Revisiting the low birthweight paradox using sibling data with implications for the classification of low birthweight

Revisiting the low birthweight paradox using sibling data with implications for the... Abstract Background We examined the birthweight threshold for increased odds of neonatal death among second births based on their elder sibling’s birthweight category. Methods This population-based cohort study included 190 575 women who delivered their first two non-anomalous singleton live births in Missouri (1989–2005). We examined the birthweight distribution and neonatal mortality curves of second births whose elder sibling had low versus adequate/high birthweight. We determined the optimal cut-off point for the classification of low birthweight among infants in each group based on the Youden index. Results Infants whose elder sibling had low birthweight had a lower mean birthweight and a higher percentage of low birthweight infants versus those whose elder sibling had adequate/high birthweight, but low birthweight infants in the former group had a lower rate of neonatal mortality. Upon standardizing the birthweight distribution to a Z-scale, neonatal mortality rates became comparable between the two groups at every rescaled birthweight for Z-scores ≥−3.7. The optimal cut-off point for low birthweight was 2500 and 3000 g among infants whose elder sibling had low and adequate/high birthweight, respectively. Conclusions Using sibling data for the classification of LBW may enable the identification of average-sized infants who may be at increased risk of neonatal mortality. birthweight paradox, low birthweight, neonatal death, sibling data Introduction While birthweight has been largely accepted as a strong indicator of morbidity and mortality, it is increasingly recognized that using the same low birthweight indicator of 2 500 g for all infants may be misleading.1 Evidence suggests that optimal birthweight—or the birthweight associated with the lowest risk of poor neonatal outcomes—is not the same for all infants but is rather determined by genetic and environmental influences.2,3 Consequently, it is desirable that the classification of low birthweight be allowed to vary as well. While we cannot directly assess each infant’s optimal birthweight, using sibling data could be of great value in that regard as the elder sibling’s birthweight is one of the strongest predictors of the birthweight of younger siblings.4–7 Siblings share 50% of their genes, experience a very similar uterine environment, and are often exposed to similar prenatal environmental conditions.8 Scandinavian studies have shown that the rate of perinatal and/or neonatal mortality among second births is conditional on the elder sibling’s birthweight, with LBW infants having lower mortality rates if their elder sibling also had low birthweight compared to those whose elder sibling had adequate/high birthweight.4–7,9,10 The ‘low birthweight paradox’, which occurs when low birthweight infants of high-risk groups exhibit lower mortality than low birthweight infants in low-risk groups, may be may be uncovered by examining birthweight on a relative (Z-) scale.5,6,11 Indeed, having a birthweight <2500 g may reflect a greater deviation from optimal birthweight among infants whose elder sibling had adequate/high versus low birthweight, thereby resulting in a higher rate of mortality among low birthweight infants in the former versus latter group.1 Moreover, the current low birthweight indicator may fail to identify infants whose birthweight exceeds 2 500 g but who are nonetheless small relative to their optimal birthweight. Therefore, the purpose of our study was to: (i) examine the birthweight distribution and weight-specific neonatal mortality curve among second births whose elder sibling had low versus adequate/high birthweight; (ii) standardize the birthweight distribution to a Z-scale and examine the weight-specific neonatal mortality curve for the rescaled distribution among second births in each group; and (iii) explore the birthweight threshold for increased odds of neonatal death among second births in each group. Methods Study design We conducted a population-based retrospective cohort study using birth certificate data from the maternally linked Missouri birth registry (1989–2005). This registry links birth certificate data of siblings to their biological mother and contains information on maternal sociodemographic characteristics, medical and obstetrical conditions, pregnancy outcomes and neonatal status at birth for each birth in Missouri, USA. The Missouri linked birth registry has been deemed to be very reliable and is used as a ‘gold standard’ for the validation of other linked vital statistics data sets in the USA.12 The methods used to link birth records of successive pregnancies and their validation have been described in detail elsewhere.12 Briefly, record linkage is based on a combination of probabilistic and deterministic approaches. First, maternal characteristics are compared across each possible pair of records and a computer-generated score is assigned for each pair, reflecting the likelihood that the two records belong to the same mother. Record linkage is then generated for the pairs with the highest scores if those pairs had exact matching on maternal name and date of birth. The linkage rate for the Missouri birth registry is 93%.12 The Saint Louis University Institutional Review Board classified this study as exempt because all information was de-identified. Study sample Non-Hispanic White and Non-Hispanic Black women were eligible for our study if they delivered their first two singleton live births at 20–44 weeks gestation in Missouri, USA, between 1989 and 2005 (n = 236 908 women, representing 473 816 births). We only included deliveries occurring between 1 January 1989, and 31 December 2005 because the recording of the variable ‘clinical estimate of gestation’ used to estimate gestational age became mandatory in 1989, and data was available for the cohort until 2005. We did not include women of ‘Other’ races due to small cell size in bivariate analysis. Women were excluded from our study if their first and/or second birth had congenital anomalies (n = 6502) because the latter exerts a strong confounding effect on our outcome of interest.13 Women were also excluded if their first pregnancy was complicated by preterm birth (n = 22 811), by a medical condition that may negatively affect birthweight (e.g. pre-existing or gestational diabetes, pre-existing or gestational hypertension, pre-eclampsia/eclampsia; n = 16 764),14,15 or by neonatal death (n = 187). Such exclusion criteria were applied in an effort to restrict our sample to women whose first birth had a relatively low likelihood of being pathologically small or pathologically large.16 We also excluded women if they had missing data on birthweight or neonatal death (n = 69), resulting in a final analytical sample of 190 575 women. Definition of variables The exposure of interest was the birthweight of the younger sibling and the outcome of interest was neonatal mortality among younger siblings, defined as death occurring within the first 28 days of life.17 The neonatal mortality rate was defined as the number of neonatal deaths per 1000 live births. The birthweight of the elder sibling was used as a stratifying variable, and was dichotomized as low birthweight (<2500 g) versus adequate/high birthweight (≥2500 g).17 Statistical analysis Birthweight distribution and neonatal mortality curve We examined the birthweight distribution and the weight-specific neonatal mortality curve among infants whose elder sibling had low versus adequate/high birthweight. We compared the mean birthweight and the percentage of infants with low birthweight in the two groups using the independent sample t-test and the Pearson chi-square (χ2) test, respectively. We also compared the neonatal mortality rate among low birthweight infants in each group using the Pearson chi-square (χ2) test. Standardized Birthweight Distribution and Neonatal Mortality Curve. We standardized birthweight among infants in each group by subtracting the sample mean from each individual birthweight and dividing the difference by the standard deviation, thereby converting each observation to a Z-score and each birthweight distribution to a Z-scale with a mean of 0 and a standard deviation