Discussion: ‘Perinatal consequences of fetal macrosomia’ by Zhang et al

Article Outline

• Abstract
• Discussion Questions
• Introduction
• Background
• Study Design
• Statistical Analyses
• Conclusions
• Reference
• Copyright

In the roundtable that follows, clinicians discuss a study published in this issue of the Journal in light of its methodology, relevance to practice, and implications for future research. Article discussed:

Zhang X, Decker A, Platt RW, Kramer MS, Michael S. Kramer, MD. How big is too big? The perinatal consequences of fetal macrosomia. Am J Obstet Gynecol 2008;198:517.e1-517.e6.

Back to Article Outline

Discussion Questions 

Back to Article Outline

Introduction 

Certainly, the dangers that are associated with low birthweight have been well-publicized.¹ But fetal macrosomia, which can be diagnosed precisely only when an infant is weighed at birth, can be perilous as well. In a new study, Zhang et al attempted to identify the weight threshold at which the risk begins to rise for stillbirth, perinatal death, and neonatal morbidity (such as, birth injury or meconium aspiration syndrome). Fetal macrosomia can result in complications for the mother, which include prolonged labor, cesarean delivery, postpartum hemorrhage, infection, third- and fourth-degree lacerations, and thromboembolic events. Here, you can follow the Journal Club members' in-depth discussion of the researchers' findings.

George A. Macones, MD, MSCE

Back to Article Outline

Background 

Scifres: What is the author's main study question? Is it an important research question?

Shroff: The author's main study question was 3-fold: at what birthweight does the mortality rate begin to rise; what are the specific causes of this increased mortality rate; and what is the timing of these causes (antenatal, neonatal, or postneonatal)?

I do believe that this is an important research question. Suspected fetal macrosomia is a common obstetric condition. It is currently defined by the American College of Obstetricians and Gynecologists as a birthweight of >4500 g, irrespective of gestational age. As the authors point out, macrosomia may lead to increases in maternal complications and infant morbidity and death. They focused on the potential adverse neonatal outcomes that are associated with macrosomia and also examine mode of delivery relative to birthweight. It is important to identify the birthweight at which morbidity and mortality rates begin to increase and the timing of adverse outcomes (in utero vs intrapartum vs neonatal) before we can begin to examine interventions to improve outcomes.

Back to Article Outline

Study Design 

Scifres: What is a cohort study? Is this an appropriate study design for this research question?

Lewkowski: A cohort is a group of individuals who share a common characteristic or exposure. In a cohort study, the participants are identified from a given population, and information is collected regarding the outcome of interest and possible predictors and risk factors that are related to this outcome. This particular study was a retrospective cohort study in which the population was identified with the use of birth certificate information from 1999-2001. Outcome information had been collected previously and was obtained from birth certificate information and from a national database that links infant births, stillbirths, and neonatal deaths. The study population included all singleton infants who weighed >2500 g at birth and who were born between 37-44 weeks of gestation to white, non-Hispanic mothers during the previously mentioned time period. The exposure was infant weight, which was further stratified to attempt to identify a population at greater risk of the development of adverse neonatal outcomes that are related to fetal macrosomia.

Cohort studies are useful for the evaluation of predictors and risk factors that may precede the outcome of interest. Although they might show an association between predictors and outcomes, cohort studies generally cannot prove a causal relationship between the 2. Zhang et al sought to identify the birthweight at which the risk of perinatal death, neonatal morbidity, and cesarean delivery rise. This study is appropriate for the determination of possible associations between these factors but cannot determine causation. One benefit of this study was that a very large population could be studied. However, given its retrospective nature, the study was bound by data limitations, because previously collected data prevented evaluation of additional variables.

Scifres: What are the linked stillbirth/live birth/infant death files compiled by the National Center for Health Statistics (NCHS)? What are some of the benefits and limitations of this data source?

Stamilio: According to the authors, the NCHS's linked stillbirth/live birth/infant death files are part of the United States Vital Statistics Cooperative Program. The NCHS is the organization that compiles, organizes, manages, and distributes national vital statistics databases that are comprised of state-level data. There are multiple databases or files (such as, the Linked Birth/Infant Death Data Set, the Fetal Death Data Set, and the Perinatal Mortality Data Set). Descriptions of the various data sets and ways to obtain the data can be found on the NCHS web site (http://www.cdc.gov/nchs/nvss.htm), but I can provide a brief description. Generally, the data are available in CD-ROM format and can be purchased through the National Technical Information Service.

