Adverse pregnancy, birth, and infant outcomes in twins: effects of maternal fertility status and infant gender combinations; the Massachusetts Outcomes Study of Assisted Reproductive Technology.

In 2015, there were 133,155 twins born in the United States, accounting for 3.3% of all live births ( 1 ). The twin birth rate in 2015 was 33.5 twins per 1,000 births, a decline from 2014, when the twin birth rate was 33.9, the highest ever reported. It has long been known that iatrogenic multiple births are the most significant complication of assisted conception ( 2 – 4 ). It is estimated that assisted conception accounts for about 40% of twin births (19% from in vitro fertilization (IVF) and 21% from non-IVF assisted conception) and 77% of triplet and higher-order births (25% from IVF and 52% from non-IVF assisted conception) ( 5 , 6 ). Internationally, there is a wide range in the twin delivery rate with IVF using fresh or frozen embryos, respectively, ranging from 5.8% and 4.7% in Sweden to 23.6% and 16.0% in Spain, 27.0% and 35.8% in Greece, 28.6% and 15.0% in Germany, and 35.8% and 21.3% in Bulgaria in 2011 ( 7 ). In comparison, the twin delivery rate in the United States from IVF during this same time period ranged from 34.9% (for fresh cycles using donor oocytes) to 21.6% (for thawed embryos) ( 8 ). Pregnancies to subfertile women, with and without infertility treatment, have been reported to have more complications, lower birthweights, and shorter gestations, even when adjusted for plurality ( 9 – 12 ). There is continued scientific debate regarding the role of parental characteristics, including the etiology of the subfertility ( 13 – 16 ), versus the effect of specific infertility treatments ( 17 – 21 ) in the suboptimal outcomes in these women. In addition, an acknowledged drawback of prior in vitro fertilization (IVF) research in the United States has been the self-reported nature of the outcomes data, which is typically provided by the patient herself or by her obstetrical provider. This study seeks to overcome these limitations by linking the Society for Assisted Reproductive Technology Clinic Outcomes Reporting System (SART CORS) data to birth certificate and hospital utilization data, as well as accounting for fertility status. This analysis is part of a larger population-based study of IVF in Massachusetts ( 15 , 22 – 38 ). In this analysis, we have increased the sample size, expanded the number of adverse outcomes, and further divided the twin group by like gender and unlike gender pairs from the original analysis, based on 2004–08 births, with four adverse outcomes ( 28 ). The objective of this current analysis is to evaluate the effect of maternal fertility status (fertile, subfertile, or IVF) on the pregnancy and birth outcomes in twin live births.

The descriptive statistics of the 10,352 study women by fertility status and twin pair gender group are shown in Table 1 . The characteristics of the subfertile and IVF groups were very similar, with women and their male partners more likely to be older, white, college educated, and have private insurance than those in the fertile group. Women in the subfertile and IVF groups averaged 4–5 years older than their fertile counterparts, and were about six times more likely to be over age 40. Likewise, their male partners also averaged 3–4 years older than partners of fertile women, and were about twice as likely to be over age 40. More than 80% of subfertile and IVF women and their partners were white, compared to about 70% in the fertile group. More than 70–75% of subfertile and IVF women and about 70% of their male partners were college graduates, compared to about 40–45% of their fertile counterparts. More than 90–95% of subfertile and IVF women had private insurance, compared to about 60% in the fertile group. The IVF treatment parameters by twin gender groups are shown in Table 2 . Autologous oocytes were used in 88% of pregnancies, and fresh embryos in 90%, with no difference by gender pair group. Like gender twins were significantly more often the result of a single embryo transferred (1.9% versus 0.2%, p<0.0001). Both gender groups of IVF pregnancies included about 5% of pregnancies with more than two fetal heartbeats at the six week ultrasound, indicating comparable levels of fetal loss.The distribution of infertility diagnoses did not differ by twin gender group; the most frequent diagnoses were male factor (32.1%), and unexplained (23.7%). The results of the bivariate unadjusted analyses of pregnancy, birth, and infant outcomes by fertility status are shown in Table 3 . Women in the subfertile and IVF groups were more likely to be nulliparous, have pre-existing chronic conditions (diabetes and chronic hypertension), develop gestational diabetes and/or pregnancy