Stillbirth represents a substantial proportion of perinatal mortality,1,2 but its causes are poorly understood.3,4 Socioeconomic status,5 including maternal education,4,6–13 is a risk factor, but little is known on how education influences stillbirth at various gestational ages, despite evidence that the influence of race/ethnicity on stillbirth14–17 and the influence of education on preterm birth,18 may vary across gestation. Even less is known about the relation between education and cause-specific stillbirth, although socioeconomic status has been established as a risk factor for hypertension and diabetes, 2 common pregnancy complications linked to stillbirth.2,3,19 Socioeconomic gradients appear to be greater for causes leading to stillbirth before labor onset8,20 and for stillbirth due to nonchromosomal congenital anomalies,21 but gestational-age-specific inequalities have not been examined.
Lack of attention to this issue may be related to the challenges involved in analyzing gestational-age-specific rates, especially at preterm gestational ages when delivery often does not result from a “normal” process. In fact, stillbirth and preterm birth may share unmeasured common causes, such as infection or rare pathologies.2,3,22,23 If education is more strongly associated with unmeasured causes of preterm birth than with stillbirth, analyses restricted to preterm gestational ages may inadvertently suggest no association or even a protective influence on stillbirth.24,25 Preterm live births may, therefore, not be a suitable referent for preterm stillbirths in analyses stratified by gestational age. This issue is important because of its implications for policy: prevention strategies require a correct understanding of the relation between risk factors and stillbirth at various gestational ages.
How to overcome the bias potentially inherent in gestational-age-specific analyses is an emerging topic. Recent developments in the causal-inference literature have proposed methods for decomposing effects of exposures on perinatal mortality outcomes into unmediated effects and effects mediated through preterm delivery.26–28 Such methods, however, may not be readily accessible to general researchers not familiar with the causal literature. In a separate literature, researchers have suggested use of a fetuses-at-risk approach (ie, fetuses not yet born, or ongoing pregnancies, as the denominator) for computing stillbirth rates.29,30 Stillbirth rates based on fetuses-at-risk were initially proposed by Yudkin et al in 1987,29 and, though controversial, use of this method has extended to other adverse birth outcomes.31 The fetuses-at-risk approach has not yet been applied to gestational-age-specific regression analysis in which the aim is to determine the association between an exposure (such as education) and an outcome (such as preterm stillbirth). Evidence from the existing literature suggests that the fetuses-at-risk approach corrects potentially biased observations, such as the well-known birth weight paradox in which infant mortality appears to be lower for smokers than nonsmokers at very preterm ages.31,32 Hence, use of fetuses-at-risk may potentially be a simple solution for researchers wishing to evaluate associations between risk factors and perinatal outcomes at preterm ages.
The objective of this study, therefore, was to evaluate the association between education and stillbirth across gestational age for all-cause and cause-specific fetal mortality and to determine whether a fetuses-at-risk approach could be used to correct potentially biased estimates at preterm gestational ages.
Data and Variables
Individual-level data spanning 1981–2006 for 2,143,134 singleton live births and 9256 stillbirths were extracted for the province of Québec, Canada. Ascertainment of live births, defined as births showing any sign of life at delivery, is complete in Québec.33 However, only stillbirths >500 g are recorded in the Québec stillbirth file (intrauterine deaths <500 g are not registered, irrespective of gestational age).33 We excluded 310 intentional pregnancy terminations with International Classification of Disease (ICD) cause of death codes 779.6 (ninth revision, 1981–1999) and P96.4 (10th revision, 2000–2006). Multiple births (N = 49,816) were not included, as biologic pathways leading to stillbirth may differ.34 The final sample consisted of 2,143,134 live-born and 8946 stillborn infants.
