Newborn weight nomograms in selected low and middle-income countries

As previously described⁷, between April 2019 and March 2020, a convenience sample of 741 newborns ≥ 2000 g at birth whose mothers were aged ≥ 18 years, intending to breastfeed for at least six months, and willing to provide informed consent, were enrolled at selected health facilities in Guinea-Bissau, Nepal, Pakistan, and Uganda. Consistent with the WHO Multicentre Growth Reference Study (MGRS) used to generate CGS charts, newborns with major congenital anomalies, danger signs, respiratory distress, or maternal or infant contraindications to breastfeeding were excluded from our study; otherwise, maternal health status did not affect eligibility either for our study or for the study population of the NEWT nomograms. In contrast to MGRS, newborns with economic or environmental constraints on growth were not excluded from our study. To improve comparability with the NEWT tool, we included newborns regardless of feeding type who were delivered at hospitals where exclusive breastfeeding was the most common initial feeding method practiced. The NEWT tool was selected for comparison because it is the most widely used newborn weight loss tool; NEWT nomograms depict quantiles of newborn weight change by hour of age, thus facilitating understanding of an individual infant’s newborn change in the context of reference norms.

Trained study staff obtained duplicate birth weights using a standardized protocol for naked newborns within 6 h of birth and at 1, 2, 3, 4 and 5 days with a Seca 334 scale (Seca Inc., Wandsbek, Germany) accurate to ± 5 g; two additional measurements were taken if the initial two measurements varied by 15 g or more. Follow-up weights were measured at the enrollment health facilities or during home visits as preferred by the study participants. Study participants were traced and located using provided contact information and maps as necessary. All enrolled infants received usual care prior to, during, and after study enrollment. No direct care was provided by the study team, and ill infants were referred. Travel reimbursement was provided; no other incentives were provided.

To generate nomograms for our cohort, we used quantile regression methods appropriate for longitudinal data to estimate the 25th, 50th (median), and 75th percentiles of weight change as a function of time after birth, separately for each country. We applied the penalized fixed-effects model in the R package “Regression Quantiles for Panel Data (rqpd)” to estimate the percentile curves⁸. The model is an extension of ordinary quantile regression methods to longitudinal settings and includes separate intercept terms for each infant, with regularization used to estimate the intercepts. The amount of regularization was controlled by a tuning parameter set to 5.

We used a natural spline with 4 degrees of freedom to estimate non-linear quantile curves as a function of time. One complicating feature of the data is that the first weight (birth weight) was recorded up to 6 h after birth. Based on our previous work, newborns typically lose weight in the first 6 h after birth, so the weight recorded may have been an underestimate of birth weight⁶. Therefore, we imputed birth weights for all newborns based on the NEWT curves prior to fitting the models; specifically, we randomly selected a percentile value to impute at 6 h from the NEWT curves. We then used linear interpolation to impute weight at time 0 based on the time from birth that the weight was recorded and the randomly selected percentile value. For example, with a first weight of 3000 g recorded at 3 h, linear interpolation based on the 95th percentile of the NEWT curves (− 0.236% at 6 h) and the 5th percentile (− 1.94% at 6 h) results in birth weights of 3004 g and 3029 g, respectively. The main assumption with our imputation approach is that the percentile estimates from our previous work reasonably approximate the weight loss in these cohorts for the first 6 h. While the randomly selected percentile is not necessarily accurate for individual newborns, we expect errors to average out across all newborns. The resulting nomograms were then superimposed on existing NEWT nomograms to allow visual comparison, since no statistical test exists to determine whether one nomogram differs from another nomogram⁹.

As a sensitivity analysis, rather than random assignment, we also imputed birth weights based on selecting the 5th percentile for all newborns and separately the 95th percentile for all newborns. The results were similar, likely because the models are primarily based on weight values at later time points and most birth weights were recorded within 4 h of birth.

This study was approved by the UCSF Institutional Review Board, the Guinea-Bissau National Committee on Ethics in Health (Comite Nacional de Etica na Saude), the Nepal Health Research Council, the Institutional Review Committee of Kathmandu University Teaching Hospital, the Ethical Review Committee at the Aga Khan University in Pakistan, the Higher Degrees, Research and Ethics Committee of Makerere University, and the Uganda National Council of Science and Technology. Informed consent for study participation was obtained from all participating mothers. Informed consent for all children was obtained from the mother of each child. All study activities were performed in accordance with relevant guidelines and regulations.