the impact of past pregnancy experience on subsequent perinatal outcomes

Methodology

The impact of past pregnancy experience on subsequentperinatal outcomesJennifer A. Hutcheona and Robert W. Platta,b

Departments of aEpidemiology and Biostatistics and bPediatrics, McGill University Faculty of Medicine, Montreal, Canada

Summary

Correspondence:Jennifer A. Hutcheon, TheMontreal Children’s HospitalResearch Institute, 4060 RueSainte Catherine OuestSuite #205, Westmount,Quebec, Canada, H3Z 2Z3.E-mail: [email protected]

Hutcheon JA, Platt RW. The impact of past pregnancy experience on subsequent peri-natal outcomes. Paediatric and Perinatal Epidemiology 2008; 22: 400–408.

In perinatal epidemiology, the basic unit of analysis has traditionally been the indi-vidual pregnancy. In this study, we sought to explore the idea of a ‘reproductivelife’-based approach to modelling the effects of reproductive exposures and outcomes,where the basic unit of analysis is a woman’s entire reproductive experience. Ourobjective was to explore whether a first pregnancy risk factor, excess gestational weightgain, has a direct effect on the birthweight outcomes of a subsequent pregnancy,independent of the weight gain and other risk factors of the second pregnancy. A studypopulation was created by linking the obstetric records of 1220 women who deliveredtheir first and second offspring at a McGill University teaching hospital in Montreal,Canada. Multivariable linear and logistic regression analyses were used to model theeffects of gestational weight gain above recommendation on the birthweight Z-scoreand risk of large-for-gestational age (LGA) subsequent offspring.

After adjusting for the risk factors of the second pregnancy, an independent effectfrom the first pregnancy was seen on the birthweight Z-score, (effect size OR 0.17 [95%CI 0.05, 0.28] but not risk of LGA of the second pregnancy 1.30 [95% CI 0.89, 1.89]). Weconcluded that a pregnancy-centred approach to research that conceptualises pregnan-cies as self-contained and interchangeable events may not always be appropriate, andpropose that analytical methods for some perinatal research questions may need toconsider a given pregnancy in the context of a woman’s past reproductive experiences.

Keywords: past obstetric history, maternal weight gain, birthweight, large-for-gestationalage, statistical methodology.

Introduction

Reproductive history matters. In some cases, thestrongest single predictor of an adverse reproductiveoutcome is a woman’s own past pregnancy experi-ences.1,2 However, despite the strong ties betweena woman’s past and current pregnancies, perinatalepidemiological research has traditionally been con-ducted on single pregnancies from a given woman. Thelink between a woman’s past and current pregnanciesis achieved, at most, through covariate adjustments forparity or past outcomes, and at worst, is ignored alto-gether. With the pregnancy as the basic unit of analysis,an implicit assumption is made that at the beginning ofthe follow-up period, conception, all pregnancies are

equal and interchangeable given a set of covariates.Although statistical techniques have been proposed toallow the analysis of more than one pregnancyoutcome per woman,3 the emphasis to date has been onthe use of these techniques as a means to correct for thestatistical nuisance of clustering in pregnancy data,rather than considering the inclusion of a woman’s pastand current pregnancies as informative or necessarydata for causal modelling.

The focus of perinatal epidemiology, with analyticalmethods that assume pregnancies to be self-containedevents, is therefore distinctly different from that ofclinical obstetrics, which tends to treat a current preg-nancy much more within the context of a woman’s

400 doi: 10.1111/j.1365-3016.2008.00937.x

Paediatric and Perinatal Epidemiology, 22, 400–408. ©2008 The Authors, Journal Compilation ©2008 Blackwell Publishing Ltd.

reproductive life experiences. Does this difference inapproach have any impact on our ability to understandreproductive exposures and outcomes? Is a woman inher fourth pregnancy, with several prior reproductiveexperiences, indeed analytically interchangeable witha nulliparous woman (so long as a covariate adjust-ment for parity is made)? Or, does a given pregnancyneed to be analysed in the context of a woman’sreproductive life? In this study, we sought to explorethe differences between a pregnancy-based and a re-productive life-based approach to perinatal researchthrough a worked example. Our goal was to examinethe implicit assumption of interchangeability ofpregnancies of the conventional pregnancy-centredapproach to research, and to begin to assess its validity.