hypertension, and to deliver by primary cesarean. Placental complications, including uterine bleeding, abruptio placenta, placenta previa, vasa previa, and other excessive bleeding at delivery were more likely in the subfertile and IVF groups. Breech or malpresentation was most likely in the IVF group: 34.0% for all twins, and highest (35.7%) among unlike gender twins. Primary cesarean delivery was greater than 60% in both subfertile and IVF groups, compared to less than 53% in the fertile group, and was highest (66.7%) among IVF unlike gender twins. The infant outcomes of very low birthweight, low birthweight, and small-for-gestation were highest in the fertile group of like gender twins and the subfertile group of unlike gender twins. Mean length of gestation was about 35 weeks for all groups, with comparable rates of early preterm and preterm births. The rate of birth defects was about 2%, with slighter higher rates for the IVF group in both like gender (2.59%) and unlike gender twin groups (2.79%). Neonatal and infant mortality were higher among the subfertile group (1.93% and 2.07%, respectively for all twins), and highest among like gender (2.30% and 2.30%) compared to unlike gender twins (1.81% and 1.51%, respectively). The risks of adverse pregnancy, birth, and infant outcomes by maternal fertility status and infant gender pair groups with the fertile group as reference are shown in Table 4 . Among all twins, the subfertile group had the highest rates of adverse outcomes, 67% (10 out of 15) compared to 20% (3 out of 15) in the IVF group, and 13% (2 out of 15) in the fertile group. Among like gender twins, the proportion of the highest rates were equally divided between the subfertile and IVF group, each with 40% (6 out of 15), compared to 20% (3 out of 15) for the fertile group. Among unlike gender twins, 73% (11 out of 15) of the highest rates of adverse outcomes were in the subfertile group, compared to 27% (4 out of 15) in the IVF group, and none in the fertile group. Among all twins, the risks for all six pregnancy outcomes were significantly increased for the subfertile and IVF groups, with highest risks for uterine bleeding (ARRs of 1.92 and 2.58, respectively), and placental complications (ARRs of 2.07 and 1.83, respectively); these risks were further magnified among like gender twins (ARRs 2.20 and 1.91, respectively). The risks of very preterm, and neonatal and infant death were increased among all twins for the subfertile group (significant ARRs of 1.36, 1.89, 1.87, respectively) and very preterm, preterm, and birth defects for the IVF group (significant ARRs of 1.28, 1.07, and 1.26, respectively).

These analyses of twin pregnancies indicate that compared to fertile women, subfertile and IVF-treated women tend to be older, have more pre-existing chronic conditions, and are at greater risk for adverse pregnancy outcomes, particularly uterine bleeding and placental complications. The twin infants of both subfertile and IVF women are at greater risk for very low birthweight, preterm and very preterm birth; in addition, IVF unlike gender twins are at higher risk for birth defects, and like gender subfertile twins are at higher risk for neonatal and infant death. Some of the difference in risk between subfertile and IVF twin pregnancies may reflect more intensive prenatal monitoring in the latter group ( 44 , 45 ). Compared to fertile women, the risk of having a primary cesarean delivery was higher in subfertile women and highest in IVF women (as was breech or malpresentation), a finding also reported in prior studies and meta-analyses of IVF pregnancies ( 16 , 46 – 49 ). Our findings of an increased risk of preterm birth for IVF twins (ARR 1.07, 95% CI 1.03, 1.12) is nearly identical to the findings from the systematic review by Helmerhorst et al ( 48 ) of 1.07, 95% CI 1.02, 1.13. The nonsignificant risk of neonatal death among IVF twins in our study has also been reported by others ( 48 , 50 ). A consistent finding in our study and prior studies is the increased risk of bleeding and placental complications in subfertile women and IVF-treated women ( 51 , 52 ). Compared to fertile women, the risk of placental complications in our study was higher among IVF women, with significant ARRs ranging from 1.82 to 1.91, and highest among subfertile women, with significant ARRs ranging from 2.00 to 2.20, depending on the gender pair group. Our adverse outcome of placental complications included abruptio placenta, placenta previa or vasa previa, but other placental and cord anomalies not assessed in our study are known to occur more frequently in twin gestations, such as single umbilical artery, velamentous or marginal cord insertion, as well as anomalies unique to twins, such as intraplacental anastomosis and cord entanglement ( 53 , 54 ). Delbaere et al ( 54 ) reported that marginal and velamentous cord insertions, and single umbilical arteries occur more frequently in twins following infertility treatment, increasing in proportion to the invasiveness of the procedure. In their analysis of spontaneous dizygotic twins versus dizygotic twins from assisted conception, the incidence of velamentous cord insertions increased from 3.6% in twins conceived spontaneously to 5% with ovulation induction, 7.4% with IVF, and 10.4% with intracytoplasmic sperm injection (ICSI). Also in their study, the incidence of single umbilical artery increased from 0.6% in spontaneous dizygotic twins to 1.9% with induction of ovulation (AOR 3.19, 95% CI 1.66, 6.11). In general, most studies show little difference in perinatal outcomes in spontaneously-conceived and subfertile or IVF twins, partially due to failure to control for monozygosity and monochorionicity ( 16 , 49 , 55 – 57 ). Two-thirds of twins are dizygotic, with 50% being like gender and 50% unlike gender. About one-third of twins are monozygotic (100% being like gender), among these about 70% are also monochorionic (sharing the chorion)—the group at highest risk for morbidity and mortality. In spontaneously-conceived twin pregnancies, the prevalence of dizygotic twinning varies with race and ethnicity, from a low of 1.3/1,000 live births in Asia to 50/1,000 live births in Africa, as well as increasing with the maternal factors of older age, taller height, higher parity, and family history of twinning ( 58 , 59 ). Prior to the advent of assisted reproductive technologies, the rate of monozygotic twinning was relatively constant worldwide at about 4/1,000 live births, regardless of maternal or familial factors ( 60 ); it has been estimated that the incidence has more than doubled with assisted conception (8–9/1,000 live births) ( 61 , 62 ). In ART treatment, the main risk factor for dizygotic twins and higher-order multiple pregnancies is the transfer of more than one embryo ( 63 , 64 ), as well as taller maternal stature (>68 inches) and higher number of oocytes retrieved (>8) ( 65 ). The risk of monozygotic twinning is increased when culture is extended to the blastocyst stage, and in cleavage stage embryos with assisted hatching ( 62 , 66 – 73 ). A recent case-control study from Canada reported that the use of ovarian stimulators alone and with intrauterine insemination greatly increased the risk of multiple births (AORs of 4.5 and 9.32, respectively) ( 74 ). Our findings in unlike gender twins are in accord with several prior studies of dizygotic or unlike gender twins, including higher rates of early preterm birth and neonatal mortality among twins conceived with ovulation induction or IVF ( 50 , 75 , 76 ). In the Dutch study of 6,964 primiparous women who delivered opposite-gender twins (dizygotic) between 2000 and 2012 ( 76 ), they also found no difference in rates of pregnancy hypertension, small-for-gestational age birthweight, but an elevated risk of perinatal mortality in the subfertile group (significant in their study, not significant in ours). The results of studies of dizygotic twin pregnancies ( 75 ) and unlike gender twin pregnancies ( 50 ), also found higher rates of early preterm birth and perinatal mortality in twin pregnancies from ART compared to those from fertile women, as did our study. Our findings of an increased risk of birth defects and the magnitude of the risk with IVF (ARR 1.26, 95% CI 1.01, 1.59) is in accord with findings from other studies: AOR 1.4, 95% CI 0.9, 2.1 ( 49 ); and AOR 1.26, 95% CI 1.14, 1.40 ( 77 ). The relatively small differences in the risks of adverse pregnancy, birth, and infant outcomes of twins by fertility status reported in our study confirm findings reported by others. Higher plurality, though, is associated with much greater risks of pregnancy, birth, and infant adverse health outcomes compared to singletons. Comparing the outcomes in this study to the analysis of 459,623 singleton births in the MOSART project during this same time period (2004–10) ( 38 ), show the manifold increased risks of twins compared to singletons. Born an average of 3 V weeks earlier and 950 grams lighter, twins were more than ten times as likely to be born very low birthweight, low birthweight, or early preterm, and eight times more likely to be born preterm. The mothers of twins were about twice as likely to develop gestational diabetes, pregnancy hypertension, or uterine bleeding, and were more than four times as likely to be hospitalized prenatally. Placental complications were more than twice as likely with twins and primary cesarean delivery three times more likely. The risk of neonatal death was more than seven times greater and the risk of infant death more than five times greater for