of 1. We then examined the weight-specific neonatal mortality curve for the rescaled birthweight distribution among infants whose elder sibling had low versus adequate/high birthweight. Birthweight threshold for increased odds of neonatal death We explored the birthweight threshold for increased odds of neonatal death among infants whose elder sibling had low versus adequate/high birthweight using the following graphical and statistical analyses. First, we graphed a two-way scatter plot of birthweight against a locally weighted smoother of predicted probabilities of neonatal death (y hat loess, ŷ) using loess, a non-parametric local regression that has been recommended when the nature of the association between two variables is not known.18 Second, we modeled birthweight in relation to neonatal death among infants in each group and examined the area under the receiver operating characteristic curve,19,20 as well as the sensitivity, specificity and Youden index (J) across the full range of possible cut-off points of birthweight, as has been previously done and recommended in clinical epidemiology.20–22 The Youden index, calculated as J = sensitivity + specificity − 1, represents the point with highest sensitivity and specificity, and has been used to determine the optimal cut-off point when sensitivity and specificity are equally desirable.23,24 In the context of our research question, we judged that low sensitivity may be more consequential than low specificity, and in turn, more important to avoid. Indeed, while low specificity may result in interventions being offered when not needed, low sensitivity may result in interventions being missed when actually needed, potentially leading to neonatal death. Thus, consistent with previous research where low sensitivity was deemed more consequential, the optimal cut-off point for birthweight was chosen as the birthweight with sensitivity that is higher than sensitivity at the birthweight with highest Youden index.22 All statistical analyses were performed using STATA (version 13.0, College Station, TX, USA). Results Figure 1 presents the birthweight distribution (top left quadrant) and the weight-specific neonatal mortality curve (lower left quadrant) among second births whose elder sibling had low birthweight (dashed line) versus adequate/high birthweight (solid line). Infants whose elder sibling had low birthweight exhibited a leftward shift in their birthweight distribution and their weight-specific neonatal mortality curve compared to those whose elder sibling had adequate/high birthweight. Infants whose elder sibling had low birthweight had a lower mean birthweight (2898.13 ± 537.05 g versus 3448.43 ± 507.51 g, P < 0.001) and a higher percentage of infants with low birthweight (19.1 versus 3.1%, P < 0.001) than those whose elder sibling had adequate/high birthweight; but low birthweight infants in the former group had a lower rate of neonatal mortality (1.3 versus 3.3%, P = 0.002; data not shown), with a crossover at around 2 000 g. Fig. 1 View largeDownload slide Actual (left panel) and standardized (right panel) birthweight distribution and weight-specific neonatal mortality curve among second births whose elder sibling had low birthweight (dashed line) or adequate/high birthweight (solid line), n = 190 575. Fig. 1 View largeDownload slide Actual (left panel) and standardized (right panel) birthweight distribution and weight-specific neonatal mortality curve among second births whose elder sibling had low birthweight (dashed line) or adequate/high birthweight (solid line), n = 190 575. The right panel of Fig. 1 shows the standardized birthweight distribution (top right quadrant) and the weight-specific neonatal mortality curve (lower right quadrant) among second births whose elder sibling had low birthweight (dashed line) versus adequate/high birthweight (solid line). The rescaled birthweight distributions of the two groups of infants looked essentially identical. The weight-specific neonatal mortality curves were comparable at every birthweight for Z-scores of −3 or greater but diverged at lower birthweight Z-scores. Infants whose elder sibling had low birthweight had significantly higher neonatal mortality rates compared to those whose elder sibling had adequate/high birthweight at birthweight Z-scores below −3.7, which corresponded to birthweight below 911 and 1570 g for the former and latter group of infants, respectively (data not shown). Figure 2 presents birthweight against a locally weighted smoother of predicted probabilities of neonatal death among infants whose elder sibling had low birthweight (dashed line) versus adequate/high birthweight (solid line). Predicted probabilities of neonatal death were close to null when birthweight was high and increased as birthweight decreased. The birthweight threshold below which predicted probabilities started to increase seemed to be around 2500 g among infants whose elder sibling had low birthweight and around 3000 g among infants whose elder had adequate/high birthweight. Figure 3 presents the receiver operating characteristic curve for birthweight in relation to neonatal death among infants whose elder sibling had low birthweight (left panel) versus adequate/high birthweight (right panel). To generate the receiver operating characteristic curve, birthweight was transformed into a discrete variable with 100 g increments, with higher values indicating lower birthweight (so that higher values reflect higher risk). The area under the receiver operating characteristic curve was 0.87 (95% confidence intervals [CI]: 0.78–0.96) and 0.82 (95% CI: 0.79–0.85) among infants whose elder sibling had low versus adequate/high birthweight, respectively (P = 0.288). Fig. 2 View largeDownload slide Birthweight against smoothed predicted probabilities of neonatal death among second births whose elder sibling had low birthweight (dashed line) or adequate/high birthweight (solid line), n = 190 575. Fig. 2 View largeDownload slide Birthweight against smoothed predicted probabilities of neonatal death among second births whose elder sibling had low birthweight (dashed line) or adequate/high birthweight (solid line), n = 190 575. Fig. 3 View largeDownload slide ROC curve for birthweight in relation to neonatal death among second births whose elder sibling had low birthweight (left panel) or adequate/high birthweight (right panel), n = 190 575. Fig. 3 View largeDownload slide ROC curve for birthweight in relation to neonatal death among second births whose elder sibling had low birthweight (left panel) or adequate/high birthweight (right panel), n = 190 575. Table 1 presents the sensitivity, specificity and Youden index at three possible cut-off points of birthweight. Among infants whose elder sibling had low birthweight, the cut-off point with the highest Youden index was 1 600 g (sensitivity = 60.0%, specificity = 98.1%, J = 0.581), while the chosen cut-off point corresponded to a birthweight of 2 500 g, as is currently used to define low birthweight (sensitivity = 73.3%, specificity = 81.1%, J = 0.545). In contrast, the cut-off point with the highest Youden index among infants whose elder sibling had adequate/high birthweight was 2 800 g (sensitivity = 61.7%, specificity = 92.0%, J = 0.537), and the chosen cut-off point with higher sensitivity was 3 000 g (sensitivity = 68.8%, specificity = 84.2%, J = 0.531). Based on the chosen cut-off points, the prevalence of low birthweight among second births whose elder sibling had low versus adequate/high birthweight was 19.1 versus 15.9%, respectively, and the rate of neonatal mortality was 13.0 versus 7.8 per 1 000 live births (P = 0.096), respectively. Table 1 Comparison of three possible birthweight cut-off points in relation to neonatal death among second births whose elder sibling had low versus adequate birthweight (n = 190 575) Birthweight cut-off points  Elder sibling with low birthweight, n = 4433  Elder sibling with adequate/high birthweight, n = 186 142  Chosena (2 500 g)  Highest Jb (1 600 g)  Current (2 500 g)  Chosena (3 000 g)  Highest Jb (2 800 g)  Current (2 500 g)  Youden index, J  0.545  0.581    0.531  0.537  0.535  Sensitivity (%)  73.3  60.0    68.8  61.7  56.4  Specificity (%)  81.1  98.1    84.2  92.0  97.0  False positive (%)  18.9  1.9    15.8  8.0  3.0  False negative (%)  26.7  40.0    31.2  38.3  43.6  Low birthweight, frequency (%)  845 (19.1)  95 (2.1)    29 558 (15.9)  15 123 (8.1)  5 753 (3.1)  Neonatal mortality, frequency (rate per 1 000)  11 (13.0)  9 (94.7)    232 (7.8)  208 (13.8)  190 (33.0)  Birthweight cut-off points  Elder sibling with low birthweight, n = 4433  Elder sibling with adequate/high birthweight, n = 186 142  Chosena (2 500 g)  Highest Jb (1 600 g)  Current (2 500 g)  Chosena (3 000 g)  Highest Jb (2 800 g)  Current (2 500 g)  Youden index, J  0.545  0.581    0.531  0.537  0.535  Sensitivity (%)  73.3  60.0    68.8  61.7  56.4  Specificity (%)  81.1  98.1    84.2  92.0  97.0  False positive (%)  18.9  1.9    15.8  8.0  3.0  False negative (%)  26.7  40.0    31.2  38.3  43.6  Low birthweight, frequency (%)  845 (19.1)  95 (2.1)    29 558 (15.9)  15 123 (8.1)  5 753 (3.1)  Neonatal mortality, frequency (rate per 1 000)  11 (13.0)  9 (94.7)    232 (7.8)  208 (13.8)  190 (33.0)  aChosen based on the Youden Index and clinical judgment. bJ, Youden index, calculated as: J = sensitivity + specificity – 1. Table 1 Comparison of three possible birthweight cut-off points in relation to neonatal death among second births whose elder sibling had low versus adequate birthweight (n = 190 575) Birthweight cut-off points  Elder sibling with low birthweight, n = 4433  Elder sibling with adequate/high birthweight, n = 186 142  Chosena (2 500 g)  Highest Jb (1 600 g)  Current (2 500 g)  Chosena (3 000 g)  Highest Jb (2 800 g)  Current (2 500 g)  Youden index, J  0.545  0.581    0.531  0.537  0.535  Sensitivity (%)  73.3  60.0    68.8  61.7  56.4  Specificity (%)  81.1  98.1    84.2  92.0  97.0  False positive (%)  18.9  1.9    15.8  8.0  3.0  False negative (%)  26.7  40.0    31.2  38.3  43.6  Low birthweight, frequency (%)  845 (19.1)  95 (2.1)    29 558 (15.9)  15 123 (8.1)  5 753 (3.1)  Neonatal mortality, frequency (rate per 1 000)  11 (13.0)  9 (94.7)    232 (7.8)  208 (13.8)  190 (33.0)  Birthweight cut-off points  Elder sibling with low birthweight, n = 4433  Elder sibling with adequate/high birthweight, n = 186 142  Chosena (2 500 g)  Highest Jb (1 600 g)  Current (2 500 g)  Chosena (3 000 g)  Highest Jb (2 800 g)  Current (2 500 g)  Youden index, J  0.545  0.581    0.531  0.537  0.535  Sensitivity (%)  73.3  60.0    68.8  61.7  56.4  Specificity (%)  81.1  98.1    84.2  92.0  97.0  False positive (%)  18.9  1.9    15.8  8.0  3.0  False negative (%)  26.7  40.0    31.2  38.3  43.6  Low birthweight, frequency (%)  845 (19.1)  95 (2.1)    29 558 (15.9)  15 123 (8.1)  5 753 (3.1)  Neonatal mortality, frequency (rate per 1 000)  11 (13.0)  9 (94.7)    232 (7.8)  208 (13.8)  190 (33.0)  aChosen based on the Youden Index and clinical judgment. bJ, Youden index, calculated as: J = sensitivity + specificity – 1. Discussion Main findings of this study We found that infants whose elder sibling had low birthweight exhibited a leftward shift in their birthweight distribution compared to those whose elder sibling had adequate/high birthweight, which reflects the strong positive correlation in birthweight between siblings.5 This was paralleled by a corresponding shift in the neonatal mortality curve such that, at the same birthweight, the rate of neonatal mortality was lower among infants whose elder sibling had low versus adequate/high birthweight, with a crossover at a birthweight of ~2000 g (Fig. 1). This ‘low birthweight paradox’ is consistent with findings from Scandinavian studies examining sibling birthweight in relation to neonatal and/or perinatal mortality, and may be uncovered by examining birthweight on a relative (Z-) scale rather than absolute scale.4–7 Upon standardization of the birthweight distribution to a Z-scale, neonatal mortality rates were comparable among second births whose elder sibling had low versus adequate/high birthweight at every rescaled birthweight despite an absolute difference of 550 g (Fig. 1). This indicates that a similar deviation from average birthweight would result in a similar risk of neonatal mortality among infants in each group. The difference in neonatal mortality rates between the two groups of infants became significant—and reversed—only for the smallest infants (Z-scores <−3.7, corresponding to birthweight <911 and <1570 g for those whose elder sibling had low versus adequate/high birthweight, respectively), when second births having a low birthweight elder sibling may be more likely to be born around the limits of viability. These results are in line with findings reported in a population-based cohort study in Norway,6 and suggest that using the same low birthweight indicator of 2500 g for infants with different birthweight distributions may be misguided. Indeed, given the strong correlation in sibling birthweight, having a birthweight <2500 g may reflect a larger deviation from optimal birthweight among infants whose elder sibling had adequate/high versus low birthweight.1,8,25 Furthermore, using the current low birthweight indicator of 2500 g may underestimate the prevalence of small infants among those whose elder sibling had adequate/high birthweight, with only 3.1% of infants in this group being classified as low birthweight in our study. It would also incorrectly classify almost half of infants who died in the neonatal period in this group as being healthy (Table 1). This may result in missed opportunities for intervention among such infants, potentially leading to neonatal mortality that could have potentially been prevented. Based on the Youden index and clinical judgment, the optimal cut-off point for defining low birthweight corresponded to a birthweight of 2500 and 3000 g among infants whose elder sibling had low and adequate/high birthweight, respectively, findings consistent with our graphical analysis. The cut-off point of 3000 g classified 15.9% of infants whose elder sibling had adequate/high birthweight as being low birthweight, making the prevalence of low birthweight comparable between the two groups of infants (Table 1). The chosen cut-off point of 3000 g could allow the identification of infants who are >2500 g but who are nonetheless small compared to the birthweight they were expected to reach and who may thus be at increased risk of neonatal mortality. What is already known on this topic Optimal birthweight may not be the same for all infants. Infants having a low birthweight exhibit lower mortality rates if their elder sibling also had low birthweight compared to those whose elder sibling had adequate/high birthweight.4–7,9,10 What this study adds This study demonstrates that a higher cut-off point for defining low birthweight may be warranted among infants whose elder sibling had adequate/high birthweight. Using the current low birthweight indicator of 2500 g would underestimate the prevalence of small infants among those whose elder sibling had adequate/high birthweight and would incorrectly classify almost half of infants who died in the neonatal period in this group as being healthy. In contrast, a cut-off point of 3000 g among these infants may enable the identification of infants >2500 g who may be at risk of poor outcomes. Limitations of this study Finings of our study should be interpreted in light of the following limitations. First, if the elder sibling has been exposed to pathological influences that adversely affect birthweight, then the elder sibling’s birthweight may be a suboptimal predictor of the younger sibling’s expected birthweight. We attempted to address this issue by excluding women whose first pregnancy was complicated by preterm birth, neonatal death, or a medical condition, thereby excluding elder siblings who may be pathologically small or large. Second, the distinction of a live birth from a stillbirth may not always be straightforward, or identified consistently by all delivery personnel, especially if some signs of respiration or movement were observed before death.26 If such misclassification were to be present, it would likely be random and would thus result in non-differential misclassification that may bias the results of our study toward the null. Third, our analysis was restricted to second births. However, given the strong correlation in birthweight between higher-order births, similar associations may be present among higher-order births as well.27 Furthermore, it has been reported that the mother’s own birthweight is a good predictor of her offspring’s expected birthweight. Thus, similar