Development of a Fetal Weight Chart Using Serial Trans-Abdominal Ultrasound in an East African Population: A Longitudinal Observational Study

Objective: To produce a fetal weight chart representative of a Tanzanian population, and compare it to weight charts from Sub-Saharan Africa and the developed world. Methods: A longitudinal observational study in Northeastern Tanzania. Pregnant women were followed throughout pregnancy with serial trans-abdominal ultrasound. All pregnancies with pathology were excluded and a chart representing the optimal growth potential was developed using fetal weights and birth weights. The weight chart was compared to a chart from Congo, a chart representing a white population, and a chart representing a white population but adapted to the study population. The prevalence of SGA was assessed using all four charts. Results: A total of 2193 weight measurements from 583 fetuses/newborns were included in the fetal weight chart. Our chart had lower percentiles than all the other charts. Most importantly, in the end of pregnancy, the 10th percentiles deviated substantially causing an overestimation of the true prevalence of SGA newborns if our chart had not been used. Conclusions: We developed a weight chart representative for a Tanzanian population and provide evidence for the necessity of developing regional specific weight charts for correct identification of SGA. Our weight chart is an important tool that can be used for clinical risk assessments of newborns and for evaluating the effect of intrauterine exposures on fetal and newborn weight. © 2012 Schmiegelow et al. The study received ethical approval from the Tanzania Medical Research Coordinating Committee (MRCC) on the 18th of April 2008 with reference number NIMR7HQ/R.8a/Vol. IX/688. MRCC is the National Regulatory Body responsible for the supervision of health research and ethical clearance in Tanzania. All procedures were conducted in accordance with the Declaration of Helsinki and Good Clinical and Laboratory Practices. All participants gave informed written consent according to Good Clinical Practice guidelines. Women residing in Korogwe District, Tanga Region, Tanzania were followed throughout pregnancy as part of the observational cohort study STOPPAM (Strategies TO Prevent Pregnancy Associated Malaria). Pregnant women attending the Reproductive and Child Health (RCH) clinic at Korogwe District Hospital (KDH) or the Lwengera, Kerenge and Ngombezi Dispensaries were included in the study from September 2008 until March 2010. Follow-up was completed in October 2010. Women with a gestational age (GA) of ≤24 weeks determined by ultrasound, having lived in Korogwe District for the past 6 months, willing to give birth at KDH and living in an accessible area were included in STOPPAM. The following conditions can affect fetal growth and BW and if present in the current pregnancy the woman/newborn was excluded from analysis; twin pregnancy [22], stillbirth, preterm delivery (GA<37 weeks) [8], multiple pregnancies, severe malformation [3], [22], maternal HIV infection [23], hypertensive disorders, malnutrition [3], diabetes, asthma, epilepsy, syphilis [24], severe anemia [25] and malaria [3], [5]. All conditions were diagnosed by the project team, except for maternal HIV and syphilis infection which were diagnosed by the governmental nurses and data extracted from the antenatal card, and asthma and epilepsy which were based on the medical history reported by the woman. Selection bias caused by an overrepresentation of women being willing to attend a RCH clinic early in pregnancy might have been introduced. Sensitization campaigns in the villages were performed to motivate all women to attend the RCH clinic and thereby reduce selection bias. The sample size for this study was not predefined, but determined by the STOPPAM projects core objective of evaluating the effect of malaria in pregnancy on the health of women and newborns requiring a sample size of 1000 women. The women were followed at the RCH clinic at KDH at 3 pre-scheduled visits at a gestational week of 26, 30 and 36. If needed, women attended extra clinic visits. If women failed to attend the clinic, home visits were performed within one week of the booking. At the first visit maternal age, obstetric history, chronic diseases, and socioeconomic status were documented. Throughout pregnancy the women were screened for the following conditions: malaria infection [positive blood smear and/or rapid diagnostic test (Parascreen™ Zephyr Biomedicals, Goa, India, Paracheck Orchid Biomedical Systems, Goa, India or ParaHIT Span diagnostics Ltd, Surat, India)], pregnancy induced hypertension (HT) [blood pressure ≥140 mmHg systolic and/or ≥90 mmHg diastolic (Digital machine, A&D Instruments, Japan and Spengler, France)] and preeclampsia [HT and ≥0,3 gr/L albumin on urine dipstick (SD Urocolour, Standard Diagnostics, Korea or Cybow, Cybow DFI, Korea)] both presenting after a GA of 20 weeks. HT diagnosed before a GA of 20 weeks was considered essential HT. Furthermore, diabetes defined as glucosuria followed by a random blood-glucose>11 mmol/L. It was considered gestational diabetes if the woman did not have pre-pregnancy diabetes. Malnutrition was defined as mid upper arm circumference<23 cm on inclusion [5] and severe anemia as a haemoglobin ≤7 g/dl (Sysmex hematological analyzer®, Kobe, Japan) at any time during pregnancy [25]. At delivery, birth weight (BW) (weighed naked, unadjusted for timespan since delivery) and sex were documented. Malformations were diagnosed with ultrasound during pregnancy