twins compared to singletons. The risk of severe maternal morbidity has also been evaluated in the MOSART study ( 34 ). Among IVF-treated women, the risk of severe maternal morbidity was more than threefold higher with twins than with singletons; among all births, the risk of severe maternal morbidity was more than fourfold higher for twins versus singletons (46.5/1,000 twin deliveries versus 10.5/1,000 singleton deliveries). In a nationwide study in the Netherlands, Witteveen et al ( 78 ) also reported a fourfold increased risk of severe maternal morbidity in multiple versus singleton births, as well a twofold increase with assisted reproductive technology. Iatrogenic multiple births are acknowledged as the most important adverse outcome of IVF treatment ( 40 , 79 , 80 ). By periodically issuing national guidelines on the number of embryos to transfer, the Society for Assisted Reproductive Technology has been able to dramatically reduce the rate of higher-order multiples, although the rate of twins remains high ( 1 ). There are a number of preventive measures that should be incorporated into prenatal care to improve the course and outcome of twin gestations. Accurate gestational dating and determination of chorionicity should be done prior to 14 weeks gestation, with more intensified monitoring and earlier planned delivery for monochorionic pregnancies ( 81 – 83 ). Supplementation with folate and multivitamins has been shown to improve outcomes in IVF pregnancies, and reduce the risk for birth defects, as well as velamentous and marginal cord insertions, and prematurity ( 84 – 89 ). Achieving BMI-specific weight gains by specific gestational periods (by 20 weeks, by 28 weeks, and at 36–38 weeks), is associated with better fetal growth, longer gestations, and fewer pregnancy-related complications in twins ( 90 – 93 ). In this study we are comparing outcomes for non-IVF fertility treatment with IVF fertility treatment vs conceptions with no fertility treatment. We don’t really know if it is a comparison of women with subfertile vs fertile women. The fertile group could include women who had subfertility and conceived without treatment. The direction of bias may likely be to lessen the magnitude of any differences seen, if the subfertility is in itself a risk factor for complications which is likely true from other data. The MOSART study, which includes linking IVF cycles to vital records and hospital utilization data, represents the first time these datasets have been linked using direct identifiers from both datasets. IVF national surveillance summaries are limited to birth outcomes reported by the patient herself or her obstetric provider ( 6 , 8 , 40 , 94 , 95 ). Prior studies ( 56 , 94 , 95 ) have relied on linkages between IVF cycles and vital records using only maternal and infant dates of birth, or probabilistic algorithms ( 40 , 95 ). Although there is a high degree of comparability between the SART CORS and vital records ( 36 ), our study design assures more accurate linkage between IVF treatment cycles, vital records, and the hospital discharge data, and a more complete picture of perinatal outcomes. Although this study has several unique advantages over prior IVF research, it is also subject to several limitations. The use of registry data carries the potential risk of misclassification and selection bias. However, the SART CORS variables undergo annual validation ( 6 , 8 , 40 ), and we have additionally validated the SART CORS variables with the MOSART study ( 36 ). This study uses retrospective data from several centralized datasets and although this is advantageous to achieve large numbers, we had the disadvantage that data entered into the SART CORS system is not as rigorously controlled as data collected for a prospective research study. Likewise, the primary purpose of vital records is civil registration, with public health research and surveillance being secondary uses. One of the limitations of comparing our results to the published literature is that the latter is often based on data spanning decades, during which time both IVF procedures and outcomes have improved. Another limitation of this analysis is that it only includes women in Massachusetts. There may be significant demographic and outcome differences in patient populations in other regions of the country and with other healthcare systems, potentially limiting the generalizability of our findings. Another limitation of this study is our inability to identify chorionicity and zygosity in our study population. By grouping twins by gender pair combinations, all monozygotic twins were included in the like gender group, and dizygotic twins were included in the unlike gender group. Approximately half of all like gender twins are monozygotic and half are dizygotic.