results may potentially be observed among primiparous women, upon stratifying by the mother’s own birthweight.1,27,28 Fourth, our findings may only be generalizable to infants born to Non-Hispanic White and Non-Hispanic Black women in Missouri, or to other populations with similar characteristics. Future studies using an independent external data set are needed to further validate our finding in different populations.29 Despite these limitations, the current study provides insight on the low birthweight paradox using sibling data in a US population and suggests the use of different cut-off points for the classification of low birthweight, based on sibling data. This may enable the identification of infants >2500 g who may nonetheless be at increased risk of neonatal mortality and who may benefit from heightened surveillance and earlier intervention, if needed. Strengths of this study include its population-based cohort design, including all eligible first and second births in Missouri that could be linked during the study period, and the large number of births included, enabling us to examine an outcome of low prevalence such as neonatal mortality. Conclusion Our findings suggest that the birthweight of the elder sibling should be taken into consideration in the classification of low birthweight. A higher cut-off point for defining low birthweight may be warranted among infants whose elder sibling had adequate/high birthweight. Such indicator may enable the identification of infants >2500 g who may be at risk of poor outcomes and who may be missed if using the same low birthweight indicator of 2500 g for all infants. Future studies should use sibling data to further validate our findings and evaluate associations between low birthweight and other neonatal and infant outcomes. Funding There were no sources of funding for the study, for the authors or for the article preparation. Conflict of interest The authors report no conflict of interest and have no financial relationships relevant to this article. Acknowledgements The authors acknowledge and appreciate the Missouri Deparment of Health and Senior Services, Section of Public Health Practice and Administrative Support as the original source of the data. The analysis, interpretations and conclusions in the present study are those of the authors and not the Missouri Department of Health and Senior Services, Section of Public Health Practice and Administrative Support. References 1 Bakketeig LS. ‘Repeater’ studies—development of a new research field. Norsk Epidemiologi  2007; 17: 153– 6. 2 Zhang X, Cnattingus S, Platt RW et al.  . Are babies born to short, primiparous, or thin mothers ‘normally’ or ‘abnormally’ small? J Pediatr  2007; 150: 603– 7. Google Scholar CrossRef Search ADS PubMed  3 Haig D. Meditations on birth weight: is it better to reduce the variance or increase the mean? Epidemiology  2003; 14: 490– 2. Google Scholar PubMed  4 Bakketeig LS, Hoffman HJ. The tendency to repeat gestational age and birthweight in successive births, related to perinatal survival. Acta Obstet Gynecol Scand  1983; 62: 385– 92. Google Scholar CrossRef Search ADS PubMed  5 Skjaerven R, Wilcox A, Russell D. Birthweight and perinatal mortality of second births conditional on weight of the first. Int J Epidemiol  1988; 17: 830– 8. Google Scholar CrossRef Search ADS PubMed  6 Melve KK, Skjaerven R. Birthweight and perinatal mortality: paradoxes, social class, and sibling dependencies. Int J Epidemiol  2003; 32: 625– 32. Google Scholar CrossRef Search ADS PubMed  7 Bakketeig LS, Jacobsen G, Skjaerven R et al.  . Low birthweight and mortality: the tendency to repeat low birthweight and its association with early neonatal and infant morbidity and mortality. Paediatr Perinat Epidemiol  2006; 20: 507– 11. Google Scholar CrossRef Search ADS PubMed  8 Lunde A, Melve KK, Gjessing HK et al.  . Genetic and environmental influences on birth weight, birth length, head circumference, and gestational age by use of population-based parent-offspring data. Am J Epidemiol  2007; 165: 734– 41. Google Scholar CrossRef Search ADS PubMed  9 Basso O, Frydenberg M, Olsen SF, Olsen J. Two definitions of small size at birth as predictors of motor development at six months. Epidemiology  2005; 16: 657– 63. Google Scholar CrossRef Search ADS PubMed  10 Pedersen CB, Sun Y, Vestergaard M et al.  . Assessing fetal growth impairments based on family data as a tool for identifying high-risk babies. An example with neonatal mortality. BMC Pregnancy Childbirth  2007; 7: 28. Google Scholar CrossRef Search ADS PubMed  11 Wilcox AJ. On the importance-and unimportance-of birthweight. Int J Epidemiol  2001; 30: 1233– 41. Google Scholar CrossRef Search ADS PubMed  12 Herman A, McCarthy B, Bakewell J 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 1): 5– 22. Google Scholar CrossRef Search ADS PubMed  13 Mathews TJ, MacDorman MF. Infant mortality statistics from the 2010 period linked birth/infant death data set. Natl Vital Stat Rep  2013; 62( 8): 1– 26. Google Scholar PubMed  14 Ananth CV, Basso O. Impact of pregnancy-induced hypertension on stillbirth and neonatal mortality. Epidemiology  2010; 21: 118– 23. Google Scholar CrossRef Search ADS PubMed  15 Kc K, Shakya S, Zhang H. Gestational diabetes mellitus and macrosomia: a literature review. Ann Nutr Metab  2015; 66( Suppl 2): 14– 20. Google Scholar CrossRef Search ADS PubMed  16 U.S. Department of Health and Human Services Child Health USA. Rockville, Maryland: US Department of Health and Human Services, 2013 17 World Health Organization (WHO). International Statistical Classification of Diseases and Related Health Problems 10th Revision (ICD-10). 2015. http://apps.who.int/classifications/icd10/browse/2015/en#/XV (April 2017, date last accessed). 18 Jacoby WG. Loess: a nonparametric, graphical tool for depicting relationships between variables. Electoral Studies  2000; 19: 577– 613. Google Scholar CrossRef Search ADS   19 DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics  1988; 44: 837– 45. Google Scholar CrossRef Search ADS PubMed  20 Fischer JE, Bachmann LM, Jaeschke R. A readers’ guide to the interpretation of diagnostic test properties: clinical example of sepsis. Intensive Care Med  2003; 29: 1043– 51. Google Scholar CrossRef Search ADS PubMed  21 Hajian-Tilaki K. Receiver operating charateristic (ROC) curve analysis for medical diagnostic test evaluation. Caspian J Intern Med  2013; 4: 627– 35. Google Scholar PubMed  22 Nanishi K, Green J, Taguri M et al.  . Determining a cut-off point for scores of the breastfeeding self-efficacy scale–short form: secondary data analysis of an intervention study in Japan. PLoS One  2015; 10( 6): e0129698. Google Scholar CrossRef Search ADS PubMed  23 Perkins NJ, Schisterman EF. The inconsistency of ‘optimal’ cutpoints obtained using two criteria based on the receiver operating characteristic curve. Am J Epidemiol  2006; 163: 670– 5. Google Scholar CrossRef Search ADS PubMed  24 Youden WJ. Index for rating diagnostic tests. Cancer  1950; 3( 1): 32– 5. Google Scholar CrossRef Search ADS PubMed  25 Dunger DB, Petry CJ, Ong KK. Genetics of size at birth. Diabetes Care  2007; 30( Suppl 2): S150– 5. Google Scholar CrossRef Search ADS PubMed  26 Kramer MS, Platt RW, Yang H et al.  . Registration artifacts in international comparisons of infant mortality. Paediatr Perinat Epidemiol  2002; 16: 16– 22. Google Scholar CrossRef Search ADS PubMed  27 Skjaerven R, Gjessing HK, Bakketeig LS. New standards for birth weight by gestational age using family data. Am J Obstet Gynecol  2000; 183: 689– 96. Google Scholar CrossRef Search ADS PubMed  28 Skjaerven R, Wilcox AJ, Oyen N et al.  . Mothers’ birth weight and survival of their offspring: population based study. Br Med J  1997; 314: 1376– 80. Google Scholar CrossRef Search ADS   29 Riley RD, Ensor J, Snell KI et al.  . External validation of clinical prediction models using big datasets from e-health records or IPD meta-analysis: opportunities and challenges. Br Med J  2016; 353: i3140. Google Scholar CrossRef Search ADS   © The Author(s) 2018. Published by Oxford University Press on behalf of Faculty of Public Health. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of Public Health Oxford University Press

Revisiting the low birthweight paradox using sibling data with implications for the classification of low birthweight

Loading next page...