Method

We examined the relationship between excessivegestational weight gain and risk of fetal overgrowth.Excessive gestational weight gain is a concern for bothmother and fetus, being linked with an increased riskof macrosomic/large-for-gestational age (LGA) babies,as well as maternal complications during pregnancyand at delivery.4–7 A steady increase over the past 2decades in the proportion of women gaining excessweight during pregnancy,7,8 combined with evidencethat excess gestational weight may contribute to post-partum and long-term obesity9,10 has increased its roleas a public health concern. In this study, we were inter-ested in exploring the direct and indirect long-termeffects of excess gestational weight gain in a firstpregnancy. We hypothesised that the metabolic stressof excessive maternal weight gain during a first preg-nancy may produce lasting changes in maternalmetabolism that could independently influence fetalgrowth in later pregnancies, regardless of risk factorsin the second pregnancy.

Our specific research objective was to determinewhether gestational weight gain above national recom-mendations for weight gain during pregnancy in a firstpregnancy has a direct effect on the birthweight andrisk of fetal overgrowth of a subsequent offspring. Ifall long-term effects of excess gestational weight gainin the first pregnancy on birthweight are mediatedthrough the risk factors of the second pregnancy [suchas pre-pregnancy body mass index (BMI) or gestationaldiabetes status in the second pregnancy, both of whichmay be influenced by retention of excess weight fromthe first pregnancy], we would conclude that pregnan-

cies can reasonably be treated as self-contained andindependent events, because the risk factors of thesecond pregnancy alone would be adequate to ex-plain variability in risk of fetal overgrowth. By self-contained, we mean that all information necessary toexplain the outcome of interest is contained within therisk factors or covariates of a given pregnancy, so that,for example, the impact of weight retention following afirst pregnancy would be entirely mediated throughthe information on pre-pregnancy BMI of a subsequentpregnancy. If, however, the experience of a first preg-nancy has a direct effect that is not channelled throughthe risk factors of the second pregnancy, then the riskof fetal overgrowth should be modelled in the contextof a woman’s past reproductive experiences; her preg-nancies would not be exchangeable.

Study population

A retrospective database analysis was conducted usingdata in the McGill Obstetric & Neonatal Database(MOND) of the Royal Victoria Hospital in Montreal,Canada. The Royal Victoria Hospital is a McGillUniversity tertiary care teaching hospital which servesa multi-ethnic population primarily of Caucasian,Middle Eastern and Asian origin. Further informationon this comprehensive clinical database, which hasbeen maintained in its present form since 1978, is avail-able elsewhere.11 The study population was drawnfrom women who delivered their first and second off-spring at the Royal Victoria Hospital between 1991 and2004. The dataset was restricted to live, singleton birthswithout congenital anomalies, and required women tohave complete obstetric data for both pregnancies,including BMI and gestational weight gain data. Whilethe term ‘reproductive life’ is recognised to include allreproductive events including early pregnancy lossesand terminated pregnancies, here a reproductive lifeapproach is used to include those events in the repro-ductive life that are of central importance in answeringthe substantive question. In this situation we havechosen to restrict analyses to pregnancies that haveprogressed to a degree that allows a reasonable oppor-tunity for excess weight gain.

Measurements

All outcome, exposure and covariate information wasobtained from obstetric chart information entered intothe MOND. Gestational age was established based on

Pregnancy outcome and past reproductive experiences 401


last menstrual period (LMP), unless the discrepancybetween LMP and early ultrasound was >7 days, inwhich case the latter was used. Birthweight was stan-dardised into a birthweight Z-score using Canadiansex- and age-specific population standards.12 Infantswere classified as LGA if their birthweight was abovethe 90th percentile for sex and gestational age.

