Bias in comparisons of mortality among very preterm births: A cohort study.

BACKGROUND: Several studies of prenatal determinants and neonatal morbidity and mortality among very preterm births have resulted in unexpected and paradoxical findings. We aimed to compare perinatal death rates among cohorts of very preterm births (24-31 weeks) with rates among all births in these groups (≥24 weeks), using births-based and fetuses-at-risk formulations.
METHODS: We conducted a cohort study of singleton live births and stillbirths ≥24 weeks' gestation using population-based data from the United States and Canada (2006-2015). We contrasted rates of perinatal death between women with or without hypertensive disorders, between maternal races, and between births in Canada vs the United States.
RESULTS: Births-based perinatal death rates at 24-31 weeks were lower among hypertensive than among non-hypertensive women (rate ratio [RR] 0.67, 95% CI 0.65-0.68), among Black mothers compared with White mothers (RR 0.94, 95%CI 0.92-0.95) and among births in the United States compared with Canada (RR 0.74, 95%CI 0.71-0.75). However, overall (≥24 weeks) perinatal death rates were higher among births to hypertensive vs non-hypertensive women (RR 2.14, 95%CI 2.10-2.17), Black vs White mothers (RR 1.86, 95%CI 184-1.88;) and births in the United States vs Canada (RR 1.08, 95%CI 1.05-1.10), as were perinatal death rates based on fetuses-at-risk at 24-31 weeks (RR for hypertensive disorders: 2.58, 95%CI 2.53-2.63; RR for Black vs White ethnicity: 2.29, 95%CI 2.25-2.32; RR for United States vs Canada: 1.27, 95%CI 1.22-1.30).
CONCLUSION: Studies of prenatal risk factors and between-centre or between-country comparisons of perinatal mortality bias causal inferences when restricted to truncated cohorts of very preterm births.

Introduction

Preterm birth is a major health problem worldwide, associated with high mortality and morbidity and life-long disability [1-3]. As a consequence, several research initiatives have targeted preterm birth. The Canadian Neonatal Network and other groups have conducted many studies among preterm populations, focusing on prenatal risk factors for perinatal outcomes, or comparing perinatal and neonatal outcomes by health centre and country, in addition to studies of interventions following preterm birth [4-13].

Studies of risk factors for adverse perinatal outcomes are typically designed to identify the potentially modifiable underlying causes of these outcomes. Similarly, comparisons of neonatal morbidity and mortality by centre or country can highlight regional and other disparities and stimulate initiatives to improve maternal and neonatal health care. Several reports document reductions in adverse neonatal outcomes at the national level following programs to foster optimal clinical practices identified through between-centre comparisons of neonatal morbidity and mortality among very preterm births [14, 15].

Despite the apparent utility of the above-mentioned investigations, many studies restricted to preterm infants have shown unexpected results. For instance, studies have reported higher neonatal mortality among very preterm infants born to normotensive mothers compared with those born to hypertensive mothers [6], and better survival among very preterm infants of older mothers than among those born to younger mothers [5]. These findings highlight the paradox of intersecting perinatal mortality curves, a phenomenon described by Yerushalmy over 50 years ago [16]: low birth weight infants of mothers who smoke in pregnancy have higher neonatal survival than low birth weight infants of non-smoking mothers, whereas the opposite is observed at higher birth weights. This paradox is a general phenomenon that occurs across diverse contrasts by risk factor (e.g., multifetal pregnancies, maternal age) and outcome (e.g., stillbirth, neonatal death, cerebral palsy, and sudden infant death syndrome), irrespective of how “maturity” is defined (by birth weight or gestational age). Despite a lack of agreement on the mechanism responsible for the paradox [16-27], the risk factors across which the paradox has been observed (e.g., maternal smoking and hypertension) are recognized as deleterious to fetal and infant health.

Although the paradox of intersecting curves has received extensive attention in the epidemiologic literature, its wider implications for etiologic studies and geographic comparisons restricted to very preterm births have often been overlooked. In this study, we analyze the effects of prenatal exposures on perinatal outcomes. To illustrate the consequences and highlight the risk of bias of restricting studies to very preterm births, we examined the association between 1) a pregnancy risk factor (i.e., hypertensive disorders in pregnancy), 2) a social determinant of health (i.e., race), or 3) regions (i.e., Canada vs. United States), and perinatal mortality.

Methods

Our cohort study was based on all live births and stillbirths in the United States and Canada for the years 2006–2015. Data on births in the United States were obtained from the period linked birth/infant death files and fetal death files of the National Center for Health Statistics [28], which together include information from live birth and stillbirth registrations in the United States. Data for births in Canada were obtained from the Discharge Abstract Database of the Canadian Institute for Health Information [29], which includes all delivery hospitalizations and represents approximately 98% of births in Canada (excluding Quebec). Gestational age in both data sources was based on the clinical estimate of gestation. Information in these databases has been validated and is routinely used in epidemiologic studies [30, 31].

