR (95% CI)
aRR[a ] (95% CI)
Singletons
21,367
151 (0.71)
2,189,757
22,307 (1.01)
0.84 (0.71–0.98)
1.27 (0.90–1.78)
< 28 wk
194
62 (31.96)
19,874
11,501 (57.87)
0.42 (0.31–0.57)
0.46 (0.26–0.85)
≥28 wk
21,173
89 (0.42)
2,169,883
10,655 (0.49)
1.02 (0.82–1.25)
1.29 (0.80–2.06)
Multiples
18,348
420 (2.29)
68,543
2,851 (4.16)
0.68 (0.59–0.78)
0.90 (0.72–1.12)
< 28 wk
856
293 (34.23)
4,380
2,132 (48.68)
0.71 (0.58–0.87)
0.64 (0.45–0.89)
≥28 wk
17,492
127 (0.72)
64,613
705 (1.09)
0.82 (0.67–1.00)
1.15 (0.77–1.72)
Abbreviations: aRR, adjusted risk ratio; ART, assisted reproductive technology; CI,
confidence interval; cRR, crude risk ratio.
a Model was adjusted for state, age, race, marital status, parity, smoking status,
and preexisting conditions of diabetes and hypertension and excluded perinatal deaths
with unknown gestational age.
Discussion
This study compared the risk of perinatal deaths between ART and non-ART conceived
pregnancies in three U.S. states. Over the study period, the perinatal mortality rate
was lower for ART than non-ART conceived perinatal deaths for both singleton and multiple
infants. After controlling for confounders, we found a protective association between
ART use and perinatal death for births occurring at a gestational age of <28 weeks
regardless of plurality. Notably, there was no difference in perinatal mortality risk
from 28 weeks of gestation, and later, suggesting that the protective effect seen
at earlier gestational ages may be due to differences in obstetric management of ART
conceived pregnancies.
Our findings are similar to the study conducted by Chughtai et al, which showed a
lower perinatal mortality rate among ART births compared with non-ART births for preterm
births (<32 weeks of gestation) from a population-based cohort in Australia.[9 ] However, studies conducted in the Nordic countries had different findings, showing
ART use to be associated with an increased risk of perinatal death and stillbirth
(particularly among singleton deliveries).[6 ]
[7 ]
[8 ] This lack of consistency may be due to the use of different definitions for perinatal
death, different ART practices for different study years, differences in health care
system or obstetric care, and variations in the study populations, such as demographic
characteristics, ART treatment characteristics, and utilization of prenatal care.
For example, we used definition III of perinatal deaths by Barfield et al, which includes
fetal deaths ≥20 weeks and infant deaths <7 days after birth, while Committee on Nordic
ART and Safety used fetal deaths ≥20 weeks to infant deaths 27 days after birth.[6 ]
[7 ]
[14 ] This difference in definition could also explain the higher perinatal mortality
rate in the non-ART group than that of the national average in the three participating
states.[17 ] Although study populations vary, the differences of certain maternal characteristics,
such as maternal age, race, education, marital status, smoking status, and parity,
were consistent between ART and non-ART perinatal deaths.[18 ]
[19 ]
[20 ] ART practice varies by country due to differences in health care access, health
insurance coverage, age limitations, procedure limitations, laboratory protocols,
and in some cases, limitations on the number of embryos that can be transferred. Thus,
ART pregnancy outcomes can vary accordingly.[2 ]
[21 ]
[22 ]
[23 ] Also, differences in infertility diagnosis, severity of diagnosis, and the presence
of other preexisting medical conditions among mothers can result in different pregnancy
outcomes. Wisborg et al concluded that the increased risk of stillbirth after IVF
may be due to the infertility treatment or other factors associated with subfertility.[7 ]
Higher proportions of adequate prenatal care among ART-conceived pregnancies may have
led to earlier detection and management of fetal and maternal conditions, and could
explain the lower rates of early perinatal deaths observed in this study.[18 ]
[19 ]
[20 ] Although NASS captures approximately 98% of all ART cycles performed in the United
States, NASS does not collect information on maternal characteristics such as pre-pregnancy
BMI, weight gain during pregnancy, or preexisting medical conditions.[2 ] By linking NASS with vital records, we were able to control for these possible confounding
factors. Additionally, using multiple imputation for missing values, we were able
to conduct the analysis without reducing our sample size.