Number of completed years of maternal education (in continuous years from 1 to 30) is routinely collected in Québec, and linkage with another data source was not necessary. Québec is, to our knowledge, the only Canadian province that collects maternal education data on live birth and stillbirth certificates. Education was expressed categorically (<10, 10–11, ≥12 years) for descriptive statistics (these cut points are commonly used in the literature),13,18 and as a continuous cumulative rank score (ranging from 0 to 1) for regression models. The cumulative rank score was computed separately for 1981–1985, 1986–1990, 1991–1995, 1996–2000, and 2001–2006 to account for distributional shifts in education over time.18 Births in the same education level were assigned the average rank for that level (rather than each woman's fractional rank, which depends on how live births and stillbirths are sorted within each level35). Education data were missing for 128,029 live births (6.0%, including 0.2% with 0 recorded years of education that may be implausible) and 2440 stillbirths (27%, all with 0 recorded years of education). Missing education data were imputed based on the distribution of observed covariates for each period using multiple imputation (5 complete datasets were generated).36
Some authors have proposed a threshold of 28 weeks to differentiate early from late stillbirth,2,3 but use of a single cut-point may mask educational differences later in pregnancy. Gestational age in completed weeks was, therefore, assessed for 4 gestational age intervals using cut-points from the preterm birth literature (<28, 28–31, 32–36, and ≥37 weeks).18 Postterm stillbirths for ≥42 gestational weeks were not separated from term births due to small sample size (n = 89 stillbirths). Gestational age was imputed 5 times for 18,225 live births (0.8%) and 315 stillbirths (3.5%) for whom data were missing, with the same multiple imputation procedure.36
Covariates included maternal age (<20, 20–24, 25–29, 30–34, 35–39, ≥40 years), marital status (legally married, not legally married), mother tongue (French, English, other language, unknown), and parity (0, 1, ≥2 previous deliveries).18 Mother tongue was used as an indicator of ethnicity. Missing maternal age was imputed for 265 cases and missing marital status for 589 cases.36 Maternal birthplace (Canadian born, foreign born, unknown) was available from 1998 onward. Birth weight was considered a potential intermediary between education and stillbirth,37 and hence was not accounted for.
Stillbirth rates30 were computed using stillbirths in a given gestational interval as the numerator. Both conventional and fetuses-at-risk approaches were used for denominators as follows: (1) total number of births in the same interval (the conventional method, which is expected to yield elevated rates early in gestation); (2) fetuses-at-risk (expected to yield rates that are higher later in gestation). Fetuses-at-risk were defined as all fetuses in utero at a given gestational interval (the sum of current and later live births and stillbirths).
We used multivariable logistic regression to compute odds ratios (ORs) and 95% confidence intervals (CIs) for the relationship between the education rank score and dichotomous stillbirth, adjusting for maternal age, marital status, mother tongue, parity, gestational age, and period. ORs for the continuous education rank score represent the odds of stillbirth for the least-educated mothers relative to the most educated. An interaction term for education rank score by gestational age was tested in a pooled model, followed by adjusted models run for each gestational age interval separately. The underlying risk for parameter estimates obtained from stratified analyses is based on the ratio of stillbirths at a given gestational interval in the numerator to live births at the same interval in the denominator (not to fetuses-at-risk). For simplicity, these estimates are hereafter denoted conventional ORs. To account for potentially biased estimates obtained with conventional ORs, associations were calculated with risks based on stillbirths at a given gestational interval in the numerator and live births in the same interval plus all subsequent live births and stillbirths (ie, fetuses-at-risk) in the denominator (hereafter denoted fetuses-at-risk ORs). Fetuses-at-risk ORs are obtained by dropping all births that occurred prior to the gestational interval of interest and recoding all subsequent stillbirths as live births. Associations for term stillbirth (≥37 weeks) are identical using either conventional or fetuses-at-risk ORs, as the denominator is the same.
To assess the relationship between education and fetal cause of death, stillbirth was expressed as a polytomous outcome with a separate category for each of 14 leading causes of death and a category for live births. ICD codes were used to specify causes that were clinically meaningful (eAppendix, http://links.lww.com/EDE/A553). We used multivariable polytomous logistic regression to estimate the OR between the cumulative education rank score and each cause of death relative to live births, adjusting for maternal age, marital status, mother tongue, parity, gestational age, and period. Models were subsequently run separately for early (<28 weeks) and late (≥28 weeks) births. Other gestational age intervals were not evaluated due to sample size restrictions. Both conventional and fetuses-at-risk ORs were computed for births <28 weeks.
In sensitivity analyses, final models were re-run, excluding missing data. Maternal birthplace was added as a covariate for models restricted to 1998 onward to determine whether immigration may have influenced the relation between education and stillbirth. Statistical analyses were undertaken with SAS 9.1 (SAS Institute Inc., Cary, NC). This study conformed to the 2010 Tri-Council Policy Statement for ethical conduct of research involving humans in Canada.