The study population was restricted to singleton births with a clinical estimate of gestation ≥24 weeks, as births at earlier gestational age are associated with wide variations in resuscitation, active treatment and birth registration practices [32-34]. Fetuses or infants with congenital or chromosomal anomalies (including anencephaly, meningomyelocele/spina bifida, cyanotic congenital heart disease, congenital diaphragmatic hernia, omphalocele, gastroschisis, limb reduction defect, cleft lip or palate, Down syndrome, hypospadias or suspected chromosomal disorder) were excluded.

Independent variables of interest

Women with hypertensive disorders in pregnancy included those with chronic (pre-pregnancy) hypertension, gestational hypertension and eclampsia, while non-hypertensive women included women without any of the aforementioned conditions. Maternal race comprised four categories: White, Black, Native American, and Asian (including Pacific Islanders). Whites were the reference group in all comparative analyses. Analyses of hypertensive disorders in pregnancy and maternal ethnicity were restricted to births in the United States. The regional comparisons contrasted births at ≥24 weeks in Canada vs. the United States.

Dependant variable

The primary outcome was perinatal death, defined as stillbirth or early neonatal death (i.e., death within 7 days of delivery). The secondary outcome was early neonatal death.

Methods for the calculation of rates

Overall and gestational age-specific rates of perinatal death were calculated using two different denominators: births-based and fetuses-at-risk [35]. Gestational age-specific rates under the births-based formulation were calculated as the proportion of perinatal deaths among total births (live births plus stillbirths) at a specific gestational week. Under the fetuses-at-risk formulation, gestational age-specific perinatal death rates were calculated using the number of perinatal deaths at each completed gestational week in the numerator, and the number of fetuses at risk for perinatal death at the beginning of the gestational week in the denominator (i.e., the number of fetuses who delivered at or after the gestational week in question). For instance, the gestational age-specific perinatal death rate at 28 weeks’ gestation was calculated using the number of perinatal deaths at 28 weeks in the numerator and the number of fetuses in utero at the beginning of the 28th week of gestation as the denominator (including those born at 28 weeks and all those delivered at a later gestational age) [35]. The gestational age-specific perinatal death rates under the fetuses-at-risk formulation can be described as a cumulative risk over a one-week period and approximate the conditional hazard rate.

Overall (≥24 weeks) births-based and fetuses-at-risk rates of perinatal death are equivalent, since both numerators include all deaths at ≥24 weeks and both denominators include all births at ≥24 weeks. Birth-based and fetuses-at-risk rates of early neonatal death were also calculated in a similar manner. Gestational age-specific rates under the birth-based approach were calculated as the proportion of early neonatal deaths among live births at a specific gestational week, while fetuses-at-risk calculations involved dividing the number of early neonatal deaths at a given gestational week by the number of fetuses at risk of birth and early neonatal death at that gestational week.

We carried out additional analyses to assess the potential role of confounding in contrasts between singletons of women with or without hypertensive disorders in pregnancy as the relation with perinatal death could be confounded by maternal age, race, comorbidity or other factors. Since the variables available for adjustment were limited, this supplementary analysis was intended to gauge the degree of potential confounding by putative confounders, and to assess whether such confounding would alter effect estimates modestly or reverse the direction of the association. Logistic regression was used to contrast hypertensive vs. non-hypertensive women with regard to perinatal death after adjusting for maternal age (indicator variables for 5-year strata), race, and diabetes.

In a final analysis, designed to illustrate differences between the denominators used in the births-based and fetuses-at-risk formulations of gestational age-specific perinatal mortality, we contrasted the compared categories (women with vs. without hypertensive disorders, White women vs. women of other race and women from Canada vs. the United States) in terms of fetuses-at-risk birth rates. These birth rates were calculated using births at any gestational week in the numerator and fetuses at risk of birth in the denominator.

The study received ethics approval from the institutional review board at the University of British Columbia. All analyses were carried out using SAS statistical software (version 9.4, SAS Institute Inc., Cary, NC, USA).

Results

The United States data sources included 39,298,721 eligible singleton live births and stillbirths ≥24 weeks’ gestation between 2006 and 2015 (Table 1). We identified 2,715,169 hospital deliveries of eligible singletons between 2006 and 2015 in Canada (excluding Quebec).

Table 1

Numbers of singleton births and perinatal deaths (excluding congenital and chromosomal anomalies), in the United States, 2006–2015.

