Summary
 Top of page
 Summary
 Introduction
 Methods
 Results
 Discussion
 Acknowledgements
 References
 Appendix
Hutcheon JA, Bodnar LM, Joseph KS, Abrams B, Simhan HN, Platt RW. The bias in current measures of gestational weight gain. Paediatric and Perinatal Epidemiology 2012; 26: 109–116.
Conventional measures of gestational weight gain (GWG), such as average rate of weight gain, are likely to be correlated with gestational duration. Such a correlation could introduce bias to epidemiological studies of GWG and adverse perinatal outcomes because many perinatal outcomes are also correlated with gestational duration. This study aimed to quantify the extent to which currently used GWG measures may bias the apparent relationship between maternal weight gain and risk of preterm birth. For each woman in a provincial perinatal database registry (British Columbia, Canada, 2000–2009), a total GWG was simulated such that it was uncorrelated with risk of preterm birth. The simulation was based on serial antenatal GWG measurements from a sample of term pregnancies. Simulated GWGs were classified using three approaches: total weight gain (kg), average rate of weight gain (kg/week) or adequacy of GWG in relation to Institute of Medicine recommendations. Their association with preterm birth ≤32 weeks was explored using logistic regression. All measures of GWG induced an apparent association between GWG and preterm birth ≤32 weeks even when, by design, none existed. Odds ratios in the lowest fifths of each GWG measure compared with the middle fifths ranged from 4.4 [95% confidence interval (CI) 3.6, 5.4] (total weight gain) to 1.6 [95% CI 1.3, 2.0] (Institute of Medicine adequacy ratio). Conventional measures of GWG introduce serious bias to the study of maternal weight gain and preterm birth. A new measure of GWG that is uncorrelated with gestational duration is needed.
Introduction
 Top of page
 Summary
 Introduction
 Methods
 Results
 Discussion
 Acknowledgements
 References
 Appendix
Maternal weight gain in pregnancy is a potentially modifiable risk factor for a number of adverse maternal and child health outcomes.^{1,2} Low gestational weight gain has been linked with increased risks of fetal growth restriction and preterm birth, while high gestational weight gain is believed to increase risks of macrosomia, caesarean delivery, maternal postpartum weight retention and longterm child obesity.^{1,2}
To establish these associations, a variety of different measures of gestational weight gain have been used.^{1} Total gestational weight gain (kg) and average rate of gestational weight gain (kg/week) are commonly used in epidemiological studies examining the relationship between maternal weight gain and health outcomes such as birthweight or neonatal morbidity,^{3–6} and are the basis for pregnancy weight gain recommendations published by the United States Institute of Medicine (IOM).^{1,7} A third commonly used measure is the adequacy of gestational weight gain in relation to IOM guidelines (IOM adequacy ratio), which is the ratio of a woman's observed (total) gestational weight gain to the gestational weight gain recommended by the IOM guidelines based on her gestational age at delivery.^{8,9}
These conventional measures of gestational weight gain may be problematic, however, because they all appear to have a builtin correlation with gestational age at delivery. Total weight gain is clearly correlated with gestational age at delivery because women who deliver at earlier gestation do not have as much time to gain weight as women who deliver at later gestational ages. The measure of average rate of weight gain (kg/week) would be appropriate if the rate of weight gain was constant throughout gestation. However, gestational weight gain follows a pattern of minimal weight gain in the first trimester and linear growth in the second and third trimesters.^{9–12} As illustrated in Figure 1, this means that the measure of ‘average rate of gestational weight gain’ becomes positively correlated with gestational age at delivery, and a woman on a steady, healthy weight gain trajectory will have a lower average rate of gestational weight gain if she delivers at 28 weeks than if she continues along the same trajectory but delivers at 40 weeks.
While the use of an IOM gestational weight gain adequacy ratio does not rely on the assumption that weight gain is linear throughout pregnancy, the measure may have limitations of its own. The IOM adequacy ratio is based on the assumption that women of healthy prepregnancy body mass index (BMI) gain 2 kg in the first trimester. However, if the average weight gain in the first trimester is higher than this amount, the adequacy ratio will become negatively associated with gestational age, with the weight above 2 kg averaged over a shorter period of time for women delivering at younger gestational ages. As estimates of the average weight gain in the first trimester are all based on data of weight gain patterns from the midlate 1980s,^{9,11,12} the validity of this assumption in contemporary cohorts of pregnant women may be questionable. As with total gestational weight gain and average rate of gestational weight gain, the IOM adequacy ratio may be correlated with gestational duration.