Gestational weight gain was calculated as last weightprior to delivery minus self-reported pre-pregnancyweight. Pre-pregnancy BMI was calculated as self-reported pre-pregnancy weight, (kg)/self-reportedheight (m2). Implausible BMI and weight gain valueswere corrected through a chart review where possible.To determine excessive gestational weight gain, abinary variable was created indicating if a woman’sweight gain during pregnancy was above Canadiannational recommendations for gestational weightgain.13 This variable was created, because it allowed usto classify the appropriateness of a woman’s weightgain in relation to her pre-pregnancy BMI, as weightgain guidelines are BMI-specific (12.5–18 kg forwomen of BMI < 20, 11.5–16.0 kg for women of BMI20–27, and 7.0–11.5 kg for women of BMI > 27). Self-reported number of cigarettes smoked per day duringpregnancy was converted to a binary variable indicat-ing whether a woman smoked during pregnancy.

Analysis

Directed acyclic graphs (DAG)14–16 were used to outlinethe hypothesised relationships between exposure,covariates and outcome, and to guide modelling deci-sions. To estimate direct effects of excessive first preg-nancy gestational weight gain on second pregnancybirthweight Z-score and risk of LGA, multiple linearand logistic regression models, respectively, were used.We repeated our analyses replacing the binary variableof ‘first pregnancy excess weight gain’ with gestationalweight gain as a continuous variable, adjusted for firstpregnancy BMI. Statistical analyses were conductedusing Intercooled stata 9.0 (College Station, TX).17

Results

The MOND database contained 6973 women who deliv-ered 2 consecutive singleton offspring of birth order 1and 2 during the study period (representing 13 946births). Removing infants with missing gestational age(n = 112), gestational age >43 or <22 weeks (n = 11),congenital anomalies (n = 747), perinatal deaths

(n = 96), and women with BMI or gestational weightgain data missing from both pregnancies (n = 2557), orthose with BMI or gestational weight gain data missingfrom one of their two pregnancies (n = 2267) left a totalof 1220 women (2440 pregnancies). The final studypopulation was restricted to women with complete BMIand gestational weight gain data from both pregnancies,because the availability of weight data from bothpregnancies (and in particular implausible changes inweight between pregnancies) allowed the detectionand correction of measurement error resulting fromincorrect units for weight in the database (kilogramsentered as pounds, or vice versa) which otherwise hadan undue influence on model results. Pregnancieswithout complete BMI and gestational weight gain datahad offspring that were significantly smaller (50 g) andyounger at birth (0.1 week) than pregnancies with com-plete data, but there were no significant differences inmaternal age, smoking status, parity, or sex. An internalreview comparing the characteristics of women withand without complete BMI data in the MOND haspreviously shown that women with missing BMIs were,on average, 1.0 kg/m2 smaller than those with completeBMI data, with other obstetric and neonatal differencessimilar to those found in the present study group(unpublished data).

Maternal and fetal characteristics of the study popu-lation are shown in Table 1 by pregnancy order. Notsurprisingly, women tended to have a slightly higherpre-pregnancy BMI at the time of their second preg-nancy and slightly lower gestational weight gain. Theproportion of women with gestational weight gainclassified as excessive decreased by over 5% from thefirst to second pregnancy. On average, offspring fromthe second pregnancy were approximately 90 g heavierat birth than their older sibling, and had correspond-ingly higher Z-scores and risk of LGA. The meanbirthweight Z-score among offspring from the secondpregnancy whose mother had excess gestationalweight gain in the earlier pregnancy was 0.55 (� 0.94),while the mean birthweight Z-score of offspring fromthe second pregnancy whose mother did not haveexcess weight gain in the earlier pregnancy was0.14 (� 0.93) (P-value < 0.001). The percentage of LGAinfants in these two groups were 21.1% and 10.2%,respectively (P-value < 0.001).