 
/lp/ou_press/revisiting-the-low-birthweight-paradox-using-sibling-data-with-RFUzu3QN89
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press on behalf of Faculty of Public Health. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com
ISSN
1741-3842
eISSN
1741-3850
D.O.I.
10.1093/pubmed/fdy087
Publisher site
See Article on Publisher Site

Abstract

Abstract Background We examined the birthweight threshold for increased odds of neonatal death among second births based on their elder sibling’s birthweight category. Methods This population-based cohort study included 190 575 women who delivered their first two non-anomalous singleton live births in Missouri (1989–2005). We examined the birthweight distribution and neonatal mortality curves of second births whose elder sibling had low versus adequate/high birthweight. We determined the optimal cut-off point for the classification of low birthweight among infants in each group based on the Youden index. Results Infants whose elder sibling had low birthweight had a lower mean birthweight and a higher percentage of low birthweight infants versus those whose elder sibling had adequate/high birthweight, but low birthweight infants in the former group had a lower rate of neonatal mortality. Upon standardizing the birthweight distribution to a Z-scale, neonatal mortality rates became comparable between the two groups at every rescaled birthweight for Z-scores ≥−3.7. The optimal cut-off point for low birthweight was 2500 and 3000 g among infants whose elder sibling had low and adequate/high birthweight, respectively. Conclusions Using sibling data for the classification of LBW may enable the identification of average-sized infants who may be at increased risk of neonatal mortality. birthweight paradox, low birthweight, neonatal death, sibling data Introduction While birthweight has been largely accepted as a strong indicator of morbidity and mortality, it is increasingly recognized that using the same low birthweight indicator of 2 500 g for all infants may be misleading.1 Evidence suggests that optimal birthweight—or the birthweight associated with the lowest risk of poor neonatal outcomes—is not the same for all infants but is rather determined by genetic and environmental influences.2,3 Consequently, it is desirable that the classification of low birthweight be allowed to vary as well. While we cannot directly assess each infant’s optimal birthweight, using sibling data could be of great value in that regard as the elder sibling’s birthweight is one of the strongest predictors of the birthweight of younger siblings.4–7 Siblings share 50% of their genes, experience a very similar uterine environment, and are often exposed to similar prenatal environmental conditions.8 Scandinavian studies have shown that the rate of perinatal and/or neonatal mortality among second births is conditional on the elder sibling’s birthweight, with LBW infants having lower mortality rates if their elder sibling also had low birthweight compared to those whose elder sibling had adequate/high birthweight.4–7,9,10 The ‘low birthweight paradox’, which occurs when low birthweight infants of high-risk groups exhibit lower mortality than low birthweight infants in low-risk groups, may be may be uncovered by examining birthweight on a relative (Z-) scale.5,6,11 Indeed, having a birthweight <2500 g may reflect a greater deviation from optimal birthweight among infants whose elder sibling had adequate/high versus low birthweight, thereby resulting in a higher rate of mortality among low birthweight infants in the former versus latter group.1 Moreover, the current low birthweight indicator may fail to identify infants whose birthweight exceeds 2 500 g but who are nonetheless small relative to their optimal birthweight. Therefore, the purpose of our study was to: (i) examine the birthweight distribution and weight-specific neonatal mortality curve among second births whose elder sibling had low versus adequate/high birthweight; (ii) standardize the birthweight distribution to a Z-scale and examine the weight-specific neonatal mortality curve for the rescaled distribution among second births in each group; and (iii) explore the birthweight threshold for increased odds of neonatal death among second births in each group. Methods Study design We conducted a population-based retrospective cohort study using birth certificate data from the maternally linked Missouri birth registry (1989–2005). This registry links birth certificate data of siblings to their biological mother and contains information on maternal sociodemographic characteristics, medical and obstetrical conditions, pregnancy outcomes and neonatal status at birth for each birth in Missouri, USA. The Missouri linked birth registry has been deemed to be very reliable and is used as a ‘gold standard’ for the validation of other linked vital statistics data sets in the USA.12 The methods used to link birth records of successive pregnancies and their validation have been described in detail elsewhere.12 Briefly, record linkage is based on a combination of probabilistic and deterministic approaches. First, maternal characteristics are compared across each possible pair of records and a computer-generated score is assigned for each pair, reflecting the likelihood that the two records belong to the same mother. Record linkage is then generated for the pairs with the highest scores if those pairs had exact matching on maternal name and date of birth. The linkage rate for the Missouri birth registry is 93%.12 The Saint Louis University Institutional Review Board classified this study as exempt because all information was de-identified. Study sample Non-Hispanic White and Non-Hispanic Black women were eligible for our study if they delivered their first two singleton live births at 20–44 weeks gestation in Missouri, USA, between 1989 and 2005 (n = 236 908 women, representing 473 816 births). We only included deliveries occurring between 1 January 1989, and 31 December 2005 because the recording of the variable ‘clinical estimate of gestation’ used to estimate gestational age became mandatory in 1989, and data was available for the cohort until 2005. We did not include women of ‘Other’ races due to small cell size in bivariate analysis. Women were excluded from our study if their first and/or second birth had congenital anomalies (n = 6502) because the latter exerts a strong confounding effect on our outcome of interest.13 Women were also excluded if their first pregnancy was complicated by preterm birth (n = 22 811), by a medical condition that may negatively affect birthweight (e.g. pre-existing or gestational diabetes, pre-existing or gestational hypertension, pre-eclampsia/eclampsia; n = 16 764),14,15 or by neonatal death (n = 187). Such exclusion criteria were applied in an effort to restrict our sample to women whose first birth had a relatively low likelihood of being pathologically small or pathologically large.16 We also excluded women if they had missing data on birthweight or neonatal death (n = 69), resulting in a final analytical sample of 190 575 women. Definition of variables The exposure of interest was the birthweight of the younger sibling and the outcome of interest was neonatal mortality among younger siblings, defined as death occurring within the first 28 days of life.17 The neonatal mortality rate was defined as the number of neonatal deaths per 1000 live births. The birthweight of the elder sibling was used as a stratifying variable, and was dichotomized as low birthweight (<2500 g) versus adequate/high birthweight (≥2500 g).17 Statistical analysis Birthweight distribution and neonatal mortality curve We examined the birthweight distribution and the weight-specific neonatal mortality curve among infants whose elder sibling had low versus adequate/high birthweight. We compared the mean birthweight and the percentage of infants with low birthweight in the two groups using the independent sample t-test and the Pearson chi-square (χ2) test, respectively. We also compared the neonatal mortality rate among low birthweight infants in each group using the Pearson chi-square (χ2) test. Standardized Birthweight Distribution and Neonatal Mortality Curve. We standardized birthweight among infants in each group by subtracting the sample mean from each individual birthweight and dividing the difference by the standard deviation, thereby