As a result of unfavorable in utero environment, the developing fetus reduces its metabolic demands as an adaptive mechanism for survival. This physiologically engineered response could be advantageous in the short-term (enhancement of in utero survival probability) but could lead to later adverse consequences. Barker and co-workers1–4 demonstrated a link between perturbation of growth in utero (in utero growth restriction) and elevated risk for adult-onset disease later in life.
However, most studies of growth restricted or small for gestational age (SGA) infants have almost exclusively considered an SGA infant as an independent and homogeneous entity. This notion may have narrowed our understanding of SGA as a marker of subsequent adult-onset disease. For instance, it is increasingly understood that a woman’s prior pregnancy experience influences the risk of adverse pregnancy outcome in subsequent pregnancy (eg, prior SGA is a risk factor for subsequent SGA).5–8 Using this model of propensity to repeat or recall prior pregnancy events, one could easily delineate two subtypes of SGA infants in a subsequent gestation. In one, an AGA-primed uterus now programs an SGA-infant, while in the other, an SGA-primed uterus now programs an SGA-infant. These two types of SGA infants may not be homogeneous entities, although data in this respect is lacking. Consequently, we undertook this study to evaluate SGA recurrence compared with SGA as a nonrecurrent phenomenon guided by the following working hypotheses:
- The initiation of pregnancy with an SGA infant (first pregnancy) enhances the likelihood of recurrence of an SGA infant in a subsequent pregnancy (second pregnancy).
- The initiation of a pregnancy with an SGA infant (first pregnancy) enhances the level of success of subsequent adaptive processes of fetal programming (SGA infant) compared with AGA in the first pregnancy with subsequent SGA events. In other words, early mortality outcomes for the second birth in an SGA–SGA sequence will be lower than for the second birth in an AGA–SGA sequence.
For the purpose of a clearer description, we define a term “fetal programming switch” to refer to nonconcordance of first compared with second pregnancy with respect to fetal growth status at birth. Thus, AGA–SGA and SGA–AGA sequences represent cases of fetal programming switch, whereas concordant infants (AGA–AGA and SGA–SGA sequences) are denoted for the purpose of this paper as “nonswitches.”
MATREIALS AND METHODS
This analysis was conducted using the Missouri maternally linked cohort data files covering the period 1978 through 1997 inclusive. The data source contains information on both live birth and fetal death for each sibling and provides a platform for a longitudinal study of birth outcomes for each pregnancy. In this data set, siblings are linked to their biologic mothers using unique identifiers. The methods and algorithm used in linking birth data into sibships and the process of validation have been described in detail previously.9 The Missouri vital record system is a reliable one that has been adopted as “gold standard” to validate U.S. national data sets that involve matching and linking procedures.10
For the purpose of our study we selected singleton pregnancies within the gestational age range of 20–44 weeks. Further inclusion and exclusion criteria were applied as detailed in Figure 1. Gestational age was largely based on the interval between the last menstrual period and the date of delivery of the neonate (95% cases). When the menstrual estimate of gestational age was inconsistent with the birth weight (eg, very low birth weight at term), a clinical estimate of gestational age on the vital records was used instead.11 The accuracy of using gestational age as reported on the U.S. birth certificate has previously been validated.12 In that validation study, the authors assessed the concordance between date of last menses reported among cases (very low birth weight infants or neonatal deaths) and randomly selected noncase population. There was a very good agreement (84.2%) with medical records. Clinical estimate of gestation in completed weeks was 79.0% concordant for cases, and 94.0% for controls. This has recently been found to be consistent with an anticipated ±2 weeks variation in date of last menses.13 Hence, gestational age as used in this study is reasonably valid.
We then categorized the study population into an exposed group consisting of those women who experienced an SGA live birth in the first pregnancy and a comparison group (those that had a non-SGA live birth in the first pregnancy). Small for gestational age was defined as less than 10th percentile of birth weight for gestational age using population-based national reference curves for singletons.14 For simplicity of expression, we denote AGA-primed uterus (first pregnancy with AGA infant) with subsequent AGA or SGA infant as AGA–AGA and AGA–SGA, respectively. In a similar fashion, SGA-primed uterus (first pregnancy with SGA-infant) with subsequent AGA or SGA infant is described as SGA–AGA and SGA–SGA, respectively.