The builtin correlation between conventional measures of gestational weight gain and gestational duration has the potential to introduce serious bias to epidemiological studies examining the effect of weight gain in pregnancy on adverse maternal and perinatal health outcomes because the vast majority of these outcomes (such as birthweight, stillbirth, neonatal mortality or serious neonatal morbidity) are also correlated with gestational age at delivery. In other words, the measures currently used in epidemiological studies may not allow the relationship between gestational weight gain and adverse pregnancy outcomes to be properly separated from the relationship between gestational age at birth and adverse pregnancy outcomes. As a result, adverse outcomes currently believed to be associated with low rate of weight gain may instead be more likely to be attributable to a young gestational age at birth. It is unclear, however, if this theoretical bias is large enough in magnitude to have a meaningful impact on our substantive understanding of the contribution of gestational weight gain to important pregnancy outcomes. The goal of this study was to quantify the degree of bias that conventional gestational weight gain measures may introduce to the apparent relationship between gestational weight gain and risk of preterm birth ≤32 weeks.
Methods
 Top of page
 Summary
 Introduction
 Methods
 Results
 Discussion
 Acknowledgements
 References
 Appendix
Pregnancy information on a large populationbased cohort of pregnant women was obtained from the British Columbia (BC) Perinatal Database Registry, which contains obstetrical and neonatal clinical chart data on all births in this Canadian province.^{13} The characteristics of these women are provided in Appendix A. Analyses were restricted to women with a normalweight prepregnancy BMI (18.5–24.9 kg/m^{2}, inclusive), and stillbirths were excluded. Very preterm birth was defined as a livebirth ≤32 completed weeks of gestation. Sensitivity analyses were conducted using an outcome of any preterm birth (livebirth <37 weeks).
We then used simulation to randomly assign a total gestational weight gain to each woman in the BC cohort that was, by design, uncorrelated with her risk of preterm birth. In the first step of this simulation, realistic estimates of gestational weight gain patterns in a contemporary cohort of pregnant women were obtained from a sample of 818 women of normalweight prepregnancy BMI who delivered a singleton birth at MageeWomens Hospital in Pittsburgh, PA, US, and for whom serial antenatal weight gain measurements were available. Stillbirths, preterm births (<37 weeks) and infants with poor perinatal outcome (defined as a 5min Apgar ≤7 or neonatal intensive care unit admission) were excluded. All women had valid prepregnancy weight measurements and at least one measured weight in the first trimester. The mean and standard deviation of total gestational weight gain in the cohort was 15.4 kg ± 6. The serial gestational weight gain measurements (an average of 11 per woman) were smoothed to obtain means and standard deviations of gestational weight gain at each week of gestation (Table 1). As all women in this cohort delivered at term, estimates of the weekspecific means and standard deviations of gestational weight gains at preterm gestational ages were independent of the risk of preterm birth.
Table 1. Smoothed weekspecific means and standard deviations of gestational weight gain as estimated from the serial antenatal weight measurements of 818 women delivering a singleton term birth at MageeWomens Hospital in Pittsburgh, PA, US Gestational age (weeks)  n  Mean (SD) gestational weight gain (kg) 

24  180  8.4 (3.9) 
25  236  9.1 (4.0) 
26  231  9.7 (4.1) 
27  232  10.3 (4.2) 
28  277  11.0 (4.3) 
29  302  11.6 (4.4) 
30  363  12.1 (4.5) 
31  324  12.6 (4.6) 
32  379  13.1 (4.7) 
33  364  13.6 (4.8) 
34  412  14.1 (4.9) 
35  424  14.8 (5.1) 
36  608  15.5 (5.2) 
37  513  16.1 (5.3) 
38  420  17.0 (5.4) 
39  298  17.7 (5.6) 
40  148  18.5 (5.7) 
41  41  19.1 (5.8) 
In the second step of our simulation, the weekspecific means and standard deviations of gestational weight gain were used to simulate a total gestational weight gain for women in our BC cohort based on their gestational age at delivery. For example, in the MageeWomens cohort, the average gestational weight gain by 28 weeks was 11.0 kg with a standard deviation of 4.3 kg. A total gestational weight gain for each woman in the BC cohort who delivered at 28 weeks was simulated by drawing a random sample from a normal distribution with a mean of 11.0 kg and a standard deviation of 4.3 kg. In other words, the gestational weight gain of a woman in our simulation who delivered at 28 weeks was drawn from the distribution of weight gains observed in the MageeWomens cohort of women at the time of their 28week antenatal visit. This was repeated for deliveries at each week of gestation using the means and standard deviations estimated from the MageeWomens cohort. We were thus able to explicitly build into the simulation that there be no true association between total gestational weight gain and gestational age at delivery [e.g. odds ratios (ORs) for preterm birth ≤32 weeks, when total weight gain was standardised to a gestational agespecific zscore (using the means and standard deviations in Table 1) and classified in quintiles, were all 1.0 or 1.1, with 95% confidence intervals [CI] that included the null].