In the Linked Birth/Infant Death Data Set, the information from the death certificate is linked to the information from the birth certificate for each infant <1 year of age who dies in the United States or its outlying territories. The purpose of the linked database is to use variables that are available from the birth certificate to conduct more detailed analyses of infant mortality patterns. Information from the birth certificate (such as age, race, birthweight, gestational age, plurality, prenatal care usage, maternal education, live birth order, marital status, and maternal smoking) is linked to information from the death certificate (such as, age at death and underlying cause of death). Linked files for a specific year are available for the years 1983-2002, except for 1992-1994.

The Fetal Death Data Set contains all fetal deaths occurred in the United States after 1993 and includes fetal deaths in the other US territories. Data are obtained from fetal death reports that are registered in each state. Most (but not all) areas require reporting fetal deaths for gestations of ≥20 weeks of gestation, but some states have varying or no lower gestational age limit for inclusion. The Perinatal Mortality Data Set contains information on a combination of births, infant deaths, and fetal deaths that occur in the United States. It appears that the investigators used this file for the study. The set is comprised of the NCHS's Linked Birth/Infant Death Data Set along with fetal death data.

Two of the more important benefits of the vital statistics data sets include the large sample size and improved generalizability that findings would offer over study samples that are not population-based. The large sample size that is contained in these data sets allows the investigation of rare outcomes and potentially rare exposures, if that information is included in the data. Because the sample is comprised of the entire national population, any results from studies that examine all groups within that population are applicable to a wider spectrum of clinical populations and generalizable to the national population.

Important limitations of vital statistics data sets include the increased potential for data or coding error, missing data, inadequate detail on exposures and outcomes, inconsistent data reporting across states, and an inconsistent ability to establish a temporal relationship between exposure and outcome. For example, a potential source of error is the method that was used for gestational age assignment. In addition, medical and obstetric complications tend to be underreported in vital statistics data. This fact tends to result in an underestimate of absolute risk and relative risk. Examples of suboptimal detail of data include the lack of maternal body mass index (BMI) information and the lack of data regarding the timing of fetal death. The inability to assess these 2 variables leaves the potential for residual confounding (by obesity) and precludes the investigators from fully assessing the relationship between fetal death and macrosomia as they relate to prenatal interventions that might have been applied in some cases.

Scifres: How was gestational age estimated in this study, and do you think this estimation was appropriate?

Shroff: In this study, gestational age was estimated by the calculation of the interval between the first day of the mother's last menstrual period (LMP) and the date of birth. If the LMP was not reported or the gestational age was not consistent with birthweight, the clinical estimate of gestation was used. Clinical estimates were used for gestational age assessment in approximately 5% of births, and most of these were due to missing LMP data. Gestational age is a very important factor in this study, because we know that adverse outcomes may be increased in either preterm or postterm pregnancies and that postterm pregnancy is also a risk factor for fetal macrosomia. Some of the difficulties with LMP dating may involve recall error and misclassification because of postconceptual bleeding. The authors also point out that, before 2003, there was no standardization of how the clinical estimate of gestation was performed in the NCHS registry.

Overall, I think this estimation is appropriate, although I did find some data to suggest that agreement between menstrual and clinical estimates of gestational age occurs most often close to term, with significant disagreement in preterm and postterm births. If postterm births were underreported, this could impact the study results, because this is a potential cause of some of the outcomes in the study, which include stillbirth, neonatal depression, and meconium aspiration.

Scifres: Were appropriate maternal demographic factors collected? Do you think any other exposures should have been included?

Shanks: There are a number of risk factors for macrosomia that include increasing maternal prepregnancy weight, weight gain during pregnancy, multiparity, male fetus, gestational age of >40 weeks of gestation, ethnicity, maternal birthweight, maternal height, maternal age <17 years, a diagnosis of diabetes mellitus, or evidence of impaired glucose tolerance on screening for gestational diabetes, even if formal diagnostic criteria are not met. The risk factors that the authors investigated were all appropriate, but information regarding previous pregnancies with macrosomia and maternal BMI were lacking. The authors address these shortcomings in the Results section, stating that the data that involved BMI were unavailable for analysis. Therefore, they were unable to assess whether adverse outcomes were the result of macrosomia or were explained partially by maternal obesity.