The hypothesised relationship between excess firstpregnancy weight gain and second pregnancy birth-weight is presented in Fig. 1. We speculated that exces-sive gestational weight gain in a first pregnancy would

402 J. A. Hutcheon and R. W. Platt


be likely to influence pre-pregnancy BMI of a woman’ssecond pregnancy, via postpartum retention of excessweight.9,18 Pre-pregnancy BMI, in turn, would influencea woman’s gestational weight gain19 and risk of gesta-tional diabetes mellitus (GDM) in her second preg-nancy,20,21 all of which would influence birthweight ofher second offspring4,6,19,22,23 (operationalised as Z-scoreor risk of LGA). Several additional pathways are alsolikely, such as a direct link from gestational weight gain

in a first pregnancy to risk of GDM in the second(independent of BMI) or a common cause of BMI andgestational weight gain within a pregnancy, such aspolycystic ovary syndrome. To avoid detracting fromthe central causal relationships under study, these havebeen omitted from the DAG in Fig. 1. We furtherhypothesised that there may be an unmeasuredcommon cause of first and second pregnancy gesta-tional weight gain patterns (such as lifestyle, care-giveradvice or cultural attitudes towards gestational weightgain, labelled as U1) as well as an unmeasured commoncause of first and second pregnancy birthweights(based on the known high correlation of sibling birth-weights,24 labelled as U2). The goal of our analysis was todetermine if there were any direct effects of excess firstpregnancy weight gain on the birthweight of the secondoffspring, indicated by the dotted arrow.

The role of birthweight of the first pregnancy is worthcareful consideration. We hypothesised that birth-weight of the first pregnancy may influence maternalweight gain of the second pregnancy, mediated, forexample, by obstetrician advice to women with ahistory of macrosomic offspring to limit gestationalweight gain. It might seem appropriate to include avariable indicating ‘prior LGA birth’ as a risk factor inour model explaining current birthweight, as this vari-able would be highly predictive of the outcome, andgreatly improve the model’s fit. Explicitly mapping thecurrent pregnancy in the context of the woman’s past

Table 1. Maternal and fetal characteristicsfor 1220 women (2440 births) with 2consecutive live births

Variable

First Offspring(n = 1220)

Mean � SD or n (%)

Second Offspring(n = 1220)

Mean � SD or n (%)

Maternal age (years) 28.6 � 4.3 31.3 � 4.4Body mass index (kg/m2) 22.7 � 4.1 23.7 � 4.4Gestational weight gain (kg) 15.5 � 6.7 14.2 � 6.2

Excessive gestational weight gaina 521 (42.7) 453 (37.1)Diabetes during pregnancyb,c 94 (7.7) 88 (7.2)Smoked during pregnancy 165 (13.5) 133 (10.9)Gestational age at birth (weeks) 39.2 � 1.6 38.9 � 1.4Sex (% female) 606 (49.7) 588 (48.2)Birthweight (g) 3422 � 505 3517 � 484

Z-score 0.03 � 0.95 0.32 � 0.95LGAd 119 (9.8) 181 (14.8)

aGestational weight gain above upper limit of Health Canada Guidelines for gestationalweight gain.bDiabetes mellitus or gestational diabetes mellitus.cIncluding 48 women with diabetes mellitus/gestational diabetes mellitus in bothpregnancies.dLarge-for-gestational age, above 90th percentile for gestational age/sex of Canadian livebirths.12

Figure 1. Hypothesised causal relationships between excessivegestational weight gain in a woman’s first pregnancy andsecond pregnancy Z-score, where BMI = body mass index;GDM = gestational diabetes mellitus; wt = weight; Interval =Interpregnancy interval; U1 & U2 = unmeasured covariates.



pregnancy showed that not only was this not necessary(as adjusting for second pregnancy weight gain wouldcontrol for any confounding effect) but, in fact, stratify-ing on the birthweight outcome of the first pregnancycould have introduced selection bias. Here, birthweightis the common effect of the excessive weight gain in thefirst pregnancy and of the unmeasured confounder U2.Stratifying on such a common effect (known as a col-lider) could create an association between its two causes(‘parents’), opening a backdoor path between exposureand outcome, and creating a spurious association.25

The results of the regression models explaining vari-ability in second pregnancy birthweight outcomes areshown in Tables 2 and 3. Five highly influential obser-vations with implausible gestational weight gains wereremoved from the models as outliers. Maternal age and

pregnancy-induced hypertension were not included inthe final models as they did not explain a significantamount of variability in outcome, and had little influ-ence on model results. Non-linear modelling of BMIthrough dummy variables or restricted cubic splines26

did not have any meaningful effect on the model.The unadjusted estimate of the effect of excessive

gestational weight gain on second pregnancy birth-weight outcomes represents the total effects of threepossible pathways based on Fig. 1: the direct causaleffect of excess first pregnancy weight gain, the indi-rect effect mediated through postpartum and secondpregnancy pre-pregnancy BMI, as well as through acommon cause (U1) which results in women tending torepeat weight gain patterns in subsequent pregnancies.As expected, the crude estimates of effect of excessive