We further determined the success of nonrecurrent SGA or AGA (a fetal programming switch) using neonatal death (death of the newborn within the first 28 days of life) as the primary outcome of interest. We also computed secondary outcomes comprising infant death (death of the infant from day 0 through 364) and postneonatal death (death of the infant from day 28 through 364). We selected neonatal death as the primary outcome because it is more closely related to pregnancy-associated events (eg, fetal programming) than the secondary outcomes (infant and postneonatal death). Interest lies in determining within a given situation where fetal programming switch occurred in the second pregnancy whether the resulting infant bears the same expected risk of mortality as a similar infant with equivalent growth profile at birth who is concordant with the first pregnancy (ie, no switch). For instance, does an AGA infant in the second pregnancy resulting from a switch (ie, first pregnancy was SGA) bear the same risk of mortality as another AGA infant of a second pregnancy without a switch (ie first pregnancy was AGA)? Similarly, does an SGA infant resulting from a switch (first pregnancy was AGA) bear the same risk level of mortality as another SGA infant of a second pregnancy that is concordant with the first pregnancy (ie, first pregnancy was SGA)?
It is well-known that certain maternal characteristics (eg, age of the mother) could influence birth outcomes in general. However, because the end point of interest in this study (namely, success of fetal programming in the second pregnancy) was more likely to be influenced by the maternal characteristics in the second pregnancy than in the first, we compared the following sociodemographic characteristics between the two groups in the second pregnancy, and adjusted for them in subsequent multivariable analyses: maternal age (younger than 35 years and 35 years or older), educational level attained (less than 12 years and 12 years of more), race/ethnicity (black, white), marital status (married, single), reported use of tobacco during pregnancy (yes, no), adequacy of prenatal care (adequate, inadequate) and body mass index (BMI, expressed as kg/m2). We categorized BMI into four groups: underweight (less than 19.80), normal (19.80–26.00), overweight (26.10–29.00) and obese (more than 29.00) based on previous publication.15 Adequacy of prenatal care was assessed using the revised graduated index algorithm.16,17 The revised graduated index assesses the adequacy of care based on three variables (trimester prenatal care began, number of visits, and the gestational age of the infant at birth).
We performed crude frequency comparisons for the presence of common obstetric complications, namely, anemia, cardiac disease, type 1 diabetes mellitus, other types of diabetes mellitus, chronic hypertension, preeclampsia, eclampsia, abruptio placenta, and placenta previa. The documentation of these morbidities on birth certificates became official as from 1989 in the United States. For this reason, comparison was restricted to the period 1989 through 1997 only.
Neonatal, postneonatal and infant mortality rates were computed by dividing the number of deaths by the total live births and multiplying by 1,000. The χ2 test was used to evaluate differences in sociodemographic characteristics and maternal pregnancy complications in the second pregnancy between the two groups (SGA compared with AGA-primed uterus). We applied the Student t test to determine differences in means between the two groups with respect to continuous outcomes. We used unconditional logistic regression to generate adjusted odds ratios to approximate relative risks. The following factors were loaded and retained in the model based on biologic plausibility and the literature: maternal age, race, smoking habits during pregnancy, maternal education, marital status, body mass index, adequacy of prenatal care, interpregnancy interval, infant gender, and year of birth. We assessed goodness-of-fit of models using the –2 log likelihood ratio test, and we estimated the significance of main effects using the Wald test.18 The inclusion of interaction terms did not improve model fit, and the model loaded with the aforementioned covariates alone were retained for the analyses.
All tests of hypothesis were two-tailed, with a type 1 error rate fixed at 5%. We used SAS 9.1 (SAS Institute, Cary, NC) to perform all analyses. This study was approved by the Office of the Institutional Review Board at the University of South Florida.
Of the total 306,114 primigravidas, 7.99% (24,453) had SGA infants in the second pregnancy. The proportion of SGA infants was significantly greater among mothers with a history of SGA in the first pregnancy (24.33% compared with 5.99%)(P<.01). The proportion of each phenotype combination (first and second pregnancy) is given in Table 1. The adjusted risk for the occurrence of an SGA infant was also about four times as likely for women with an SGA infant in the first pregnancy (odds ratio [OR] 4.10; 95% confidence interval [CI] 4.00–4.20). This adjusted risk estimate remained unchanged even after controlling for the effects of pregnancy-associated complications (OR 4.10; 95% CI 3.90–4.30).
Table 2 provides a summary of frequency comparison between mothers who initiated their reproductive function with AGA infants compared with those who initiated theirs with an SGA infant with respect to sociodemographic characteristics in the second pregnancy. Mothers who experienced an SGA infant in their first pregnancy were more likely to be African American and unmarried and to have attained a lower educational level as compared with mothers who initiated their reproductive function with an AGA infant. Mothers with SGA infants in the first pregnancy were also more likely to be smokers and to have received a lower than expected level of prenatal care. However, mothers who initiated pregnancy with an AGA infant tended to be older and of higher body mass index than their counterparts with SGA experience in the first pregnancy.