	Overall	Hypertensive disorders in pregnancy		Maternal race
		No	Yes	White	Black	Native American	Asian
Women ≥35 years old at delivery	5,698,555 (14.5)	5,255,189 (14.3)	423,858 (18.7)	4,385,076 (14.6)	684,922 (11.0)	40,046 (8.8)	588,511 (23.4)
Chronic hypertension	524,870 (1.3)	0 (0.0)	524,870 (23.2)	334,853 (1.1)	162,594 (2.6)	7,424 (1.6)	19,999 (0.8)
Diabetes	2,051,552 (5.2)	1,733,255 (4.7)	318,297 (14.1)	1,500,801 (5.0)	293,453 (4.7)	34,446 (7.6)	222,852 (8.9)
Total births ≥24 weeks	39,298,721	36,890,944	2,265,317	30,102,675	6,223,981	456,164	2,515,901
Total births 24–31 weeks	480,471 (1.2)	382,470 (1.0)	91,024 (4.0)	299,506 (1.0)	151,347 (2.4)	5,765 (1.3)	23,853 (0.9)
Live births ≥24 weeks	39,161,645	36,777,532	2,249,673	30,011,079	6,187,092	454,391	2,509,083
Live births 24–31 weeks	422,756 (1.1)	336,003 (0.9)	83,511 (3.7)	263,039 (0.9)	133,594 (2.2)	5,098 (1.1)	21,025 (0.8)
Stillbirth ≥24 weeks	137,076	113,412	15,644	91,596	36,889	1,773	6,818
Stillbirths 24–31 weeks	57,715 (42.1)	46,467 (41.0)	7,513 (48.0)	36,467 (39.8)	17,753 (48.1)	667 (37.6)	2,828 (41.5)
Early neonatal deaths ≥24 weeks	44,555	39,635	4,424	31,373	10,383	593	2,206
Early neonatal deaths 24–31 weeks	21,855 (49.1)	18,770 (47.4)	2,813 (63.6)	14,323 (45.7)	6,257 (60.3)	267 (45.0)	1,008 (45.7)
Perinatal deaths ≥24 weeks	181,631	153,047	20,068	122,969	47,272	2,366	9,024
Perinatal deaths 24–31 weeks	79,570 (43.8)	65,237 (42.6)	10,326 (51.5)	50,790 (41.3)	24,010 (50.8)	934 (39.5)	3,836 (42.5)

Numbers in parentheses represent proportions (%); e.g., overall 14.5% of women were ≥35 years old at delivery, 1.3% had chronic hypertension, 5.2% had diabetes, and 1.2% of total births were 24–31 weeks’ gestation.

Hypertensive disorders in pregnancy

Table 1 presents the numbers of singleton live births, stillbirths and perinatal deaths among women with and without hypertensive disorders in the United States. Among births at 24–31 weeks’ gestation, the births-based perinatal death rate was among women with hypertensive disorders (113.4 per 1,000 total/live births) compared with women without hypertensive disorders (170.6 per 1,000 total/live births; P<0.001; Table 2). On the other hand, the overall (≥24 weeks) births-based perinatal death rate among women with hypertensive disorders (8.86 per 1,000 total/live births) was significantly than among women without hypertensive disorders (4.15 per 1,000 total/live births; P<0.001; Table 2).

Table 2

Comparisons of rates of perinatal death at 24–31 weeks’ gestation and overall among singletons with no congenital or chromosomal anomaly, 2006–2015.

	Perinatal death rate at 24–31 weeks (95% CI), Births-based calculation (per 1,000 total births)a	RR (95% CI)	Perinatal death rate at 24–31 weeks (95% CI), Fetuses-at-risk calculation (per 1,000 fetuses-at-risk)b	RR (95% CI)	Perinatal death rate overall (95% CI), (per 1,000 total births) c	RR (95% CI)
Hypertensive disorders in pregnancy d
No	170.6 (169.4 to 171.8)	Ref	1.77 (1.75 to 1.78)	Ref	4.15 (4.13 to 4.17)	Ref
Yes	113.4 (111.4 to 115.5)	0.67 (0.65 to 0.68)	4.56 (4.47 to 4.65)	2.58 (2.53 to 2.63)	8.86 (8.74 to 8.98)	2.14 (2.10 to 2.17)
		Adj: 0.63 (0.62 to 0.64)		Adj: 2.41 (2.36 to 2.46)		Adj 1.92 (1.89 to 1.95)
Maternal race/ethnicityd
White	169.6 (168.2 to 170.9)	Ref	1.69 (1.67 to 1.70)	Ref	4.08 (4.06 to 4.11)	Ref
Black	158.6 (156.8 to 160.5)	0.94 (0.92 to 0.95)	3.86 (3.81 to 3.91)	2.29 (2.25 to 2.32)	7.60 (7.53 to 7.66)	1.86 (1.84 to 1.88)
Native American	162.0 (152.5 to 171.5)	0.96 (0.90 to 1.01)	2.05 (1.92 to 2.18)	1.21 (1.14 to 1.30)	5.19 (4.98 to 5.40)	1.27 (1.22 to 1.32)
Asian	160.8 (156.2 to 165.5)	0.95 (0.92 to 0.98)	1.52 (1.48 to 1.57)	0.90 (0.88 to 0.93)	3.59 (3.51 to 3.66)	0.88 (0.86 to 0. 09)
Country of birth
United States	165.6 (164.6 to 166.7)	Ref	2.02 (2.01 to 2.04)	Ref	4.62 (4.60 to 4.64)	Ref
Canada	226.0 (220.1 to 231.9)	1.36 (1.33 to 1.40)	1.61 (1.56 to 1.65)	0.79 (0.77 to 0.82)	4.29 (4.22 to 4.37)	0.93 (0.91 to 0.95)

