Objective
Using a cohort of 110,447 singleton, term pregnancies, we aimed to validate the previously proposed customized standard of large-for-gestational-age (LGA) birthweight, derive an additional customized LGA model excluding maternal weight, and evaluate the association between differing definitions of customized LGA and perinatal morbidities.
Study Design
Three customized LGA classifications, in addition to a population-based 90th percentile, were made according to the principals described by Gardosi: (1) customized LGA using Gardosi’s previously published coefficients (LGA-Gardosi), (2) customized LGA using coefficients derived by a similar method but from our larger cohort, and (3) derived without customization for maternal weight. Associations between the LGA classifications and various perinatal morbidity outcomes were evaluated.
Results
Coefficients derived here for physiologic and pathologic effects on birthweight were similar to those previously reported by Gardosi. Customized LGA (any method) generally identified more births to younger, nonwhite, nulliparous mothers with female neonates of lower birthweight compared with population-based LGA. Rates of maternal and neonatal morbidity were greatest in births classified by both population-based LGA and customized LGA (any method). However, the model that excluded customization for maternal weight, revealed a greater proportion of women previously unidentified by population-based LGA who were more frequently black (40% vs 25%) and obese (30% vs 5.1%), along with greater rates of shoulder dystocia, neonatal intensive care unit admission and neonatal respiratory complications, than with LGA-Gardosi.
Conclusion
The use of customized methods of defining LGA was not decisively superior compared with population-based LGA, but custom LGA may be improved by modification of the parameters included in customization.
High birthweight may occur because of multiple physiologic (eg, genetics) or pathologic (eg, excess adipose deposition) factors. Excessively high birthweight is associated with increased neonatal intensive care unit (NICU) admission rate, hypoglycemia, shoulder dystocia, delivery injury, postpartum maternal hemorrhage, and cesarean delivery, as well as long-term child health risks related to obesity, cardiovascular, and metabolic disease. Despite this knowledge of heightened risks for potentially serious maternal and/or neonatal complications with overly large birth size, defining overgrowth in clinical practice and epidemiologic research remains challenging. Clinical definitions to describe fetal overgrowth and high birthweight include large-for-gestational-age (LGA), using a cutoff such as the 90th percentile of birthweight in a reference population, or macrosomia defined as birthweight ≥4000 or 4500 g. Using the 90th percentile, by definition, results in approximately 10% of births being classified as LGA, but the proportion of neonates with overgrowth varies depending on the prevalence of pathologies such as maternal obesity and diabetes in the population. Furthermore, such cutoffs will not identify neonates with a “normal” population-based birthweight percentile (eg, 60th), but having a fetal growth trajectory greater than their genetic potential due, for example, to excess fetal nutrient supply.
Customized birthweight percentiles have been proposed to help differentiate infants born large, but healthy from those with intrauterine overgrowth by accounting for inherent differences in birthweight attributable to presumed nonpathologic factors including maternal race, stature, parity, and neonatal sex, according to gestational age. Prior studies have typically examined customization for small-for-gestational-age (SGA) citing improvement in prediction of perinatal morbidity and mortality, but the few studies investigating customized LGA have reported either maternal delivery outcomes only or limited neonatal health data in smaller cohorts. Understanding the relationship between customized LGA and neonatal health outcomes is important for clinical practice, however. For example, changing the definition of LGA will affect which infants are screened for hypoglycemia as prescribed by current guidelines. Furthermore, debate continues regarding which factors contribute to physiologic vs pathologic variation in birthweight and whether a more complicated approach to identifying excessively large birthweight adds value to clinical care and research. Given the prevalence of maternal obesity and its perinatal comorbidities and the known relationship between increased maternal weight and LGA, customization for maternal weight, as has been done previously, may be inappropriate for assessment of LGA. Therefore, the aims of the present study, using a large US birth cohort, were to (1) validate the previously proposed customized standard of LGA, (2) derive an additional customized LGA model excluding maternal weight to determine its effect on predicting perinatal health outcomes, and (3) evaluate the role of these methods of customized LGA in predicting maternal and neonatal health outcomes compared with the conventional population-based 90th percentile birthweight cutoff.