Risks of all maternal and most infant adverse outcomes are increased for subfertile and IVF twins. Among all twins, the highest risks were for uterine bleeding and placental complications for the subfertile and IVF groups, and neonatal and infant death in the subfertile group. These findings provide further evidence supporting single embryo transfer and more cautious use of ovulation induction.

We used similar methods to that described in the singleton analysis using this same study population ( 38 ). The study design and setting, data sources, variables, and statistical analysis are similar, except for the additional provisions made for categorizing and analyzing twin pregnancies and twin infant pairs. This longitudinal cohort study included all women with twin live births (both liveborn) of ≥22 weeks gestation and both twins ≥350g birthweight in Massachusetts from July 1, 2004 through December 31, 2010. As a project within the Massachusetts Department of Public Health, the Pregnancy to Early Life Longitudinal (PELL) system links records from birth certificates, hospital discharges, birth defects registry, and program data from child health and development programs. The PELL system has linked information on more than 99% of all births and fetal deaths in Massachusetts from 1998–2010 to corresponding hospital utilization data (hospital admissions, observational stays, and emergency room visits) for individual women and their children, including 1,004,320 deliveries. The Massachusetts Department of Public Health (MDPH) and the Massachusetts Center for Health Information and Analysis are the custodians of the PELL data system, composed of individual databases linked together by randomly-generated unique IDs for mother and infant. The data source for IVF data for this study was the SART CORS, which contains comprehensive data from more than 83% of all clinics performing IVF and more than 91% of all IVF cycles in the United States ( 39 ). Data are collected and verified by SART and reported to the Centers for Disease Control and Prevention in compliance with the Fertility Clinic Success Rate and Certification Act of 1992 (Public Law 102–493). SART maintains HIPAA-compliant business associates agreements with reporting clinics. In 2004, following a contract change with CDC, SART gained access to the SART CORS data system for the purposes of conducting research. The national SART CORS database for 2004–10 contains 930,957 IVF treatment cycles. The data in the SART CORS are validated annually ( 40 ) with some clinics having on-site visits for chart review based on an algorithm for clinic selection. The Massachusetts Outcome Study of Assisted Reproductive Technology (MOSART) project links data from the SART CORS with the PELL data system to evaluate pregnancy and child health outcomes on a population basis. Human subjects approval was obtained from Boston University, Massachusetts Department of Public Health, Dartmouth College, and Michigan State University. The study also had the approval of the SART Research Committee. We constructed the MOSART database by linking the SART CORS and PELL data systems for all Massachusetts births to Massachusetts resident women between July 1, 2004 and December 31, 2010. The starting date was chosen based on the availability of SART CORS data (January 1, 2004) to allow us to capture any births associated with IVF and the end date reflected the latest available linked data of the SART CORS to PELL. A deterministic five phase linkage algorithm methodology was implemented ( 22 ) using mother’s first and last name, mother’s date of birth, father’s name, race of both parents, date of delivery, and number of babies born per delivery. Linked files were later identified by use of a linkage ID from which identifiers were removed. The linkage rate was 89.7 % overall and 95.0 % for deliveries in which both zip code and clinic were located in Massachusetts. The linkage yielded deliveries identified for this study as IVF deliveries . We identified a subfertile group as previously described ( 24 ). Briefly, all Massachusetts deliveries were reviewed for the answer to two questions on the Massachusetts birth certificate about use of fertility drugs and assisted reproduction. Those who answered “yes” to either or both of these questions and had not been identified in the SART CORS linkage