a Births-based death rates represent proportions, with the number of perinatal deaths at 24–31 weeks in the numerator and the number of total births at 24–31 weeks in the denominator.

b Fetuses-at-risk rates represent cumulative incidence rates with the number of perinatal deaths at 24–31 weeks in the numerator and the number of fetuses at risk of perinatal death at 24 weeks (i.e., fetuses who were delivered at 24 weeks or later) in the denominator.

c This calculation is identical for births-based and fetuses-at risk formulations.

d Based on births in the United States.

Adj Adjusted for maternal age (using indicator variables for 15–19, 20–24, 25–29, 30–34, 35–39, 40–44, 45–49, 50–54 year age categories), maternal race (using indicator variables for White, Black, American Indian/Alaskan Native, Asian/Pacific Islander), and diabetes; CI denotes confidence intervals; RR denotes rate ratios.

Fig 1 shows gestational age-specific perinatal death rates among women with vs. without hypertensive disorders calculated using births-based denominators. This contrast illustrates the paradox of intersecting perinatal mortality curves: death rates among women with hypertensive disorders were lower at early gestation but higher at later gestation than among women without hypertensive disorders.

Fig 1

Gestational age-specific perinatal death (A), early neonatal death (B) and stillbirth (C) rates of singletons with no congenital or chromosomal anomalies among women with and without hypertensive disorders using a births-based denominator, United States, 2006–2015. The yellow area highlights the restricted subpopulation at 24–31 weeks’ gestation.

The perinatal death rate at 24–31 weeks’ gestation among women with hypertensive disorders (4.56 per 1,000 fetuses-at-risk, respectively) was significantly than the same rate among women without hypertensive disorders (1.77 per 1,000 fetuses-at-risk; P<0.001) when calculated using fetuses-at-risk denominators (Table 1). Fig 2 shows gestational age-specific perinatal mortality rates calculated based on fetuses-at-risk denominators: death rates were higher among births to women with hypertensive disorders at all gestational ages.

Fig 2

Gestational age-specific perinatal death (A), early neonatal death (B) and stillbirth (C) rates of singletons with no congenital or chromosomal anomalies among women with or without hypertensive disorders using a fetuses-at-risk denominator, United States, 2006–2015.

Adjusted odds ratios expressing the association between maternal hypertension and perinatal death differed only modestly from the unadjusted estimates (Table 2).

Maternal racial groups

Comparisons of perinatal mortality rates among singleton births by maternal race showed the same paradox of intersecting perinatal mortality curves when calculated using births-based denominators. Birth-based perinatal death rates were significantly among singletons of Black mothers and non-significantly lower among births to Native American mothers compared with White mothers (158.6, 162.0, and 169.6 per 1,000 births, respectively; P<0.001; S1 Fig). The fetuses-at-risk formulation showed significantly perinatal death rates among Black and Native American mothers compared with White mothers (3.86, 2.05, and 1.69 per 1,000 fetuses-at-risk, respectively; P<0.001; Table 2; S1 Fig). Overall rates also showed perinatal death rates among singletons of Black and Native American mothers (Table 2). Asian mothers had significantly lower perinatal death rates than White mothers, but the difference was underestimated by births-based calculations. Comparisons of early neonatal death rates are provided in S1 Table.

Canada vs. the United States

Among singleton births at 24–31 weeks’ gestation, the births-based perinatal death rate was significantly in Canada than in the United States (226.0 vs. 165.6 per 1,000 total births; P<0.001; Table 2, S2 Fig). However, the perinatal death rate at 24–31 weeks’ gestation in Canada was substantially than that in the United States when calculated using fetuses-at-risk denominators (1.61 vs. 2.02 per 1,000 fetuses-at-risk; P<0.001; Table 2, S2 Fig). Differences in the overall perinatal death rate (≥24 weeks) were similar to differences in fetuses-at-risk perinatal death rates at 24–31 weeks’ gestation: overall perinatal mortality rates were significantly in Canada than in the United States (4.29 vs. 4.62 per 1,000 total births; P<0.001; Table 2).