For the entire study sample the mean interpregnancy interval was 929 days (2.5 years). Mothers with SGA-primed uterus had a higher mean interpregnancy intervals than did mothers with AGA-primed uterus (mean±standard deviation 942±819 days compared with 928±763 days, respectively; P<.01). The prevalence of anemia, chronic hypertension, preeclampsia, eclampsia, and placental abruption was greater in the second pregnancy among women with SGA-primed than AGA-primed uterus (Table 3). On the other hand, women who had AGA infants in the first pregnancy had a higher prevalence of other types of diabetes during pregnancy. The levels of type 1 diabetes mellitus and placenta previa were similar in the two groups.
A total of 2,065 cases of infant mortality occurred among infants born to both groups of mothers in the second pregnancy, representing an infant mortality rate of 6.75 per 1,000. This could be segregated into neonatal death counts of 1,124 (equivalent to a neonatal mortality rate of 3.67 per 1,000) and postneonatal death counts of 941 (equivalent to a postneonatal mortality rate of 3.07 per 1,000). A breakdown of these rates according to whether a shift in the intrauterine fetal growth experience occurred (AGA–SGA, SGA–AGA sequences) or not (AGA–AGA, SGA–SGA sequences) is illustrated in Figure 2. The figure is so constructed as to compare mortality outcomes among infants with nonrecurrent fetal growth pattern (in the second pregnancy) compared with those with recurrent fetal growth experience (namely, SGA–AGA compared with AGA–AGA and AGA–SGA compared with SGA–SGA).
We provide adjusted estimates for the association between the occurrence of fetal programming switch and likelihood of death in the first year of life. Model 1 provides adjusted odds ratio with AGA–AGA infants as referent category for all estimates. By contrast, model 2 compares outcomes of the second pregnancy of AGA-primed compared with SGA-primed uterus based on whether a switch occurred or not in the second pregnancy. The referent in each case is the corresponding nonswitch category with equivalent programming status as the nonrecurrent pattern for the second pregnancy. We further designated a model as “B” if mothers with obstetric and medical complications were excluded from the analysis; otherwise the model bears the designation “A.” Because information on obstetric and medical complications were only routinely collected in the United States as from 1989, the analysis for models B were restricted to that period only.
In model 1A (Table 4), the greatest risk of mortality was among AGA–SGA infants, which was most pronounced for neonatal mortality (OR 5.45; 95% CI 4.68–6.34). When women with complications during the second pregnancy were excluded from the analysis (model 1B), the risk for neonatal mortality among AGA–SGA infants rose to more than eightfold (OR 8.40; 95% CI 6.16–11.46).
Model 2A (Table 5) illustrates mortality risk differential between infants of second pregnancy with associated fetal programming switch in comparison with their counterparts who did not experience switch but were otherwise comparable in growth status. Appropriate for gestational age infants from SGA-primed uterus had a 19% and 29% greater likelihood of infant and neonatal mortality, respectively, when compared with AGA-infants from AGA-primed uterus. Approximately the same magnitude of risk elevation for neonatal and infant mortality was noted among SGA infants resulting from AGA-primed as compared with SGA infants from SGA-primed uterus. When pregnancy-associated complications were accounted for by excluding women with complications of pregnancy from the adjusted model 2B, the results were approximately same.
We initiated this study with two main hypotheses. Our first hypothesis posits that the initiation of pregnancy with an SGA infant enhances the likelihood of another SGA infant to occur in a subsequent pregnancy. In this study we observed a fourfold elevation in the risk of SGA in the second pregnancy given that the first pregnancy was an SGA infant (5.99% compared with 24.33%). In a study encompassing 3,505 sib pairs of full-term singleton neonates, the risk for SGA delivery after adjusting for potential confounders was four times greater when an older sibling was also SGA, as compared with an older sibling of normal size.6 This is in perfect agreement with the findings in this study.
The repetition of SGA may result from several factors, including pregnancy-associated complications, prenatal smoking, or genetic determination.7,8 As we recently showed, the familial aggregation of SGA is also a product of the contribution of maternal and paternal factors.19 In that study, the risk of SGA offspring was 4.7 times greater for mothers and 3.5 times greater for fathers who were SGA themselves, as compared with appropriate for gestational age parents. The combined maternal and paternal effect on subsequent SGA in the offspring was equivalent to the product of the independent parental effects on a multiplicative scale. Our results in this article confirm that even after taking into account the role of maternal risk factors and pregnancy complications, the risk for SGA repetition remained unchanged. However, a limitation in our analysis is that we were unable to assess parental genetic contribution to SGA repetition due to unavailable information on the relevant variable.