There were 4.2 stillbirths per 1000 total births in the population overall (Table 1). The rate decreased from 5.4 in 1981–1985 to 3.4 in 2001–2006. Mean maternal education for live births (13.1 years) was higher than for stillbirths (12.5 years), and this trend was reflected by a higher rate of stillbirth among mothers with <10 years of education (4.7 per 1000 total births) compared with those with ≥12 years (2.9 per 1000).
Mean gestational age of stillbirth was 32.5 weeks (standard deviation = 6.1; median = 34; interquartile range = 28–38), whereas that of live births was 39.1 weeks (1.8; 39; 38–40). As expected, conventional stillbirth rates decreased progressively with greater gestational age (Table 2). In contrast, stillbirth rates using fetuses-at-risk were highest at term and lower early in gestation. Stillbirth rates, nonetheless, rose progressively with fewer years of maternal education, irrespective of the method used to calculate gestational-age-specific rates.
The adjusted OR in Table 3 indicated that women with the lowest education had two-times higher odds of overall stillbirth relative to those with the highest education. An interaction term for cumulative-education-rank score by gestational age suggested associations differed by gestational age (X2 = 26.1, P < 0.0001; first imputation). Models run with data stratified, according to gestational age (conventional denominator), indicated that odds of stillbirth were greatest for term births (OR = 1.82 [95% CI = 1.54–2.15]), and somewhat lower but still elevated for births at 28–31 and 32–36 gestational weeks. The OR at <28 weeks was not elevated (odds were 13% lower for the least, relative to most, educated mothers), but the association overestimated the risk, as the rare disease assumption did not hold.38 ORs computed using fetuses-at-risk in the denominator were higher at every preterm gestational interval than at term (Table 3). These findings contrast with conventional ORs that suggested education was more strongly associated with term stillbirth.
Education was most strongly associated with diabetes-related stillbirth, especially at ≥28 weeks (OR = 5.04) (Table 4). Hypertensive disorders, premature rupture of membranes, placental abruption, slow fetal growth/short gestation, and hypoxia were other causes associated with odds of 2 or more for stillbirth related to low (relative to high) education among births ≥28 weeks. Slightly weaker associations were present for cord compression, placenta previa, nonchromosomal congenital anomalies, unexplained and residual causes. Education was not associated with any particular cause of death among births <28 weeks based on conventional denominators; however, for results based on fetuses-at-risk, low education was associated with higher odds of stillbirth for most causes, especially premature rupture of membranes, placental abruption, nonchromosomal congenital anomalies, and unexplained causes.
Sensitivity analyses excluding missing data and adjusting for maternal birthplace resulted in similar findings.
This study investigated the association between education and stillbirth across gestation and cause of death. It is the first, to our knowledge, to assess how fetuses-at-risk denominators performed relative to conventional denominators in regression models to obtain gestational-age-specific estimates. Using conventional denominators, low education was associated with stronger odds of stillbirth relative to high education at term, weaker odds at 28–36 gestational weeks, and protective odds at <28 weeks. In contrast, fetuses-at-risk denominators yielded associations that were stronger for all preterm gestational intervals compared with term. Similar patterns were observed for cause-specific stillbirth—most causes of fetal death were not associated with education at <28 gestational weeks using conventional denominators, but associations based on fetuses-at-risk indicated that education was associated with up to 3 times greater odds of stillbirth for several causes (eg, premature rupture of membranes, hypertension). These findings illustrate the usefulness of using preterm live births combined with ongoing pregnancies as the referent for preterm stillbirths in regression models, especially at very early gestational ages; educational inequalities at <28 weeks' gestation were masked when the rate of stillbirths was computed with live births <28 weeks as the denominator, but not when the underlying risk had fetuses-at-risk in the denominator.