The simulated gestational weight gains were classified in each of three different ways: (a) total gestational weight gain (kg), (b) rate of gestational weight gain (total gestational weight gain in kg/gestational duration in weeks), and (c) 2009 IOM adequacy ratio [observed (total) gestational weight gain divided by the weight gain recommended by the IOM guidelines based on her gestational age at delivery]. The IOM recommended weight gain for a woman of normal prepregnancy BMI is calculated as: 2 kg average weight gain in the first trimester + [recommended rate of second and third trimester weight gain of 0.42 kg per week × (gestational duration in weeks − 13 weeks for first trimester)]. The recommended weight gain for a woman delivering at 40 weeks would therefore be calculated as: 2 kg for first trimester + 0.42 kg/week × (40 − 13 weeks) = 13.3 kg. Logistic regression was used to examine the relationship between each gestational weight gain measure and preterm birth ≤32 weeks. Gestational weight gain was modelled using restricted cubic splines,^{14} which allowed continuous, curvilinear relationships to be explored. Each measure of gestational weight gain was also grouped into fifths (i.e. based on quintile cutoff values), and the risk of preterm birth ≤32 weeks in each fifth was compared with that of the middle fifth (reference category). To avoid the possibility that the results obtained from our simulation were unrepresentative, we repeated these analyses on a total of 1000 randomly generated gestational weight gains for each woman.
Results
 Top of page
 Summary
 Introduction
 Methods
 Results
 Discussion
 Acknowledgements
 References
 Appendix
There were 128 371 women of normalweight prepregnancy BMI in the BC Perinatal Database Registry between 2000 and 2009, including 901 infants born ≤32 weeks. The smoothed weekspecific means and standard deviations of gestational weight gain used to simulate a total gestational weight gain for each woman in the BC population are provided in Table 1.
The relationships between (simulated) total gestational weight, as classified by each gestational weight gain measure, and risk of preterm birth ≤32 weeks, are shown in Figure 2. Although, by design, no association should have been present, all three measures indicated an apparent association with gestational age. Total gestational weight gain and average rate of weight gain were observed to have an inverse association with risk of preterm birth, while the association between IOM adequacy ratio and risk of preterm birth appeared to be more Ushaped in nature. As shown in Table 2, the magnitude of these associations was both statistically significant and clinically meaningful: for women in the lowest fifth of IOM adequacy ratio, a 60% increase in risk of preterm birth ≤32 weeks was estimated, while women in the lowest fifth of average rate of weight gain and total weight gain were estimated to be at 1.8 and 4.4fold increased risk, respectively. An apparent protective effect of higher gestational weight gain was observed when using total gestational weight gain and average rate of gestational weight gain (OR in highest fifths of 0.2 [95% CI 0.1, 0.3] and 0.7 [95% CI 0.6, 0.9], respectively), while the use of IOM adequacy ratio showed a trend towards a Ushape in risk of preterm birth (OR in highest fifth of 1.3 [95% CI 1.0, 1.6]). Repeating the simulation 1000 times did not alter our findings (data available upon request). Sensitivity analyses with the outcome of preterm birth <37 weeks showed similar trends, although the magnitude of the ORs was somewhat attenuated (e.g. OR for first fifth of average rate of weight gain was 1.5 [95% CI 1.4, 1.6], compared with 1.8 [95% CI 1.5, 2.2] for the outcome of very preterm birth) (Table 2).