Table 2. Effects of first pregnancyexcessive gestational weight gain onsecond pregnancy birthweight Z-score

Variable

Unadjusted effect of excessfirst pregnancy weight gain

Adjusted for secondpregnancy risk factors

beta [95% CI] beta [95% CI]

First pregnancyExcessive weight gaina

(yes/no)0.41 [0.31, 0.52] 0.17 [0.05, 0.28]

Second pregnancyBMI (kg/m2) – – 0.05 [0.04, 0.06]Smoker (yes/no) – – -0.32 [-0.48, -0.15]Gestational weight gain (kg) – – 0.04 [0.03, 0.05]Diabetes in pregnancyb – – 0.19 [-0.01, 0.39]

aGestational weight gain above upper limit of Health Canada guidelines for gestationalweight gain.bDiabetes mellitus or gestational diabetes mellitus.BMI, body mass index.

Table 3. Effects of first pregnancyexcessive gestational weight gain on secondpregnancy risk of LGA

Variable

Unadjusted effect of firstpregnancy weight gain

Adjusted for secondpregnancy risk factors

Odds ratio [95% CI] Odds ratio [95% CI]

First pregnancyExcessive weight gaina 2.33 [1.69, 3.23] 1.30 [0.89, 1.89]

Second pregnancyBMI (kg/m2) – – 1.12 [1.08, 1.17]Smoker (yes/no) – – 0.54 [0.29, 1.01]Gestational weight gain (kg) – – 1.10 [1.06, 1.14]Diabetes in pregnancyb – – 1.83 [1.07, 3.13]

aGestational weight gain above upper limit of Health Canada guidelines for gestationalweight gain.bDiabetes mellitus or gestational diabetes mellitus.BMI, body mass index.



gestational weight gain in a first pregnancy were con-siderably attenuated once the risk factors of the secondpregnancy were incorporated into the linear andlogistic regression models, and the effects of U1 wereblocked by controlling for second pregnancy weightgain. Nevertheless, there remained a significant directeffect on birthweight Z-score of the second offspring of0.17 of a standard deviation, which would correspondto an increase in birthweight of 82 g in this population.Restricting the analysis to term births (as preterm preg-nancies would have less opportunity to gain excessweight) did not have a major impact on this finding(coefficient of 0.19 [95% CI 0.07, 0.31]). The point esti-mate for the direct effect of excessive gestationalweight gain on risk of LGA was suggestive of a 30%increased risk, but the corresponding confidence inter-val was also compatible with a null effect.

A linear regression model in which the binaryvariable for excessive weight gain was replaced with acontinuous variable for first pregnancy weight gain(additionally adjusted for first pregnancy BMI) alsoproduced a significant direct effect, suggesting an0.017 birthweight Z-score increase for every 1 kgincrease in gestational weight gain [95% CI 0.008, 0.28].This corresponded to half the effect of the samemodel’s coefficient for second pregnancy gestationalweight gain of 0.036 [95% CI 0.25, 0.047] on Z-score.

Discussion

In perinatal epidemiology, analysis has conventionallybeen at the level of the individual pregnancy, ratherthan at the level of the mother’s reproductive life. Suchan approach implicitly assumes pregnancies to beindependent and therefore analytically exchangeableevents (or able to be made exchangeable throughcovariate adjustments). Our goal in this study was toexplore the validity of this conceptual approach. Wewere interested in examining the extent to which theevents of a first pregnancy have direct influences onthe outcomes of subsequent ones and, in turn, theextent to which it may not be appropriate to analysepregnancies as interchangeable events. In this study,we found that excessive gestational weight gain in afirst pregnancy had a direct effect on the birthweightoutcome of a subsequent pregnancy, independent ofthe risk factors of the second pregnancy. While thiseffect was modest from a clinical perspective, from amethodological perspective it was large enough tosupport the conclusion that an analytical approach in

which the basic unit of analysis is at the level of thewoman, rather than a single pregnancy, may berequired to fully understand the causal relationshipsof reproductive exposures and outcomes.