Materials and Methods
Subjects
The Consortium on Safe Labor (CSL) was a retrospective cohort study conducted by the Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health. The population included deliveries ≥23 weeks’ gestation at 12 clinical study centers (including 19 hospitals) across the US from 2002 to 2008 (n = 228,562 births). All study procedures were reviewed and approved by each participating sites’ institutional review board. For the present analysis, pregnancies with multiple gestation (n = 5050), preterm birth (<37 weeks; n = 29,612), and/or fetal anomalies (n = 17,127) were excluded to be consistent with previous studies. In women contributing more than 1 delivery during the study period (n = 19,867), only the first pregnancy recorded in the study database was included, leaving 168,945 eligible pregnancies. Pregnancies with data missing for modeled variables included in the customized LGA equations (maternal race/ethnicity [n = 7545], height/weight [n = 53,980], infant sex [n = 162], birthweight [n = 1503]), as well implausible birthweight (<615 g; n = 7), were also excluded, resulting in a final sample size of 110,447 births.
Maternal and neonatal morbidity and mortality outcomes
Data on delivery and neonatal outcomes were ascertained from electronic medical records supplemented with discharge International Classification of Diseases , ninth revision codes (ICD9). Previous work comparing detailed chart review to ICD9 data demonstrated good agreement; these data collection and quality control methods for the CSL have been described in full detail elsewhere. If a particular outcome was recorded as present in either chart or ICD9 data, the outcome was considered to be present. For outcomes ascertained through ICD9 only (neonatal hypoglycemia, neonatal jaundice), the code utilized is given. The following LGA-related delivery outcomes were assessed: 3rd and 4th degree perineal lacerations, cervical laceration, shoulder dystocia, cesarean delivery and postpartum maternal hemorrhage. The following neonatal outcomes were also assessed: neonatal hypoglycemia (ICD9 775.6, neonatal hypoglycemia), respiratory complications (respiratory distress syndrome, meconium aspiration, transient tachypnea of the newborn and/or resuscitation greater than giving oxygen only), neonatal jaundice (ICD9 774.6, unspecified fetal and neonatal jaundice), 5-minute Apgar score <4, admission to NICU, and perinatal mortality (antepartum, intrapartum or other unspecified stillbirth and/or neonatal death).
Definitions of LGA and analysis
Each neonate was classified as LGA according to each of 4 LGA definitions. Infants were first classified as LGA if birthweight was greater than a previously published population-based 90th percentile according to week of gestational age (LGA Pop ). Three customized LGA classifications were then made according to the principles of Gardosi. The first was determined using Gardosi’s previously published customized standard for the US population derived from a cohort of 30,837 singleton term births from the National Institutes of Health First- and Second-Trimester Evaluation of Risk Research Consortium (LGA Gard ), and 2 other customized LGA classifications were derived from the present CSL study population. Specifically, 2 predictive models of birthweight were generated using multiple linear regression analysis with backward elimination. Regression models included significant predictors of birthweight considered to be either physiologic (gestational age, neonatal sex, maternal height, race, parity and, for the first model only, maternal prepregnancy weight) or pathologic (smoking, prepregnancy underweight [body mass index {BMI} <18.5 mg/kg 2 ], obesity [BMI ≥30], chronic or gestational diabetes, gestational hypertensive disease and, for the second model only, maternal history of other chronic diseases). The inclusion of maternal history of chronic diseases in the final model was determined using stepwise linear regression; maternal history of heart, renal, gastrointestinal, and thyroid diseases were dropped from the models because of lack of significance ( P > .05), whereas maternal depression, asthma, and anemia were included ( P < .05). A customized calculation of optimal term birthweight for each woman was then calculated using the coefficients for the physiologic predictors of birthweight only, separately for each of the 3 custom LGA classifications: (1) using the coefficients previously published by Gardosi (as indicated above: LGA Gard ), (2) using the coefficients derived from the first predictive model of birthweight in our CSL population based on an analogous model to that reported by Gardosi (LGA CSLGard ), and (3) using the coefficients derived from the second predictive model of birthweight in our CSL population with the addition of maternal health history data and the exclusion of maternal weight (as a continuous physiologic variable) from the model (LGA CSLcust ). From the calculated optimal birthweight, a customized cutoff for LGA was calculated for each woman using the coefficient of variation of actual measured birthweight for the entire study population (coefficient of variation = 13.3%) and a multiplier of 1.28, the z-score used to extrapolate to the 90th percentile from the mean:
Custom 90 th percentile LGA cutoff = calculated optimal weight + ( calculated optimal weight × 0.133 × 1.28 )
Customized LGA classification was made if the actual birthweight was greater than the calculated cutoff. This procedure was done separately for each model of customized LGA. As a sensitivity analysis, the procedures were repeated in a subsample of pregnancies with no pathology (ie, no presence of any of the pathologic variables) and the resulting beta-coefficients were nearly identical (data not shown).