converting each observation to a Z-score and each birthweight distribution to a Z-scale with a mean of 0 and a standard deviation of 1. We then examined the weight-specific neonatal mortality curve for the rescaled birthweight distribution among infants whose elder sibling had low versus adequate/high birthweight. Birthweight threshold for increased odds of neonatal death We explored the birthweight threshold for increased odds of neonatal death among infants whose elder sibling had low versus adequate/high birthweight using the following graphical and statistical analyses. First, we graphed a two-way scatter plot of birthweight against a locally weighted smoother of predicted probabilities of neonatal death (y hat loess, ŷ) using loess, a non-parametric local regression that has been recommended when the nature of the association between two variables is not known.18 Second, we modeled birthweight in relation to neonatal death among infants in each group and examined the area under the receiver operating characteristic curve,19,20 as well as the sensitivity, specificity and Youden index (J) across the full range of possible cut-off points of birthweight, as has been previously done and recommended in clinical epidemiology.20–22 The Youden index, calculated as J = sensitivity + specificity − 1, represents the point with highest sensitivity and specificity, and has been used to determine the optimal cut-off point when sensitivity and specificity are equally desirable.23,24 In the context of our research question, we judged that low sensitivity may be more consequential than low specificity, and in turn, more important to avoid. Indeed, while low specificity may result in interventions being offered when not needed, low sensitivity may result in interventions being missed when actually needed, potentially leading to neonatal death. Thus, consistent with previous research where low sensitivity was deemed more consequential, the optimal cut-off point for birthweight was chosen as the birthweight with sensitivity that is higher than sensitivity at the birthweight with highest Youden index.22 All statistical analyses were performed using STATA (version 13.0, College Station, TX, USA). Results Figure 1 presents the birthweight distribution (top left quadrant) and the weight-specific neonatal mortality curve (lower left quadrant) among second births whose elder sibling had low birthweight (dashed line) versus adequate/high birthweight (solid line). Infants whose elder sibling had low birthweight exhibited a leftward shift in their birthweight distribution and their weight-specific neonatal mortality curve compared to those whose elder sibling had adequate/high birthweight. Infants whose elder sibling had low birthweight had a lower mean birthweight (2898.13 ± 537.05 g versus 3448.43 ± 507.51 g, P < 0.001) and a higher percentage of infants with low birthweight (19.1 versus 3.1%, P < 0.001) than those whose elder sibling had adequate/high birthweight; but low birthweight infants in the former group had a lower rate of neonatal mortality (1.3 versus 3.3%, P = 0.002; data not shown), with a crossover at around 2 000 g. Fig. 1 View largeDownload slide Actual (left panel) and standardized (right panel) birthweight distribution and weight-specific neonatal mortality curve among second births whose elder sibling had low birthweight (dashed line) or adequate/high birthweight (solid line), n = 190 575. Fig. 1 View largeDownload slide Actual (left panel) and standardized (right panel) birthweight distribution and weight-specific neonatal mortality curve among second births whose elder sibling had low birthweight (dashed line) or adequate/high birthweight (solid line), n = 190 575. The right panel of Fig. 1 shows the standardized birthweight distribution (top right quadrant) and the weight-specific neonatal mortality curve (lower right quadrant) among second births whose elder sibling had low birthweight (dashed line) versus adequate/high birthweight (solid line). The rescaled birthweight distributions of the two groups of infants looked essentially identical. The weight-specific neonatal mortality curves were comparable at every birthweight for Z-scores of −3 or greater but diverged at lower birthweight Z-scores. Infants whose elder sibling had low birthweight had significantly higher neonatal mortality rates compared to those whose elder sibling had adequate/high birthweight at birthweight Z-scores below −3.7, which corresponded to birthweight below 911 and 1570 g for the former and latter group of infants, respectively (data not shown). Figure 2 presents birthweight against a locally weighted smoother of predicted probabilities of neonatal death among infants whose elder sibling had low birthweight (dashed line) versus adequate/high birthweight (solid line). Predicted probabilities of neonatal death were close to null when birthweight was high and increased as birthweight decreased. The birthweight threshold below which predicted probabilities started to increase seemed to be around 2500 g among infants whose elder sibling had low birthweight and around 3000 g among infants whose elder had adequate/high birthweight. Figure 3 presents the receiver operating characteristic curve for birthweight in relation to neonatal death among infants whose elder sibling had low birthweight (left panel) versus adequate/high birthweight (right panel). To generate the receiver operating characteristic curve, birthweight was transformed into a discrete variable with 100 g increments, with higher values indicating lower birthweight (so that higher values reflect higher risk). The area under the receiver operating characteristic curve was 0.87 (95% confidence intervals [CI]: 0.78–0.96) and 0.82 (95% CI: 0.79–0.85) among infants whose elder sibling had low versus adequate/high birthweight, respectively (P = 0.288). Fig. 2 View largeDownload slide Birthweight against smoothed predicted probabilities of neonatal death among second births whose elder sibling had low birthweight (dashed line) or adequate/high birthweight (solid line), n = 190 575. Fig. 2 View largeDownload slide Birthweight against smoothed predicted probabilities of neonatal death among second births whose elder sibling had low birthweight (dashed line) or adequate/high birthweight (solid line), n = 190 575. Fig. 3 View largeDownload slide ROC curve for birthweight in relation to neonatal death among second births whose elder sibling had low birthweight (left panel) or adequate/high birthweight (right panel), n = 190 575. Fig. 3 View largeDownload slide ROC curve for birthweight in relation to neonatal death among second births whose elder sibling had low birthweight (left panel) or adequate/high birthweight (right panel), n = 190 575. Table 1 presents the sensitivity, specificity and Youden index at three possible cut-off points of birthweight. Among infants whose elder sibling had low birthweight, the cut-off point with the highest Youden index was 1 600 g (sensitivity = 60.0%, specificity = 98.1%, J = 0.581), while the chosen cut-off point corresponded to a birthweight of 2 500 g, as is currently used to define low birthweight (sensitivity = 73.3%, specificity = 81.1%, J = 0.545). In contrast, the cut-off point with the highest Youden index among infants whose elder sibling had adequate/high birthweight was 2 800 g (sensitivity = 61.7%, specificity = 92.0%, J = 0.537), and the chosen cut-off point with higher sensitivity was 3 000 g (sensitivity = 68.8%, specificity = 84.2%, J = 0.531). Based on the chosen cut-off points, the prevalence of low birthweight among second births whose elder sibling had low versus adequate/high birthweight was 19.1 versus 15.9%, respectively, and the rate of neonatal mortality was 13.0 versus 7.8 per 1 000 live births (P = 0.096), respectively. Table 1 Comparison of three possible birthweight cut-off points in relation to neonatal death among second births whose elder sibling had low versus adequate birthweight (n = 190 575) Birthweight cut-off points  Elder sibling with low birthweight, n = 4433  Elder sibling with adequate/high birthweight, n = 186 142  Chosena (2 500 g)  Highest Jb (1 600 g)  Current (2 500 g)  Chosena (3 000 g)  Highest Jb (2 800 g)  Current (2 500 g)  Youden index, J  0.545  0.581    0.531  0.537  0.535  Sensitivity (%)  73.3  60.0    68.8  61.7  56.4  Specificity (%)  81.1  98.1    84.2  92.0  97.0  False positive (%)  