As with any methodological exercise, the conclu-sions drawn will be dependent to a certain extent onthe substantive example chosen. Though the impact ofexcessive weight gain in a first pregnancy on the out-comes of subsequent pregnancies led us to concludethat it may not be appropriate to model pregnancies asisolated and interchangeable events, it is possible thatsuch an approach may be quite valid with differentexposures and outcomes. As always, the need to con-sider reproductive life courses will depend on the spe-cific research question and the goals of the analysis.Many research questions do not require considerationof reproductive history, but we argue that this shouldnot be routinely dismissed.

With the individual pregnancy as the unit of analy-sis, multiple pregnancies per woman within a singledataset may be considered to add unnecessary statisti-cal complexity. From a statistical perspective, the preg-nancies are now no longer independent events andrequire special analytical or study design consider-ations to address the clustering. Several authors haveused approaches such as generalised estimating equa-tions (GEE)27,28 to address the statistical correlation ofsuccessive pregnancies. Although accounting for statis-tical non-independence is an important step in pre-venting bias, it does not incorporate the idea that theorder of pregnancies may be informative; in fact theuse of any multi-level model (either GEE or mixedrandom effects model) to account for successive preg-nancies is based on an assumption that units of analy-sis within the clusters (the pregnancies of an individualwoman) are exchangeable given covariates and corre-lation structures. In other words, these statisticaltechniques assume that if order of the pregnancieswithin the cluster were changed, it would not changethe inference (given covariates). Whether the modelprovides an estimate of the effect of an exposure at apopulation level (as in a GEE) or an individual subject(as in a mixed model), both methods require anassumption that the units in the cluster are exchange-able (given covariates and correlation structure), andthat the position of a pregnancy in a woman’s repro-ductive life does not need to be considered.

From an epidemiological perspective, a focus onindividual pregnancies results in a woman’s priorreproductive experiences (or lack therefore) being



treated primarily as potential confounders. It is notuncommon to see analyses that report covariate adjust-ments for a woman’s parity or past reproductiveoutcomes, but often with little indication as to theirhypothesised causal relationship with exposures andoutcomes under study. Several authors have illustratedthe bias that may be introduced by inappropriateadjustment or stratification on past pregnancyoutcome29–31 or parity,32 and highlight the need forcareful consideration of the links between past andcurrent pregnancies. Building on this work, Howardset al.33 recently reviewed the use of DAGs of causalpathways that include past pregnancy exposures andoutcomes as a means to guide analyses. They demon-strated that conventional regression models mayproduce biased results if the exposures or outcomes ofa first pregnancy influence the exposures or outcomesof a subsequent one, and provided guidance on theappropriate use of alternative analytical methods suchas marginal structural models for such cases. In thepresent study, the use of a DAG was able to maketransparent the appropriate analytical model to use,as well as establish which variables to include in thismodel. Careful consideration of the causal questionusing tools such as DAGs can be helpful in making thedecision to include or disregard reproductive history.

The limitations of this study’s substantive analysesshould be noted. Most importantly, there were a largenumber of potentially eligible women excluded fromthe analysis because of missing BMI or gestationalweight gain. Selection bias could have been intro-duced, if the relationship between excess weight gainand risk of fetal overgrowth was different amongthese women than among those included in the finalstudy population. Perinatal deaths were excluded inan attempt to limit analyses to pregnancies withoutserious complications; however, it is possible thatselection bias could have been introduced if excessweight gain itself was an independent cause of thedeaths. As our primary goal for these analyses was toillustrate a methodological issue rather than to assessan aetiological relationship, further research with astudy sample that has a lower potential for selectionbias because of missing data is needed before substan-tive conclusions should be drawn.