or at birth. At the hospital, BW was measured using a spring scale (Fazzini®, Italy) to nearest 50 gr (until July 2009) or a digital strain gauge scale (ADE®, Germany) to nearest 10 gr (after July 2009). At home deliveries BW were measured using the Fazzini® spring scale. BW measured >24 hours after delivery were excluded from analyses [26], but FWs were still included from these newborns. At the inclusion visit, GA was estimated using ultrasound and considered reliable until a GA of 24 weeks [27]. A new estimation was done within two months if the GA was <11 weeks at inclusion. The ultrasound based estimate was used to define GA for all women, and the GA was not changed at later visits. Variation in biometric measurements due to ethnic group is limited in early pregnancy, and biometric references for white populations were used [28], [29]. GA was estimated using crown-rump length (CRL) until at length of ≤75 mm (13 weeks and 4 days) and the Hadlock algorithm [29]. If CRL>75 mm head circumference (HC) was measured and converted according to Chitty et al [30]. HC is less affected by head shape and parity [28], [31] and was preferred to biparietal diameter. At the visit at 26, 30 and 36 weeks of gestation HC, abdominal circumference (AC), and femur length (FL) were measured using techniques as described elsewhere [13] and recorded in millimeters. For each parameter, a mean of two measurements was used. If only one acceptable measurement was obtained a single measurement of the parameter was used. FWs were estimated (EFW) using the Hadlock algorithm [32]: Log10(EFW) = 1.326+0.0107*HC+0.0438*AC+0.158*FL – 0.00326*AC*FL. If it was not possible to obtain an acceptable HC, EFW was estimated using the Hadlock algorithm [32]: Log10(EFW) = 1.304+0.05281*AC+0.1938*FL – 0.004AC*FL. To assess the accuracy of the Hadlock algorithm to predict FW in this population, BW estimates based on a projection of the last FW, assuming a weight gain of 24.2 g/day [33], was calculated (method A). For women with a FW measured within 35 days of delivery, BW was also estimated by applying the Hadlock proportionality formula [34], using the ratio between the individuals last EFW and the population median FW to predict the BW at term. As population reference the median FW and BW were extracted from the modified Hadlock chart developed using the method by Mikolajzcyk et al [20] (described below) (method B). BW had a non-parametric distribution and the estimated and the observed BW were compared using median error in grams and median percentage error. The percentage of BW estimates that were predicted accurately to within ±10% and ±15% of the observed BW was calculated [35]. The estimated BWs were only used for comparison and were not included in the development of the weight chart. Ultrasound investigations were done at the RCH clinic at KDH by the first author and a local midwife trained for the study using a Sonosite TITAN®, US High resolution ultrasound system with a 5-2 MHz C60 abdominal probe. A few investigations were performed by a trained Tanzanian medical doctor. To evaluate and diminish inter-observer variability, randomly selected fetuses were measured by two investigators and measurements were compared. All investigations were stored as still pictures using SiteLink Image Manager 2.2. Hybrid weight charts using a combination of EFW and observed BW were produced including only healthy pregnancies. The general weight chart was compared to the Congolese chart by Landis et al [10] and the chart by Hadlock et al [21]. Due to the closer geographic location of Congo to Tanzania, this chart was preferred over the chart from Burkina Faso [11]. The general weight chart was also compared to a modified Hadlock weigth chart using the web-based program by Mikolajzcyk et al [20]. The mean BW and variance (as a percentage) from newborns delivered at a GA of 40 to 40 weeks and 6 days in our cohort were imputed into the program. Using the ratio between the mean BW and the mean weight at term from the Hadlock chart the percentiles at all GA were calculated assuming a constant ratio and variance of the mean throughout pregnancy. The prevalence of SGA in the cohort (weight below the 10th percentile) [3], was evaluated by superimposing the observed BW on all the charts. Data were double entered and validated using Microsoft Access 2007. Growth charts were developed using R 2011, and other statistical analyses performed in STATA 10. SigmaPlot 9.0 was used for graphical presentation. The reference curves were constructed using local linear smoothing techniques [36] on a log-transformed version of the EFW/BW that lead to approximate normality of the residuals from the mean curve. The GA dependent variance was estimated using local linear smoothing of the squared residuals. Subsequently, we constructed the reference curves using the GA dependent mean and variance curves. This simple smoothing approach ignored the dependence in the repeated measurements within each subject, but in reality the smoothing primarily used independent measurements due to the somewhat regular pattern of the sampling ages for each woman. The bandwidth for the smoothing was selected by visual inspection. We further validated the results by random effect modeling using splines to fit the data. For this type of modeling the variance structure was derived from the specified random effects structure. The smoothing based technique and the random effects approach gave very similar results, but we preferred the simple non-parametric approach because of the full flexibility in mean and variance structure.