Conventional ORs indicated that low education was more strongly associated with live birth than stillbirth at extremely early gestational ages, whereas associations were stronger with stillbirth later in gestation. Stillbirth and preterm live birth are competing outcomes ideally to be avoided until the fetus reaches maturity when delivery would be appropriate. Preterm delivery is also downstream of the education-stillbirth causal pathway (or an intermediate in the pathway if fetal death occurs secondary to forces of preterm labor). Furthermore, preterm delivery is often used to avoid potential stillbirth in pregnancies complicated by maternal comorbidity. Given these pathways, education may be more strongly associated with preterm live birth at early gestational ages if severe comorbidities leading to iatrogenic live birth are more strongly associated with education than risk factors in remaining pregnancies that led to stillbirth. In other words, the associations observed with conventional ORs may be related partly to shifting of potential stillbirths to the preterm live birth category (resulting in attenuated or reversed associations). At later gestational ages, other causes of preterm birth not as strongly linked to education (eg, spontaneous preterm labor) may be more common, resulting in conventional ORs that are in the expected direction. In the context of these findings, it can be inferred that preterm birth behaves as an effect modifier of the relation between education and stillbirth, and therefore that conventional ORs conditioned on preterm delivery yield biased associations at early gestational ages.
Using fetuses-at-risk as the denominator of the stillbirth rate, however, appears to eliminate the bias introduced when ORs are calculated based on the conventional denominator for the rate. There is controversy in the literature regarding use of fetuses-at-risk over conventional methods for calculating stillbirth rates. This stems from challenges in determining which method is more useful for clinical care and which is more useful for understanding the etiology of stillbirth.39 Conventional methods typically yield higher stillbirth rates at preterm gestational intervals, whereas fetuses-at-risk yield rates higher at term, which may be confusing from a clinical standpoint.39 Use of fetuses-at-risk has, however, resolved several paradoxes in which vulnerable populations expected to have high risks of perinatal mortality (eg, smokers) unexpectedly have lower mortality relative to the less vulnerable group (eg, nonsmokers).31,32 Such studies have stopped short of using a fetuses-at-risk approach in regression modeling. Thus, our findings suggest that the use of fetuses-at-risk could likely be extended beyond the association between education and preterm stillbirth, to encompass multiple situations where preterm delivery modifies the relation between a risk factor and early perinatal outcome. Preeclampsia is a good example—studies suggest that preeclampsia is protective against outcomes such as infant mortality, retinopathy, and cerebral palsy at preterm gestational ages, but not at term (based on stratified regression analyses).22,40–42 Use of a fetuses-at-risk approach may resolve these paradoxical observations.31,32,43 Furthermore, fetuses-at-risk could be applied to situations where other mediators are involved, such as birth weight, where intersecting weight-specific mortality curves for risk factors, such as smoking and race, have been observed (ie, the birth weight paradox).37,44,45
Another area of contention is the point beyond delivery at which a fetuses-at-risk approach should no longer be used. Although stillbirth may be relatively easily evaluated using fetuses-at-risk, its use may be more controversial for neonatal and postneonatal mortality where the contribution of preterm delivery as a mediator becomes more questionable with time since birth. Some evidence suggests that conventional methods of determining gestational-age-specific associations between neighborhood deprivation and postnatal outcomes are biased at preterm gestational intervals for neonatal mortality (deaths occurring <28 days after live birth), but that bias is less pronounced for postneonatal mortality.12 Although more research is needed, these findings12 suggest that the debate on use of fetuses-at-risk after delivery should include the first month of life as well as beyond.
In practical terms, education is increasingly considered an important determinant of health.46 Although maternal education has been associated with stillbirth,4–13,47 studies that evaluate its influence at various gestational ages have, to our knowledge, not been undertaken, and cases <28 weeks are often excluded.6,8–10,13 Furthermore, many studies have relied on data linkage to obtain measures of education6,7,9,10 or evaluated education as a covariate rather than a main effect.4,6,8,10,11,47 Cause-specific analyses indicated that educational inequalities tended to be stronger for diabetes compared with other causes, especially at ≥28 gestational weeks. Diabetes, an established risk factor for stillbirth,48 can be either preexisting or gestational, and socioeconomic status is potentially associated with both. Although it is reassuring that diabetic stillbirth was rare during this study period, rising rates of obesity may result in a greater population impact in the future, given its relation with both diabetes and stillbirth.49,50 Similarly, throughout gestation education was associated with hypertension—another common cause of stillbirth51 that may be increasing in populations.33 The quality of recorded cause of death on stillbirth certificates is, however, not known, and results should be interpreted with caution in light of potential nondifferential misclassification of cause of death (and attenuated associations).