Early neonatal death rates at 24–31 weeks in Canada and the United States similarly displayed opposite associations when calculated using births-based and fetuses-at-risk formulations (S1 Table).

Birth rates

The birth rate at 24–31 weeks’ gestation was higher among women with hypertensive disorders compared with women without hypertensive disorders (S3 Fig), among Black women compared with White women (S4 Fig) and in the United States compared with Canada (S5 Fig).

Discussion

Our study shows that, in research with a cause-and-effect focus, comparisons of perinatal mortality between fetuses of mothers with and without hypertensive disorders of pregnancy, between mothers of different races, and between mothers in Canada and the United States, are seriously biased when the study population is restricted to very preterm births and analyses are based on births-based denominators. Births-based perinatal death rates were significantly lower among women with hypertensive disorders, among Black mothers, and among births in Canada when the study populations were restricted to very preterm births. Inferences made from such truncated births-based analyses of very preterm births conflicted starkly with those obtained from analyses of the whole cohort of fetuses (births at ≥24 weeks) and analyses based on fetuses-at-risk calculations at 24–31 week’s gestation. The latter analyses showed significantly higher mortality among births to women with hypertensive disorders compared with births to non-hypertensive women, among Black and Native American mothers, and among births in the United States.

Our gestational-age specific analysis of perinatal mortality by exposure to hypertensive disorders in pregnancy, maternal race and country confirm the paradox of intersecting perinatal mortality curves. The fetuses-at-risk approach resolves the paradox by using a fetal perspective and a survival analysis formulation [35, 36]. A recent mechanistic explanation that reconciles the births-based and fetuses-at-risk approaches suggests that the lower births-based mortality experienced by higher-risk populations at earlier gestation is the consequence of an accelerated birth rate, which leads to more non-compromised fetuses being born at early gestation [37]. The problem of the paradox of perinatal mortality curves has also been framed as a collider stratification bias [19] (i.e., restricting by birth weight or gestational age, which are common effects of the perinatal determinants and unmeasured confounders, introduces a bias), and as effect-modification due to cortisol-mediated intrauterine adaptation resulting from the chronic stress associated with preterm birth [38]. However, the paradox of intersecting perinatal mortality curves manifests across a range of contrasts (including maternal age, smoking, hypertension, race, and country), and although it is possible to postulate a mechanism for the reduced mortality at preterm gestation associated with one factor or other, the need to propose a mechanism for each of these diverse contrasts supports the more parsimonious, singular explanation that the birth-based analysis is flawed.

Whereas the calculation of stillbirth rates has changed from a births-based formulation to a fetuses-at-risk denominator in recent decades (and thereby eliminated the paradox with regard to stillbirths), the situation with regard to the appropriate denominator for calculating neonatal death rates remains a topic of debate [39-44]. Some epidemiologists argue that live births (but not fetuses) experience neonatal death, and hence neonatal mortality rate calculations must use live births (and not fetuses) as the denominator [39, 41]. However, the alternative viewpoint, whereby both stillbirths and neonatal deaths are deemed to be closely related outcomes amenable to obstetric intervention, has been a traditional construct of modern obstetrics. Thus, clinical trials with a perinatal intervention (whether prenatal iron supplementation, antenatal corticosteroid therapy, magnesium sulphate for preterm birth at <31 weeks or labour induction at post-term gestation) involve the randomization of pregnant women, with effect assessment based on outcomes observed among all the randomized fetuses after birth.

The bias illustrated in this study, which arises from restricting analyses to a very preterm population, underscores the importance of methodologic rigour in non-experimental epidemiologic research with a causal focus. The recent creation of research networks of neonatal intensive care units (NICUs) in several countries has facilitated the reporting and benchmarking of neonatal outcomes, and research on neonatal health [45-55]. These networks have collaborated in NICU studies comparing neonatal outcomes by centre and country, and also studies of prenatal determinants [4–13, 56]. Such studies are particularly vulnerable to bias and can obscure health disparities. With study populations restricted to very preterm live births and births-based calculations of rates, NICU studies have shown lower mortality, necrotizing enterocolitis and sepsis rates among very preterm infants born to women of advanced maternal age [5]. Similarly, births-based analyses of very preterm infants have shown that hypertension in pregnancy is associated with a lower risk of death, severe brain injury and retinopathy [6]. A recent study of severe neonatal morbidity among very preterm infants also highlighted the underestimation of racial disparities in analyses that rely on births-based calculations [57]. Biased inferences from studies restricted to preterm births may be even more insidious in regional comparisons of perinatal mortality or morbidity without prior expectations regarding differences [4, 7, 58]. Although studies may be restricted to very preterm infants with the intent of comparing NICU practices (S6a Fig), the contrast of delivery hospital or regional rankings with regard to fetal and neonatal mortality at early gestational age can be similarly biased by differential distributions of gestational age at birth (S6b Fig), which is an intermediate factor between region or hospital admission and perinatal death. Distributions of gestational age at birth could differ between regions due to variations in obstetrical and neonatal practices and access to care for example. NICU comparisons can also be biased because they are limited to live-born infants who survive until admission to the NICU. Careful consideration should be given to such potential sources of bias in the interpretations of findings.