Scifres: Do you think the authors' defined set of outcomes are clinically significant?

Lewkowski: The outcomes that were evaluated in this study were fetal death, infant death, cause-specific death, and neonatal morbidity. Information was also collected regarding mode of delivery (spontaneous vaginal, instrumental vaginal, and cesarean delivery). Infant death was categorized on the basis of the International Classification of Diseases-10 codes with the following categories: immaturity-related conditions, congenital anomalies, asphyxia-related conditions, sudden infant death syndrome, infectious disease, and external causes.

Neonatal morbidity outcomes included Apgar score <4 at 5 minutes, need for mechanical ventilation, neonatal seizures, birth injury (any impairment of the infant's body function or structure because of adverse events that occurred at birth), and meconium aspiration syndrome (MAS).

These outcomes are significant because fetal and neonatal death are devastating pregnancy complications; outcomes such as neonatal depression, need for ventilation, seizures, MAS, and birth injury contribute significantly to neonatal health and healthcare costs. One limitation of the study's outcomes is that stillbirth was not classified further into antepartum vs intrapartum. The mechanisms behind these 2 types of stillbirth and the potential interventions to improve outcomes are obviously very different.

Back to Article Outline

Statistical Analyses 

Scifres: What is statistical imputing, and how might it affect the data?

Allsworth: Statistical imputation is a broad term for multiple techniques that are used to handle missing data. Common imputation techniques include hot-deck imputation and regression imputation. In hot-deck imputation, missing data are filled in with the use of observed data from similar but complete participant records. Regression imputation involves the simulation or estimation of “likely” values on the basis of observed data. For example, if BMI was missing for a particular patient, a reasonable estimate for this value could be made with the use of observed data on sex, age, height, and comorbid conditions. This imputation procedure may be repeated multiple times, with estimates combined to produce estimated values that incorporate missing data uncertainty.

In this article by Zhang et al, 2 variables were imputed when maternal age and marital status were missing. Maternal age was assigned the value of the age of the mother from the previous birth record of the same race and birth order. This is a form of hot-deck imputation. Marital status, on the other hand, was assumed to be “married.” For both of these variables, the prevalence of missing data was <0.5%.

Imputation techniques must be applied thoughtfully and carefully, because they may distort estimates and/or standard errors if naively performed. However, they are useful techniques and are potentially superior to complete case analysis, which could result in reduced precision and biased estimates.

Scifres: The authors restricted their analysis to singleton live births of white, non-Hispanic mothers. Do you think they provided adequate justification for this decision? How might this affect the results of the study?

Shanks: The authors restricted their analysis to singleton gestations of white, non-Hispanic mothers because “plurality and maternal ethnicity are associated with birthweight and perinatal death.” Inclusion of both risk factors could have made the findings more robust and certainly would assist in the applicability of the results to a more heterogeneous patient population. The other option would have been to include women of different races or parity and to establish whether there were significant differences among these factors with regards to macrosomia. If differences existed, these differences would then require further evaluation in the logistic regression model.

Scifres: Can you describe the chi-square test and 1-way analyses of variance (ANOVA)? When are these tests used?

Stamilio: In the unadjusted analysis, the investigators used the chi-square test for trend and 1-way ANOVA to compare the birthweight groups with regard to characteristics and outcomes. The key to understanding their selection of statistical tests is that the study includes >2 comparison groups; namely, they compared characteristics across 4 birthweight or exposure groups.

There are several versions or applications of the chi-square test, but all are used as a test of statistical significance in comparing ≥2 study groups for a categoric outcome variable or for a characteristic that has ≥2 categories. For example, in this study, the chi-square test was used to determine whether there was a difference in the infant gender, which is a binary characteristic variable, across the 4 birthweight groups. Of note, this chi-square test of trend assesses for differences among all groups, not pairwise comparisons.