Associations between education and medically preventable causes of stillbirth varied across causes. No relationship was present between education and isoimmunization-related stillbirth, possibly due to universal screening of maternal blood type that has been successfully implemented independent of social status in Québec (through universal public health insurance). This public health success contrasts, however, with educational inequalities that were present for cord compression and hypoxia-related stillbirth, 2 causes potentially preventable through intrapartum external fetal monitoring and cesarean delivery.2 Our findings suggest that fetal monitoring may be better implemented in mothers with high socioeconomic status. Interestingly, other studies have not observed socioeconomic inequalities in stillbirth occurring during labor (inequalities were present for stillbirths that occurred prior to labor, however).8,20 Congenital-anomaly-related stillbirth, which is potentially preventable through folic acid food fortification (implemented in Québec in 1998),52 was associated with education but only for stillbirths of nonchromosomal origin—which is in line with European data showing an association between area deprivation and nonchromosomal congenital anomaly stillbirth.21
This study was limited by the change in ICD coding from the ninth to 10th revision in year 2000, or by other potential changes in coding through time that we were unaware of. Ultrasound dating of gestational age may not have been completely implemented in the 1980s.18 In addition, the date of delivery but not the actual date of death was known, which may have resulted in overestimation of stillbirth gestational age. These misclassification issues were, however, unlikely to be differential across education. Similarly, potential misclassification of education was likely similar for stillbirth and live births, and it would be expected to have attenuated associations. We used ICD causes of death, and other potentially useful classification systems were not available in the data.53,54 We could not evaluate educational inequalities in infection-related stillbirth,55 an important cause that may have been grouped with unexplained causes. We did not have data on maternal characteristics, such as smoking, that are linked with both stillbirth and socioeconomic status, although the literature suggests that tobacco does not fully account for socioeconomic inequalities in stillbirth.56
This study found that potentially biased estimates between education and preterm stillbirth obtained with conventional ORs from stratified regression could be corrected with a fetuses-at-risk approach. Use of fetuses-at-risk merely requires that the comparison group for preterm stillbirth (or the denominator in the underlying stillbirth risk) include all live births in the gestational interval as well as fetuses still in utero. We suspect this simple method can be adapted to multiple situations in which preterm delivery (or birth weight) modifies the relationship between a risk factor and perinatal outcome. In addition, this study reinforces the importance of maternal education as a risk factor for stillbirth, including at preterm gestational intervals.
1. Fretts RC. The study of stillbirth. Am J Obstet Gynecol. 2009;201:429–430.
2. Smith GC, Fretts RC. Stillbirth. Lancet. 2007;370:1715–1725.
3. Fretts RC. Etiology and prevention of stillbirth. Am J Obstet Gynecol. 2005;193:1923–1935.
4. Froen JF, Arnestad M, Frey K, Vege A, Saugstad OD, Stray-Pedersen B. Risk factors for sudden intrauterine unexplained death: epidemiologic characteristics of singleton cases in Oslo, Norway, 1986–1995. Am J Obstet Gynecol. 2001;184:694–702.
5. Jorgensen T, Mortensen LH, Andersen AM. Social inequality in fetal and perinatal mortality in the Nordic countries. Scand J Public Health. 2008;36:635–649.
6. Arntzen A, Moum T, Magnus P, Bakketeig LS. Marital status as a risk factor for fetal and infant mortality. Scand J Soc Med. 1996;24:36–42.
7. Hemminki E, Merilainen J, Malin M, Rahkonen O, Teperi J. Mother's education and perinatal problems in Finland. Int J Epidemiol. 1992;21:720–724.
8. Little RE, Weinberg CR. Risk factors for antepartum and intrapartum stillbirth. Am J Epidemiol. 1993;137:1177–1189.
9. Olsen O, Madsen M. Effects of maternal education on infant mortality and stillbirths in Denmark. Scand J Public Health. 1999;27:128–136.
10. Rasmussen S, Albrechtsen S, Irgens LM, et al.. Risk factors for unexplained antepartum fetal death in Norway 1967–1998. Early Hum Dev. 2003;71:39–52.
11. Kiely JL, Paneth N, Susser M. An assessment of the effects of maternal age and parity in different components of perinatal mortality. Am J Epidemiol. 1986;123:444–454.
12. Luo ZC, Wilkins R, Kramer MS. Effect of neighbourhood income and maternal education on birth outcomes: a population-based study. CMAJ. 2006;174:1415–1420.