Conclusions

The pregnancy-centred analyses commonly used inperinatal epidemiology require an assumption that

pregnancies can be treated as self-contained andinterchangeable events. Such an assumption may notalways be valid, and instead, an approach in which awoman’s reproductive life is the focus of analyses maybe more appropriate. The difference between the twoapproaches can best be illustrated in the context oftime-to-event (survival) analyses, where the ‘time zero’and subsequent follow-up period must be madeexplicit. Instead of pregnancy survival analysis modelswith a ‘t = 0’ of conception and follow-up to perinataloutcome,34 we suggest here that the ‘time zero’ shouldperhaps instead be set to the start of a woman’s repro-ductive life, with follow-up for the duration of herpregnancy experiences.

The use of pregnancies, rather than women, as theunit of analysis in perinatal research is arguably drivenlargely by pragmatic concerns. It is clearly cheaper andeasier to conduct a study following a woman for the9 months of pregnancy than a decade or more of re-productive experiences. In many cases, the longerfollow-up may be neither feasible nor warranted. Nev-ertheless, researchers should consider the followingrecommendations in the design and analysis of repro-ductive exposures and outcomes:1 Directed acyclic graphs in which hypothesised

causal mechanisms are reproductive life-based,rather than pregnancy-based, are a valuable tool toensure analytical methods do not introduce bias byincorrectly incorporating obstetric history.

2 In this example, the experiences of an earlier preg-nancy appeared to have a direct influence on theoutcomes of a subsequent one. When studying therelationship between risk factors and outcomes inperinatal epidemiology, it is important for research-ers to consider the potential impact that each ofthe pregnancy-based and reproductive life-basedapproaches will have for their specific research ques-tion, and design their research studies accordingly.

We hope this project will stimulate further work todetermine the impact that pregnancy-centred andreproductive-life-centred approaches have on ourunderstanding of perinatal research and populationhealth.

Acknowledgements

We would like to acknowledge the work of the late DrRobert Usher in establishing and maintaining theMOND database, and thank Ms Danielle Vallerand forher assistance in extracting the data used for this study.



JAH was supported by Canadian Institutes of HealthResearch (CIHR) Doctoral Research Award, RWPholds a Chercheur-Boursier award from the Fonds dela Recherche en Santé du Québec (FRSQ). Parts of thiswork were presented at the 2006 Joint SPER/SER-SC/ASA Statistics in Epidemiology Section Workshop onAdvanced Methods in Reproductive and PerinatalEpidemiology.

References

1 Mueller-Heubach E, Guzick DS. Evaluation of risk scoringsystem in a preterm birth prevention study of indigentpatients. American Journal of Obstetrics and Gynecology 1989;160:829–835.

2 Galbraith RS, Karchman EJ, Piercy WN, Low JA. Theclinical prediction of intrauterine growth retardation.American Journal of Obstetrics and Gynecology 1979;133:231–236.

3 Buck Louis G, Dukic V, Heagerty PJ, Louis TA, Lynch CD,Ryan LM, et al. Analysis of repeated pregnancy outcomes.Statistical Methods in Medical Research 2006; 15:103–126.

4 Johnson JW, Longmate JA, Frentzen B. Excessive maternalweight and pregnancy outcome. American Journal ofObstetrics and Gynecology 1992; 167:353–370.

5 Stotland NE, Hopkins LM, Caughey AB. Gestational weightgain, macrosomia, and risk of cesarean birth in nondiabeticnulliparas. Obstetrics and Gynecology 2004; 104:671–677.

6 Thorsdottir I, Torfadottir JE, Birgisdottir BE, Geirsson RT.Weight gain in women of normal weight before pregnancy:complications in pregnancy or delivery and birth outcome.Obstetrics and Gynecology 2002; 99(5 Pt 1):799–806.

7 Helms E, Coulson CC, Galvin SL. Trends in weight gainduring pregnancy: a population study across 16 years inNorth Carolina. American Journal of Obstetrics and Gynecology2006; 194:e32–e34.

8 Kramer MS, Morin I, Yang H, Platt RW, Usher R,McNamara H, et al. Why are babies getting bigger?Temporal trends in fetal growth and its determinants.Journal of Pediatrics 2002; 141:538–542.

9 Linne Y, Dye L, Barkeling B, Rossner S. Weightdevelopment over time in parous women – the SPAWNstudy – 15 years follow-up. International Journal of Obesityand Related Metabolic Disorders 2003; 27:1516–1522.

