A customised fetal growth and birthweight standard for Qatar: a population-based cohort study

Study design and participants

This study used anonymised data from the PEARL-Peristat Study, a population-based retrospective cohort registry containing maternal and neonatal demographics and outcomes collected from all women delivering in Qatar between 2017 and 2018. The registry is funded by the Qatar National Research Fund (Grant number: NPRP 6-238-3-059) and approved by the Hamad Medical Corporation (HMC) Institutional Review Board, with a waiver of informed consent (HMC-IRB 13064/13).

We included women delivering a singleton livebirth of more than 24 weeks gestation between April 2017 and March 2018 at Women’s Wellness and Research Centre (WWRC) in Doha, Qatar, which is the largest national maternity hospital and leading provider of secondary and tertiary health care facilities in the country (averaging 18,000 deliveries annually, and nearly 70 % of births in Qatar). For calculating the coefficients, all preterm births (<37 weeks gestation) and babies with congenital anomalies were excluded. There were no other exclusion criteria.

Data source and variables

The registry data was collected by independent data collectors from Cerner Millenium^® electronic patient medical records. After applying the inclusion criteria, we extracted anonymised data and cleaned the dataset by removing outliers for gestational age, birth weight, maternal weight and height (<1 % of the total). Patient-identifying variables such as name, date of birth, hospital registration number, and date of delivery were retained in the registry, accessible only by the registry owners, and stored in password-protected databases saved on the hospital network.

The maternal variables included age in completed years, height and weight documented at the first antenatal visit, parity – defined as any previous birth >24 weeks gestational age, medical conditions like preexisting diabetes, hypertension (including essential and pregnancy-induced) and gestational diabetes – defined as abnormal 75 g glucose tolerance test performed during the pregnancy between 16 and 32 weeks gestation according to patient risk factors [20], and self-reported current smoking status (yes/no). Maternal age was categorised into four groups: <20 years, 20–28 years, 29–34 years and ≥35 years. The body mass index (BMI) was generated from maternal weight and height (kg/m²) and categorised according to the WHO BMI categories [21]. Maternal nationality was extracted from the government health card details.

The gestational age (GA) in days was determined by a first-trimester dating ultrasound scan within 14 weeks of pregnancy measuring the crown-rump length or the last menstrual period (if a dating scan was unavailable). Other variables included preterm delivery – defined as any delivery before 37 completed weeks of gestation (259 days) [22], congenital anomalies – including any babies diagnosed with any chromosomal or congenital anomalies diagnosed antenatally or at birth by a neonatologist and biological sex of the baby at birth. The birthweight was documented in the hospital records as the newborn’s weight measured in grams, using calibrated hospital scales, immediately after birth.

Statistical analysis

Continuous variables were recorded as mean ± standard deviation (SD) and median ± interquartile range (IQR). Frequencies and percentages of the total cohort were used to describe categorical variables. After eliminating outliers, the data from 12,845 women were used for univariate analysis. The regression model was based on 11,693 cases after excluding preterm births and babies with congenital anomalies.

The statistical methods used to derive the birthweight coefficients were as described previously [10]. A stepwise multiple linear regression with backward elimination (outcome – birthweight in grams) was performed, including patient characteristics such as GA at birth, maternal height, weight, ethnicity, parity, biological sex of the baby, maternal preexisting illness such as hypertension and diabetes, and smoking status. Quadratic and cubic terms for GA, weight and height were included in the model in addition to the linear terms to account for the non-linearity of the associations. The model was centred on the median GA at birth for the cohort. Variables with a p value less than 0.05 were retained in the model.

To allow comparison with models from other countries, the analysis was centred on a “standard” mother – a woman 163 cm tall weighing 64 kg, nulliparous (parity 0), with the baby’s sex averaged between male and female. This is based on findings from the initial study performed in the UK population, defining an average-sized European mother, which was used as the standard for comparison with other models [10]. The term optimal birthweight (TOW) constant was then derived by adjusting the constant to 280 days gestation (40 weeks – considered full term) for a standard mother, excluding the effect of preexisting and gestational diabetes, hypertension, and smoking.

Weight for each gestational day was calculated using the proportionality equation [10], based on Hadlock’s fetal weight standard [23]. Polynomial regression equations to the third order were used to plot the 10, 50 and 90 centile GROW curves, with the standard 11 % coefficient of variation of the model [10]. Based on these calculations, the proportions of SGA babies (<10 %) and LGA babies (>90 %) in the study cohort were determined. The birthweight distribution at 280 days according to the Hadlock fetal weight standard [23] and the Fenton neonatal weight standard [24], currently in clinical use in the country, was compared with the predicted birthweight distribution using GROW coefficients. This comparison was done for the entire cohort and for mothers of Qatari nationality separately. All analyses were performed using Stata release 17, StataCorp, TX, USA [25].

Main results

This is the first study generating customised birthweight standards and coefficients in a multi-ethnic population in Qatar, determining the TOW using the maternal physiological variables. The gestational age, maternal height and weight at booking were significantly associated with birthweight, the relationship assuming a nonlinear parabolic curve due to the significant quadratic terms for gestational age and weight. Additionally, the biological sex of the baby and maternal parity significantly influenced birthweight.