The t-test is used generally to compare 2 study groups for continuous variables that are normally distributed; however, because there are >2 study groups, the t-test cannot assess differences across all groups simultaneously. The application of several pairwise t-tests increases alpha error to an unacceptable level because of multiple comparisons. Thus, ANOVA, which is a method in which variance is compared across all study groups, is the statistical test of choice when ≥3 study groups are compared for a continuous variable, such as gestational age. Pairwise comparisons can then be explored with the t-test only if the ANOVA reveals a statistically significant difference among all groups.

Scifres: Describe the adjusted odds ratio and the 95% confidence interval. How do you interpret these numbers?

Lewkowski: The odds ratio is defined as the ratio of the odds of an event that occurs in 1 group (exposed) to the odds of it occurring in another group (unexposed). It is a measure of risk. If the odds ratio is >1, the event is more likely to occur than not. Odds ratios are difficult to interpret, however, because they indicate only whether something is likely to happen. In situations in which the outcome is rare, the odds ratio is very similar to the relative risk, which are sometimes interpreted interchangeably. An adjusted odds ratio describes the results of a logistic regression, which adjusts for the simultaneous effect of other confounding variables.

The CI gives an estimated range of values that are likely to include the population parameter of interest. For example, if a range represents the 95% CI, one can be 95% confident that the true value lies within it. In interpreting an odds ratio when the 95% CI includes an odds ratio of 1, the exposed group is no more or less likely than the unexposed group to experience the outcome.

For example, the authors report an odds ratio of 13.2 for risk of stillbirth in infants who weigh >5000 g. The 95% CI is 9.8-17.7. Given that this is a rare event (0.67% of patients in this group experienced this outcome), one can approximate the odds ratio to the relative risk. This means that fetuses who weighed >5000 g were 13.2 times more likely to be stillborn than fetuses who weighed 3500-4499 g. One can be 95% confident that the true value is 9.8-17.7 times more likely. This is an adjusted odds ratio that was derived from a logistic regression that controlled for gestational age, gender, parity, maternal age, maternal education, maternal marital status, maternal diabetes mellitus, and smoking.

Scifres: Speaking of the adjusted odds ratio, the authors adjusted for a number of variables in their multivariable model. How do authors decide what to adjust for?

Allsworth: Adjustment for confounding is an essential issue in any multivariable model. A confounder is a variable that is associated with both the exposure and outcome of interest, but it is not part of the causal pathway. If important confounders are ignored in analyses, estimates of the association between exposure and outcome might be biased.

The current study adjusted for gestational age and a number of maternal characteristics, but it did not specify how the confounders were chosen. In general, researchers use a number of considerations when selecting confounders. First, authors review the literature to establish which variables have been found to be associated with the exposure and outcome in independent studies. This step includes careful consideration of whether a variable might be part of the causal pathway or an intermediate step in the physiologic process, because adjustment for intermediate factors may result in biased estimates of the exposure-outcome association.

Once a list of candidate confounders is established, each variable is tested for association with exposure and outcome in the current study. This information often guides which variables are to be included in multivariable analyses, but authors might override a lack of association in the current study if there is strong evidence of confounding in previous research.

The final step includes the construction of the multivariable model. A common rule of thumb is the 10% rule. That is, if the addition of the potential confounder to the multivariable model changes the estimate of the association by ≥10%, it is considered a confounder.

The approach I have described is a common one and seeks to incorporate prior knowledge, statistical properties, and researcher judgment, but it is not the only method. Some researchers prefer a statistics-based technique that relies solely on hypothesis testing, such as what you would see in forward or backward stepwise regression.

Sample size is the last consideration when building a multivariable model. A general rule of thumb for logistic regression is that 10 events were variable are needed[should this read “10 events PER confounding variable are needed”?] for appropriate estimates and CIs, which is a criterion that was met in most, but not all, of the models that were presented here.

Scifres: The issue of maternal obesity is an interesting one. Specifically, how do you think that the absence of this information might influence the results of the study?

Shroff: Maternal obesity, weight gain during pregnancy, and diabetes mellitus all have been associated with macrosomia. Maternal obesity has also been linked to stillbirth and may be a predictor of shoulder dystocia, which is a potential cause of both birth trauma and asphyxia, 2 of the outcomes in this study.