Table 2. Observed associations between gestational weight gain measures and risk of preterm birth when, by design, gestational weight gain has been simulated to be independent of risk of preterm birth (see text for details) Quintile of gestational weight gain  Total gestational weight gain (kg)  Average rate of gestational weight gain (kg/week)  IOM adequacy ratio^{a} 


First  OR^{b} preterm birth ≤32 weeks [95% CI]  4.4 [3.6, 5.4]  1.8 [1.5, 2.2]  1.6 [1.3, 2.0] 
OR^{b} preterm birth <37 weeks [95% CI]  2.1 [1.9, 2.2]  1.5 [1.4, 1.6]  1.4 [1.3, 1.5] 
Median value of weight gain measurement  10.3  0.27  0.81 
Second  OR^{b} preterm birth ≤32 weeks [95% CI]  2.0 [1.6, 2.5]  1.0 [0.8, 1.2]  1.1 [0.9, 1.3] 
OR^{b} preterm birth <37 weeks [95% CI]  1.4 [1.3, 1.5]  1.1 [1.0, 1.2]  1.1 [1.0, 1.2] 
Median value of weight gain measurement  14.6  0.38  1.14 
Third  OR^{b} preterm birth ≤32 weeks [95% CI]  1.0 Reference  1.0 Reference  1.0 Reference 
OR^{b} preterm birth <37 weeks [95% CI]  1.0 Reference  1.0 Reference  1.0 Reference 
Median value of weight gain measurement  17.6  0.45  1.37 
Fourth  OR^{b} preterm birth ≤32 weeks [95% CI]  0.4 [0.3, 0.6]  0.8 [0.6, 1.0]  1.0 [0.8, 1.2] 
OR^{b} preterm birth <37 weeks [95% CI]  0.6 [0.6, 0.7]  0.9 [0.8, 1.0]  1.0 [0.9, 1.0] 
Median value of weight gain measurement  20.6  0.53  1.60 
Fifth  OR^{b} preterm birth ≤32 weeks [95% CI]  0.2 [0.1, 0.3]  0.7 [0.6, 0.9]  1.3 [1.0, 1.6] 
OR^{b} preterm birth <37 weeks [95% CI]  0.3 [0.3, 0.4]  0.8 [0.7, 0.8]  1.0 [0.9, 1.0] 
Median value of weight gain measurement  25.0  0.63  1.92 
In posthoc analyses, comparisons of the weight gain trajectories in the Magee cohort with IOM recommendations (Figure 3) showed that the extent to which IOM recommendations underestimated the actual weight gain trajectories of women with good perinatal outcomes became more pronounced with increasing gestational age [i.e. the rate of second and third trimester weight gain in the cohort was faster than the rate of weight gain expected by the IOM guidelines (0.42 kg per week)]. The difference between observed and expected weights therefore became greater as the pregnancy progressed, leading to a positive correlation between IOM adequacy ratio and gestational duration. For example, a woman with gestational weight gain pattern that followed that of the population average throughout her pregnancy would have had an IOM gestational weight gain adequacy ratio of 1.3 at 32 weeks (average weight gain 12.7 kg), but an IOM adequacy ratio of 1.5 by 40 weeks (average weight gain 18.5 kg). This correlation was likely to be responsible for the bias introduced into estimates of risk of preterm birth ≤32 weeks at lower IOM adequacy ratios.
Discussion
 Top of page
 Summary
 Introduction
 Methods
 Results
 Discussion
 Acknowledgements
 References
 Appendix
In this study, we demonstrated that conventional measures of gestational weight gain introduce a nontrivial degree of bias to epidemiological studies assessing the relationship between gestational weight gain and risk of preterm birth ≤32 weeks. Using a populationbased pregnancy cohort typical in size of those used for many studies of gestational weight gain and adverse pregnancy outcomes, statistically significant and clinically meaningful associations were observed even though by design, no true associations existed. This work raises serious concerns about our current understanding of the effect of gestational weight gain on preterm birth. Two studies commissioned by the 2009 IOM Guidelines Committee on the relationship between gestational weight gain and preterm birth both used average rate of weight gain in their analyses, and found a modest Ushaped relationship.^{1} Although differences in study populations, sources of gestational weight gain data, and calendar time, limit our ability to compare results, our study findings of an increased risk of preterm birth associated with lower average rates of gestational weight gain (OR = 1.5 [95% CI 1.4, 1.6] in lowest fifth compared with middle fifth) suggest that the earlier studies' findings may be partly an artefact of the measure of gestational weight gain chosen. Residual differences between our study and those in the literature are likely to reflect a true biological relationship between gestational weight gain and preterm birth.