Odds ratios and associated c-statistics for the association between each of the LGA classifications and maternal and neonatal outcomes were calculated using logistic regression analysis. The c-statistic is a measure of the area under the receiver operator characteristic curve and in the present analysis was used to compare the overall performance of different definitions of LGA in association with maternal and neonatal morbidities. Sensitivity, specificity, positive predictive values (PPV), and negative predictive values were also calculated. Occurrence of all morbidity outcomes was also determined in subgroups of births excluded from, shared with or additionally captured by LGA Gard and LGA CSLcust compared with LGA Pop. These subgroups were compared using contrast statements within either logistic or linear models (for categorical or continuous variables, respectively) to evaluate the changes in subject characteristics and morbidity prevalence that occurred with the adoption of the different custom methods of LGA in lieu of a population-based cutoff. All analyses were conducted using SAS 9.3 (SAS Institute, Cary, NC).
Results
Women included in the analysis (n = 110,447) were 18% obese, 43% nulliparous, 55% non-Hispanic white, 19% non-Hispanic black, and 20% Hispanic ( Table 1 ). The average birthweight was (mean ± standard deviation [SD]) 3371 ± 447 g and gestational age was 39.2 ± 1.1 weeks. The method with the highest rate of LGA classification was LGA Pop (n = 9,423; 8.5% of total sample), followed by LGA CSLcust , (n = 8859; 8.0%), LGA CSLGard (n = 7783; 7.0%), and LGA Gard (n = 6207; 5.6%). Actual mean (± SD) birthweight within each of the 4 LGA groups was 4218 ± 248 g in LGA Pop , 4147 ± 325 g in LGA CSLcust , 4166 ± 334 g in LGA CSLGard , and 4150 ± 376 g in LGA Gard .
| Characteristic | n (%) or mean ± SD |
|---|---|
| Maternal age, y a | 27.6 ± 6.1 |
| Maternal height, cm | 163.5 ± 7.3 |
| Maternal weight, kg | 67.7 ± 17.3 |
| BMI, kg/m 2 | |
| <18.5 | 5933 (5.4) |
| 18.5 to <25 | 59,919 (54.3) |
| 25 to <30 | 24,686 (22.4) |
| ≥30 | 19,909 (18.0) |
| Parity | |
| 0 | 47,722 (43.2) |
| 1 | 31,819 (28.8) |
| 2 | 17,593 (15.9) |
| 3 | 7968 (7.2) |
| 4+ | 5345 (4.8) |
| Race/ethnic group | |
| Non-Hispanic white | 60,239 (54.5) |
| Non-Hispanic black | 21,457 (19.4) |
| Hispanic | 21,982 (19.9) |
| Asian/Pacific Islander | 4160 (3.8) |
| Other or multiracial | 2609 (2.4) |
| Smoking | 7193 (6.5) |
| Maternal diabetes | |
| Preexisting | 1279 (1.2) |
| Gestational | 5067 (4.6) |
| Maternal hypertensive disease | |
| Chronic hypertension | 1824 (1.7) |
| Gestational hypertension | 3058 (2.8) |
| Mild preeclampsia | 2875 (2.6) |
| Severe preeclampsia | 678 (0.6) |
| Superimposed preeclampsia | 546 (0.5) |
| Eclampsia | 67 (0.1) |
| Unspecified hypertension | 391 (0.4) |
| Antepartum bleeding or abruption | 2616 (2.4) |
| Gestational age at delivery, wk | 39.2 ± 1.1 |
| Infant birthweight, g | 3370.7 ± 447.1 |
Comparison of coefficients across models
Predicted term optimal birthweight for a standard mother (non-Hispanic white, 163 cm height, 64 kg weight, nulliparous, 280 d gestation) and coefficients for gestational age, neonatal sex, parity, and pathologic conditions (diabetes, hypertensive diseases) were similar in the 2 predictive models of birthweight generated in the CSL population ( Table 2 ), as those reported for LGA Gard , thus validating Gardosi’s previously published coefficients. However, both LGA CSLGard and LGA CSLcust yielded a positive coefficient for the effect of Hispanic ethnicity on birthweight, whereas LGA Gard had a negative coefficient of nearly twice the magnitude. Also, in the LGA CSLcust model, which excluded customization for maternal weight, the coefficients for maternal height and the (pathologic) effect of obesity and underweight on birthweight were markedly greater in magnitude than models that customized for weight.