Our findings are subject to several limitations. The adoption of the 2003 standard
by the states in the study was not uniform across the study period. This limited the
analysis to vital statistics variables, which were collected uniformly by the three
states for the full study period with exceptions as noted in methods, such as pre-pregnancy
BMI and gestational weight gain. The validity and quality of fetal death records in
the United States are inconsistent because there are no standard reporting requirements.[11 ]
[12 ]
[14 ]
[24 ]
[25 ] Fetal deaths can be misclassified as spontaneous abortion or vice versa, particularly
for loss that occur during early pregnancy (20–27 weeks); similarly, misclassification
between stillbirth, and infant deaths can occur among births of late gestational age
(≥28 weeks). The study excludes pregnancies reaching 20 weeks, which can introduce
selection bias if ART pregnancies are more likely to miscarry in early pregnancy.[26 ] However, previous data show no difference in miscarriage rates for ART versus non-ART
births, suggesting that the bias may be minimal but cannot be discounted.[2 ]
Due to missing or inconsistent information in vital records, the use of a probability
linkage algorithm can result in possible mismatches and lower linkage rates.[11 ] Certain variables on the fetal death reports or infant death certificates have consistently
high levels of missing data, such as maternal BMI, pregnancy weight gain, and utilization
of prenatal care, as some of these variables are self-reported and parents may be
reluctant to share demographic or pregnancy information following a fetal death.[11 ]
[27 ]
[28 ] While there is no clear evidence that ART-conceived pregnancies are more likely
to deliver before 28 weeks than spontaneously conceived pregnancies, stratifying by
gestational age may have resulted in collider bias.[29 ] As such, the stratified estimates should be interpreted with caution, and causation
should not be inferred. Information on cause of death, placentation, or chorionicity
for multiple gestations; prenatally diagnosed birth defects; and subfertility factors,
although not available for the current analysis, could potentially explain the differences
of perinatal death rates by gestational age (<28 vs. ≥28 weeks) and use of ART. We
are also unable to determine the effects of non-IVF fertility treatments that women
may have received, as these data are not contained in the SMART linked database. Today,
there is no scientific way to differentiate the absolute exposure and effects of ART
treatments among the pregnant population at a certain time; residual effects from
previous fertility treatments for previous pregnancies, births, or failed attempts
at pregnancy are also possible.[30 ] Due to limited sample size, we were unable to compare rates of perinatal mortality
stratified by shorter intervals of gestational age; however, we have no reason to
suspect our findings are inaccurate. Likewise, due to the small numbers of perinatal
deaths among triplets and higher order gestations, we analyzed a combined population
of twins and higher order gestations. Nevertheless, the results remained unchanged
when we restricted the study population to either twin (results not shown). Some nonresident
patients were likely included in our dataset, but such numbers are thought to be small
as the participating states provided birth certificate data only from mother's resident
state for majority of the study period. In addition, numbers without known state of
residence were very small and therefore not likely to bias our results. Finally, SMART
Collaborative data, because it only encompasses three states, may not be generalizable
to the national population.
Perinatal death is an important public health concern.[18 ]
[19 ] This study used linked surveillance and vital records data to examine associations
between use of ART and perinatal deaths at the population level. Our findings suggest
that ART use may be associated with a decreased risk of perinatal deaths prior to
28 weeks of gestation, which may be explained by earlier detection and management
of fetal and maternal conditions among ART-conceived pregnancies. This information
adds to the growing body of research on ART-associated perinatal outcomes and can
inform patient counseling on treatment risks.