13. Rom AL, Mortensen LH, Cnattingius S, Arntzen A, Gissler M, Nybo Andersen AM. A comparative study of educational inequality in the risk of stillbirth in Denmark, Finland, Norway and Sweden 1981–2000. J Epidemiol Community Health. 2010 . [Epub ahead of print].
14. Allen CL, Hulsey TM, Hulsey TC. The influence of race on fetal outcome. Am J Perinatol. 2005;22:245–248.
15. Balchin I, Whittaker JC, Patel RR, Lamont RF, Steer PJ. Racial variation in the association between gestational age and perinatal mortality: prospective study. BMJ. 2007;334:833.
16. Willinger M, Ko CW, Reddy UM. Racial disparities in stillbirth risk across gestation in the United States. Am J Obstet Gynecol. 2009;201:469–468.
17. Cai J, Hoff GL, Okah F, et al.. Fetal mortality: timing of racial disparities. J Natl Med Assoc. 2007;99:1258–1261.
18. Auger N, Roncarolo F, Harper S. Increasing educational inequality in preterm birth in Québec, Canada, 1981–2006. J Epidemiol Community Health. 2010;65:1091–1096.
19. Simpson LL. Maternal medical disease: risk of antepartum fetal death. Semin Perinatol. 2002;26:42–50.
20. Parsons L, Duley L, Alberman E. Socio-economic and ethnic factors in stillbirth and neonatal mortality in the NE Thames Regional Health Authority (NETRHA) 1983. Br J Obstet Gynaecol. 1990;97:237–244.
21. Neasham D, Dolk H, Vrijheid M, Jensen T, Best N. Stillbirth and neonatal mortality due to congenital anomalies: temporal trends and variation by small area deprivation scores in England and Wales, 1986–96. Paediatr Perinat Epidemiol. 2001;15:364–373.
22. Mann JR, McDermott S, Griffith MI, Hardin J, Gregg A. Uncovering the complex relationship between pre-eclampsia, preterm birth and cerebral palsy. Paediatr Perinat Epidemiol. 2011;25:100–110.
23. Basso O, Wilcox AJ. Might rare factors account for most of the mortality of preterm babies? Epidemiology. 2011;22:320–327.
24. Whitcomb BW, Schisterman EF, Perkins NJ, Platt RW. Quantification of collider-stratification bias and the birthweight paradox. Paediatr Perinat Epidemiol. 2009;23:394–402.
25. Platt RW, Joseph KS, Ananth CV, Grondines J, Abrahamowicz M, Kramer MS. A proportional hazards model with time-dependent covariates and time-varying effects for analysis of fetal and infant death. Am J Epidemiol. 2004;160:199–206.
26. Ananth CV, Vanderweele TJ. Placental abruption and perinatal mortality with preterm delivery as a mediator: disentangling direct and indirect effects. Am J Epidemiol. 2011;174:99–108.
27. VanderWeele TJ, Hernández-Diaz S. Is there a direct effect of pre-eclampsia on cerebral palsy not through preterm birth? Paediatr Perinat Epidemiol. 2011;25:111–115.
28. Vanderweele TJ, Vansteelandt S. Odds ratios for mediation analysis for a dichotomous outcome. Am J Epidemiol. 2010;172:1339–1348.
29. Yudkin PL, Wood L, Redman CW. Risk of unexplained stillbirth at different gestational ages. Lancet. 1987;1:1192–1194.
30. Cheung YB. On the definition of gestational-age-specific mortality. Am J Epidemiol. 2004;160:207–210.
31. Joseph KS, Demissie K, Platt RW, Ananth CV, McCarthy BJ, Kramer MS. A parsimonious explanation for intersecting perinatal mortality curves: understanding the effects of race and of maternal smoking. BMC Pregnancy Childbirth. 2004;4:7.
32. Joseph KS. Commentary: exegesis of effect modification—biological or spurious? Paediatr Perinat Epidemiol. 2009;23:417–420.
33. Institut national de santé publique du Québec et ministère de la Santé et Services sociaux du Québec. Portrait de santé du Québec et de ses régions 2006: les statistiques—Deuxième rapport national sur l′état de santé de la population du Québec. Gouvernement du Québec, 2006. Available at: http://www.inspq.qc.ca/pdf/publications/portrait_de_sante.asp?E=p
34. Lee YM, Wylie BJ, Simpson LL, D'Alton ME. Twin chorionicity and the risk of stillbirth. Obstet Gynecol. 2008;111:301–308.