18.9  1.9    15.8  8.0  3.0  False negative (%)  26.7  40.0    31.2  38.3  43.6  Low birthweight, frequency (%)  845 (19.1)  95 (2.1)    29 558 (15.9)  15 123 (8.1)  5 753 (3.1)  Neonatal mortality, frequency (rate per 1 000)  11 (13.0)  9 (94.7)    232 (7.8)  208 (13.8)  190 (33.0)  Birthweight cut-off points  Elder sibling with low birthweight, n = 4433  Elder sibling with adequate/high birthweight, n = 186 142  Chosena (2 500 g)  Highest Jb (1 600 g)  Current (2 500 g)  Chosena (3 000 g)  Highest Jb (2 800 g)  Current (2 500 g)  Youden index, J  0.545  0.581    0.531  0.537  0.535  Sensitivity (%)  73.3  60.0    68.8  61.7  56.4  Specificity (%)  81.1  98.1    84.2  92.0  97.0  False positive (%)  18.9  1.9    15.8  8.0  3.0  False negative (%)  26.7  40.0    31.2  38.3  43.6  Low birthweight, frequency (%)  845 (19.1)  95 (2.1)    29 558 (15.9)  15 123 (8.1)  5 753 (3.1)  Neonatal mortality, frequency (rate per 1 000)  11 (13.0)  9 (94.7)    232 (7.8)  208 (13.8)  190 (33.0)  aChosen based on the Youden Index and clinical judgment. bJ, Youden index, calculated as: J = sensitivity + specificity – 1. Table 1 Comparison of three possible birthweight cut-off points in relation to neonatal death among second births whose elder sibling had low versus adequate birthweight (n = 190 575) Birthweight cut-off points  Elder sibling with low birthweight, n = 4433  Elder sibling with adequate/high birthweight, n = 186 142  Chosena (2 500 g)  Highest Jb (1 600 g)  Current (2 500 g)  Chosena (3 000 g)  Highest Jb (2 800 g)  Current (2 500 g)  Youden index, J  0.545  0.581    0.531  0.537  0.535  Sensitivity (%)  73.3  60.0    68.8  61.7  56.4  Specificity (%)  81.1  98.1    84.2  92.0  97.0  False positive (%)  18.9  1.9    15.8  8.0  3.0  False negative (%)  26.7  40.0    31.2  38.3  43.6  Low birthweight, frequency (%)  845 (19.1)  95 (2.1)    29 558 (15.9)  15 123 (8.1)  5 753 (3.1)  Neonatal mortality, frequency (rate per 1 000)  11 (13.0)  9 (94.7)    232 (7.8)  208 (13.8)  190 (33.0)  Birthweight cut-off points  Elder sibling with low birthweight, n = 4433  Elder sibling with adequate/high birthweight, n = 186 142  Chosena (2 500 g)  Highest Jb (1 600 g)  Current (2 500 g)  Chosena (3 000 g)  Highest Jb (2 800 g)  Current (2 500 g)  Youden index, J  0.545  0.581    0.531  0.537  0.535  Sensitivity (%)  73.3  60.0    68.8  61.7  56.4  Specificity (%)  81.1  98.1    84.2  92.0  97.0  False positive (%)  18.9  1.9    15.8  8.0  3.0  False negative (%)  26.7  40.0    31.2  38.3  43.6  Low birthweight, frequency (%)  845 (19.1)  95 (2.1)    29 558 (15.9)  15 123 (8.1)  5 753 (3.1)  Neonatal mortality, frequency (rate per 1 000)  11 (13.0)  9 (94.7)    232 (7.8)  208 (13.8)  190 (33.0)  aChosen based on the Youden Index and clinical judgment. bJ, Youden index, calculated as: J = sensitivity + specificity – 1. Discussion Main findings of this study We found that infants whose elder sibling had low birthweight exhibited a leftward shift in their birthweight distribution compared to those whose elder sibling had adequate/high birthweight, which reflects the strong positive correlation in birthweight between siblings.5 This was paralleled by a corresponding shift in the neonatal mortality curve such that, at the same birthweight, the rate of neonatal mortality was lower among infants whose elder sibling had low versus adequate/high birthweight, with a crossover at a birthweight of ~2000 g (Fig. 1). This ‘low birthweight paradox’ is consistent with findings from Scandinavian studies examining sibling birthweight in relation to neonatal and/or perinatal mortality, and may be uncovered by examining birthweight on a relative (Z-) scale rather than absolute scale.4–7 Upon standardization of the birthweight distribution to a Z-scale, neonatal mortality rates were comparable among second births whose elder sibling had low versus adequate/high birthweight at every rescaled birthweight despite an absolute difference of 550 g (Fig. 1). This indicates that a similar deviation from average birthweight would result in a similar risk of neonatal mortality among infants in each group. The difference in neonatal mortality rates between the two groups of infants became significant—and reversed—only for the smallest infants (Z-scores <−3.7, corresponding to birthweight <911 and <1570 g for those whose elder sibling had low versus adequate/high birthweight, respectively), when second births having a low birthweight elder sibling may be more likely to be born around the limits of viability. These results are in line with findings reported in a population-based cohort study in Norway,6 and suggest that using the same low birthweight indicator of 2500 g for infants with different birthweight distributions may be misguided. Indeed, given the strong correlation in sibling birthweight, having a birthweight <2500 g may reflect a larger deviation from optimal birthweight among infants whose elder sibling had adequate/high versus low birthweight.1,8,25 Furthermore, using the current low birthweight indicator of 2500 g may underestimate the prevalence of small infants among those whose elder sibling had adequate/high birthweight, with only 3.1% of infants in this group being classified as low birthweight in our study. It would also incorrectly classify almost half of infants who died in the neonatal period in this group as being healthy (Table 1). This may result in missed opportunities for intervention among such infants, potentially leading to neonatal mortality that could have potentially been prevented. Based on the Youden index and clinical judgment, the optimal cut-off point for defining low birthweight corresponded to a birthweight of 2500 and 3000 g among infants whose elder sibling had low and adequate/high birthweight, respectively, findings consistent with our graphical analysis. The cut-off point of 3000 g classified 15.9% of infants whose elder sibling had adequate/high birthweight as being low birthweight, making the prevalence of low birthweight comparable between the two groups of infants (Table 1). The chosen cut-off point of 3000 g could allow the identification of infants who are >2500 g but who are nonetheless small compared to the birthweight they were expected to reach and who may thus be at increased risk of neonatal mortality. What is already known on this topic Optimal birthweight may not be the same for all infants. Infants having a low birthweight exhibit lower mortality rates if their elder sibling also had low birthweight compared to those whose elder sibling had adequate/high birthweight.4–7,9,10 What this study adds This study demonstrates that a higher cut-off point for defining low birthweight may be warranted among infants whose elder sibling had adequate/high birthweight. Using the current low birthweight indicator of 2500 g would underestimate the prevalence of small infants among those whose elder sibling had adequate/high birthweight and would incorrectly classify almost half of infants who died in the neonatal period in this group as being healthy. In contrast, a cut-off point of 3000 g among these infants may enable the identification of infants >2500 g who may be at risk of poor outcomes. Limitations of this study Finings of our study should be interpreted in light of the following limitations. First, if the elder sibling has been exposed to pathological influences that adversely affect birthweight, then the elder sibling’s birthweight may be a suboptimal predictor of the younger sibling’s expected birthweight. We attempted to address this issue by excluding women whose first pregnancy was complicated by preterm birth, neonatal death, or a medical condition, thereby excluding elder siblings who may be pathologically small or large. Second, the distinction of a live birth from a stillbirth may not always be straightforward, or identified consistently by all delivery personnel, especially if some signs of respiration or movement were observed before death.26 If such misclassification were to be present, it would likely be random and would thus result in non-differential misclassification that may bias the results of our study toward the null. Third, our analysis was restricted to second births. However, given the strong correlation in birthweight between higher-order births, similar associations may be present among higher-order births as well.27 Furthermore, it has been reported that the mother’s own birthweight