In the second hypothesis, we proposed that the initiation of a pregnancy with an SGA infant enhances the level of success of subsequent adaptive processes of fetal programming (SGA infant) as compared with AGA initiation with subsequent SGA events. We could not reject this hypothesis because our results showed that AGA-primed mothers who switched to an SGA programming in the second pregnancy experienced a 29% and 16% higher likelihood of neonatal and infant demise. Although our original hypothesis in this respect was to examine the success of program switch and nonswitch for SGA infants only in the second pregnancy (ie, AGA–SGA compared with SGA–SGA respectively) further analysis revealed that the occurrence of a switch in the second pregnancy as a universal phenomenon was less advantageous than a nonswitch of equivalent growth profile. This is because even among AGA infants in the second pregnancy one could delineate two subtypes; switches (SGA–AGA) and nonswitches (AGA–AGA). The risk differential for neonatal and infant mortality was also 29% and 19% greater among AGA switch (SGA–AGA) than AGA nonswitch (AGA–AGA), exactly the same magnitude as for SGA subtypes mentioned earlier.
A possible explanation for the observed relative failure of a program switch is the preponderance of adverse maternal risk factors and pregnancy complications in the second pregnancy associated with the switch. It is likely that a woman who initiated her reproductive function with an AGA infant but experienced a switch in the second pregnancy (now changed to an SGA program) probably had more pregnancy complications in the second pregnancy than an SGA–SGA nonswitch. The greater frequency of the pregnancy-associated complications in such a case would explain both the switch and the subsequent switch failure. However, when we controlled for the confounding effects of pregnancy complications by restricting our analyses to women without complications of pregnancy, the risk of infant mortality remained the same, whereas that of the primary indicator of programming failure (neonatal mortality) increased to 55% among SGA switches (AGA–SGA) as compared with nonswitches (SGA–SGA). A similar albeit less pronounced elevation in neonatal risk estimate was also noted among AGA switches (SGA–AGA) as compared with nonswitches (AGA–AGA) after adjustment for pregnancy-associated complications. Thus, the heightened risk of neonatal and infant mortality among fetal program switches is not explained by adverse maternal characteristics and pregnancy complications in the second pregnancy.
An SGA–SGA phenotype is more likely to be constitutional or genetically determined than a sporadic phenotype resulting from a switch (AGA–SGA), and the latter could perhaps be more likely to be associated with underlying fetal morbidity or pregnancy complications. Although this could easily explain an SGA switch, the notion may not be applicable to an AGA switch.
A third explanation is the concept of event memory hypothesis, a theory recently advanced to explain the tendency of pregnancy outcomes (eg, fetal loss repetition) to recur.20 Using molecular analogous models the theory postulates that when an event (eg, fetal death) occurs during pregnancy, that event is retained (memorized) as a program that is replayed in future pregnancies. Assuming this theory to be correct and because in real life memorized events tend to be recalled more easily and with greater efficiency, one would, therefore, expect an AGA-primed uterus to be apt at programming a subsequent AGA-fetus as compared with an SGA-primed uterus. By the same token, one would expect an SGA-primed uterus to be more efficient in programming a subsequent SGA fetus as compared with an AGA-primed uterus. Hence, the event memory hypothesis/theory explains the relative failure of switches as compared with nonswitches in this study. The increased risk of mortality among switches as observed in our study is therefore, attributable to suboptimal event memory recall leading to less than expected efficiency in fetal programming.
An important limitation of the data to bear in mind in making interpretations is the long period of follow-up of these women, which spanned almost 20 years. Different infant cohorts were aggregated and analyzed together. Because these neonates were exposed to varying obstetric practices across the period of study, the results we have presented might have been biased by this cohort effect. However, by controlling for year of birth in computing adjusted relative risk estimates the effect of this potential source of bias on our results must have been minimized considerably. Additionally, subanalyses for the period cohorts 1978–1988 and 1989–1997 yielded similar results. Another important issue is the exclusion of nonviable fetuses from the analyses. Gestational age estimation of nonviable infants may be problematic, and if these were to be included in this study, they would have induced biases that could be complicated to decipher as a result of the unstable and high rates of mortality coupled with elevated potential for erroneous estimation of gestational age in this high-risk group.21
A strength of this study is that it is population wide, and the results are therefore minimally affected by selection biases (eg, referrals, etc), a source of concern in data derived from individual health facilities. The advantage is that the findings are reasonably generalizable. Another merit of this work is that it adds new information to a domain that is new and poorly understood (the concept of fetal programming switch is novel). Nevertheless, our findings should not be construed as definite but rather as impetus for more refined studies that will potentially offer answers to many questions emanating from this study.