This correlation between conventional measures of gestational weight gain and gestational age at delivery may affect our understanding of the relationship between gestational weight gain and many outcomes besides risk of very preterm birth. As gestational age at delivery is one of the strongest predictors of a large number of maternal and perinatal health outcomes (including birthweight, stillbirth, neonatal mortality and serious neonatal morbidity),^{1,2} the correlation between gestational weight gain and gestational age at delivery could introduce bias to studies of these outcomes as well. Risks of outcomes such as smallforgestationalage or largeforgestationalage infants may also be affected, because the vast majority of these infants are born at term gestational ages. The biased nature of contemporary gestational weight gain measures means that the effects of gestational weight gain become intermixed with the effects of gestational age at delivery, thereby obscuring our understanding of the true contribution of gestational weight gain to maternal and perinatal health outcomes. Restriction of study populations to term gestational ages would help minimise this bias, as the relationship between adverse perinatal outcomes and gestational age is weaker within term ages, but would not eliminate it.
The use of serial antenatal weight gain measurements represents an alternative approach that could be used to resolve the bias in conventional measures of gestational weight gain. In studies of preterm birth, for example, weight gain among women with a preterm birth could be compared with weight gain up to the same time in gestation in women who subsequently delivered at term ages. The small number of studies that have used serial weight gain data (or first trimester weight gain data) to examine patterns of weight gain and risk of preterm birth^{15–18} are not prone to the same bias as studies using only total gestational weight gain. However, difficulties in obtaining large, populationbased data on gestational weight gain patterns have limited such studies to date, and total gestational weight gain continues to be widely used in epidemiological studies.
While the limitations of using total gestational weight gain as a measure of weight gain in pregnancy have been previously recognised,^{7} the limitations of measures such as average rate of weight gain or IOM gestational weight gain adequacy ratio do not appear to be well appreciated. In their 2009 recommendations, for example, the IOM Committee described the measure of average rate of gestational weight gain as ‘the only proper measure to compare pregnancies of varying duration’^{1} (p. 212). Our findings suggest the builtin correlation with gestational age of these latter two measures will still introduce bias that is large enough in magnitude to affect substantive conclusions.
A new and methodologically improved approach to the assessment of gestational weight gain is needed. One potential approach could be the calculation of maternal weight gainforgestational age zscores or percentiles, similar to the approach used in the study of fetal growth.^{19,20} Gestational agestandardised zscores or percentiles would, by definition, be independent of gestational duration. While the means and standard deviations of gestational weight gain produced in this study can be used by the researcher to produce an unbiased estimate of the relationship between total gestational weight gain and pregnancy outcomes, ideally, maternal weight gain zscore reference values should be based on a populationbased, representative sample that is also able to take into account longerterm maternal and child health outcomes.
It is only with the development of new methods to assess total gestational weight gain that are uncorrelated with gestational age that we will be able to obtain an unbiased understanding of the contribution of maternal weight gain in pregnancy to a number of important maternal and perinatal health outcomes. Meanwhile, the use of conventional measures of gestational weight such as average rate of weight gain or IOM adequacy ratio should be avoided when studying outcomes that are correlated with gestational age because results may reflect bias rather than true biological associations.
Acknowledgements
 Top of page
 Summary
 Introduction
 Methods
 Results
 Discussion
 Acknowledgements
 References
 Appendix
The authors thank Terri Pacheco, Perinatal Services BC, for her assistance with BC Perinatal Database Registry data extraction and coding.
This work was supported by postdoctoral fellowship awards from the Canadian Institutes of Health Research and the Michael Smith Foundation for Health Research (J. A. H.), a career scholar award from the Child and Family Research Institute (Vancouver, Canada) (K. S. J.), and a Chercheurboursier and core support to the Research Institute of the McGill University Health Centre from the Fonds de la Recherche en Santé du Québec (R. W. P.).