is a good predictor of her offspring’s expected birthweight. Thus, similar results may potentially be observed among primiparous women, upon stratifying by the mother’s own birthweight.1,27,28 Fourth, our findings may only be generalizable to infants born to Non-Hispanic White and Non-Hispanic Black women in Missouri, or to other populations with similar characteristics. Future studies using an independent external data set are needed to further validate our finding in different populations.29 Despite these limitations, the current study provides insight on the low birthweight paradox using sibling data in a US population and suggests the use of different cut-off points for the classification of low birthweight, based on sibling data. This may enable the identification of infants >2500 g who may nonetheless be at increased risk of neonatal mortality and who may benefit from heightened surveillance and earlier intervention, if needed. Strengths of this study include its population-based cohort design, including all eligible first and second births in Missouri that could be linked during the study period, and the large number of births included, enabling us to examine an outcome of low prevalence such as neonatal mortality. Conclusion Our findings suggest that the birthweight of the elder sibling should be taken into consideration in the classification of low birthweight. A higher cut-off point for defining low birthweight may be warranted among infants whose elder sibling had adequate/high birthweight. Such indicator may enable the identification of infants >2500 g who may be at risk of poor outcomes and who may be missed if using the same low birthweight indicator of 2500 g for all infants. Future studies should use sibling data to further validate our findings and evaluate associations between low birthweight and other neonatal and infant outcomes. Funding There were no sources of funding for the study, for the authors or for the article preparation. Conflict of interest The authors report no conflict of interest and have no financial relationships relevant to this article. Acknowledgements The authors acknowledge and appreciate the Missouri Deparment of Health and Senior Services, Section of Public Health Practice and Administrative Support as the original source of the data. The analysis, interpretations and conclusions in the present study are those of the authors and not the Missouri Department of Health and Senior Services, Section of Public Health Practice and Administrative Support. References 1 Bakketeig LS. ‘Repeater’ studies—development of a new research field. Norsk Epidemiologi  2007; 17: 153– 6. 2 Zhang X, Cnattingus S, Platt RW et al.  . Are babies born to short, primiparous, or thin mothers ‘normally’ or ‘abnormally’ small? J Pediatr  2007; 150: 603– 7. Google Scholar CrossRef Search ADS PubMed  3 Haig D. Meditations on birth weight: is it better to reduce the variance or increase the mean? Epidemiology  2003; 14: 490– 2. Google Scholar PubMed  4 Bakketeig LS, Hoffman HJ. The tendency to repeat gestational age and birthweight in successive births, related to perinatal survival. Acta Obstet Gynecol Scand  1983; 62: 385– 92. Google Scholar CrossRef Search ADS PubMed  5 Skjaerven R, Wilcox A, Russell D. Birthweight and perinatal mortality of second births conditional on weight of the first. Int J Epidemiol  1988; 17: 830– 8. Google Scholar CrossRef Search ADS PubMed  6 Melve KK, Skjaerven R. Birthweight and perinatal mortality: paradoxes, social class, and sibling dependencies. Int J Epidemiol  2003; 32: 625– 32. Google Scholar CrossRef Search ADS PubMed  7 Bakketeig LS, Jacobsen G, Skjaerven R et al.  . Low birthweight and mortality: the tendency to repeat low birthweight and its association with early neonatal and infant morbidity and mortality. Paediatr Perinat Epidemiol  2006; 20: 507– 11. Google Scholar CrossRef Search ADS PubMed  8 Lunde A, Melve KK, Gjessing HK et al.  . Genetic and environmental influences on birth weight, birth length, head circumference, and gestational age by use of population-based parent-offspring data. Am J Epidemiol  2007; 165: 734– 41. Google Scholar CrossRef Search ADS PubMed  9 Basso O, Frydenberg M, Olsen SF, Olsen J. Two definitions of small size at birth as predictors of motor development at six months. Epidemiology  2005; 16: 657– 63. Google Scholar CrossRef Search ADS PubMed  10 Pedersen CB, Sun Y, Vestergaard M et al.  . Assessing fetal growth impairments based on family data as a tool for identifying high-risk babies. An example with neonatal mortality. BMC Pregnancy Childbirth  2007; 7: 28. Google Scholar CrossRef Search ADS PubMed  11 Wilcox AJ. On the importance-and unimportance-of birthweight. Int J Epidemiol  2001; 30: 1233– 41. Google Scholar CrossRef Search ADS PubMed  12 Herman A, McCarthy B, Bakewell J 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 1): 5– 22. Google Scholar CrossRef Search ADS PubMed  13 Mathews TJ, MacDorman MF. Infant mortality statistics from the 2010 period linked birth/infant death data set. Natl Vital Stat Rep  2013; 62( 8): 1– 26. Google Scholar PubMed  14 Ananth CV, Basso O. Impact of pregnancy-induced hypertension on stillbirth and neonatal mortality. Epidemiology  2010; 21: 118– 23. Google Scholar CrossRef Search ADS PubMed  15 Kc K, Shakya S, Zhang H. Gestational diabetes mellitus and macrosomia: a literature review. Ann Nutr Metab  2015; 66( Suppl 2): 14– 20. Google Scholar CrossRef Search ADS PubMed  16 U.S. Department of Health and Human Services Child Health USA. Rockville, Maryland: US Department of Health and Human Services, 2013 17 World Health Organization (WHO). International Statistical Classification of Diseases and Related Health Problems 10th Revision (ICD-10). 2015. http://apps.who.int/classifications/icd10/browse/2015/en#/XV (April 2017, date last accessed). 18 Jacoby WG. Loess: a nonparametric, graphical tool for depicting relationships between variables. Electoral Studies  2000; 19: 577– 613. Google Scholar CrossRef Search ADS   19 DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics  1988; 44: 837– 45. Google Scholar CrossRef Search ADS PubMed  20 Fischer JE, Bachmann LM, Jaeschke R. A readers’ guide to the interpretation of diagnostic test properties: clinical example of sepsis. Intensive Care Med  2003; 29: 1043– 51. Google Scholar CrossRef Search ADS PubMed  21 Hajian-Tilaki K. Receiver operating charateristic (ROC) curve analysis for medical diagnostic test evaluation. Caspian J Intern Med  2013; 4: 627– 35. Google Scholar PubMed  22 Nanishi K, Green J, Taguri M et al.  . Determining a cut-off point for scores of the breastfeeding self-efficacy scale–short form: secondary data analysis of an intervention study in Japan. PLoS One  2015; 10( 6): e0129698. Google Scholar CrossRef Search ADS PubMed  23 Perkins NJ, Schisterman EF. The inconsistency of ‘optimal’ cutpoints obtained using two criteria based on the receiver operating characteristic curve. Am J Epidemiol  2006; 163: 670– 5. Google Scholar CrossRef Search ADS PubMed  24 Youden WJ. Index for rating diagnostic tests. Cancer  1950; 3( 1): 32– 5. Google Scholar CrossRef Search ADS PubMed  25 Dunger DB, Petry CJ, Ong KK. Genetics of size at birth. Diabetes Care  2007; 30( Suppl 2): S150– 5. Google Scholar CrossRef Search ADS PubMed  26 Kramer MS, Platt RW, Yang H et al.  . Registration artifacts in international comparisons of infant mortality. Paediatr Perinat Epidemiol  2002; 16: 16– 22. Google Scholar CrossRef Search ADS PubMed  27 Skjaerven R, Gjessing HK, Bakketeig LS. New standards for birth weight by gestational age using family data. Am J Obstet Gynecol  2000; 183: 689– 96. Google Scholar CrossRef Search ADS PubMed  28 Skjaerven R, Wilcox AJ, Oyen N et al.  . Mothers’ birth weight and survival of their offspring: population based study. Br Med J  1997; 314: 1376– 80. Google Scholar CrossRef Search ADS   29 Riley RD, Ensor J, Snell KI et al.  . External validation of clinical prediction models using big datasets from e-health records or IPD meta-analysis: opportunities and challenges. Br Med J  2016; 353: i3140. Google Scholar CrossRef Search ADS   © The Author(s) 2018. Published by Oxford University Press on behalf of Faculty of Public Health. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

Journal of Public HealthOxford University Press

Published: May 19, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off