were included as subfertile . In addition, any woman who at delivery, or in the 5 years previous to delivery, had been hospitalized with a discharge code of female infertility (ICD diagnosis code 628.0, Infertility-Anovulation, 628.2, Infertility-Tubal Origin, 628.3, Infertility-Uterine Origin, 628.8, Female Infertility of other specified origin, 628.9, Female Infertility of unspecified origin or CPT procedural code V230, Pregnancy With Diagnosis of Infertility) was also included as part of the subfertile group if they were not in the SART CORS linkage. Deliveries not in either the subfertile or ART groups were listed as fertile . In addition, twin pregnancies were classified by the genders of the twin pair as all twins, unlike genders (male and female), or like genders (both male or both female). Independent variables included parental ages, race and ethnicity, education, and payor status at delivery; parity (nulliparous and parous), smoking, maternal pre-pregnancy medical conditions (chronic hypertension and diabetes mellitus); and repeat cesarean delivery, and infant gender ( Table 1 ). Dependent variables included gestational diabetes, pregnancy hypertension, uterine bleeding, placental complications (abruptio placenta, placenta previa, or vasa previa), prenatal hospitalizations, breech/malpresentation at delivery, cephalopelvic disproportion at delivery, other excessive bleeding at delivery, primary cesarean delivery, very low birthweight (VLBW, <1,500 grams), low birthweight (LBW, <2,500 grams), small-for-gestation birthweight (SGA, Z-score ≤−1.28), large-for-gestation (LGA, Z-score ≥1.28), very preterm (<32 weeks), preterm (<37 weeks), birth defects, neonatal death (0–27 days), and infant death (0–364 days). We created composite variables for gestational diabetes, diabetes mellitus, chronic and pregnancy hypertension, and placental complications (placenta previa, abruptio placenta, and vasa previa) using data from the birth certificate and hospital discharge delivery records, using ICD-9 648.8 for gestational diabetes, ICD-9 648.0 or 250 for diabetes mellitus, chronic hypertension as ICD-9 401, 402, 403, 404, or 405, pregnancy-related hypertension as ICD-9 642, placenta previa as ICD-9 641.0 or 641.1, abruptio placenta as ICD-9 641.2, and vasa previa as ICD-9 663.5. The variables of uterine bleeding, breech/malpresentation at delivery, cephalopelvic disproportion at delivery, other excessive bleeding at delivery were derived from birth certificate records in PELL. Factors obtained from the birth certificate included parental ages at delivery, race/ethnicity, and education. Parental age was evaluated as a continuous variable. Parental race/ethnicity was categorized as white, black, Asian, Hispanic, and other. Parental education was categorized as ≤high school or GED (General Education Development diploma), some college or Associate degree, or Bachelor degree or graduate school. Payor status at delivery was a composite of the payor source as reported on the birth certificate and the hospital discharge record. In the multivariate analyses, payor status was categorized as private or public (composite measure of public, self-pay, and free care). Length of gestation was calculated by using the birth certificate delivery date minus date of last menstrual period (LMP) corrected for clinical estimate at early ultrasound. Deliveries prior to 32 weeks gestation were classified as very early preterm, those less than 37 completed weeks gestation were classified as premature, and those which were 37 weeks or greater were classified as term. Birthweight was obtained from the birth certificate. Birthweights at each gestational age are normally distributed, and a z-score (or standard deviation score) is the deviation of the value for an individual from the mean value of the reference population divided by the standard deviation for the reference population ( 41 ). Birthweight z-scores were calculated to evaluate adequacy of weight-for-age using population-based standards, as recommended by Land ( 42 ) and modeled as continuous and categorical variables. We generated gender-, race/ethnicity-, and gestation-specific birthweight means and standard deviations using Massachusetts data for all live births from 1998–2010. Infants with birthweight z-scores ≤−1.28 (below the 10 th percentile for gestation and gender) were