For a study like this, which attempts to establish an association between macrosomia and an outcome such as stillbirth, we would like to have information on maternal obesity because of the issue of confounding. A confounding variable is a variable that has a linear relationship (either positive or negative) with the dependent and independent variable in a study, in this case, stillbirth and birthweight. If maternal obesity and fetal weight increase together, if obese women are more likely to have diabetes mellitus, and if macrosomic fetuses are more likely to be stillborn, it would be incorrect to assume that macrosomia caused stillbirth without performing the appropriate statistical tests to address the issue of confounding.

Back to Article Outline

Conclusions 

Scifres: How do you interpret the study results? How are they limited?

Stamilio: The primary conclusion is that birthweight of ≥4500 g is associated with increased risks for neonatal death and morbidity relative to birthweights between 3500 and 4499 g [said 3500 -4000, but 4499 was end of reference category]. The increases in risk that are associated with high birthweight persisted when adjusted for known confounding variables in logistic regression models. The main results are in Table 2. Birthweight in the range of 4500-4999 g is associated with a statistically significant 2- to 3-fold increase in stillbirth and early neonatal death rate relative to those with a birthweight of 3500-4499 g, which is the reference category. The increase in risk is even more pronounced with a birthweight of ≥5000 g; this very high birthweight is associated with a 5- to 13-fold increase in risks for stillbirth, early neonatal death, and late neonatal death and is associated with a 2-fold increase in the postneonatal death rate. In addition, birthweight in the range of 2500-3499 g was associated with close to a 2-fold increase in rates of all death outcomes, relative to the reference group.

One thing to consider is that all the estimates that I discussed earlier are relative risks or estimates of relative risk, but we should also consider absolute risk. In this case, the absolute risk for neonatal or perinatal death is actually quite low, so the population attributable risk for infant or neonatal death because of fetal macrosomia is <1% in almost all cases. Although an odds ratio of 13 sounds quite impressive, we need to keep in mind that. because the baseline risk is very low, even an odds ratio of 13 results in a small absolute risk increase of generally <1%; the only exception to this is the increase in the risk for stillbirth with a birthweight <5000 g. In the study, the population attributable risk was just <8%.

Table 3 depicts the analysis of the cause of neonatal death. When compared with the reference birthweight category, a birthweight of ≥4500 g was associated with an increased rate of asphyxia as a cause of death, and a birthweight of ≥5000 g was associated with increased rates of infection and sudden infant death syndrome as causes of death. Table 4 depicts increasing rates of neonatal morbidity with increasing birthweight; neonatal morbidity measures in this study include a low 5-minute Apgar score, seizures, ventilation requirement, birth injury, and MAS.

One limitation of the study is that only white non-Hispanic mothers with delivery gestational ages of 37-44 weeks were included. Thus, the results may not be generalizable to other races or gestational ages. Given the previously described limitations of the data source, which may result in nondifferential misclassification bias, it is possible that the magnitudes of the odds ratios are underestimated. It is also possible that residual confounding by unmeasured factors (such as obesity) could exist, which can result in over- or underestimation of odds ratios. Last, it is unclear from this study how knowledge of high estimated fetal weight could or did affect perinatal outcomes. Using the vital statistics data, one cannot discern whether clinicians altered monitoring or therapy based on a suspicion of high birthweight. Prudent clinical practice may have resulted in an underestimate of stillbirth that was associated with high birthweight.

Scifres: Fetal macrosomia is clearly an important topic. What further research do you think is needed in this area?

Shanks: Although the results of this study are interesting, it is important to remember that observational studies like this one are hypothesis-generating, not practice-changing. The optimal management of macrosomia remains a clinical challenge. One difficulty in the management of macrosomia is that our ability to assess fetal weight accurately is limited. In addition, although the increased incidence of stillbirth is striking, we are limited in our knowledge of whether these deaths were antepartum or intrapartum, and the absolute risks are still low. The benefit and cost-effectiveness of cesarean delivery for the prevention of birth injury in pregnancies that are complicated by macrosomia remains controversial. Also, better delineation of the antepartum risk of stillbirth could lead us to investigate whether heightened surveillance of these pregnancies improves outcomes.

Back to Article Outline

Reference 

1. American College of Obstetricians and Gynecologists. Fetal Macrosomia: ACOG Practice Bulletin No.: 22. Washington (DC): The College; 2000;
   • View In Article