| Variable | LGA Gard a | LGA CSLGard | SE | P value | LGA CSLcust | SE | P value |
|---|---|---|---|---|---|---|---|
| Physiologic variables | |||||||
| Intercept | 3453.4 | 3481.4 | 2.7 | < .001 | 3452.8 | 2.6 | < .001 |
| Gestational age (from 280 d) | |||||||
| Linear term | 22.86 | 17.4 | 0.26 | < .001 | 17.6 | 0.27 | < .001 |
| Quadratic term | −0.311 | −0.35 | 0.02 | < .001 | −0.35 | 0.02 | < .001 |
| Cubic term | −0.007 | −0.004 | 0.001 | < .001 | 0.00 | 0.001 | < .001 |
| Sex | |||||||
| Male | 66.0 | 65.8 | 1.2 | < .001 | 65.6 | 1.2 | < .001 |
| Female | −66.0 | −65.8 | 1.2 | < .001 | −65.6 | 1.2 | < .001 |
| Maternal height (from 163 cm) | |||||||
| Linear term | 6.398 | 7.33 | 24.1 | < .001 | 11.7 | 21.3 | < .001 |
| Quadratic term | 5.40 | 141.2 | .0001 | 3.59 | 141.6 | .01 | |
| Cubic term | −0.003 | −2.72 | 579.3 | < .001 | −3.72 | 580.9 | < .001 |
| Maternal prepregnancy weight (from 64 kg) | |||||||
| Linear term | 7.578 | 6.20 | 0.17 | < .001 | |||
| Quadratic term | −0.087 | −0.10 | 0.01 | < .001 | |||
| Cubic term | 0.0005 | 0.001 | 0.0001 | < .001 | |||
| Race | |||||||
| Non-Hispanic black | −161.0 | −164.2 | 3.2 | < .001 | −155.9 | 3.2 | < .001 |
| Hispanic | −38.6 | 14.8 | 3.3 | < .001 | 20.9 | 3.3 | < .001 |
| Asian/Pacific Islander | −27.8 | 6.4 | < .001 | −44.6 | 6.4 | < .001 | |
| Other or multiracial | −140.8 | −45.7 | 7.9 | < .001 | −42.0 | 7.9 | < .001 |
| Parity | |||||||
| 1 | 96.2 | 79.5 | 2.9 | < .001 | 87.0 | 2.9 | < .001 |
| 2 | 121.9 | 91.2 | 3.6 | < .001 | 102.3 | 3.6 | < .001 |
| 3 | 125.9 | 110.1 | 4.9 | < .001 | 122.9 | 4.9 | < .001 |
| 4+ | 122.7 | 124.4 | 5.8 | < .001 | 138.1 | 5.8 | < .001 |
| Pathologic variables | |||||||
| Smoking b | −148.9 | 4.8 | < .001 | −145.5 | 4.9 | < .001 | |
| Underweight (BMI <18.5 kg/m 2 ) | −20.4 | −32.3 | 6.2 | < .001 | −149.9 | 5.3 | < .001 |
| Obese (BMI ≥30 kg/m 2 ) | −63.4 | −22.4 | 5.5 | < .001 | 95.3 | 3.2 | < .001 |
| Maternal diabetes | |||||||
| Preexisting | 241.7 | 200.0 | 11.1 | < .001 | 216.5 | 11.2 | < .001 |
| Gestational | 114.3 | 5.7 | < .001 | 129.8 | 5.7 | < .001 | |
| Maternal hypertensive disease | |||||||
| Chronic hypertension | −14.4 | 9.5 | .13 | ||||
| Gestational hypertension | −26.8 | −25.4 | 7.3 | .001 | −13.8 | 7.3 | .06 |
| Mild preeclampsia | −60.7 | −36.1 | 7.5 | < .001 | −29.4 | 7.5 | < .001 |
| Severe preeclampsia | −156.3 | 15.2 | < .001 | −155.2 | 15.2 | < .001 | |
| Superimposed preeclampsia | −67.9 | 16.9 | < .001 | −55.2 | 17.0 | .001 | |
| Eclampsia | −136.6 | 47.9 | .004 | −143.9 | 48.2 | .003 | |
| Unspecified hypertension | −20.3 | 19.9 | .31 | −5.0 | 20.1 | .80 | |
| Antepartum bleeding or abruption | −41.2 | −62.4 | 7.8 | < .001 | −60.5 | 7.8 | < .001 |