As illustrated in our study, births-based analyses restricted to very preterm births bias causal inference, as they are restricted to the left-end of the gestational age distribution. Such restriction results in a right truncation and analysis of incomplete data from the original cohort of individuals exposed or unexposed to a risk factor in pregnancy (see the S1 Appendix for an illustrated example). Births-based rates are not inherently problematic as they can be appropriate for prognostication (since they quantify the mortality experience of very preterm infants of women who are older, smoke, or have hypertension, for example). However, such rates are a poor basis for causal inference regarding prenatal factors (such as older maternal age, smoking and hypertension), addressing health inequities and evaluating neonatal health services (through between-centre and between-region comparisons). The causation vs prediction dichotomy may be best illustrated by the effect of maternal smoking on neonatal mortality [16]: it is preferable to frame the lower mortality rates among low birth weight and preterm infants of mothers who smoke in pregnancy in prognostic (predictive) terms. It is well-understood that maternal smoking has deleterious effects on the fetus and infants, irrespective of the gestational age at birth. Cohorts of very preterm births represent a select, truncated subpopulation, and studies restricted to very preterm births are unsuitable for non-experimental epidemiologic studies that attempt to estimate the causal effect of prenatal exposures using birth-based denominators.

Strengths and limitations

Information for our study was obtained from two large data sources that have been validated for epidemiologic research purposes [30, 31]. However, some misclassification of gestational age at birth and underreporting of hypertensive disorders is likely. Hypertensive disorders in pregnancy in our study cohort may have occurred at a gestation later than 24 weeks, but in absence of information on the time of diagnosis, we were not able to treat this exposure as a time-dependent variable. However, separate analyses of chronic hypertension and other hypertensive disorders showed the same risk of bias. We obtained a measure of the degree of confounding of the hypertension-perinatal death association by adjusting for maternal age, race and diabetes. It is possible, though not likely, that other factors such as parity, chronic diseases and socioeconomic status could have confounded the association to a greater extent since adjustment for maternal age and diabetes, two well-known confounders of the relationship between hypertensive disorders and outcomes, resulted in a 5 to 11% change in the rate ratios. Strong unmeasured confounders would be required to nullify or change the direction of the association. Information on Canadian births was obtained from hospitalization records, and early neonatal deaths that occurred after discharge could have led to a slight underestimation of the frequency of this outcome. Furthermore, data constraints led us to use information on the gestational age at birth for stillbirths, whereas gestational age at the time of fetal death would have been preferable. Although such systematic errors could have affected absolute rate estimates, they are unlikely to have occurred differentially in the contrasted populations and would not have had a major impact on our conclusions.

Conclusion

Our analyses highlight the bias inherent in studies restricted to very preterm birth subpopulations that aim to estimate the causal effect of prenatal risk factors, social determinants of health and between-country and between-centre comparisons. Although the adverse consequences on child development and lifelong health make continued research on very preterm birth a priority, non-experimental studies on this subpopulation require to be designed and analyzed with appropriate care and attention in order to avoid erroneous inferences. A careful consideration of the study question is paramount in identifying the appropriate study population and methods of analysis.

Gestational age-specific perinatal death rates of singletons with no congenital or chromosomal anomalies by maternal race using a births-based denominator (A) or using a fetuses-at-risk denominator (B), United States, 2006–2015. The yellow area highlights the restricted subpopulation at 24–31 weeks’ gestation.

(TIF)

Click here for additional data file.

Gestational age-specific perinatal death rates of singletons with no congenital or chromosomal anomalies in Canada or in the United States using a births-based denominator (A) or using a fetuses-at-risk denominator (B), 2006–2015. The yellow area highlights the restricted subpopulation at 24–31 weeks’ gestation.

(TIF)

Click here for additional data file.

Gestational age-specific birth rates of singletons with no congenital or chromosomal anomalies exposed or not to hypertensive disorders in pregnancy, United States, 2006–2015.