The model allows adjusting the predicted birthweight according to the physiological characteristics of each mother. The centile calculators are available via the Gestation network (www.gestation.net), and they currently include coefficients from at least 100 different ethnic or country-of-origin groups [8]. The model generated in this study will make the charts accessible and customised for the multi-ethnic Qatar maternity population. The GROW charts can be produced early in pregnancy (after the dating scan or the first antenatal visit) and attached to the antenatal sheet of patients; the fundal height measurements during antenatal visits and estimated fetal weight from growth scans can be plotted onto the GROW charts, which can tremendously improve abnormal fetal growth detection.

After applying the model, the SGA (<10 centile) for mothers of Qatari nationality was 11.2 %. Previous studies report an SGA rate of 6.7 % for babies of Qatari nationality from the same dataset – SGA defined using the Intergrowth 21st 10th centile universal standards [26]. The customised growth charts successfully reclassified a minimum of 4 % of babies as SGA. The risk of LGA in the previous study was 13.6 % compared to 12.2 % for the entire cohort in this study. Figure 1 compares the Hadlock and Fenton standard distributions to the Qatari nationality birthweights observed in the cohort. Using these standards resulted in an overestimation of the SGA rate (28.5 vs. 11.2 %) and an underestimation of the risk of LGA (1.4 vs. 11.8 %), highlighting the issues of applying these universal standards to the maternal population in Qatar.

The incidence of diabetes affecting pregnancies in Qatar is nearly 25 % [27]. These high rates likely contributed to the high proportion of LGA identified by the GROW curves and missed by the other standardized curves. Hence, GROW curves can determine SGA/LGA rates customized to the true population distribution.

A 2023 study, including 2.2 million live births in UK, compared the customized centiles to the Intergrowth 21, Hadlock and WHO population standards [28]. The GROW centiles produced similar rates of SGA when comparing ethnicities, whereas the other standards produced a wide variation. Moreover, GROW was able to pick up the increase in stillbirths in SGA babies in mothers with higher BMI. Similar results were shown in a French study reporting increased perinatal mortality in the babies reclassified as SGA by GROW [14]. Future studies need to assess the impact of customized charts on FGR and stillbirth risk in the population in Qatar.

Comparison with previous models

Table 3 compares the coefficients generated in other populations [8, 11], [12], [13, 15], [16], [17], [18, 29, 30] with those in this dataset using a ‘standard mother’ (height 164, weight 63 kg, ‘average’ fetal sex, and gestational age adjusted to 280 days) which gives a population TOW of 3,228.4 g. This TOW is at least 250 g less than that in Europe, USA, and Australia. The pattern of the other coefficients is very similar across all populations, highlighting that the differences noted are part of a natural variation between populations. As previously reported, fetal growth is slower between 26 and 36 weeks gestation and then increases linearly till term [31]–this phenomenon is reflected in the coefficients as GA has a nonlinear parabolic relationship with birthweight with significant quadratic terms. The rationale behind not using BMI as a predictor is that maternal height and weight have additive associations with birthweight, which would not be picked up by BMI alone. The maternal age was not significant in most models after adjusting for parity, although age more than 30 years was included in the Slovenian model [29], having up to 28.4 g impact on birthweight (age ≥40 years).

Ethnicity is an important predictor of birthweight. The UK, Australia and Ireland models adjust for Middle-Eastern ethnicity and report a decrease in TOW from 90 to 170 g, much less than the 250 g difference noted in our model. This highlights the importance of this study and generating coefficients specific to the population in Qatar rather than extrapolating from previous models. We have included several other nationalities in our model and report the ones that significantly impact birthweight, which has not been reported in previous models. Some models include other constitutional variables like paternal height and weight, maternal age and even BMI – however, the similarity of the coefficients shows that these variables probably had a minimum additional impact on birthweight.

The pathological variables included in most models are similar, such as hypertension, preexisting and gestational diabetes and smoking. The coefficients are similar in direction and magnitude for these variables, with all models reporting the biggest coefficients for preexisting diabetes and the largest negative coefficients for smoking. Some models show a dose-response effect for smoking based on the number of cigarettes [10, 16, 29], which was not possible in our cohort due to the unavailability of this data.

The R² for our model is 0.24, which is similar to those from other models (Table 3) and seems modest despite customisation. This is because the model predicts birthweight in pregnancies with normal outcomes only rather than the variations at upper and lower extremes, which are primarily due to pathological factors. The effect of customisation on the predictive ability of the significant coefficients being adjusted for is illustrated when excluding upper and lower extremes by assessing the mid-tertile part of the distribution [8], where the R² value increases stepwise with each variable added (sex, parity, ethnicity, maternal height, maternal weight) up to a more respectable R² of 0.76.

Strengths and limitations

A strength of this study is that we used an ethnically heterogeneous population for our analysis, allowing customised growth charts to be created for multiple ethnicities for women in Qatar. Additionally, nearly 70 % of deliveries in the country occur at the study site; hence, the study cohort is representative of the population. Nearly 12,000 women were included in the model, further increasing the representativeness and generalizability of the results. The PEARL registry provides quality information, with minimum missing data, collected by well-trained independent data collectors based on pre-defined standardized definitions.

Some limitations need to be kept in mind. Previous reports document a smoking rate of 3 % in women of Qatar in 2021 [32]. This study showed a self-reported smoking rate of 0.5 %, the accuracy of which needs to be questioned as expectant mothers might deny this history due to social pressures. Regardless, smoking had the highest detrimental impact on birthweight, with coefficients similar to the moderate-to-high levels of smoking reported in other models [10, 16, 29]. There could be human errors in the documentation of height and weight; however, the data entries were always double-checked, and the outliers were removed before analysis.