| History of preterm birth c | −67.9 | 5.4 | < .001 | −67.8 | 5.5 | < .001 | |
| Maternal health history d | |||||||
| Depression | −23.1 | 6.1 | < .001 | ||||
| Asthma | −24.0 | 4.7 | < .001 | ||||
| Anemia | 46.4 | 4.1 | < .001 |
a Coefficients reported from similar analysis of previous cohort of 30,837 births by Gardosi
b Smoking was reported by Gardosi in number/day (coefficients: 1-9 n/d, −99.2; 10-19, −174.9; ≥20, −246.3)
c Maternal birth history was reported separately for history of miscarriage (coefficient: 12.9) a single preterm birth (−55.1) or ≥2 preterm births (−77.6) by Gardosi
d Maternal history of heart, renal, gastrointestinal and thyroid diseases were also evaluated, but were dropped from the models because of lack of significance ( P > .05).
Deliveries with reclassified LGA status
Although there was overlap between LGA Pop and customized LGA, LGA Gard excluded 4796 (51% of LGA Pop ) from LGA classification, whereas adding 1580 (1.4% of total population) new cases of LGA ( Figure , A). Likewise, LGA CSLcust excluded fewer at 2680 (28%) of LGA Pop , whereas adding 2116 (1.9% of total population) LGA cases ( Figure , B).
Customized LGA classification (any method) generally identified more births to younger, nonwhite, nulliparous mothers with female neonates of lower birthweight compared with LGA Pop . However, LGA CSLcust revealed a greater proportion of previously unidentified women who were more frequently black (40% vs 25%), obese (30% vs 5.1%), and with preexisting (3.5% vs 1.8%) or gestational (10.0% vs 7.1%) diabetes compared with LGA Gard ( Table 3 ).
| Variable | LGA Pop vs LGA Gard | LGA Pop vs LGA CSLcust | ||||
|---|---|---|---|---|---|---|
| LGA Pop-only | LGA Pop+Gard | LGA Gard-only | LGA Pop-only | LGA Pop+CSLcust | LGA CSLcust-only | |
| n = 4796 | n = 4627 | n = 1580 | n = 2680 | n = 6743 | n = 2116 | |
| Maternal age, y | 29.2 ± 5.7 | 29.0 ± 6.0 | 26.7 ± 6.5 a | 29.6 ± 5.6 | 28.9 ± 6.0 | 27.1 ± 6.4 a |
| BMI, kg/m 2 | ||||||
| <18.5 | 1.2 | 2.9 | 11.1 a | 2.8 | 1.8 | 2.3 a |
| 18.5 to <30 | 66.6 | 75.4 | 83.8 a | 77.1 | 68.5 | 67.7 a |
| ≥30 | 32.3 | 21.7 | 5.1 a | 20.2 | 29.8 | 30.0 a |
| Parity | ||||||
| 0 | 28.7 | 40.3 | 63.2 a | 25.8 | 37.8 | 56.6 a |
| 1 | 32.5 | 30.3 | 22.2 a | 34.0 | 30.4 | 25.2 a |
| 2 | 20.8 | 16.0 | 9.1 a | 19.9 | 17.9 | 11.3 a |
| 3 | 10.6 | 7.8 | 3.3 a | 12.2 | 8.1 | 4.4 a |
| 4+ | 7.4 | 5.6 | 2.2 a | 8.1 | 5.9 | 2.6 a |
| Race/ethnic group | ||||||
| White | 70.2 | 50.0 | 26.0 a | 74.5 | 54.6 | 32.5 a |
| Black | 10.3 | 16.5 | 24.9 a | 4.4 | 16.9 | 40.3 a |
| Hispanic | 17.4 | 25.8 | 27.9 a | 18.3 | 22.8 | 19.0 a |
| Asian/Pacific Islander | 0.7 | 4.3 | 14.6 a | 1.1 | 3.0 | 5.1 a |