35. Chen Z, Roy K. Calculating concentration index with repetitive values of indicators of economic welfare. J Health Econ. 2009;28:169–175.
36. Sterne JA, White IR, Carlin JB, et al.. Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls. BMJ. 2009;338:b2393.
37. Hernandez-Diaz S, Schisterman EF, Hernan MA. The birth weight “paradox” uncovered? Am J Epidemiol. 2006;164:1115–1120.
38. Cummings P. The relative merits of risk ratios and odds ratios. Arch Pediatr Adolesc Med. 2009;163:438–445.
39. Wilcox AJ, Weinberg CR. Invited commentary: analysis of gestational-age-specific mortality–on what biologic foundations? Am J Epidemiol. 2004;160:213–214.
40. Chen XK, Wen SW, Smith G, Yang Q, Walker M. Pregnancy-induced hypertension is associated with lower infant mortality in preterm singletons. BJOG. 2006;113:544–551.
41. Fortes Filho JB, Costa MC, Eckert GU, Santos PG, Silveira RC, Procianoy RS. Maternal preeclampsia protects preterm infants against severe retinopathy of prematurity. J Pediatr. 2011;158:372–376.
42. Piper JM, Langer O, Xenakis EM, McFarland M, Elliott BD, Berkus MD. Perinatal outcome in growth-restricted fetuses: do hypertensive and normotensive pregnancies differ? Obstet Gynecol. 1996;88:194–199.
43. Hutcheon JA, Lisonkova S, Joseph KS. Epidemiology of pre-eclampsia and the other hypertensive disorders of pregnancy. Best Pract Res Clin Obstet Gynaecol. 2011;25:391–403.
44. Basso O, Wilcox AJ. Intersecting birth weight-specific mortality curves: solving the riddle. Am J Epidemiol. 2009;169:787–797.
45. Parker JA, Klebanoff MA. Invited commentary: crossing curves—it's time to focus on gestational age-specific mortality. Am J Epidemiol. 2009;169:798–801.
46. Baker DP, Leon J, Greenaway EGS, Collins J, Movit M. The education effect on population health: a reassessment. Popul Dev Rev. 2011;37:307–332.
47. Froen JF, Arnestad M, Vege A, et al.. Comparative epidemiology of sudden infant death syndrome and sudden intrauterine unexplained death. Arch Dis Child Fetal Neonatal Ed. 2002;87:F118–F121.
48. Macintosh MC, Fleming KM, Bailey JA, et al.. Perinatal mortality and congenital anomalies in babies of women with type 1 or type 2 diabetes in England, Wales, and Northern Ireland: population based study. BMJ. 2006;333:177.
49. Chu SY, Callaghan WM, Kim SY, et al.. Maternal obesity and risk of gestational diabetes mellitus. Diabetes Care. 2007;30:2070–2076.
50. Chu SY, Kim SY, Lau J, et al.. Maternal obesity and risk of stillbirth: a metaanalysis. Am J Obstet Gynecol. 2007;197:223–228.
51. Allen VM, Joseph K, Murphy KE, Magee LA, Ohlsson A. The effect of hypertensive disorders in pregnancy on small for gestational age and stillbirth: a population based study. BMC Pregnancy Childbirth. 2004;4:17.
52. De Wals P, Tairou F, Van Allen MI, et al.. Reduction in neural-tube defects after folic acid fortification in Canada. N Engl J Med. 2007;357:135–142.
53. Froen JF, Pinar H, Flenady V, et al.. Causes of death and associated conditions (Codac): a utilitarian approach to the classification of perinatal deaths. BMC Pregnancy Childbirth. 2009;9:22.
54. Reddy UM, Goldenberg R, Silver R, et al.. Stillbirth classification–developing an international consensus for research: executive summary of a National Institute of Child Health and Human Development workshop. Obstet Gynecol. 2009;114:901–914.
55. Goldenberg RL, McClure EM, Saleem S, Reddy UM. Infection-related stillbirths. Lancet. 2010;375:1482–1490.
56. Gray R, Bonellie SR, Chalmers J, et al.. Contribution of smoking during pregnancy to inequalities in stillbirth and infant death in Scotland 1994–2003: retrospective population based study using hospital maternity records. BMJ. 2009;339:b3754.
We acknowledge Jacques Rivard for assistance with data preparation.