(TIF)

Click here for additional data file.

Gestational age-specific birth rates of singletons with no congenital or chromosomal anomalies by maternal race, United States, 2006–2015.

(TIF)

Click here for additional data file.

Gestational age-specific birth rates of singletons with no congenital or chromosomal anomalies in Canada and in the United States, 2006–2015.

(TIF)

Click here for additional data file.

Directed acyclic graph of relationships between NICU practices and perinatal deaths (A) and between hospital or region of residence and perinatal death (B).

(TIF)

Click here for additional data file.

Comparisons of rates of early neonatal death rates at 24–31 weeks’ gestation and overall among singletons with no congenital or chromosomal anomaly, 2006–2015.

(PDF)

Click here for additional data file.

Example of an antenatal exposure associated with gestational age at birth, resulting in a paradoxical association with mortality among very preterm births.

(PDF)

Click here for additional data file.

5 Mar 2021

PONE-D-20-40633

Bias in comparisons of mortality among very preterm births: a cohort study

PLOS ONE

Dear Dr. Boutin,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Apr 19 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

We look forward to receiving your revised manuscript.

Kind regards,

Kelli K Ryckman

Academic Editor

PLOS ONE

Journal Requirements:

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For information on unacceptable data access restrictions, please see http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions.

In your revised cover letter, please address the following prompts:

a) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially identifying or sensitive patient information) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent.

b) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. Please see http://www.bmj.com/content/340/bmj.c181.long for guidelines on how to de-identify and prepare clinical data for publication. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories.

We will update your Data Availability statement on your behalf to reflect the information you provide.

3.We note that the grant information you provided in the ‘Funding Information’ and ‘Financial Disclosure’ sections do not match.

When you resubmit, please ensure that you provide the correct grant numbers for the awards you received for your study in the ‘Funding Information’ section.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: No

Reviewer #2: No

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: This is a well-written manuscript that examines the differences in how 3 risk factors (maternal hypertension, maternal race, and U.S. vs Canada country of birth) are associated with perinatal and neonatal mortality when analyzed using different subsets of births (limited to births at 24-31 weeks gestation or to all births ≥24 weeks) and with different denominators (among births at the same gestational age week or by a fetuses-at-risk approach). Although it is a methodology-focused manuscript, it is written in a language that should be accessible to both clinicians and clinical researchers. Study data are from population-based datasets based on vital statistics in the U.S. and Canada. The authors show that the methodologic choices described above (subsetting births by a gestational age range and using a denominator based on births or potential births [fetuses-at-risk]) provide paradoxical-seeming results. The findings are important, particularly, as the authors point out, because of the common use of cohort studies based on subsets of births in perinatal/neonatal medicine and epidemiology (example cohorts with hundreds of papers between them that have influenced modern obstetric/neonatology practice: Canadian Neonatal Network, NICHD Neonatal Research Network, iNEO, Vermont Oxford Network, etc). The authors should be commended for a simple demonstration of a complex subject.

MAJOR COMMENTS:

The conclusion "Studies of perinatal risk factors and between-centre or between-country comparisons of

perinatal mortality lead to biased inferences when restricted to very preterm births" seems overly simplistic. Similarly, there are several instances of this language in the Discussion (e.g., "Our study shows that comparisons of perinatal mortality between fetuses of mothers with and without hypertensive disorders of pregnancy, between mothers of different ethnicity, and between mothers in Canada and the United States, were seriously biased when the study population was restricted to very preterm births and analyses were based on births-based denominators.")

The authors might further explore the reasons for the paradoxical associations that they observe. They acknowledge the rich literature on "collider stratification bias" and the theoretical mechanistic model of the "accelerated birth rate" whereby infants in some subgroups who are at lower risk are born earlier and so these subgroup-defining traits appear protective at low gestational ages.

However, there are alternative explanations for the authors findings that do not fall squarely in the realm of bias. As a clinical neonatologist, I care for infants who are born at early gestational ages. From my point of view, among the very preterm babies admitted to our unit in any given week, birth secondary to maternal hypertensive disease is a positive prognostic sign (compared to the alternative, which is usually birth secondary to some maternal inflammatory/infectious process). Is that point of view (from the neonatologist or perinatologist who delivers or cares for the baby) not valid? The authors allude to the difference between prognostic (my perspective on how the baby will do) and causal models (concerned with etiologic risks), but this might be further explored. From the perspective of the neonatologist/perinatologist, the comparison for births born due to maternal hypertension is not healthy normal pregnancies/births - rather, the comparison is to other births due pathology.