| Other or multiracial | 1.4 | 3.4 | 6.7 a | 1.7 | 2.6 | 3.2 a |
| Smoking (any) | 4.7 | 4.0 | 1.9 a | 4.3 | 4.4 | 3.1 |
| Maternal diabetes | ||||||
| Preexisting | 1.6 | 4.2 | 1.8 a | 0.8 | 3.7 | 3.5 a |
| Gestational | 5.9 | 10.6 | 7.1 a | 3.9 | 9.9 | 10.0 a |
| Maternal hypertensive disease | ||||||
| Chronic hypertension | 1.9 | 2.3 | 1.7 | 1.1 | 2.5 | 3.0 a |
| Gestational hypertension | 3.1 | 3.0 | 2.6 | 2.2 | 3.4 | 3.0 a |
| Mild preeclampsia | 2.3 | 3.3 | 3.3 a | 1.7 | 3.2 | 4.3 a |
| Severe preeclampsia | 0.3 | 0.6 | 1.0 a | 0.2 | 0.5 | 1.1 a |
| Superimposed preeclampsia | 0.4 | 0.6 | 0.4 | 0.1 | 0.7 | 1.1 a |
| Eclampsia | 0.1 | 0.1 | 0.1 | 0 | 0.04 | 0.1 |
| Unspecified hypertension | 0.5 | 0.5 | 0.4 | 0.5 | 0.5 | 0.5 |
| Antepartum bleeding or abruption | 1.5 | 2.0 | 2.0 | 1.4 | 1.9 | 2.5 a |
| Gestational age at delivery, wk | 39.8 ± 1.0 | 39.1 ± 1.1 | 38.3 ± 1.0 a | 39.9 ± 1.0 | 39.3 ± 1.1 | 38.6 ± 1.1 a |
| Infant birthweight, g | 4138 ± 150 | 4300 ± 298 | 3711 ± 188 a | 4101 ± 116 | 4264 ± 271 | 3775 ± 166 a |
| Infant sex (% m) | 73 | 56 | 31 a | 83 | 57 | 27 a |
a Custom-only group (LGA Gard-only or LGA CSLcust-only ) significantly different than in LGA Pop-only ( P < .05).
Maternal and neonatal morbidity by differing LGA definitions overall
Overall, there were 37,843 (34.3%) deliveries complicated by 1 or more maternal delivery complications (severe perineal or cervical laceration, shoulder dystocia, cesarean delivery, or postpartum hemorrhage), 24,364 (22.1%) deliveries with 1 or more neonatal complications (hypoglycemia, respiratory problems, jaundice, low Apgar, NICU admission, or perinatal mortality), resulting in a total of 51,105 (46.3%) deliveries having at least 1 complication. In comparison to LGA Pop , customized LGA (any method) identified births with generally higher rates (PPV) and odds of severe perineal and cervical lacerations, cesarean delivery, and shoulder dystocia, yet the sensitivity and c-statistic for these outcomes was overall similar across LGA methods and was generally lowest in LGA Gard ( Table 4 ). Differences in the rates and odds of neonatal morbidities were more attenuated, compared to maternal morbidities, among custom LGA definitions vs LGA Pop , with similar or greater PPV using custom methods, but often lower sensitivity than LGA Pop ( Table 5 ). However, all of the customized definitions of LGA detected a slightly higher rate of perinatal mortality compared with LGA Pop , as seen by increased PPV with odds ratio point estimates about 30% greater in LGA Gard and LGA CSLcust than LGA Pop . Yet, the odds of mortality among LGA by any definition was statistically nonsignificant (95% confidence intervals all including 1.0).