10 Rossner S. Pregnancy, weight cycling and weight gain inobesity. International Journal of Obesity and Related MetabolicDisorders 1992; 16:145–147.

11 Smith LP, DeLeon A, Funnell Y-IR, Lalonde ABM,McLean FH, Usher RH. A research-oriented system forMcGill Obstetrical and Neonatal Data (MOND). ActaObstetricia et Gynecologica Scandinavica 1982; 109:49–50.

12 Kramer MS, Platt RW, Wen SW, Joseph KS, Allen A,Abrahamowicz M, et al. A new and improvedpopulation-based Canadian reference for birth weight forgestational age. Pediatrics 2001; 108:e35.

13 Health Canada. Nutrition for a Healthy Pregnancy: NationalGuidelines for the Childbearing Years. Ottawa: Minister ofPublic Works and Government Services Canada, 1999.

14 Greenland S, Pearl J, Robins JM. Causal diagrams forepidemiologic research. Epidemiology 1999; 10:37–48.

15 Greenland S, Brumback B. An overview of relations amongcausal modelling methods. International Journal ofEpidemiology 2002; 31:1030–1037.

16 Robins JM. Data, design, and background knowledge inetiologic inference. Epidemiology 2001; 12:313–320.

17 Intercooled STATA 9, Version 9. College Station, TX:StataCorp, 2005.

18 Ohlin A, Rossner S. Maternal body weight developmentafter pregnancy. International Journal of Obesity 1990;14:159–173.

19 Cedergren M. Effects of gestational weight gain and bodymass index on obstetric outcome in Sweden. InternationalJournal of Gynaecology and Obstetrics 2006; 93:269–274.

20 MacNeill S, Dodds L, Hamilton DC, Armson BA,VandenHof M. Rates and risk factors for recurrence ofgestational diabetes. Diabetes Care 2001; 24:659–662.

21 Lu GC, Luchesse A, Chapman V, Cliver S, Rouse DJ.Screening for gestational diabetes mellitus in the subsequentpregnancy: is it worthwhile? American Journal of Obstetricsand Gynecology 2002; 187:918–921.

22 Langer O, Mazze R. The relationship betweenlarge-for-gestational-age infants and glycemic control inwomen with gestational diabetes. American Journal ofObstetrics and Gynecology 1988; 159:1478–1483.

23 Kieffer EC, Tabaei BP, Carman WJ, Nolan GH, Guzman JR,Herman WH. The influence of maternal weight and glucosetolerance on infant birthweight in Latino mother-infantpairs. American Journal of Public Health 2006; 96:2201–2208.

24 Beaty TH, Skjaerven R, Breazeale DR, Liang KY. Analyzingsibship correlations in birth weight using large sibshipsfrom Norway. Genetic Epidemiology 1997; 14:423–433.

25 Cole SR, Hernan MA. Fallibility in estimating direct effects.International Journal of Epidemiology 2002; 31:163–165.

26 Harrell FE. Regression Modeling Strategies. New York:Springer-Verlag, 2001.

27 von Ehrenstein OS, Guha Mazumder DN, Hira-Smith M,Ghosh N, Yuan Y, Windham G, et al. Pregnancy outcomes,infant mortality, and arsenic in drinking water in WestBengal, India. American Journal of Epidemiology 2006;163:662–669.

28 Misra DP, Astone N, Lynch CD. Maternal smoking andbirth weight: interaction with parity and mother’s ownin utero exposure to smoking. Epidemiology 2005;16:288–293.

29 Weinberg CR. Toward a clearer definition of confounding.American Journal of Epidemiology 1993; 137:1–8.

30 Weinberg CR. Should we adjust for pregnancy historywhen the exposure effect is transient? Epidemiology 1995;6:335–337.

31 Hernan MA, Hernandez-Diaz S, Werler MW, Mitchell AA.Causal knowledge as a prerequisite for confoundingevaluation: an application to birth defects epidemiology.American Journal of Epidemiology 2002; 155:176–184.



the impact of past pregnancy experience on subsequent perinatal outcomes

Documents