classified as small-for-gestational age (SGA) and infants with birthweight z-scores ≥1.28 (above the 90 th percentile for gestation and gender) were classified as large-for-gestation (LGA). Birthweights less than 1,500 grams were classified as very low birthweight (VLBW), and less than 2,500 grams were classified as low birthweight (LBW). The Massachusetts Birth Defects Monitoring Program (BDMP) conducts statewide, population-based active surveillance of birth defects among Massachusetts residents through 1 year of age. The primary focus of the state surveillance system is the identification of major structural birth defects that occur with or without a chromosomal abnormality or other non-chromosomal malformation syndrome. The program’s active surveillance system uses multiple sources of ascertainment, including delivery and specialty care hospitals, and birthing centers. Vital records serve as an additional source of information, providing demographic and clinical information on cases, and acting as an additional source of case-finding. Potential birth defect cases, identified through these varied sources, are assigned to medical record abstractors who review maternal and infant medical records. All cases are coded according to the International Classification of Diseases, Ninth Revision, Clinical Modification, modified British Pediatric Association (ICD-9-CM/BPA) system. Complex cases and cases in which the infant died are reviewed by a clinical geneticist. The birth defects included in the Massachusetts surveillance are ICD-9 CM codes ranging from 740.0 to 759.9 and several other selected codes outside this range for defects such as DiGeorge syndrome, Pierre Robin sequence and amniotic bands. The birth defects included in this analysis have been identified through the BDMP system and linked to each child’s birth data. For women in the IVF group, the frequency of infertility diagnoses and IVF treatment parameters was summarized from the SART CORS ( Table 2 ). Infertility diagnoses included male factor, endometriosis, ovulation disorders, diminished ovarian reserve, tubal factors, uterine factors, other factors, and unexplained infertility. IVF treatment parameters included oocyte source (autologous, donor), embryo state (fresh, thawed), number of embryos transferred (1, 2, or >2), and number of fetal heartbeats at the six week ultrasound exam (1 or >1). We compared maternal and paternal demographic characteristics, pre-pregnancy diagnoses, and perinatal outcomes across fertility groups (fertile, subfertile, and IVF) using generalized linear regression for continuous variables and χ 2 for categorical variables ( Tables 1 and 3 ). The association between fertility status and the six adverse pregnancy outcomes and nine adverse infant outcomes were computed as adjusted relative risk ratios and 95% confidence intervals from multivariate log binomial regression models adjusted for parental ages, race and ethnicity, and education; maternal payor status, pre-existing conditions (diabetes mellitus and chronic hypertension), and parity ( Table 4 ). In a few instances where the models didn’t converge, log-Poisson models were used ( 43 ). Our statistical models were computed using generalized estimating equations (GEE) in order to account for the correlation resulting from the inclusion of outcomes for each twin. Due to the small sample size within twins (and clustered data), and rare occurence of outcome (death), we decided to remove the covariates that resulted in the least change in the strength of association between fertility status and neonatal and infant death. Chronic diabetes was the least significant covariate for deaths and when we tested for the effect of association by introducing potential confounders one at a time, chronic diabetes did not alter the risk ratios of either of the fertility group variables by 10% or more, and therefore was not retained in the model. Models were computed separately using the fertile group as the reference, and the subfertile group as the reference, for all twins, and for unlike gender pairs and like gender pairs. Results were considered significant with p values <0.05 for bivariate unadjusted analyses, and when the 95% confidence intervals did not include 1 in the multivariate analyses. All analyses were performed using the SAS software, version 9.3 (SAS Institute).