Another alternative point of view to consider is that of effect modification. Consider the example of a persistent patent ductus arteriosus (PDA), a common presentation in premature infants that may result in morbidity. In the aggregate, persistent PDA is associated with morbidity. But, in the particular, there are conditions (e.g., transposition of the great arteries) that affect a minority of the group in which the PDA confers a major protective advantage. In a similar way, is it not possible that while, in the aggregate, birth in Canada confers a lower risk of mortality than in the U.S. -- at the same time, care for very preterm births (due to a more aggressive/intensive/expensive approach) in the U.S. (versus Canada) confers a lower risk of mortality for this minority of patients?

The authors should explore reasons for their findings that are not simply the result of bias. Bias implies that these other perspectives are erroneous, which is not clearly the case. They might consider a more nuanced elaboration of how to interpret their findings in light of the research/clinical question being asked.

MINOR COMMENTS

Ln 33 - specify that this is the perinatal (or neonatal) death rate

Ln 86 - consider removing phrase “very poor survival.” In some cohorts, survival exceeds >50% at 22 and 23 weeks. The other reasons are valid enough without this one.

Table 1 - some of the cell totals for no hypertensive disorders in pregnancy are greater than the overall totals for those rows

Table 2 - I found the inclusion of "Births- based/fetuses- at risk calculation (per 1,000 total births)" under "overall" perinatal death rate confusing. (Same applies to S1 Table.) I think that the authors are trying to demonstrate that these formulations are the same. But, because this is already specified in the text, perhaps they should simply write "Calculation per 1,000 total births" (easier to read/less confusing) and, if needed, use a footnote to remind the reader that "for the overall calculation, births-based and fetuses-at-risk formulations are equivalent."

Ln 283 - why write "though not likely"? Is that quantifiable (as in an E-value?) or is it an opinion?

Ln 293 - as noted above, perhaps "potential bias"? Or some more nuanced interpretation?

S3Fig - should the y-axis be for "fetuses-at-risk" (same as S4 and S5Fig)?

Reviewer #2: This paper is extremely well written and tackles a very important issue around bias in perinatal mortality and neonatal mortality rates. It takes the research around analysis of births versus fetuses at risk in an important direction and is written in an easy to understand style with clear diagrams.

I have only a few minor points:

1. A definition of early neonatal death and perinatal death used here would be extremely helpful as there some slight variation in the use of these terms.

2. Some of the graphs may benefit from alternative plotting shapes to add differentiation of the lines as the colours are hard to differentiate e.g. triangles and squares to differentiate between the ethnic group data that is very closely aligned

3. More detailed description of why the biases might arise would benefit the discussion and make it easier to interpret with perhaps examples.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: Yes: Lucy Smith

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

18 May 2021

The response to the reviewer has been uploaded as a word document.

Submitted filename: PONE-D-20-40633_Response to Reviewers_20210509.docx

Click here for additional data file.

16 Jun 2021

Bias in comparisons of mortality among very preterm births: a cohort study

PONE-D-20-40633R1

Dear Dr. Boutin,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Kelli K Ryckman

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: No

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: This manuscript is important and well-written. The authors quantitatively demonstrate a common and poorly understood cause of bias (at least among clinicians) pervasive in cohort studies in neonatology/perinatology. The authors should be commended on this work.

All of my comments and questions from the first review were well addressed.

A few minor things, in case they are of any value to the authors:

Ln 120 - I understand what this is intending to say, but wonder if the word “overall” is confusing. It may read to some as if there is supposed to be a comma after it. Consider alternative wording - maybe adding “The" as in "The overall…” would fix this.

Ln 245 - perhaps “effect modification, SUCH AS due …” would be more appropriate. Other arguments about mechanistic effect modification could be made.

Ln 284 - Some large cohort studies do take this into account. The authors' point is valid but doesn’t seem quite on-point and it isn't clear how this is related to the analysis.

If the authors wish to explore other common sources of biases in this literature of neonatal cohort studies limited to gestational age, they might consider also noting issues such as: non-consideration of the intention to treat (when non-treatment results in death); mixing of inborn and outborn infants without consideration of those who die prior to transfer; limitation to live births where the exposure (e.g., region/SES) may be related to classification as live birth or stillbirth. All of these (and surely others) have potentially severe results on causal inference.

I wonder if the authors would rather drop this extra comment about an unrelated source of bias to their analysis (left truncation to NICU admission) -- or, alternatively, provide a more detailed elaboration. The listing of one unrelated bias sometimes observed just seemed out of place.

Ln 328 - “require to be” = “should be”?

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: Yes: Matthew Rysavy

21 Jun 2021

PONE-D-20-40633R1

Bias in comparisons of mortality among very preterm births: a cohort study

Dear Dr. Boutin:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Kelli K Ryckman

Academic Editor

PLOS ONE