In summary, we found fetal programming switch in subsequent gestation to be inefficient in terms of early survival of affected infants. This novel finding should incite more in-depth research regarding the mechanisms involved, because enhanced understanding of the phenomenon could potentially assist in future counseling and in formulation of appropriate strategies to enhance survival.
1. Hales CN, Barker DJ, Clark PM, Cox LJ, Fall C, Osmond C, et al. Fetal and infant growth and impaired glucose tolerance at age 64. BMJ 1991;303:1019–22.
2. Ravelli AC, van der Meulen JH, Michels RP, Osmond C, Barker DJ, Hales CN, et al. Glucose tolerance in adults after prenatal exposure to famine. Lancet 1998;351:173–7.
3. Barker DJ, Osmond C. Infant mortality, childhood nutrition, and ischaemic heart disease in England and Wales. Lancet 1986;1:1077–81.
4. Barker DJ, Osmond C, Golding J, Kuh D, Wadsworth ME. Growth in utero, blood pressure in childhood and adult life, and mortality from cardiovascular disease. BMJ 1989;298:564–7.
5. Salihu HM, Sharma PP, Aliyu MH, Kristensen S, Grimes-Dennis J, Kirby RS, et al. Is small for gestational age a marker of future fetal survival in utero? Obstet Gynecol 2006;107:851–6.
6. La Batide-Alanore A, Tregouet DA, Jaquet D, Bouyer J, Tiret L. Familial aggregation of fetal growth restriction in a French cohort of 7,822 term births between 1971 and 1985. Am J Epidemiol 2002;156:180–7.
7. Wollmann HA. Intrauterine growth restriction: definition and etiology. Horm Res 1998;49 suppl:1–6.
8. Williams RL. Intrauterine growth curves: intra- and international comparisons with different ethnic groups in California. Prev Med 1975;4:163–72.
9. Herman AA, McCarthy BJ, Bakewell JM, Ward RH, Mueller BA, Maconochie NE, et al. Data linkage methods used in maternally-linked birth and infant death surveillance data sets from the United States (Georgia, Missouri, Utah and Washington State), Israel, Norway, Scotland and Western Australia. Paediatr Perinat Epidemiol 1997;11 suppl:5–22.
10. Martin J, Curtin S, Saulnier M, Mousavi J. Development of the matched multiple birth file. In: 1995–1998 matched multiple birth dataset. NCHS CD-ROM series 21, no. 13a. Hyattsville (MD): National Center for Health Statistics; 2003.
11. Taffel S, Johnson D, Heuser R. A method of imputing length of gestation on birth certificates. Vital Health Stat 2 1982;(93):1–11.
12. Piper JM, Mitchel EF Jr, Snowden M, Hall C, Adams M, Taylor P. Validation of 1989 Tennessee birth certificates using maternal and newborn hospital records. Am J Epidemiol 1993;137:758–68.
13. Wingate MS, Alexander GR, Buekens P, Vahratian A. Comparison of gestational age classifications: date of last menstrual period vs. clinical estimate. Ann Epidemiol 2007;17:425–30.
14. Alexander GR, Kogan M, Martin J, Papiernik E. What are the fetal growth patterns of singletons, twins, and triplets in the United States? Clin Obstet Gynecol 1998;41:114–25.
15. Kuczmarski RJ, Flegal KM. Criteria for definition of overweight in transition: background and recommendations for the United States. Am J Clin Nutr 2000;72:1074–81.
16. Alexander GR, Kotelchuck M. Quantifying the adequacy of prenatal care: a comparison of indices. Public Health Rep 1996;111:408–18.
17. Alexander GR, Cornely DA. Prenatal care utilization: its measurement and relationship to pregnancy outcome. Am J Prev Med 1987;3:243–53.
18. Clayton D, Hills M. Statistical models in epidemiology. Oxford (UK): Oxford University Press; 1993.
19. Jaquet D, Swaminathan S, Alexander GR, Czernichow P, Collin D, Salihu HM, et al. Significant paternal contribution to the risk of small-for-gestational age. BJOG 2005;112:153–9.
20. Salihu HM. Fetal death repetition: the event memory hypothesis. Med Hypotheses 2007 Aug 3;[Epub ahead of print].
21. Salihu HM, Emusu D, Aliyu ZY, Kirby RS, Alexander GR. Survival of “pre-viable” infants in the United States. Wien Klin Wochenschr 2005;117:324–32.